Understanding the Conditions of Bias: Essays on Gender Differences in Evaluation Outcomes across A Three Empirical Contexts MASACHUSETTS INSTITUTE OF TECHNOLOLGY by JUN 02 2015 Mabel Lana Botelho Abraham B.A., Mathematics, Providence College, 2003 S.M., Management Research, Massachusetts Institute of Technology, 2013 LIBRARIES Submitted to the Sloan School of Management in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the Massachusetts Institute of Technology June 2015 Massachusetts Institute of Technology 2014. All Rights Reserved. Signature of redacted Author: Mabel Abraham Sloan School of Management Signature redacted May 7, 2015 Certified by: -to erto Fernandez illiam F. Pounds Professor of Management Professonr of O~r anizatnion St1-udies L11V g Thesis Supervisor I> Signature redacted Accepted by: Ezra Zuckerman Sivan Nanyang Technological University Professor Professor of Technological Innovation, Entrepreneurship, and Strategic Management Chair MIT Sloan PhD Program 1 Understanding the Conditions of Bias: Essays on Gender Differences in Evaluation Outcomes across Three Empirical Contexts by Mabel Lana Botelho Abraham Submitted to the Sloan School of Management on May 6, 2015 in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Management Abstract This dissertation contributes to our understanding of when and how gender is incorporated into the evaluation of individuals, leading to unequal outcomes for similar men and women. Prior research has shown that because ascriptive characteristics, such as gender, are associated with widely-held performance expectations, evaluators often rely on gender as an indicator of quality, particularly when quality is uncertain or indeterminate. Whereas existing research has importantly documented that gender differences in evaluation outcomes exist, this dissertation shifts the focus to uncovering the conditions under which this is the case as well as the underlying mechanisms driving these observed gender differences. Specifically, the three papers in this dissertation contribute to our understanding of the evaluative mechanisms perpetuating gender inequality by answering the following overarching research question: Under what conditions and how do evaluation processes lead to different outcomes for comparable men and women, particularly when more relevant indicators of quality are available to evaluators? I draw on data from three distinct empirical contexts to examine when and how evaluations of similar men and women vary within social networks, a financial market setting, and an organization. I pay particular attention to the often levied criticism of gender inequality research, namely failure to adequately account for underlying quality or performance differences. I show that the gender of the evaluatee, or the individual being evaluated, plays a role beyond serving as a proxy for missing quality information and that male and female evaluators incorporate gender differently under certain conditions. Thesis Committee Roberto Fernandez William F. Pounds Professor of Management Emilio Castilla Associate Professor of Management Ray Reagans Alfred P. Sloan Professor of Management Susan Silbey Leon and Anne Goldberg Professor of Sociology and Anthropology 2 TABLE OF CONTENTS ACKNOW LEDGEM ENTS..............................................................................................................................................5 I INTRODUCTION.................................................................................................................................................9 1.1 2 OUTLINE .......................................................................................................................................... 10 EXPLAINING UNEQUAL RETURNS TO SOCIAL CAPITAL AMONG ENTREPRENEURS.......................................12 2.1 INTRODUCTION...................................................................................................................................................... 12 2.2 THEORY AND HYPOTHESES ....................................................................................................................................... 15 2.2.1 2.2.2 2.2.3 2.2.4 2.3 3 DISSERTATION Generating network advantages through activation.............................................................................. The role of resource-holders' gender beliefs in evaluations ..................................................................... The role of third-parties in evaluations.................................................................................................... Gendered occupations and bias................................................................................................................... DATA AND M ETHODS ............................................................................................................. ,............................... 16 17 19 21 23 2.3.1 Research Setting and Data........................................................................................................................... 2.3.2 Em pirical Strategy......................................................................................................................................... 2.3.3 Variables and Empirical Model.................................................................................................................... 2.4 RESULTS: W OMEN GET LOWER RETURNS TO SOCIAL CAPITAL ....................................................................................... 2.4.1 Replacem ent events ..................................................................................................................................... 2.4.2 Robustness checks and ruling out alternatives........................................................................................ 23 26 30 32 33 39 2.5 DISCUSSION .......................................................................................................................................................... 42 2.6 FIGURES AND TABLES..............................................................................................................................................47 NAMING YOUR DAUGHTER JACK: THE EFFECT OF GENDER IN ATTENTION AND EVALUATION .................... 3.1 INTRODUCTION ...................................................................................................................................................... 3.2 THEORY AND HYPOTHESES....................................................................................................................................... 3.2.1 3.2.2 3.2.3 3.2.4 3.3 3.3.1 3.3.2 3.3.3 3.3.4 3.3.5 3.3.6 3.4 3.4.1 3.4.2 3.4.3 3.4.4 3.4.5 The Role of Status-based Mechanisms of Discrimination for Gender Inequality .................................... Double Standards Theory and Gendered Outcom es ................................................................................. Stages of Evaluation..................................................................................................................................... Search Costs.................................................................................................................................................. DATA AND M ETHODS ............................................................................................................................................. Em pirical Context........................................................................................................................................... Sam ple .......................................................................................................................................................... Dependent Variables: The Evaluation Process........................................................................................... Independent Variable: Gender..................................................................................................................... Control Variables.......................................................................................................................................... Empirical M odel............................................................................................................................................ RESULTS ............................................................................................................................................................... Is there a Gender effect?. .............................................................................................................................. Ruling out the alternative that wom en are doing things differently? .................................................... Status and Increased Search Costs........................................................................................................... Unobserved gender differences ................................................................................................................... Robustness Checks........................................................................................................................................ 58 58 62 62 62 66 68 69 69 75 75 76 76 79 79 79 82 86 87 88 3.5 DISCUSSION .......................................................................................................................................................... 88 3.6 FIGURES AND TABLES.............................................................................................................................................. 92 4 FORMALIZATION REVISITED: CONSIDERING THE EFFECTS OF MANAGER GENDER AND DISCRETION ON CLOSING THE W AGE GAP.......................................................................................................................................100 4.1 INTRODUCTION.................................................................................................................................................... 4.2 THEORY AND HYPOTHESES..................................................................................................................................... 103 Form alization of Pay and Gender Inequality ............................................................................................. 103 4.2.1 3 100 4.2.2 Considering Manager Gender and Differences in Gender Biases using Social Identity Theory ............... 106 4.2.3 Explaining Variation in Manager Preferences Using Status Characteristics Theory ................................ 108 4.3 DATA AND M ETHODS ........................................................................................................................................... 110 111 Research Setting ......................................................................................................................................... 4.3.1 4.3.2 Participants................................................................................................................................................. 112 4.3.3 Com pensation Structure and Form alization .............................................................................................. 113 4.3.4 M easures .................................................................................................................................................... 115 4.3.5 Analytical Strategy ..................................................................................................................................... 119 4.4 RESULTS ............................................................................................................................................................. 120 4.4.1 Descriptive Statistics...................................................................................................................................120 4.4.2 Overall Effect of Form alization on Gender Pay Inequality ........................................................................ 122 4.4.3 Form alization and M ale versus Fem ale M anager Impact on Pay ............................................................ 123 4.5 DIScUSSION ........................................................................................................................................................ 127 4.5.1 Importance of Considering Characteristics of Manager and Formalization............................................. 128 4.5.2 Im portance of Accounting for Context-specific Status Ordering............................................................... 129 4.5.3 Lim itations and Future Research................................................................................................................ 130 4.5.4 Im plications for Organizations ................................................................................................................... 132 4.6 5 6 FIGURES AND TABLES............................................................................................................................................ CO NCLUSION 134 ................................................................................................................................................. 143 5.1 CONTRIBUTIONS................................................................................................................................................... 143 5.2 FUTURE RESEARCH............................................................................................................................................... 145 REFERENCES .................................................................................................................................................. 4 147 ACKNOWLEDGEMENTS The adage that it takes a village has never rung truer for me than it does today. The intellectual transformation from student to scholar is simply not possible in solitude. I am indebted to so many for their immeasurable contributions to both my intellectual development and personal growth leading up to this point. It is a privilege to have this opportunity to express my gratitude to those who have supported me in reaching this milestone. I owe tremendous thanks to each of my committee members. I would like to begin by thanking the chair of my dissertation committee, Roberto Fernandez, for his selfless dedication to my development. Roberto has been, and I am sure will continue to be, an extraordinary advisor to me. I credit him for guiding me through the transition from a consumer to a producer of knowledge. I fondly recall spending countless hours in his office working on every stage of our first project together, from idea generation through drafting a working paper. It was through this apprenticeship that I developed a taste for good research and learned to be meticulous in my own work. Importantly, Roberto also recognized when to let go. He slowly stepped back, while remaining accessible, and forced me to become an independent thinker. At the time this shift felt challenging, but as I reflect on my experience it is clear that this balanced support enabled me to launch my own research program and prepared me to enter the next phase of my career. In addition to teaching me to conduct strong research, Roberto also provided love and support along the way, often reminding me that he was a "full-service advisor". Every time I have thanked Roberto, he simply reminds me to pay-it-forward. I will certainly do my best and can only hope to be as generous an advisor to my future students as he has been to me. Roberto, I am indebted to you for being so thoughtful and patient with me. For this and so much more, I thank you. Susan Silbey has been an endless source of knowledge and inspiration. Not only has she taught me how to be a thoughtful researcher, but she has also helped me to define my identity as a scholar and to consider how this identity fit into the whole. After the birth of my daughter, it was Susan who helped me navigate those early months of balancing motherhood and my work, a much more challenging transition than I had anticipated. As I prepared for the job market, Susan told me a personal story where she, as the mother of two young daughters, was able to accomplish what seemed like an insurmountable task by remaining calm and focused. I carried her words with me through the following months as I slowly, but steadily, refined my research and put together the intricacies of my application packets. Her words were profound, but what was even more impactful was the subtle way that she reminded me of these same principles every time we interacted. She always made herself available when I needed her most and simultaneously gave me much needed positive reinforcement while pushing me along every dimension that needed improving. Thank you, Susan. While I have far from mastered the multiple facets of this craft, I am, without doubt, a better scholar because of you. I have also benefited greatly from working with Ray Reagans. He has often given me a renewed sense of direction about the appropriate theoretical framework or methodology, but even more 5 importantly he has consistently provided positive reinforcement, saying things like "your parents must be so proud". I am not sure he realizes that these positive words were often what I needed most to keep pushing forward when things became challenging. I am especially grateful to Ray for helping me work through many phases of the job market process. He really listened closely (several times, even when I may not have been making complete sense) and helped me to remain calm and confident during an otherwise stressful time. Ray, I thank you for showing me by example that it is possible to be a successful academic, while remaining true to myself. Emilio Castilla has always provided careful attention to detail, particularly during my early years in the program when my uncertainties were heightened. On numerous occasions, he willingly spent hours talking through my analytical questions and helping me to identify the appropriate methods. Emilio also taught me an important lesson about being a constructive critic of other's research - to always highlight what is done well, in addition to offering criticism. Emilio, in a world where we quickly learn to be critics, thank you for reminding me how to be most constructive in the process. I have also benefited immensely from the guidance and mentorship of many other MIT faculty. While this list is certainly not exhaustive, I owe special thanks to Ezra Zuckerman, Kate Kellogg, and Lotte Bailyn, who generously provided feedback on much of my work. Ezra tirelessly contributes to the development of the members of the MIT PhD community, always willing to engage deeply and being generous with his feedback. My first interactions with Ezra were in the classroom, where he helped me understand what Economic Sociology meant and sparked my interest in engaging with scholars in this area through my own research. He is a masterful teacher, with the uncanny ability to help students learn to identify theoretical gaps and develop clear theoretical deductions. I am humbled by his willingness to always make time to attend my presentations or provide feedback on my drafts, despite the many demands on his time. Thank you, Ezra, for helping to build our academic community and for demonstrating what it really means to be a contributing member. Kate and Lotte were my first points of contact at MIT and I am so grateful for their constant investment in me. Despite not having any formal obligations to provide support and feedback, I have always been able to rely on both of them. As a rising junior faculty member, I hope to emulate Kate, who has always made time for me despite facing inevitable time constraints associated with the tenure clock. I thank both of you for your endless support over the years. Not only is my research better as a result of your thoughtful feedback, but you have taught me how to remain balanced in a profession that can sometimes feel all-consuming. I owe great thanks to both The Kauffman Foundation and The American Association of University Women (AAUW) for the generous funding they provided me. Relatedly, I also thank my dear friend, Rosa Taormina, for providing incomparable research support. I cannot thank you enough for internalizing my need for accuracy and doing everything possible to help me meet these goals. You went above and beyond to help and I truly value our friendship. I have had the privilege of sharing my time at MIT with a tremendous group of fellow students, many of whom have become close friends. To Julia and Aruna, having both of you to laugh and cry 6 with over the years has been so important to me. I truly value our friendship and look forward to continuing to support each other as we enter the next chapter. I also thank Ben Rissing, Phech Colatat, Oliver Hahl, Eunhee Sohn, Roman Galperin, Josh Krieger, Jason Greenberg, JP Ferguson, Melissa Mazmanian, Joelle Evans, and Michael Bikard for providing advice and support through the years. I am also grateful for the broader MIT community. I owe a special thank you to Matthew Bidwell and Susan Perkins for always providing thoughtful, practical advice about navigating this career. Susan, I am especially thankful for all of the support you offered during the job market and for always being available to chat. To Hillary Ross and Sarah Massey, thank you for helping me stay (close to) on time for deadlines over the years and for always offering a warm smile. Having my brother, Tristan Botelho, as a colleague over the past four years has greatly contributed to my success and, even more importantly, my happiness in the PhD program. It is a rare opportunity to get to share in professional triumphs and failures with someone you also consider a best friend. I feel deeply blessed to be in that minority. Tristan, my most treasured memories as a PhD student involve you and I certainly cannot imagine this process without you in it. You have always been my best critic, while also being my biggest supporter, continuously pushing me just hard enough to help me inch towards the best possible version of myself. Most importantly, you have always been there to celebrate my successes and work through my failures as if they were your own. I often worry that I take far more than I give, but I hope I am half the support system for you that you have always been for me. It is without question that our collaboration has been the most fun project that I have worked on and I hope it is the first of many. Thank you for being the most amazing brother I could ask for and for giving me a wonderful sister and goddaughter. Kate, I cherish your love and support, and Rowan, Dinha loves you so much. I am most indebted to my wonderful parents, Manuel and Gloria Botelho, for teaching me the importance of education, supporting me every step of the way, and never second-guessing my ability to make it. You have always encouraged me to pursue my dreams and to seek happiness above all else, while never pushing me towards some preconceived notion of success. While you both continuously tell me how proud you are of me, I want you to know that I owe each of my successes to you. Above all else, I simply thank you for giving selflessly of yourselves and for loving me unconditionally. Mom, you are the most loving, patient, and supportive person and I am endlessly grateful for being on the receiving end of our relationship. Among so many things, you have taught me to value all people and recognize that each person has a contribution to make. You are the definition of authenticity, always acting in a manner that is consistent with your words and encouraging me to be myself through honesty and openness. I am especially thankful to you for being the same amazing source of love and support for Emery that you are for me. Thank you for spending several nights per week in our small townhouse and for doing so with so much love, expecting nothing in return. I am grateful for the opportunity to spend so much quality time with you during this period. I cannot 7 thank you enough. As I begin the next phase of my career, it brings me tremendous joy to know that my role as teacher also indirectly fulfills your own dreams. Dad, you have always worked tirelessly to be sure I was able to follow my dreams. Watching you work with determination and persistence, always committing to everything that you do, taught me the path to success. Though our work is quite different, our approach is so similar. Thank you for teaching me never to be afraid of hard work and that putting in some sweat and tears is often the only way. These lessons have been my source of motivation during times that I was most unsure of myself. To my in-laws, the Abrahams, I am blessed to have a family by marriage that is as incredible as anyone could ask for. To my mother-in-law, Linda, I am especially grateful to you for making the weekly trek to Arlington to care for Emery. Not only did you provide me with the support that I needed to reach this milestone, but you also gave Emery the gift of your time and love. You are an amazing Nana and I am forever grateful for your relationship with her. Last, but far from least, my deepest love and appreciation goes to my husband, Joe, and my daughter, Emery. Joe, I cannot imagine any part of this journey without you by my side, and honestly, I simply do not want to. Thank you for (almost) always knowing exactly what I needed, even when I did not know myself, and for always being a source of calmness and stability. Most of all, I thank you for being an equal partner in everything we do. Leaving Emery repeatedly for travel was hard on me, but I was able to do it and remain focused because I knew you were working twice as hard to keep things together at home. You never wavered in your faith in me, even when I questioned my own ability to do this. You gave me the confidence that I needed to persevere, even when insecurity set in. I owe my ability to "wrap it up", and the fact that I remained sane through the process, in no small part to you. I am forever grateful that I get to share my life with you. Emery, you bring me so much happiness and inspire me every day through your endless curiosity and unconditional love. You remind me to remain true to what is really important and to take time to enjoy every moment. In ways I hope you someday understand, you reignited my passion for my work and gave me the strength to cross the finish line. I strive to do right by you and, if nothing else, I hope to lead by example in showing you that you can accomplish anything. Joe and Emery, I dedicate this dissertation to you. As I look forward to our next adventure, I am filled with excitement and anticipation about what our future holds. I love you both more than you will ever know. Mabel Abraham May 4, 2015 Arlington, MA 8 1 INTRODUCTION Gender inequality is a persistent feature of the U.S. economy, with women commonly earning lower wages and being underrepresented in elite occupations and organizational positions (cf. Elvira & Graham, 2002; Petersen & Morgan, 1995). This dissertation focuses on one potential source of unequal outcomes for men and women: the evaluation process. Across a wide range of organizational and market settings, an individual's occupational and financial success is at least partly contingent on the assessment of an evaluator or audience (Fernandez & Fernandez-Mateo, 2006; Zuckerman, Kim, Ukanwa, & von Rittman, 2003). In making these assessments, evaluators often lack objective information about quality or performance, forcing them to rely on status signals as indicators of expected performance (Kovacs & Sharkey, 2014; Podolny, 1994, 2005; Simcoe & Waguespack, 2011). Prior research suggests that because ascriptive characteristics, such as gender, are associated with widely-held. performance expectations, evaluators often rely on gender as an indicator of quality, & particularly when quality is uncertain or indeterminate (Becker, 1957; Berger et al., 1977; Correll Ridgeway, 2003). This reliance on gender often leads to a female disadvantage since, as the lowerstatus gender, women are commonly expected to be worse performers (Correll & Ridgeway 2003). There is even some evidence that gender is incorporated into evaluations despite more relevant indicators of quality being available (e.g. Foschi, 1989). To fully understand the effect of gender in evaluative processes, it is necessary to move beyond simply documenting that men and women receive different evaluations, and to instead disentangle the conditions under which this is the case and the underlying mechanisms. By more carefully examining how the evaluations of similar men and women differ across distinct evaluative processes, this dissertation first identifies scope conditions for when assessments of similar men and women vary, and second identifies how evaluative processes contribute to these observed gender differences. Specifically, across the three essays of this dissertation I show that gender plays a role beyond serving as a proxy for missing quality information and that male and female evaluators incorporate gender differently under certain conditions. 9 This dissertation fills this gap by answering the following central research question: Under what conditions and how do evaluation processes lead to different outcomes for comparable men and women, particularly when more relevant indicators of quality are available to evaluators? To address this question, I draw on data from three unique empirical contexts to examine when and how gender affects: resource allocation within social networks, attention and ratings within a financial market context, and wages among employees within an organization. I pay particular attention to the often levied criticism of gender inequality research, namely failure to adequately account for underlying quality or performance differences by gender. Across the three essays of this dissertation, using fieldbased data and employing both econometric and qualitative research methods, I remedy this by comparing similar men and women. 1.1 DISSERTATION OUTLINE The next three chapters of this dissertation each present independent research papers related to the central question outlined in this introductory chapter. In Chapter 2, I examine whether resource exchange within social networks leads to unequal benefits for male and female entrepreneurs. For networks to yield benefits, resource-holders, as evaluators, decide with whom to share resources, or activate social ties. I study business referral networks where non-competing male and female entrepreneurs exchange business referrals, or connections to potential new clients. I find that despite having access to the same social ties, women still receive fewer referrals than similar men in the same social structure. In addition to identifying this gender imbalance, this paper suggests a new network mechanism for gender inequality: anticipatory third-party bias. I show that evaluators do not favor men when deciding whether to hire a fellow male versus female network group member for their own personal use. However, they do exhibit a male preference when connecting their outside contactsclients, family members, and friends-to their fellow network group members. In Chapter 3, together with my coauthor Tristan Botelho, I examine whether there are differences in the attention and subsequent ratings men and women receive in a financial market setting comprised of investment professionals. This research assesses whether gender remains salient 10 in the evaluation process within a competitive market setting, despite the presence of more prevalent and freely available performance information. We study an online platform where investment professionals openly share and view investment recommendations. Because there is a visible marketbased performance metric assigned to each recommendation and investment professionals strive to identify the most lucrative recommendations, we would not expect gender to play a role in the decision making process about which investment recommendations to view. However, we find that recommendations posted by women receive less attention than do recommendations posted by men, though there is no evidence of a gender-related performance difference. We find evidence that a driving force of this bias stems from attempts to reduce search costs. This gender difference in attention is most prevalent when evaluators are selecting recommendations from a large set of possible options where search costs are higher. Furthermore, we do not find a gender difference in subsequent stages of the evaluative process where the evaluator is focused on a single recommendation. Chapter 4 shifts the focus to the evaluator. Specifically, I examine whether male and female evaluators, in this setting managers, differ in how they use the discretion afforded by less formalized pay systems to affect gender pay inequality among employees. Existing research in management, economics, and sociology has largely argued that allowing managerial discretion through less formalized pay systems perpetuates inequality. However, implicit in this argument is that organizational decision makers uniformly adopt these policies, and specifically adopt these in a way that leads to a male advantage. I draw on social identity theory and status characteristics theory to argue that this is unlikely to be the case because it is important to consider how male and female managers may use discretion differently. Among 857 employees in 120 retail branches of a financial services firm, I find evidence of less gender pay inequality in terms of less formalized components of pay for employees in the lowest organizational ranks reporting to a female, relative to a male, manager. Chapter 5 concludes by summarizing the core contributions of this dissertation and offering a few suggestions for future research. 11 2 EXPLAINING UNEQUAL RETURNS TO SOCIAL CAPITAL AMONG ENTREPRENEURS 2.1 INTRODUCTION Despite fifty years of political mobilization and dedicated public policy, American working women maintain a consistently lower economic position relative to similarly qualified men (e.g., Bird & Sapp, 2004; Elvira & Graham, 2002; Petersen & Morgan, 1995). Women are particularly disadvantaged in terms of entrepreneurial outcomes. While there have been recent increases in the rate at which women establish new businesses (U.S. Small Business Administration, 2001), women own less than 30 percent of all U.S. firms (U.S. Department of Commerce, Economics and Statistics Administration, 2010, p. 1) and women-owned businesses continue to be smaller and less lucrative than firms owned by men (U.S. Department of Commerce, Economics and Statistics Administration, 2010; see also Bird & Sapp, 2004; Hundley, 2001; Loscocco, Robinson, Hall, & Allen, 1991; Merrett & Gruidl, 2000; Robb & Coleman, 2009). Since individuals commonly choose entrepreneurship as a means for career attainment and economic growth (cf. Sorensen & Sharkey, 2014), uncovering the mechanisms contributing to gender differences in entrepreneurial outcomes is of particular importance. These gender differences, coupled with the central role of social networks for entrepreneurs (e.g., Hallen, 2008), have resulted in a burgeoning literature focused on the role of social ties for the pervasiveness of gender inequality in entrepreneurship. As Stewart (1990: 149) stated, "For many entrepreneurs, their most significant resource is a ramifying personal network." But like the gender differences in the social networks of men and women more broadly, it has been argued that female entrepreneurs have less valuable networks (e.g., Aldrich, 1989; Aldrich, Elam, & Reese, 1997; Aldrich, Reese, & Dubini, 1989; Ruef, Aldrich, & Carter, 2003). The sorting of women into lower status occupations and organizational positions embeds them in networks with fewer high status and influential contacts than those of men (Brass, 1985; Campbell, 1988; Ibarra, 1997; Marsden, 1987; McGuire, 2000, 2002; Moore, 1990; Smith, 2000) and, as a consequence, women generate lower 12 returns from social capital than do men (Braddock & McPartland, 1987). Because high status actors occupy central positions that facilitate access to network resources (Lincoln & Miller, 1979), people with higher status contacts receive greater advantages when using their social ties. Thus, women's differential network structures and composition is posited as a key factor in the pervasive under& representation and lower success rates of women among entrepreneurs (Aldrich, 1989; Cromie Birley, 1992; Katz & Williams, 1997; Renzulli, Aldrich, & Moody, 2000). Although lacking access to others' resources constrains women's opportunities, it is unclear whether men and women with the same access to resources through social ties would receive equal benefits. The mere presence of social ties does not necessarily guarantee that resources will be exchanged. For networks to yield benefits, actors possessing resources must decide to share those resources with network contacts, in effect, behaviorally activating their social ties (Lin, 2001; cf. Smith, 2005). In other words, resource-holders serve as evaluators deciding with whom to share resources, or activate a social tie. Despite the central role of resource exchange within networks for generating benefits from social ties, we do not have a clear understanding of how gender affects resource sharing, and whether this resource allocation process has implications for gender inequality. Scholars have argued that in many situations diffuse status characteristics such as gender influence & evaluations of actors in a way that disadvantages women (e.g., E. J. Castilla, 2008; Castilla Benard, 2010; Correll & Benard, 2006; Correll, Benard, & Paik, 2007; Ridgeway, 2011; Turco, 2010), suggesting that gender may also play a role in how resources are exchanged among network contacts. These gender-based expectations have been found to be particularly salient in male-dominated areas, such as entrepreneurship (cf. Gupta, Turban, Wasti, & Sikdar, 2009). Female entrepreneurs are commonly perceived to be less competent (Buttner & Rosen, 1988; Thebaud, 2010) and to lack credibility (Carter and Cannon, 1992; Moore & Buttner, 1997). Thus, to develop a more complete understanding of how networks contribute to gender inequality among entrepreneurs, it is critical to examine how the process of resource allocation within networks impacts gender inequality. This research moves beyond structural network accounts of gender inequality to determine whether men and women with the same access to social ties receive equal benefits, or returns to social capital. 13 In this paper, I show that women receive fewer benefits through social ties than do similar men in the same social structure. I study business referral network groups to examine how referrals, or connections to potential new clients, are exchanged among entrepreneur-members.' A comparison of the referrals accruing to men and women with the same access to social ties reveals that female entrepreneurs receive fewer business referrals. Consequently, this study identifies that access to valuable social ties alone will not necessarily eliminate gender differences in outcomes produced through social ties. Resource allocation plays a role in perpetuating gender inequality as women receive fewer benefits through social ties relative to similar men with access to the same social ties. Further, I posit a new network-based mechanism to explain this gender inequality, which I term anticipatory third-party bias. I show that while members do not favor men when directly connecting to other members in their network group, they do exhibit a male preference when connecting these fellow network group members to their outside contacts-clients, friends, and family members. Where entrepreneur-members expect that a client, friend, or family member has a preference for men over women, they disproportionately give referrals to male rather than female network group members. In other words, men and women are equally likely to receive business directly from their network contacts, but women receive far fewer connections to the clients, family, and friends of their network contacts, or third-party referrals. Further, I find that this gender difference in receiving third-party referrals is most pronounced among women in male-dominated occupations. Thus, despite having equal access to network resources through social ties, women receive lower returns to their social capital because resource-holders respond to what they believe are the gender preferences of others, or third-parties. To demonstrate how resource allocation, rather than differential network composition, contributes to gender inequality, this study compares the resources acquired through social ties by similar men and women embedded in the same network groups. It is only possible to identify whether and how resource exchange within networks affects gender inequality by holding network I use entrepreneur-members, members, network group members, and network contacts interchangeably to refer to members of these business referral groups. To differentiate between the person deciding whether to give a referral to a fellow member and the potential recipient of these referrals I use resource-holder and resource-seeker respectively. 14 structure constant. In this respect, the tendency for networks to be gender-segregated poses an empirical challenge. To overcome this challenge, I collected six years of referral data from a popular business networking club organized to help entrepreneurs generate new business through social ties. This setting offers a unique opportunity to observe how resources are exchanged among male and female entrepreneurs in the same networks. Entrepreneurs in these clubs represent a wide range of occupations and meet weekly to exchange business referrals that connect recipients with potential clients. I unobtrusively collected all referral data, observed three groups for two years, and interviewed 18 entrepreneur-members. In my analysis, I focus on 416 cases where a member exits a group and is replaced by a new member in the same occupation. Leveraging these replacements, particularly those involving a change in gender, allows for a comparison of men and women in the same occupations, with access to the same social ties, thus providing better identification of any gender difference in returns to social capital. I proceed as follows. First, I develop hypotheses about whether and why women receive fewer resources than do men embedded in the same social networks. I explain how gender preferences, or performance expectations, may prevent women from generating the same returns to social capital, despite having the same access to social ties, as men. Second, I introduce the research setting and analytical method used to test whether women receive lower returns to social capital and to identify the role of expectations about third-party preferences in perpetuating gender inequality. Third, I present results which demonstrate that actors' anticipation about the gender preference of their contacts prevents sharing referrals with women under certain conditions. This finding suggests a new mechanism by which networks perpetuate gender inequality: anticipatory third-party bias in social capital activation. Therefore, access to valuable social ties alone will not necessarily mitigate gender differences in outcomes from using social ties. I conclude with a discussion of the theoretical and organizational implications of the research findings for our understanding of how social networks contribute to gender inequality among entrepreneurs and more broadly. 2.2 THEORY AND HYPOTHESES 15 2.2.1 Generating network advantages through activation Social networks are often conduits of valuable resources, such as information and influence, which can be leveraged to achieve desirable outcomes (Burt, 1995; Granovetter, 1995; Lin, 2001; Podolny & Baron, 1997). As such, social capital refers to the potential and actual set of resources made available through an actor's direct and indirect social ties to others (Coleman, 1988; Lin, 2001). The advantages associated with social ties, however, are not automatic. An actor's network structure and composition provides the potential set of resources available through social ties, but & within this structure actors exercise agency, and preferences (Emirbayer & Goodwin, 1994; Gulati Srivastava, n.d.; Zeng & Xie, 2008). The mere presence of social ties alone does not necessarily mean that resources will be exchanged. In his theory of social capital, Lin (2001: 29) makes a clear distinction between resources that are available through social ties and resources that are "accessed and/or mobilized in purposive action". Therefore, for networks to yield benefits, social capital must be both available and activated such that a resource-holder shares resources with a network contact(Lin, 2001; cf. Marin, 2012; Smith, 2005). In this way, network structure and composition constrain choice (Zeng & Xie, 2008) by providing the potential social capital, or pool of resources, to which actors have access and the opportunity to exploit. While we still lack a comprehensive understanding of how people use social ties to generate benefits (e.g., Ibarra, Kilduff, & Tsai, 2005), recent research has shown that individuals often activate only a subset of the social relations to which they have access. For example, limitations in the cognitive recall of relationships (Smith, Menon, & Thompson, 2012) and reputation concerns (Marin, 2012; Smith, 2005) may prohibit actors from activating social ties. In considering factors that may prevent resource-holders 2 from giving resources to particular network contacts, labor market studies examining when people provide their social ties with job information are particularly illustrative. For example, in a study of the black urban poor, decisions about whether to share job2 Choice can be exercised by either the resource holder in deciding whether or not to share resources (Marin, 2012; Smith, 2005) or the resource seeker in deciding whether or not to express needs (Hurlbert, Haines, & Beggs, 2000). In my setting, all actors are similarly seeking and expressing needs for resources, which I discuss further in the results section. Thus, I focus on the role of the resource holder in deciding whether to share resources with others and how this process impacts gender differences in the benefits men and women receive through social ties. 16 related information with potential job-seekers are largely influenced by expectations of how doing so will impact the information holder's reputation (Smith, 2005). Similarly, among professional workers, information holders protected their reputation with their social ties by refraining from sharing information about job opportunities unless their contacts had explicitly asked or the job was a clear match (Marin, 2012). The importance of activation for understanding how social ties generate benefits has been clearly established. Some scholars have begun to examine how activation contributes to inequality in benefits (e.g. Smith, 2005), however, extant research has not focused on gender inequality. 2.2.2 The role of resource-holders' gender beliefs in evaluations In social networks, resource-holders act as evaluators assessing the quality of network contacts before deciding whether to share sought-after resources (cf. Smith, 2005). Ascriptive characteristics, such as gender, influence the evaluation of actors in ways that generally disadvantage women (E. Castilla, 2008; Castilla & Benard, 2010; Correll & Benard, 2006; Correll et al., 2007; Ridgeway, 2011; Turco, 2010). Commonly, evaluators are unable to observe the actual quality of the individuals being assessed, forcing them to rely on more accessible indicators of expected quality (Podolny, 1993, 2005), often historically marked by gender, race, religion and social class. Both status-based and statistical theories of discrimination posit that, because gender is conventionally understood to convey information about an actor's quality, or expected performance, evaluators incorporate gender into their assessments to remedy this problem of uncertainty, that is, lacking knowledge of actual quality. Status-based preferences stem from beliefs that gender is a status characteristic, where men are more highly valued, and generally perceived as more competent, than are women (Berger, Fisek, Norman, & Zelditch, 1977; Correll & Ridgeway, 2003). These performance expectations are often incorporated into evaluations of male and female actors (Correll and Benard, 2006) such that men's performances are evaluated more positively than those of similar women (e.g., Ridgeway & Correll, 2004; Ridgeway & Smith-Lovin, 1999; Wagner & Berger, 1993) leading to a male social advantage (Webster & Hysom, 1998). Similarly, economic theories of statistical discrimination posit that if quality or performance is difficult to observe directly and has 17 different distributions for men and women, it may be rational for an evaluator to prefer men (for reviews, see Correll and Benard 2006; England 1994, pp. 60-63). Therefore, if an evaluator has had more positive experiences with men than with women, it would be logical to use this information when comparing a male and female candidate to conclude that the man is more likely to be of higher quality. However, reliance on gender as signal of quality may persist even when actual, more relevant performance information is freely available (Berger et al., 1977; Correll & Ridgeway, 2003). Double standards theory, a particular status-based account of discrimination, suggests that gender serves as a distinguishing characteristic in assessments of competence even when men and women with equal performance are compared (e.g., Foschi, 1989). According to this perspective, discrimination arises because stereotypes about the relative expected performance of men and women influence evaluations beyond freely available evidence of equal past performance, particularly when the evaluator is sorting through a large number of candidates (e.g. Botelho and Abraham, 2014). While the role of gender for network activation remains unclear, these theories of discrimination suggest that resource-holders may favor men when assessing whether or not to share resources with network contacts. In job search, for example, using networks tends to channel women into low-wage, female-dominated jobs because women commonly receive job information from female contacts (Drentea, 1998; Hanson & Pratt, 1991; Mencken & Winfield, 2000). Based on this observed association, it is often assumed that women lack access to social contacts possessing more valuable job information. It is plausible, however, that despite having access to high status network contacts with information about better jobs, women are not receiving information from these contacts. In other words, if individuals with valuable job information are choosing to share this information with their male, but not female, contacts, women may be relegated to relying on a subset of their network for job information. Thus, resource-holders may take gender into account when assessing whether or not to share resources with other entrepreneurs in their network such that: 18 Hypothesis 1: Female entrepreneurs will receive fewer resources from network contacts than will similar male entrepreneurs in the same network. 2.2.3 The role of third-parties in evaluations The arguments presented so far highlight how resource-holders' gender beliefs or stereotypes may drive them to disproportionately favor male network contacts over comparable women. This challenges the assumption that men and women with access to the same social ties will receive the same benefits. Alternatively, expectations about the gender preferences of relevant others, irrespective of an evaluator's personal beliefs or preferences, may also lead an evaluator to favor men. Networks are social structures where actors may be influenced not only by their personal beliefs, but also by how they believe their actions will be perceived by a third party. At the most basic level, the exchange of resources between network contacts, or social capital activation, involves at least two social actors. In many cases, however, the sharing of resources within networks is a triadic process (cf. Rubineau & Fernandez, 2014). In addition to a resource-holder and a resourceseeker, there are often third-parties such as other actors or organizations (e.g. employers), involved in, or observing, resource exchanges between network contacts. For example, the exchange of job information involves an information holder who knows of a vacancy, a job seeker receiving the information, and the employing organization looking to fill that vacancy. In this example, the employer represents a third-party involved in the direct sharing of job information between two network contacts. Generally, third-party observers, or audiences, play an important role in shaping actor behavior. The presence of an audience serves to impose expectations of what is considered appropriate behavior in that particular context (Zuckerman, Kim, Ukanwa, & James von Rittmann, 2003). This is especially pronounced when the consequences, success or failure, of an actor's decision is not her personal quality criteria, but rather the expectations of her audience (Clark, Clark, & contingent on the reaction of others. Under these conditions, a decision maker is challenged to satisfy Polborn, 2006; Emerson, 1983; Jensen, 2006; Ridgeway & Correll, 2006). For example, employment agencies, acting as hiring intermediaries, are strongly motivated to anticipate their client's---the 19 hiring firm's---ideal candidate in order to select the most appropriate applicants from the overall pool (Fernandez-Mateo & King, 2011). Specifically, if decision makers know the conclusion that their audience wants them to reach, they tend to bias their choices in an effort to satisfy the audience (Lerner & Tetlock, 1999). When there is uncertainty about a third-party's preferences, third-order inferences about status, or assessments of who or what most other people believe to be of higher status, suggest that it is likely for this actor to prefer men (Correll, Ridgeway, Zuckerman, Jank, & Jordan-Block, 2014). Thus, insofar as a decision-maker is dependent on approval from a third-party and anticipates that the third-party is aware of the status structure (i.e. men higher status), it may be rational for a decisionmaker to favor high-status options, even when the audience's true preferences are unknown and the decision-maker personally does not endorse these beliefs. There is evidence suggesting that anticipating the preferences of an audience, or third party, may serve to perpetuate gender inequality. Economists have long posited that inequality in labor market outcomes is at least partly a function of customer discrimination (Neumark, Bank, & Nort, 1996). In employees leads hiring agents to disproportionately select male candidates (Fernandez-Mateo & terms of gender inequality in hiring, expectations of customer or client preferences for male King, 2011). Similarly, the likelihood that a female attorney is promoted to partner is significantly greater when the law firm has women-led clients (Beckman & Phillips, 2005). Therefore, in social networks, expectations about the gender preferences of relevant third-parties is proposed to play a role in resource-holders decisions of whether to share resources, such that: Hypothesis 2: Female entrepreneurs will receive fewer resources from network contacts than will similar male entrepreneurs in the same network when the exchange of resources involves a third-party. Importantly, the proposed role of third-parties and the accountability they evoke can exist both under conditions where the resource-holder has personal status beliefs or biases and under conditions where the resource-holder has more egalitarian preferences (c.f. Correll et al. 2014). For anticipatory 20 third-party biases to play a role, the success of the resource-holder's social capital activation decision must simply be contingent on approval from a third-party. 2.2.4 Gendered occupations and bias The degree to which a third-party has a preference for men is likely to be context dependent. More generally, the potential for bias is magnified when an actor occupies a role that is incongruent with his or her gender (Foschi, 1989). The notion of the ideal worker (Acker, 1990; Williams, 1999) offers an explanation for the heightened use of gender in evaluations of actors in gender incongruent roles. The ideal worker image defines the characteristics of the individuals who are expected to be strong performers (Gorman, 2005). As such, individuals are commonly evaluated against this norm(Heilman, 1983). To the extent that the ideal worker image is gendered, it is increasingly likely that evaluations favor members of one gender group over another as gender is particularly salient in these contexts (cf. Turco, 2010). Consistent with this view, female disadvantage has been posited to be most pronounced when women are in roles or perform tasks typed as male (Foschi, 1989) or where they are the numerical minority (Kanter, 1977). The numerical representation of men and women in a particular occupation is a primary driver of perceptions that the role is typed as male or female. For example, masculine personality and physical attributes are perceived as essential for success in occupations historically dominated by men (Cejka & Eagly, 1999). Some research has shown that evaluators tend to give lower assessments to women occupying an occupational role that is gender incongruent, or maletyped (Eagly, 2004). As compared to their male counterparts, women tend to be evaluated as less competent in roles traditionally held by men (Davison & Burke, 2000; Eagly & Karau, 2002), with their successes commonly attributed to luck rather than skill (Swim & Sanna, 1996). Therefore, as related to hypothesis 2, the degree to which a female network contact occupies a role that is gender incongruent is likely to have implications for resource-holder's decisions of whether to share resources when the information sharing involves a third-party such that: 21 Hypothesis 3a: When the exchange of resources involves a third-party, there will be a greater female penalty in receiving resources from network contacts among women in maledominated occupations, such that female entrepreneurs will receive even fewer resources relative to similar male entrepreneurs in the same network when in male-dominated occupations. It is less clear whether men occupying gender incongruent roles face similar penalties. Male minorities have been posited to reap benefits from their position as tokens ascending to high status positions within these occupations more quickly than do women (Kmec, 2008; Williams, 1992). Despite these advantages, men are posited to encounter discrimination from people outside of their profession (Williams, 1992). Therefore, it may be reasonable for a resource-holder to anticipate that third-party contacts have a preference for women in female-typed occupations. For example, a resource-holder may expect that his third-party contact has reservations about a male child-care provider, an occupation traditionally held by women, and as a result, refrain from making referrals in this case. Further, recent audit studies have suggested that in occupations where women are the clear majority, such as secretarial positions, applications from women are favored over applications from men (Booth & Leigh, 2010; Riach & Rich, 2006). Therefore, the degree to which a male network contact occupies a role that is gender incongruent is likely to have implications for resourceholder's decisions of whether to share resources when the information sharing involves a third-party such that: Hypothesis 3b: When the exchange of resources involves a third-party, there will be a male penalty in receiving resources from network contacts among men in female-dominated occupations, such that male entrepreneurs will receive fewer resources relative to similar female entrepreneursin the same network when in female-dominated occupations. One possible explanation for these mixed findings regarding the outcomes for men in femaledominated roles stems from the overarching status ordering of gender. In general, gender is a status characteristic, where men are higher status and, as a result, generally more highly valued and perceived as more competent, than are women (Berger et al., 1977; Correll & Ridgeway, 2003). Since gender is a diffuse status characteristic where men are typically seen as the higher status gender, even when men are in female-typed roles, there may be a higher tolerance for this inconsistency. Thus, when a resource-holder is assessing the gender preferences of a third-party, the threshold for 22 assuming a female preference is likely higher. For example, a resource-holder may be more apt to take a chance and refer his client to a male child-care provider than to a female contractor. As a result, the degree to which a network contact occupies a role that is gender incongruent is likely to have greater implications for resource-holder's decisions of whether to share resources for women than for men such that: Hypothesis 4: There will be a greater penalty in receiving resources from network contacts for women in male-dominated occupations than for men in female-dominated occupations when the exchange of resources involves a third-party. DATA AND METHODS 2.3 2.3.1 Research Setting and Data Data for this study were collected from "RefClubs", a popular organization providing a forum for entrepreneurs to exchange information and business referrals. Entrepreneurs pay annual dues to become members of RefClubs and gain access to a networking group of other entrepreneurs in the same geographic region. The primary purpose of the organization is to bring together individuals seeking to grow their businesses and provide them with a forum for sharing information about potential new clients. These are instrumental networking groups where individuals join with the explicit goal of establishing and leveraging social ties with other members of the group to generate new business. I have studied the exchange of referrals among entrepreneur-members within 37 individual RefClubs network groups using archival records including each and all individual exchanges. I collected, coded, and analyzed records of referrals from these network groups for the years 20072013. To deepen my understanding of this empirical context, between October 2011 and October 2013 I spent 40 hours observing individual weekly group meetings across 10 unique groups and conducted 18 semi-structured interviews with individual entrepreneur-members. The quantitative data I analyze include complete information on the referrals generated and received by each of the 2,310 members in these 37 unique network groups. These data offer two key 23 advantages for testing the hypotheses I have outlined. First, through analysis of network records, I unobtrusively observed the actual exchange of resources among entrepreneurs allowing me to move subjective perceptions of survey respondents (Bernard, Kilworth, & Sailer, 1981; Quintane & beyond common approaches relying on self-reports. Self-reports have been found to be tainted by the Kleinbaum, 2011) calling into question the reliability of network surveys (Bernard et al., 1981), although recently, studies have made progress in avoiding respondent biases by using email exchange records (e.g., Kleinbaum, 2012; Kleinbaum, Stuart, & Tushman, 2013). Second, in these referral networks, men and women are co-located within the same network groups and thus have the same structural opportunity to access resources. This allows me to hold network structure constant and identify how resource exchange within networks affects gender disparities. Without knowledge of the opportunity structure for accessing resources, or the set of social ties that could have been activated, it is not possible to distinguish between structure- and activation-based explanations. Self-reports, for example, do not capture the entire network, or the opportunity structure, from which resources are generated. Without holding constant network structure, observations of women's reduced benefits could be the result of women's differential access to valuable contacts or of women's inability to mobilize social ties (see McDonald 2011 for a similar critique questioning whether feminized networks are less valuable). [INSERT FIGURE 1] Figure 1 depicts the structure and composition of these network groups. Within each group there are an average of 40 entrepreneur-members. The gender composition of these network groups mirrors that of women's presence in entrepreneurship more generally (e.g., U.S. Department of Commerce, Economics and Statistics Administration, 2010, p. 1), with approximately 35 percent of members being female. Entrepreneur-members work in a wide range of occupations, as illustrated in Figure 2A. Within each network group, there is only one member from each occupational specialty, thus reducing the role of competition among members in referral exchange. The specification of occupational specialty, as opposed to broader occupation or profession categories depicted in Figure 24 2A, is deliberate as members may overlap in the latter. For example, a group may have two members from the legal occupation, both of whom are attorneys, where one specializes in estate planning and the other in family law. While these members are in the same broad occupation, they are not in competition as they are not vying for the same customer base. The non-competing nature of these groups reduces the likelihood that observed gender differences in referral patterns result from men and women being differentially impacted by competition. Weekly group meetings have a routinized format for entrepreneur-members to learn about each other's businesses and to exchange business referrals. Meetings run from 90 to 120 minutes and are scheduled outside of normal business hours, typically held either in the early mornings or evenings. This time is largely structured with an opportunity for informal interaction among members before the start of the meeting. During meetings each member is allotted a time slot to speak publicly to the group about specific business and referral needs. This short presentation provides each member the opportunity to educate other members about their business in order to facilitate referral exchange. Since all members partake in this request for referrals, coupled with the instrumental nature of these groups, this setting minimizes concerns that observed gender differences are due to women's tendency not to ask for help (e.g., Babcock & Laschever, 2003). It is also during these weekly meetings that business referrals, or information about potential new clients, is exchanged among members. Referrals generated in these groups provide a non-trivial source of revenues for members. On average, each member receives approximately 23 referrals per year contributing to approximately $10,000 in new business per member per year. Each member has the opportunity to activate a tie with, or pass information about a potential new client to, another member during the meetings. Making a referral to another member involves completing a form where a copy of this information goes directly to the member receiving the referral and a second copy is entered into the groups' records by an administrator. In addition, all business referrals exchanged between members and successful business transactions that resulted from earlier referrals are announced verbally to the group. Through this process of publicizing resource exchange, 25 members may become aware of the relative value that each other member brings to the group in terms of generating referrals and business revenues. There are two distinct types of business referrals exchanged among entrepreneur-members, which I refer to as direct use and third-party referrals. In direct use referrals, a member expresses interest in directly hiring another member in the group to provide a service. For example, the accountant in Group A (Figure 1) indicates to the real estate agent, also in Group A, that she would like to hire him to help her find new office space. Since direct use referrals reflect the personal consumption of members, these referrals represent the referring member's, or resource-holder's, personal gender preferences or beliefs. Alternatively, third-party referrals involve a member providing another member with information about a third-party contact---client, family member, or friend---who is not a member of the group but may be interested in hiring the member. In third-party referrals, the accountant activates the tie to the real estate agent by connecting the real estate agent to one of her clients, whom the real estate agent would not have access to otherwise. Therefore, third-party referrals are subject to expectations of the gender preferences of a referring member's third-party contact. [INSERT FIGURE 2A] [INSERT FIGURE 2B] 2.3.2 Empirical Strategy To identify whether women receive fewer referrals than men with access to the same social ties, the natural approach would be to count the number of referrals male and female entrepreneurs receive and estimate a negative binomial regression with gender as the main predictor variable of interest. By design, the empirical setting I use in this study provides an opportunity to compare men and women co-located within the same networks. In this setting all members within an individual networking group have the same opportunity to obtain business referrals. As a result, I am able to hold network structure constant by including network group fixed effects allowing for a comparison of the number of referrals men and women within the same network group receive. In addition, to 26 account for potential gender differences in contributions to other members and participation in the group, this model would include relevant controls including number of referrals given to others, attendance, and tenure for each entrepreneur-member. However, this empirical approach would fail to consider the possibility of occupational sorting. It & is well-established that men and women tend to be sorted into different occupations (Fernandez Sosa, 2005), particularly among entrepreneurs where men and women historically own businesses in different industries (Rosenfeld, 2002). As Figure 2B reveals, the representation of female entrepreneur-members varies across occupations with women being underrepresented in some and overrepresented in others. Since approximately 35 percent of members are female, female members are underrepresented in occupations at the far bottom-left of Figure 2B and overrepresented in occupations at the top-right. Thus it is plausible that, despite having access to the same social capital, observations that women receive fewer referrals than do men in the same group are a function of occupational sorting. For example, entrepreneur-members may refer to female members in their group less frequently not because they are women but rather because there is less demand for the particular businesses that women tend to represent, such as education and training, than for businesses dominated by men, such as construction and trades. As aforementioned, within any single group in my setting there can only be one member from each detailed occupational category. Therefore, simply comparing the number of referrals men and women receive does not directly allow for a comparison of men and women in the same occupations who also have access to the same social ties. One way to begin to take occupational sorting into account is to compare men and women in the same occupations across network groups using occupation fixed effects. While this approach accounts for occupational heterogeneity, and reveals a similar female penalty in referrals received (see Table 4, Model 3), it does not compare men and women with the same access to social ties. To more carefully compare similar men and women, I use a unique identification strategy that simultaneously accounts for occupation and network group. I analyze cases where a member exits from a group and is replaced by a member in the same detailed occupation ("replacers"). This offers 27 a natural laboratory for examining my research question. Leveraging these replacers allows for a comparison of men and women in the same occupations, with access to approximately the same social ties, providing controlled conditions for better identification of the gender differences in returns to social capital where they exist. While at any given time each group has only one person in each detailed occupation (e.g. one real estate attorney), once a member decides to exit the group (i.e. leaver) it is possible for another person in that same occupation to join (i.e. replacer). As Figure 1 reveals, without focusing on these replacers, achieving either of these objectives would come at the expense of the other. Therefore, comparing the returns to social capital for members involved in a replacement event provides an opportunity for cleaner identification of the gender differences in returns to social capital. During my study window, I observed 416 of these replacement events where a member exits a group and is replaced by another member in the same occupational specialty. To ensure that the replacers and the leavers had access to approximately the same social capital and were employed in the same detailed occupation, I used strict boundaries for defining replacement events. First, I define replacements using the most narrow definition of occupational specialty available. For example, a replacement event is defined as a real estate attorney leaving and being replaced by another real estate attorney, as opposed to an attorney being replaced by an attorney. Since men and women are sorted into different detailed occupations, using this more fine-grained occupational definition accounts for gender differences even within broad occupation categories. Second, I include only replacement events where the replacers enter a group within 12 months of the leaver exiting the same group, with a mean length of 133 days between the exit and replacement, to maximize the overlap in social ties available to each of the involved members. In any 12 month period, greater than 70 percent of members remain, 3 resulting in the leaver and the replacer having approximately identical access to resources through social ties in this context. 3 One way to improve the overlap in the social ties available to leavers and replacers is to shorten the period between the leaver exiting and the replacer entering the group. For example, within a six month window approximately 87 percent of the group remains intact. But it takes groups several months (approximately four months on average) to replace an exiting member. Therefore, shortening this window results in a drastic reduction in the number of cases of 28 [INSERT TABLE 1] Table 1 illustrates the types of replacement events that I observed. Of the 416 replacement events meeting the criteria outlined above, approximately 60 percent involved a same-gender replacement: While these cases alone do not provide a means for comparing the relative number of referrals women receive relative to comparable men, these provide a baseline for comparing the replacement cases involving a gender switch enabling me to rule out a key alternative explanation. Without these same-gender replacements, it would not be possible to disentangle whether an observation that women replacing men received fewer referrals was due to the replacer's gender or to a newcomer effect. Comparing differences between the referrals received by replacers and leavers across the different categories provides a means for isolating gender effects from the effects of being a new member to the group.4 [INSERT TABLE 2] Importantly, the subsample of entrepreneur-members involved in a replacement event are representative of the study population in a number of key ways. Replacements occurred in each of the 37 chapters in my study population and represent most occupation specialties present. Unsurprisingly, the occupation specialties not represented among replacement events are those that are least common in this setting, each representing less than five percent of all members. The gender composition of the members involved in the replacements is identical to the study population, and to women's presence among entrepreneurs more generally, with 35 percent being female. Furthermore, a leaver being replaced by a new member in the same detailed occupation. The results are substantively robust to the following alternative specifications: shortening the window to a six month period while broadening the definition of same occupation and shortening the window to a nine month period while using the same fine-grained occupational definitions without the detailed occupation fixed effects. Because the goal of this empirical strategy is to more carefully compare men and women within the same occupations, the analyses focuses on replacements within the same detailed occupation and occurring within 12 months of a leaver's exit. 4 This does not rule out the possibility that there may be differences in the tenure and years in a member's life cycle with the group between the leavers and the replacers. For these differences to impact my results, the differences between leavers and replacers would need to vary across replacement types in a way that led to underestimating the referrals that replacers received in the male-to-female group but not in the other groups. These differences do not appear to be sizable nor to differ across groups. To minimize this concern, I also estimate models limited to replacement events where the absolute difference in tenure for the replacer and leaver was less than six months. The results are robust to this alternate specification as presented in the appendix, Table Al. 29 the distribution of men and women to occupations among this subsample mirrors the broader pattern, with most occupations being within two percentage points. As shown in Table 2, the pattern of referrals received overall and by referral type (i.e. direct versus third-party referrals) is largely identical among the replacements and the study population. Replacers, on average, give slightly fewer referrals to, and generate less money for, other entrepreneur-members in their group per year.' All models control for the referral giving behavior of the replacer and the leaver to be sure that these differences are not affecting the results. 2.3.3 Variables and Empirical Model The analysis in this study examines whether there are gender differences in the number of referrals each member received per year among male and female entrepreneurs in the same network groups. This analysis uses three dependent variables - direct use referrals received, or the total number of times a member receives requests for services directly from other members in the network group for their own consumption (i.e. personal or business use) per year, third-party referrals received, or the total number of times a member receives connections to the third-party contacts (e.g. clients) of other members in the network group per year, and total referrals received, or the total number of direct and third-party referrals a member receives per year. Differentiating between direct referrals received and third-party referrals received allows for testing the mechanism that resourceholder's expectations of the gender preferences of others drive observed gender differences, as stated in hypothesis 2. Whereas third-party referrals received involve a third-party, the decision to make a direct referral is based only on the preferences of the resource-holder. Thus, comparing the referrals women receive relative to men across these two types of referrals provides a way to differentiate whether anticipated third-party gender preferences or personal preferences are driving observed gender differences. s The observed difference in referrals given and dollars generated by the replacer group relative to the broader study population is the result of slight tenure differences between these groups. Replacers have nearly one year shorter tenure on average. Limiting the study population to those with the same tenure as the replacer group eradicates this difference. 30 All models include controls for the following individual characteristics---the number of total referrals given by a member per year; the total number of additional meetings a member partook in with other members in addition to weekly meetings per year; tenure with the group; the total number of weekly meetings a member missed, or absences, per year; and the total dollar amount a member generated for other members in thousands, or dollars generated, per year. To examine gender differences within network group and occupation, I also include fixed-effects for network group and occupation, using the Bureau of Labor Statistics Standard Occupational Classification (SOC) codes, in some models as noted.' The first set of models leverage the full study population to establish whether there is evidence of an overall female disadvantage. I estimate negative binomial regression models to predict total referrals received, which is a skewed count variable. The remaining analyses focus on the replacement events to more carefully isolate how resource exchange among social ties differentially benefits men and women. In these models, I compare men and women in the same detailed occupations, with access to approximately the same social ties, to determine whether there are gender differences in the number of referrals men and women receive. Limiting the analysis to these replacement events helps to rule out the alternative that observed gender differences are due to occupational sorting. This analysis of the entrepreneur-members involved in replacement events focuses on the difference in the average number of referrals received by a replacer and the leaver that was replaced in terms of each of the three measures of referrals. For each measure - the average number of total referrals received by a member, the average total number of direct use referrals received by a member, and the average total number of third-party referrals received by a member - I calculated the difference in the number of referrals received between the replacer and the leaver that they replaced as follows, Y, 1= Y, - Y 1, 6 where Y1 represents each of the three types of referrals, r indexes In my main analyses I use the SOC major groups codes. My main results are robust to using the more fine-grained SOC minor group codes. Because the intercoder reliability for coding occupations into the major codes was higher, 97 percent versus 88 percent, I report results based on models using the SOC major codes. 31 the replacer and 1 indexes the leaver. As such, a negative value in these dependent variables indicates that the replacer received fewer referrals on average than did the leaver whom he replaced. The key independent variables are the four types of replacement events - male-to-male, femaleto-female, female-to-male, and male-to-female. I created four dummy variables to capture each of these categories with male-to-male as the reference group. The effect of each of the three remaining categories represents how the difference between the referrals received by a replacer and leaver in the given category compares to the difference between male replacers and the male leaver they replace in the reference category. All models in the analysis of replacements include the aforementioned controls for individual characteristics at both the level of the leaver and the replacer. While replacers are compared directly to the leaver that they replace, which by definition is within network group and detailed occupation, comparisons across categories of replacement are not necessarily within group or occupation. Therefore, in a final model for each measure of referrals received I include fixed-effects for network group and occupation, using Bureau of Labor Statistics SOC codes, to account for the effect of potential variation across chapters and occupations in these comparisons. I estimate two sets of ordinary least squares (OLS) regressions predicting the difference between - what the replacer received relative to the leaver he replaced: first overall - total referrals received and then by referral type - direct referrals received and third-party referrals received. 2.4 RESULTS: WOMEN GET LOWER RETURNS TO SOCIAL CAPITAL Given that the focus of this study is to uncover the network mechanisms that contribute to gender differences in the resources male and female entrepreneurs receive, Table 3 compares male and female members in the study population in terms of each of the relevant variables. The motivation for this comparison is to determine whether male and female entrepreneurs studied differ in substantive ways that could contribute to differences in the number of referrals they receive. Table 3 reveals that, on average, male and female members do not differ in terms of the number of 32 referrals they generate for other members, the amount of money they generate for other members through referrals, or their tenure with the group. Women engage in more additional meetings and have fewer absences from the weekly meetings relative to men. Despite evidence that women contribute at least comparably to men, they receive fewer referrals. This gender differences in referrals seems to be largely driven by third-party referrals, with women receiving over 26 percent fewer third-party referrals than men. While this comparison does not include any controls, it indicates that female entrepreneurs receive fewer resources through their network ties than do male entrepreneurs on average, despite being more involved. [INSERT TABLE 3] [INSERT TABLE 4 Table 4 presents results from negative binomial regression models estimating the effect of member gender on the total number of referrals received to determine whether women receive fewer total referrals than do men. Model 1 contains only control variables, revealing that on average, members receive more referrals when they contribute more to the group. Specifically, the number of additional meetings a member engages in, the number of referrals a member gives to others, and the amount of money a member generates for others, are each strong predictors of the benefits that they receive. In line with the comparison of means, Model 2 reveals that women received approximately 25 percent fewer referrals per year than men in the same network group, conditional on the covariates. Model 3 introduces occupation fixed effects, and removes network group fixed effects, to compare men and women across network groups who are in the same broad occupations. Accounting for potential gender differences in occupations, women still received nearly 15 percent fewer referrals than men. The reduction in the magnitude of the gender difference suggests that occupational sorting is a contributing factor to observed differences. Thus, despite having access to the same social ties, women receive fewer referrals than men who are contributing similarly to the group. 2.4.1 Replacement events 33 Total referrals. To provide a precise test of hypothesis 1, this next set of analyses focuses on the subsample of members involved in a replacement event to compare men and women in the same detailed occupations within the same networking group. In Table 5, the constant term provides the estimated difference in the number of referrals that the replacer received, on average, relative to what the leaver received for the omitted category, male-to-male replacements, conditional on the covariates. In other words, the constant tells us how many more or fewer referrals a male replacing a male receives relative to the man he replaced. The coefficients on male-to-female, female-to-male, and female-to-female dummy variables provide estimates of whether the difference in what the replacer received relative to the leaver he or she replaced in each respective replacement category is greater or less than observed differences in the male-to-male replacement category. As Model 2 in Table 5 shows, the gap between what a replacer receives relative to the leaver being replaced is far wider in the male-to-female category than in the male-to-male category. Whereas a male in the male-to-male category receives approximately the same number of total referrals as the leaver he replaced, as evidenced by the non-significant constant term, a female replacer in a male-to-female replacement event receives five, or 23 percent, fewer referrals than the man whom she replaced. The introduction of occupation fixed effects in Models 3 and 4 does not lessen this observed gender difference. The penalty for female replacers in the male-to-female category is even greater when more fine-grained occupation categories are included (Model 4), with women generating nearly seven fewer referrals than the men they replace. [INSERT TABLE 5] - The coefficients of the other two replacement categories - female-to-female and female-to-male are not significant indicating that replacers in these categories, like the male replacers in the maleto-male category, receive a similar number of total referrals as the person whom they replaced. I also conducted a post-estimation F test to determine whether the effect for the male-to-female category differs significantly from the female-to-female and female-to-male categories. The results of this test reveal that these differences are also statistically significant. In other words, with the exception of female replacers in the male-to-female category, there is no penalty or advantage for replacers, on 34 average, for entering a network group relative to the person they replaced. Taken together, these results provide support for hypothesis 1 revealing that resource exchange directly contributes to gender inequality in outcomes. Type of referral. The previous section showed that female entrepreneurs receive fewer total referrals than their male counterparts with access to the same social ties and in the same detailed occupation. This section presents additional analyses aimed at uncovering the mechanism driving this gender difference and testing hypothesis 2. I find that expectations that a client, family member, or friend, prefers to work with male entrepreneurs, or what I call anticipatory third-party bias, lead members in these network groups to disproportionately give referrals to male group members under certain conditions. To disentangle the anticipatory third-party bias mechanism proposed in this study, Table 6 presents estimates from models predicting the difference in number of referrals replacers receive relative to the referrals received by the leaver they replaced by referral type---direct use referrals (Model 1) and third-party referrals (Model 2). These models reveal that the gender difference observed is solely driven by women receiving fewer third-party referrals than their male counterparts. Women receive as many direct use referrals as the men whom they replace, as indicated by the nonsignificant coefficient of male-to-female in model 1, but over five fewer third-party referrals as shown by the coefficient of male-to-female in model 2. Since male leavers replaced by women receive approximately 17 third-party referrals (presented below in Table 8), this difference translates to women receiving 30 percent fewer third-party referrals than the men they replace. [INSERT TABLE 6] Whereas third-party referrals involve one member giving the information of his outside contact to another member, direct use referrals only involve the two members in the group. Thus, unlike direct use referrals, the success of third-party referrals requires that the outside, or third-party, contact be satisfied with the referrer's decision to make a referral. Members commonly mentioned taking the preferences of their clients, family members, and friends into account. When discussing 35 the criteria used to decide whether to make third-party referrals to fellow entrepreneur-members in interviews, members commonly stressed the importance of a good match between their fellow contacts in their network group and their third-party contacts outside of the group. For example, one entrepreneur-member shared: "You can be a pretty good judge of who is going to hit it off with who and certain things like that. It really comes down to a fit." Together, these results provide support for hypothesis 2 suggesting that expectations about the gender preferences of third-party contacts not in the group serve to restrict the number of referrals women receive relative to men. Gender composition of occupation. To provide additional support that anticipatory thirdparty biases are leading to fewer referrals for women, this section tests hypotheses 3a, 3b, and 4, by exploring variation in observed gender differences in third-party referrals based on the gender composition of the occupation an entrepreneur-member occupies. To the extent that members are anticipating the gender preferences of a third-party when deciding whether or not to make referrals, it is reasonable for members to expect that their contacts are particularly likely to have a preference for men in roles that are traditionally performed by men, as posited in hypothesis 3a. Similarly, to the extent that third-party contacts have a preference for women, this is most likely in female-dominated occupations, as posited in hypothesis 3b. Figures 3A and 3B graph the difference in the predicted number of third-party referrals received by replacers relative to the leaver they replaced when the replacer's gender is congruent versus incongruent with the gender-type of the occupation. As in the models presented in Table 6, the underlying OLS models used to generate Figures 3A and 3B predict the number of third-party referrals received based on the type of replacement and introduce dummy variables for the sex composition of the occupation - male occupation for Figure 3A, to test hypothesis 3a, and female occupation for Figure 3B, to test hypothesis 3b. These models include both the main effect of male or female occupation and interactions between replacement type and these dummy variables. These interaction terms estimate how differences in the referrals received by replacers versus leavers varies based on the 36 degree to which the occupation that the men and women being compared occupy is gendered. The estimates from these models were used to calculate the predicted difference in the number of referrals that a replacer received relative to the person they replaced based on the gender of both actors and the gender composition of the occupation. Figure 3A introduces male occupation, which takes the value of one for occupations where 70 percent or more of occupants are male, and compares the relative referrals received by replacers versus leavers in the male-to-female category to the difference for the relevant baseline group, maleto-male. Consistent with hypothesis 3a, Figure 3A reveals a clear pattern: the penalty in third-party referrals received by a female replacer relative to male leaver she replaced is most pronounced in occupations that are predominantly male. Whereas women in occupations that are more than 70 percent male receive 4.6, or 25 percent, fewer referrals than the men they replaced, there is no gender difference in referrals received by men and women in occupations that are less than 70 percent male. Figure 3B introduces female occupation, which takes the value of one for occupations where 70 percent or more of occupants are female, and compares the relative referrals received by replacers versus leavers in the female-to-male category to the difference for the relevant baseline group, female-to-female. This figure depicts that there is no significant difference between these two categories of replacements suggesting that men are not penalized for entering female-typed occupations in this setting. As a result, I do not find support for hypothesis 3b. Rather, when the man replaces a female leaver he is not penalized in terms of third-party referrals, even when occupying a role traditionally held by women. [INSERT FIGURE 3A] [INSERT FIGURE 3B] Together these results provide support for hypothesis 4, that the degree to which a network contact occupies a role that is gender incongruent has greater implications for resource-holder's decisions of whether to share resources with women than with men. Even stronger, these results 37 suggest that this penalty only impacts women. Among members involved in a replacement, the only notable gender difference is when a woman replaces a man in a male-typed occupation, where thirdparty contacts are likely to have the strongest gender preferences. Moving from replacement events to the overall study population, Figure 4 graphs the effect of the gender composition of the occupation a member occupies on the predicted number of third-party referrals received by men versus women. The underlying model for Figure 4 predicts the number of third-party referrals received based on gender using a negative binomial regression. As in the models presented in Table 4, the underlying model includes the same set of controls and introduces percent male occupation as an independent variable, which is an ordinal variable ranging from 1-7 and measuring the degree to which actors within each occupation tend to be men. This model includes both the main effect of percent male occupation and interactions between female and percent male occupation. This interaction term estimates how gender differences in third-party referrals received varies based on the degree to which the occupation that the men and women being compared occupy. To generate Figure 4, the estimates from this model were used to calculate the predicted number of referrals that a man and woman will receive based on the gender composition of their occupation. [INSERT FIGURE 4] [INSERT FIGURE 5] Consistent with hypothesis 4, Figure 4 reveals that women receive fewer third-party referrals than their male counterparts in male-dominated occupations. Whereas women in occupations that are more than 70 percent male receive 63.6 percent fewer referrals than their male counterparts, there is no gender difference in referrals received by men and women in occupations that are less than 50 percent male. To confirm that this pattern is due to anticipated third-party biases, Figure 5 depicts the effect of the gender composition of the occupation a member occupies on the predicted number of direct-use referrals received by men and women. This figure reveals that there are no significant differences in the number of direct use referrals men and women receive. Thus, women 38 only receive fewer third-party referrals than do men and this gender difference is limited to comparisons of men and women in male-typed occupations. This finding provides additional support for my proposed mechanism, namely anticipatory third-party bias. Gender differences in the referrals men and women receive result from expectations about the gender preferences of others. Taken together, the results presented in this paper reveal that despite having access to the same social ties as male entrepreneur-members, women receive fewer referrals. Further, this study provides support that this gender differences is driven by anticipated third-party biases. It is only when entrepreneur-members anticipate that their clients, friends and family members are likely to prefer to work with men over women that they disproportionately give referrals to male members within their network group. 2.4.2 Robustness checks and ruling out alternatives Screening effect. One possible alternative mechanism for the finding that women receive fewer third-party, but not direct use, referrals relates to potential unobserved gender differences in on-thejob performance or competency. The argument would be that members hire female members (i.e. direct use referrals) and discover that these women are poor performers. As a result, they do not connect their third-party contacts to the female members in their group. While this is a plausible alternative explanation, neither my quantitative nor qualitative data provide evidence that supports this alternative explanation. To the extent that actors in these groups are not referring to women because they hire them and realize that they are poor performers, we would expect that as a woman receives more direct use referrals she would receive fewer third-party referrals. The logic here is that, if women are lower quality performers, the more direct use referrals they receive the more likely members in the group will learn of their lower quality through informal mechanisms. Based on this logic, as other members learned about a woman's lower quality, either by hiring the female member themselves of through word-of-mouth from other members who had hired the female member, they would be less likely to connect their outside contacts to the female member in the group. 39 [INSERT TABLE 7] Table 7 presents the results from a model estimating the number of third-party referrals a replacer received as a function of the number of direct use referrals he or she received. If this alternative screening mechanism is driving observed gender differences, we would expect that the number of direct use referrals a woman received would, at least partly, account for observed gender differences in third-party referrals received. In other words, by introducing the number of direct use referrals a woman receives as a predictor in this model, we would expect to see a reduction in the main effect of the male-to-female coefficient. This is not the case, however, as the results are identical to those presented in Table 6. On average, women receive approximately six, or 35 percent, fewer third-party referrals than the men that they replace, even after controlling for the number of direct use referrals they received in that year. In line with this quantitative evidence, my interviews do not suggest that entrepreneur-members are using direct use referrals to screen women. Generally, competency does not seem to be a major concern among members. When asked about experiences with, or knowledge of, poorly performing others, members typically stated that those who were not strong in their area of specialty typically did not last long in the group. Furthermore, they had a difficult time recalling cases where a member was a poor performer. It was often the case that multiple members within a networking group offered the same story of a poor performer, suggesting that this was in fact an uncommon occurrence. As a result, members did not center their decisions of whether to refer on competence. As one entrepreneur-member stated: "Competency is important, but most of the people that I've seen are competent in their field. I haven't seen anyone in our [group] that I don't feel can do the job." Even when asked about the criteria used for assessing others in the group, members often did not mention assessing quality or on-the-job performance. Rather, members discussed the central importance of learning about other members in the group as people. As one member shared: "Unless you get to know the people in the group, it's very hard to give that referral. You have to learn who they are". Another member stresses the importance of the characteristics of the person over their 40 performance stating: "I'm only going to make referrals if I know the person I'm referring to. I know the quality of who they are as people because my prejudice is that if they are a good quality of person the quality of their work will be fine." Taken together, both my quantitative and qualitative evidence cast doubt on the alternative that direct use referrals serve as a screening mechanism excluding women from the consideration set for third-party referrals. Thus, it is unlikely that members are directly hiring female members, realizing that they are poor performers, and consequently not referring their outside contacts to the women in the group. Differences in the quality of leavers. A second possible alternative mechanism for the finding that women receive fewer third-party referrals than the men they replace relates to potential heterogeneity in the leavers by replacement category. Specifically, if women are replacing top performing men, but men are replacing lower performing men, it would be possible that women receive fewer referrals than the men they replace because they are replacing the stars in the group. Similarly, if women are replacing very poor performers they may receive less not because they are women, but rather because other members in the group do not value the particular occupation they represent. In either case, my results could be biased, such that a female penalty would be exaggerated. While this is unlikely, it is also falsifiable. To examine whether this could be the case, Table 8 depicts the mean number of third-party referrals received by leavers in each of the four replacement categories. As this table shows, male leavers receive on average approximately 16 to 17 third-party referrals and female leavers receive approximately 13. Importantly, the mean number of referrals received by a leaver does not differ significantly based on whether the male leaver is replaced by a same or opposite gender member. This captures potential differences due to the quality of the leaver as well as potential differences in the opportunity for generating referrals for a given occupation. Thus, it is unlikely that differences in the benefits received by the leavers account for the finding that women receive less than the men that they replace. 41 [INSERT TABLE 8] Women do not ask. Another plausible explanation for the observed difference in the number of referrals received by women versus men is that women are not asking or expressing a need for referrals. Research has often argued that women are less likely to make requests than their male counterparts (Babcock & Laschever, 2003). To the extent that women do not make their needs known, they could receive fewer referrals not because other members in the group are favoring men, but rather because men ask, it is easier to identify opportunities to refer to them. In this empirical context, however, this is an unlikely explanation for two key reasons. First, by simply joining these instrumental business networking groups, members are all seeking to grow their businesses by gaining connections to new potential clients through business referrals. Second, by the design and structure of the weekly meetings, all members in these groups publicly state their needs during the meeting. As I observed, all members leveraged the opportunity to discuss their business and express the types of referrals they were seeking at that time. Furthermore, members can sign up for a longer presentation to pitch their work and educate the other members on how to make referrals to them during the weekly meeting. A comparison of the propensity for men and women to sign up for these slots reveals that both genders are equally likely to do so. Therefore, in this setting I do not find evidence of gender differences in requesting referrals from fellow network contacts. 2.5 DIsCUSSION Uncovering the sources of gender differences in the benefits men and women receive through social ties has been the focus of an extensive body of research. Lower returns to social capital for women are commonly attributed to gender differences in the composition and structure of social networks (McGuire, 2000). Undoubtedly, lack of access to valuable social ties contributes to women's inability to generate benefits from social ties. Existing research is less clear, however, about whether men and women with access to the same social ties receive equal returns. This study addresses this gap in our understanding of how resource exchange, or the activation of social ties, contributes to 42 gender inequality by comparing the relative benefits received by male and female entrepreneurs with the same access to social ties. Using a unique empirical setting where male and female entrepreneurs are members of the same business networking groups, I find that women receive fewer referrals through social ties than do similar men in the same network group, but this female penalty is not ubiquitous. The results presented in this study indicate that this female disadvantage is limited to cases where a resourceholder deciding whether to share resources with a network contact expects -that a relevant thirdparty---such as a client, friend, or family member---has a preference for men over women, which I call anticipatory third-party bias. Whereas men and women are equally likely to receive business directly from their network contacts, women receive far fewer connections to the clients, family, and friends of their network contacts, or third-party referrals. Further, I show that this gender difference in referrals only exists among women in occupations traditionally occupied by men. This finding provides further support for the central role third-parties play in the relative benefits men and women receive through their social ties. This paper makes two main contributions to the study of the network sources of gender inequality. First, by showing that women receive fewer benefits through social ties than men despite having the same access, this study provides evidence that access to social ties is insufficient for eradicating gender differences in the benefits of networks. Resource allocation also plays a role in perpetuating gender inequality. Network scholars have long argued that specifying individual-level differences in how actors use social ties provides an opportunity for making theoretical progress (e.g., Ibarra et al., 2005; Renzulli & Aldrich 2005). This study reinforces the importance of focusing on how resources are exchanged among network contacts. To gain a deeper understanding of how networks affect gender inequality in particular, future research must carefully account for both the structure and composition of networks and the process of resource exchange, or activation, within networks. 43 Second, this paper identifies a new network-based mechanism explaining gender inequality, which I call anticipatory third-party bias. The finding that women only receive lower returns to their social capital when resource-holders anticipate the gender preferences of others, or third-parties, provides support for this mechanism. Expectations about the gender preferences of others, or anticipatory third-party bias, influence resource-holders' decisions of whether to share resources with network contacts in a way that limits the benefits women receive through social ties. The fact that resourceholders only exhibit a male preference in occupations that are traditionally held by men provides further support for this mechanism. To the extent that a resource-holder is anticipating the preferences of a third-party, it is reasonable to expect that the third-party prefers men in occupations that are male-typed. In other words, the assumption that a third-party contact prefers a male electrician is more likely to be accurate than the assumption that a third-party contact prefers a male nurse. Future research could further uncover the conditions that make third-party biases more or less prevalent in evaluations. Resource-holders may differ in their propensity to incorporate gender into their decisions of whether to share resources based on their own gender or on the gender of their outside contact, or third-party. Understanding whether anticipated third-party biases operate differently for certain actors or under certain conditions would provide a more nuanced understanding of how and when resource exchange within networks disadvantages women relative to men. This study also makes a contribution to research on how ascriptive characteristics affect evaluations more generally by providing empirical evidence for the role of third-parties in perpetuating gender inequality. A challenge in identifying how anticipatory third-party beliefs contribute to gender inequality is disentangling this mechanism from other competing explanations. Specifically, existing evidence consistent with the notion that anticipating the preferences of others leads actors to disproportionately favor men over women has been unable to account for personal gender beliefs, or preferences (e.g. Fernandez-Mateo & King, 2011). To develop comprehensive theories about sources of gender inequality, and to redress these inequities, it is necessary to more carefully identify not only the potential mechanisms, but also the extent to which these mechanisms 44 are responsible for observed gender differences in outcomes. This study offers a first attempt to disentangle these competing mechanisms by comparing the relative returns men and women receive from social ties in cases where decisions to share resources are contingent on a third-party versus cases where decisions are based on personal preferences. Future work could further disentangle these processes by comparing decisions to share resources under conditions with and without a third-party made by the same evaluator. This will allow for further identification of how anticipating the preferences of a third-party impacts gender inequality in the benefits received through social ties by accounting for each individual's personal gender preferences. In terms of identifying when third-parties are more apt to perpetuate versus reduce inequality, this study provides evidence consistent with the importance of visibility, or transparency. Acting in accordance with the anticipated preferences of others is particularly likely when the presence of bias in decisions is not observable, or visible, to others. Of course, third party audience expectations can serve to mitigate reliance on ascriptive characteristics in evaluations (Salancik & Pfeffer, 1978; P. Tetlock, 1992; Valian, 1999). By imposing expectations, audiences serve to create accountability, or impose sanctions for behavior that is in violation of established norms (Adut, 2005; Jensen, 2006). Therefore, concerns that the third-party will sanction biased behavior may discourage reliance on gender in decisions of whether or not to share resources. But an audience is only able to impose its expectations, tacitly or otherwise, onto a decision maker if this bias is visible to the evaluating audience (see Castilla (2008) for a similar discussion about the importance of transparency). This study provides support for the presence of bias under conditions where a third-party is unable to perceive whether a decision-maker is exercising bias. Since the third-party contacts of entrepreneurmembers are unlikely to know that a member refrained from connecting them to another member of the group, bias is not visible to third-parties in this setting. In conclusion, my findings help to explain how networks contribute to the underrepresentation and lower success of women among entrepreneurs. Even if female entrepreneurs have access to the same networks, and engage similarly within these networks, as their male counterparts, I find that they may still receive fewer benefits. Thus, to the extent that entrepreneurs strategically shape their 45 networks (Burt, 2007; Stuart & Sorenson, 2007), simply gaining access to valuable social ties may not lead to the same benefits for men and women. Further, even if network contacts are not exercising overt bias, or male preference, in deciding with whom to share resources, expectations about collective gender beliefs of others may serve to constrain actors in a way that perpetuates gender inequality. The gender preferences of third-parties are likely to be particularly salient in assessments of entrepreneurs because entrepreneurship is a male-dominated area (Gupta et al., 2009). Therefore, redressing gender inequality in the benefits that male and female entrepreneurs receive through social ties requires more than providing women with access to better networks or eliminating personal gender stereotypes or beliefs. Equipping entrepreneurs with knowledge about the role of third-parties for gender inequality provides a first step toward diminishing gender differences in the benefits available through social ties. Since actors are not bound to anticipating the gender preferences of third-parties, more equitable network-based outcomes are possible if actors avoid basing decision on these expected gender preferences, or at least learn the true preferences or expectations of others where possible. 46 2.6 FIGURES AND TABLES Figure 1. Structure of groups Group B Group A IT Consultant Real Estate Agent Real Estate Agent T stw - IT Consultant Figure 2A. Distribution of Entrepreneur-members to Occupation Categories based on Bureau of Labor Statistics Standard Occupational Classification (SOC) Codes Business and Finaidal Specialists Construction sind Trades Office and Administrative Support Retail and Services Sales Arts, Design. Entertainnent, and Mediai Legal Computer Healthbare Pramctitioiners Cmornunity and Soial Services Building and Grounids MITaintenan ce -IFPrint and Production Personal Care and Service Installation and Repair Architecture and Engineering Food Preperation and Service Advertising, Marketing, and Promotions Healthcare Support Transportationi and Material Moving Protectivo Services Education and Training 0 so 100 150 47 200 250 300 350 Figure 2B. Gender Composition of Occupation Categories based on Bureau of Labor Statistics Standard Occupational Classification (SOC) Codes 0.9 0. 0.6 0.2 0.1 0 N x , -P> &, NC4 t VZN < 44kO' &- Figure 3A. Predicted Difference in Third-party Referrals Received by Replacer Relative to Leaver in Male to Female Replacements versus Male to Male replacements by Gender Composition of Occupation 7E a) -0- E z - =19 Ct at 80; 0 1 Male-dominated Occupation =1 - Male-to-Male - -f - - Male-to-Female Error bars represent 95% confidence intervals 48 Figure 3B. Predicted Difference in Third-party Referrals Received by Replacer Relative to Leaver in Female to Male Replacements versus Female to Female replacements by Gender Composition of Occupation w z 0 Female-dominated 0 - ,D Occupation Female-to-Female Error bars represent 95% confidence - =I +--Female-to-Male intervals .S- CL 4. Predicted Overall Gender Difference in Third-party Referrals Figure CQ by Gender Composition of Occupation Received E <30% 30 -40% 50 - 60% 40 -50% Percent Male in Occupation -t-Fem - - 60 -70% - F Error bars represent 95% confidence intervals 49 > 70% Figure 5. Predicted Overall Gender Difference in Direct Use Referrals ) Received by Gender Composition of Occupation 0) "0 E) .zD < 3 % 3 '0 0-5% 0-6% PretMln a) 6 '0 Ocpto MN40 Errr brs9% epeset cnfdene iteral 50 0 TABLE 1. Replacement Events by Type % of Total Number of Unique Number of Replacement Persons Replacements Events Male to Male 376 188 45.19% Male to Female 176 88 21.25% Female to Female 132 66 15.87% Female to Male 148 74 17.79% Total 832 416 100% TABLE 2. Comparing Participation and Referral Behavior of Replacement Sample to Study Population, Yearly Averages Study Population Replacement Sample Mean s.d. Min Max Mean s.d. 23.58 19.68 0.00 196.00 23.35 20.29 0.00 Min Max Referrals Received Total Referrals Received 196.00 Direct Use Referrals Received 7.69 9.73 0.00 73.00 7.37 11.20 0.00 73.00 Third-party Referrals Receive 15.90 15.06 0.00 188.00 15.99 15.20 0.00 151.00 24.18 20.30 0.00 264.00 22.42 18.97 0.00 264.00 8.20 15.98 13,045.00 8.87 16.44 30,410.57 0.00 0.00 0.00 85.00 248.00 789,598.40 7.07 15.35 9,935.09 8.23 15.89 31,500.74 0.00 0.00 0.00 73.00 248.00 661,666.00 14.49 3.03 4.04 15.62 2.89 4.89 0.00 0.15 0.00 75.00 16.83 52.00 12.75 2.17 4.42 13.36 2.12 5.31 0.00 0.15 0.00 75.00 14.41 52.00 Referrals Given Total Referrals Given Direct Use Referrals Given Third-party Referrals Given Dollars Generated Participationin Group Additional Meetings Tenure Absences N 2,310 832 51 TABLE 3. Basic Descriptive Statistics for Key Variables by Gender, First Year in Dataa Female Members Male Members Mean s.d. Min Max Mean s.d. Total Referrals Received 20.22 17.99 0.00 137.00 24.74 22.84 0.00 Direct Use Referrals ReceivE 7.94 10.74 0.00 68.00 8.23 10.52 0.00 73.00 Third-party Referrals ReceiN 12.28 11.80 0.00 98.00 16.51 17.85 0.00 188.00 23.28 18.77 0.00 160.00 22.40 20.83 0.00 205.00 7,528.50 18,235.82 0.00 257,744.00 8,072.73 19,517.00 0.00 389,108.00 Additional Meetings 15.25 15.87 0.00 75.00 10.95 12.91 0.00 74.00 Tenure 1.41 1.75 0.15 12.35 1.56 2.04 0.15 12.83 Absences 3.84 4.84 0.00 40.00 4.41 5.55 0.00 46.00 First Year Observed 2.36 1.40 1.00 5.00 2.47 1.46 1.00 5.00 Dollars Generated t-testb Max 196.00 * Total Referrals Given Min 1488 822 persons 'Comparisons robust for each nt'year in the data and by calendar year. Use first year because captures all members. b indicates whether differences in values for male and female entrepreneurs are statistically significant based on two-sided t-tests *p 0.05, ** p 0.01, *** p 0.001 52 Table 4. Negative Binomial Regressions Predicting Total Referrals Received per Year Additional Meetings 0.009 * (0.001) 0.003 ** (0.001) Tenure Absences Dollars Generated (in 000's) 0.011 -0.149 (0.032) 0.010 *** (0.023) (0.001) (0.001) 0.004 *** (0.001) 0.013 *** 0.003 (0.001) 0.010 0.030 (0.006) -0.006 (0.006) -0.006 (0.004) (0.003) (0.003) (0.002) 0.000 * * (0.000) Constant -0.251 * 2.656 *** 0.000 (0.000) 2.735 *** * Total Referrals Given Model 3 * Female Model 2 * Model 1 -0.007 ** 0.001 0.000 2.991 *** (0.110) (0.118) (0.051) Year Yes Yes Yes Group Yes Yes No No No Yes 485.320 548.460 1479.670 Fixed Effects Occupation Wald Chi-squared DF 45 46 30 Persons 2,310 2,310 2,310 Person-years 5,588 5,588 5,588 Notes: Standard errors, in parentheses, are clustered by individual member. Models including occupation fixed effects use BLS SOC major codes. Results are robust to more finegrained occupational fiexed effects. * p 0.05, ** p 0.01, *** p 0.001 53 Table 5. OLS Regressions Predicting Difference in Total Referrals Received by Replacer Relative to Exiter 1 Model Model 2 Model 3 Model 4 Replacement Type Female-to-Female Female-to-Male -5.257 * * -5.305 (2.462) -6.832 * Male-to-Female (2.631) (2.719) -3.639 -1.511 -1.062 (2.661) (3.157) (3.329) 0.437 1.377 1.633 (2.478) (2.724) (2.813) Controls for Replacer Tenure 0.302 (0.056) 0.213 (0.058) (0.090) (0.092) 0.170 (0.085) 3.150 (1.520) Absences Dollars Generated (in 000's) * 0.287 0.219 * 0.161 (0.085) *** * * * 0.307 (0.055) * Additional Meetings 0.310 (0.055) * Total Referrals Given 2.962 2.041 1.972 (1.516) (1.585) (1.620) -0.104 -0.117 -0.117 -0.126 (0.206) (0.206) (0.213) (0.217) -0.023 -0.019 -0.028 -0.017 (0.000) (0.050) (0.051) (0.052) Controls for Exiter Additional Meetings -0.177 -0.202 (0.124) Tenure -0.870 -0.944 Dollars Generated (in 000's) -0.224 ** (0.083) (0.125) (0.520) (0.521) -0.204 (0.085) -0.199 -0.191 (0.130) (0.135) -0.864 -0.967 (0.538) 0.395 0.387 0.297 (0.548) 0.299 (0.187) (0.188) (0.193) (0.195) * Absences -0.251 ** (0.080) * -0.264 ** (0.080) 0.145 (0.000) ** * Total Referrals Given 0.150 (0.053) ** 0.141 (0.054) ** 0.150 (0.054) -6.186 -0.545 4.507 21.596 (10.132) (10.419) (11.455) (14.280) Yes Yes Yes Yes Occupation - SOC Major No No Yes No Occupation - SOC Minor No No No Yes 0.235 0.242 0.244 416 416 416 Constant Fixed Effects Group Adj R-squared Observations Notes: Standard errors in parentheses. * p! 0.05, ** p 0.01, *** p 0.001 54 0.250 416 ** Table 6. OLS Regressions Predicting Difference in Direct Use versus Thirdparty Referrals Received by Replacer Relative to Leaver Model 1: Difference in Model 2: Difference Direct Use Referrals in Third-party Replacement Type -0.005 -5.231 (1.251) (2.221) 0.445 -1.610 (1.295) (2.665) Female-to-Female Female-to-Male * Male-to-Female 0.107 0.934 (1.501) (2.299) 0.066 0.237 Total Referrals Given * Controls for Replacer (0.027) Additional Meetings Tenure Dollars Generated (in 000's) (0.048) 0.092 0.122 (0.043) (0.076) 0.080 1.946 (0.754) (1.338) 0.064 -0.181 (0.101) (0.185) -0.035 0.007 (0.024) (0.043) Absences *** Controls for Exiter Additional Meetings -0.201 (0.070) -0.062 -0.137 (0.062) (0.109) (0.554) -0.309 (0.256) (0.454) * Tenure -0.024 (0.039) Absences * Total Referrals Given 0.120 0.178 (0.092) (0.163) 0.022 0.119 (0.026) (0.045) Dollars Generated (in 000's) 3.590 0.865 * Constant ** (9.668) 5.447 Fixed Effects Group Yes Yes Occupation Yes Yes 0.1739 0.193 416 416 Adj R-squared Observations Notes: Standard errors in parentheses. Occupation fixed effects for SOC major codes used. Results robust to more fine-grained occupation fixed effects. *p - 0.05, ** p 0.01, *** p 0.001 55 Table 7. OLS Regressions Predicting Difference in Third-party Referrals Received based on Difference in Direct Use Referrals Received by Replacer Relative to Leaver Replacement Type Male-to-Female * -5.772 (2.531) Female-to-Female -0.918 (3.108) Female-to-Male 0.573 (2.932) Screening Effect Direct Referrals Received 0.106 (0.141) MFxDirect 0.124 (0.258) FFxDirect -0.154 (0.287) FMxDirect -0.008 (0.274) Controls for Replacer 0.234 Total Referrals Given *** (0.049) Additional Meetings 0.129 (0.077) Tenure 1.742 (1.362) Absences -0.168 (0.181) Dollars Generated (in 000's) 0.009 (0.044) Controls for Exiter Total Referrals Given -0.202 ** (0.071) Additional Meetings -0.129 (0.110) Tenure -0.335 (0.461) Absences 0.193 (0.165) Dollars Generated (in 000's) 0.121 ** (0.046) Constant -0.438 (9.791) Fixed Effects Group Yes Occupation Yes Adj R-squared Observations Notes: Standard errors in parentheses. 0.1875 Uccupation tixed effects robust to more fine-grained occupation fixed effects. ' * p for 416 SOU major codes used. Results 0.05, ** p 56 0.01, *** p 0.001 Table 8. Comparing Third-party Referrals Received by Leavers Involved in a Replacement Event, by Replacement Type Third-party Referrals Mean s.d. Min Max Male-to-Male 15.93 12.96 0.00 69.00 Male-to-Female 17.19 14.63 0.00 83.00 Female-to-Male 13.74 11.04 0.00 53.00 Female-to-Female 13.41 12.81 0.00 97.00 57 3 NAMING YOUR DAUGHTER JACK: THE EFFECT OF GENDER IN ATTENTION AND EVALUATION 3.1 INTRODUCTION There are a wide range of organizational and market settings where an evaluator is tasked with assessing a set of candidates with the explicit goal of selecting the most qualified from this set. For example, this exercise is common in labor markets, where job applicants are evaluated by a hiring manager (Fernandez & Fernandez-Mateo, 2006); financial markets, where securities and firms are assessed by investors and analysts (Zuckerman, 1999, 2004); and cultural markets, where books, films, and music are appraised by consumers and critics (Kovacs & Sharkey, 2014; Salganik, Dodds, & Watts, 2006). A persistent challenge facing evaluators across these contexts is that the actual quality of the options being compared is not easily observable, forcing them to rely on accessible indicators of expected quality. To minimize the effect of signals, and to achieve unbiased results, in taste tests, for example, the evaluator is often blinded, or made unaware of the identity of the producer of the product that she is evaluating. However, the exclusion of signals is difficult, and in some cases their absence may result in suboptimal outcomes (Spence, 1973), therefore, as uncertainty increases signals are commonly relied upon (Podolny, 1994, 2005; Simcoe & Waguespack, 2011). This central role of signals turns the evaluation process into more of an art than a science, with recent causal evidence showing that ratings may stem from factors unrelated to underlying quality (Muchnik, Aral, & Taylor, 2013). The art of the evaluation process, along with the fact that the outcomes of these evaluations have significant economic impact, has spurred a long tradition of research questioning whether evaluation processes contribute to systematic disadvantage for some groups (Bertrand & Mullainathan, 2004; Foschi, 1996, 2009: 200; Foschi, Sigerson, & Lembesis, 1995; Foschi & Valenzuela, 2012; Pager, Bonikowski, & Western, 2009; Reskin, 1988). How ascriptive characteristics affect the multiple stages of the evaluation process, especially in competitive markets, where more objective quality or performance information is available, however, remains unclear. In 58 this paper, we aim to understand how gender affects the evaluation process by focusing on two stages: first, the selection stage, and second, the rating and feedback stage. We unpack this question using a competitive market setting where performance is important, several relevant and objective screening criteria are available to the audience of evaluators, and gender is easily inferred, but not needed. In trying to understand the persistence and effect of ascriptive characteristics on the evaluation process, early economic theories of discrimination would suggest that evaluations based on attributes not directly related to productivity are costly and thus should be competed away in an efficient market (Altonji & Blank, 1999; Arrow, 1971; Gary Stanley Becker, 1957; Charles & Guryan, 2008). Based on this perspective, because of competitive pressures, we would not expect evaluators seeking to maximize performance to base their assessments of candidates on ascriptive characteristics. However, research has documented that the reliance on ascriptive characteristics, such as gender, in evaluation processes serves to perpetuate inequality (Bielby & Baron, 1986; Perry, Alison Davis-Blake, & Kulik, 1994; Reskin & Roos, 1990), as a result of either unconscious evaluator stereotypes (Heilman, 1980), or more deliberate efforts to maintain majority group privilege (Reskin, 1988). Status-based theories of discrimination posit that a preference for members of a particular group emerges because of higher performance expectations for members of higher status groups (Ridgeway & Correll, 2004; Ridgeway & Smith-Lovin, 1999). While it is well established that these micro-level selection and evaluative processes contribute to disadvantage for certain ascriptive groups, identifying whether and the conditions under which ascriptive characteristics lead to an advantage for higher status groups is less clear (Correll & Benard, 2006). Additionally, we lack an understanding of how these characteristics affect outcomes for the candidates that move beyond the initial selection stage of evaluation. Undoubtedly, a primary explanation for our limited understanding of the extent to which status based mechanisms contribute to persistent inequality is the challenge of collecting appropriate data. Isolating the role of status-based discrimination for gender inequality is particularly difficult as it is necessary to examine gender inequality in a context where statistical discrimination is not likely 59 at play. Statistical discrimination posits that differences in the treatment of groups is a logical response to the problem of limited information (Arrow, 1971; Phelps, 1972). According to this perspective, if a valued characteristic is difficult or costly to observe, and has different distributions for men and women, treating these groups differently would be rational. Using unique data from an online platform where investment professionals openly share relevant investment recommendations, this study addresses this challenge by providing direct evidence that status-based mechanisms of discrimination play an important role for perpetuating gender inequality. Because there is a universal, unbiased, market-based performance metric assigned to each investment recommendation, which is visible to all members of the platform, evaluators in this setting do not need to rely on gender as a proxy for performance. We find that evaluators are more likely to view investment recommendations linked to male investment professionals than those linked to female investment professionals. Specifically, this study provides support for a particular form of status-based discrimination, namely double standards theory, which posits that even when performance information is available, biased evaluators will favor men over equally qualified women because generally men are perceived as the higher status actor (Berger, 1977; Correll & Ridgeway, 2003). This study also demonstrates the conditions under which status-based discrimination is most prevalent. While we find that that evaluators prefer male candidates over women in the initial selection stage, once men and women are selected into the consideration set we do not find evidence of a female penalty in the subsequent rating and feedback stage. Furthermore, we find that evaluators are particularly likely to invoke double-standards favoring men in the initial selection stage when they are faced with high search costs. By isolating the role of status-based mechanisms of discrimination for perpetuating gender inequality, this study identifies not only how, but also when evaluators use gender as a criterion despite having more relevant indicators of performance with which to assess candidates. We posit that together these findings provide evidence that double standards are applied, such that women are held to a higher standard, as a means for sifting through a high volume of information, but not when search costs decrease. Furthermore, this strategically 60 selected context allows us to rule out a strong alternative explanation that these observed gender differences are associated with female non-conformity. In the financial markets, an actor's propensity to take risk, or risk-appetite, is a salient characteristic. Importantly for our research question, women are commonly perceived as being more risk-averse (Alden, 2013). To the extent that women in this setting are more risk-averse, it would be plausible that statistical discrimination plays a prominent role with evaluators using gender as a proxy for this trait; therefore, it is important to rule out this alternative. This study does not, nor does it claim to, falsify or rule out the possibility that statistical discrimination also plays a significant role in perpetuating inequality in some cases. Since statistical explanations are based on evaluators relying on gender as a proxy for desirable characteristics that they are not able to directly observe, statistical discrimination would be least likely in our setting. Therefore, using the selected context, this study adjudicates between these two competing explanations by providing confirming evidence that status-based mechanisms of discrimination, particularly double standards, contribute to gender inequality in outcomes under certain conditions. Additionally, the ability to follow a subject through the evaluation process is rare, therefore, we are also able to identify how the use of gender changes from the selection stage to the rating and feedback stage of the evaluation process. We proceed as follows. First, we develop hypotheses about whether and when we should expect evaluators to exhibit a preference for male candidates by drawing on theories of status-based discrimination and double standards. We explain how the impact of evaluators may differ at different stages of evaluation and based on the number of options available for consideration. Second, we introduce the research setting and analytical method used to test whether status-based mechanisms of discrimination account for gender inequality in the attention or ratings candidates receive from evaluators. Third, we present results which demonstrate whether and when double standards lead to a female penalty in this setting. We conclude with a discussion of the theoretical and managerial implications of the research findings for our understanding of gender inequality resulting from evaluative processes. 61 3.2 3.2.1 THEORY AND HYPOTHESES The Role of Status-based Mechanisms of Discrimination for Gender Inequality Status-based theories of discrimination posit that a preference for the members of a particular group emerges because of higher performance expectations for members of higher status groups. A status characteristic emerges when an observable social distinction is attributed to a widely held cultural belief that actors with one category of the characteristic have greater competence and social value than actors with an alternate category (Berger, 1977; Berger, Rosenholtz, & Zelditch, 1980; Wagner & Berger, 1993). Because status characteristics carry these performance expectations, it is posited that evaluators, either consciously or otherwise, use these characteristics when assessing candidates (Correll & Benard, 2006). As a result, actors with the positive state of the characteristic, or higher-status, benefit from social advantages relative to actors possessing lower-status states of the characteristic (Webster & Hysom, 1998). In the case of gender, status beliefs about expected quality or performance lead evaluators to favor men, the higher status actors, because they are expected to outperform women. For example, as the higher-status gender, men are expected to be more competent, have more influence over others in a group, and have their performances evaluated more positively than their female counterparts (Ridgeway & Correll, 2004; Ridgeway & Smith-Lovin, 1999). To the extent that ascriptive characteristics become imbued with status (Ridgeway & Correll, 2000), it is not surprising that evaluators form these expectations that subsequently inform evaluations about candidates. Furthermore, this is in line with status theories more broadly predicting that status is used as an indicator of quality, especially in times of uncertainty. 3.2.2 Double Standards Theory and Gendered Outcomes Despite this consensus that status plays a central role by signaling quality, the importance of status varies. Podolny (1994) argued that status is most important when uncertainty makes it difficult to evaluate the quality of a product prior to its exchange. For example, younger firms receive greater benefit from relationships with high status partners than do older firms because the 62 quality of younger firms is more uncertain (Stuart, Hoang, & Hybels, 1999). Similarly, more experienced evaluators may rely on status less because they are more capable of discerning quality (Jensen, 2006). These studies suggest that the significance of status for evaluations declines as evaluators are able to assess candidate quality more precisely and directly. Alternatively, double standards theory, a particular status-based account of discrimination, suggests that gender7 serves as a distinguishing characteristic in assessments of competence even when men and women with equal performance are compared (e.g., Foschi, 1989). According to this perspective, discrimination arises because stereotypes about the relative expected performance of men and women influence evaluations. Standards are defined as, the "norms defining requirements for the inference of an attribute, such as the level of performance considered necessary to conclude that a person is competent" (Foschi, 2000). Importantly for this study, double standards theory centers on situations where these standards are not established a priori by a third party, but rather are formulated and applied by the same evaluator (e.g., Foschi, 1989). Double standards exist when, instead of the same set of norms, or requirements, being applied to all performers, there is more than a single performance-requirement for the inference of levels of competence (Foschi, 2000). A useful distinction for understanding how differences in standards contribute to unequal outcomes for men and women is to classify these in terms of strictness, where a strict standard for ability requires additional evidence of competence than does a lenient standard (Foschi & Foddy, 1988). At the core of this theory is the notion that if members of lower status groups (e.g., women) are systematically evaluated using stricter standards, these lower status actors will be perceived as being less competent than similarly performing members of the higher status group (e.g., men) who are evaluated using less strict standards (Foschi, 1989). Because strong performances are unexpected from the lower status group, when a man and woman exhibit an equal, good performance, the female performance is assessed using a stricter standard. While double standards theory has provided insight on the role of status characteristics more broadly, we will center our discussion on gender as this is the focus of this paper. 63 These assessments of competence are proposed to contribute to gender differences in outcomes for men and women, thus serving as a potential source of inequality. For example, a double standard would exist if evaluations on an exam deemed a score of 70 as passing for some, but a score of 80 was necessary to pass for others. In relation to gender, these standards would perpetuate inequality if men were held to the former, while women the latter. Alternatively, women may be held to a universalistic standard, whereas men are advantaged by particularistic standards where exceptions to this standard occasionally make requirements more lenient for men (Lorber, 1984). This is not to say, however, that universal standards applied similarly to men and women may not disadvantage women. For example, one explanation for why few women advance to equity partner level in law firms is that both men and women understand this role as requiring them to privilege work considerations over family (Pinnington & Sandberg, 2013). To the extent that both men and women who place priority on family obligations are disadvantaged, this would be considered a universal standard, even if women are disproportionately affected due to their tendency to place primary attention on family. Double standards theory, and this study, is centered on the formercases where different standards are used to evaluate men and women. Extant experimental research has been foundational in establishing underlying theory and providing empirical evidence for the existence of double standards, yet the extent to which double standards shape gender inequality in organizational and market settings remains unclear. Numerous laboratory studies, testing whether, and under which conditions, double standards are activated, have found support for the theory-even when male and female subjects perform similarly, women are evaluated using stricter standards and thus perceived as less competent (Foschi, 1996; Foschi, Lai, & Sigerson, 1994; Foschi et al., 1995). These studies have found that, in many cases, minority groups face more stringent evaluation criteria (Foschi et al., 1994). For example, when paired with a partner in laboratory studies, subjects set higher standards for female higher performing partners than they did for male partners (Foschi, 1996). Further, audit studies comparing the likelihood that racial minorities, as compared to whites, progress in the hiring process show that evaluators prefer higher-status candidates (e.g., white applicants) over lower-status applicants (e.g., black applicants), 64 even when comparing resumes for equally qualified applicants from each status category (Bertrand & Mullainathan, 2004; Neumark et al., 1996; Pager et al., 2009). However, more recent examinations offer conflicting predictions about the role of gender in & eliciting lower competence assessments for women altogether (Foschi & Lapointe, 2002; Foschi Valenzuela, 2008, 2012; Jasso & Jr., 1999; Okamoto & Smith-Lovin, 2001). For example, Foschi and Lapointe (2002) found that subjects were similarly influenced to update their response (i.e., reject influence) upon learning that their partner disagreed with their response, irrespective of whether the partner was male or female, suggesting that male and female partners were equally influential. In line with this perspective, some recent studies attempting to identify whether double standards exist in hiring processes have found that subjects evaluating similarly qualified male and female resumes did not show a preference for the male applicant (Foschi & Valenzuela, 2008, 2012). In some cases, research has found that while men were neutral, women exhibited reverse double standards, thus positing that male participants show no bias and female participants may be overcompensating (Foschi & Valenzuela, 2008). Furthermore, by design, these lab experiments are not subject to the competitive pressures that commonly influence behavior in organizational and market settings. To the extent that evaluations based on attributes not directly related to productivity are competed away in efficient markets (Altonji & Blank, 1999; Arrow, 1971; Gary Stanley Becker, 1957; Charles & Guryan, 2008), we would not expect evaluators seeking to maximize performance to base their assessments of candidates on ascriptive characteristics. Therefore, the role of ascriptive characteristics, such as gender, are unclear when a competitive market is analyzed. Further, we lack an understanding of how these ascriptive characteristics affect multiple stages of the evaluation process. Thus, the prevalence of double standards for perpetuating gender inequality remains unclear. Conflicting evidence stemming from experimental studies coupled with the absence of field studies examining the role of double standards in the presences of competitive forces, leaves a gap in our understanding of whether, and under what conditions, double standards are actually activated in 65 practice. Understanding whether male and female actors are evaluated using different standards is critical because we know that such standards are norms specifying the level and type of outcome required to infer ability (Foschi, 1996). To the extent that women are systematically evaluated using more stringent standards, women will be viewed as having lower ability levels than similarly performing men. Therefore, the standards used to judge ability may play a key role in perpetuating social and economic inequalities. In order to develop a complete theory about the factors contributing to gender inequality, and specifically the role of double standards in perpetuating this inequality, it is necessary to address this gap. It is only through this form of scientific inquiry that we will be able to develop effective solutions to redress gender inequality in groups, organizations, and the economic sector. Given that the finance industry, particularly investment management (e.g., hedge fund, mutual fund), is male dominated (Alden, 2013), it presents a well-suited setting for testing whether, and under what conditions, double standards contribute to gender inequality. A key scope condition of the theory is that double standards disadvantaging women are expected for tasks that are either male, or ambiguous, in terms of gender association (Foschi, 1989). The ideal worker (Acker, 1990; Williams, 1999) defines the characteristics of the individuals who are expected to be strong performers (Gorman, 2005). This image of the ideal worker shapes evaluative processes as candidates are commonly evaluated against this norm (Heilman, 1983). The investment industry's ideal worker image has been found to be gender-typed male "because [the ideal worker image] conflicts directly with macro-cultural stereotypes about women and defines commitment to the occupation as incompatible with motherhood" making gender particularly salient in this context (Turco, 2010). Hypothesis 1: Evaluators will use stricter standards when evaluating investment recommendations made by female investment professionals such that recommendations submitted by women will be less likely to be viewed than recommendations submitted by men. 3.2.3 Stages of Evaluation The process of evaluation is commonly a two-stage process: the selection stage, where candidates are first selected to identify the subset that will be given further consideration, and 66 subsequent assessment, where members of this reduced set are evaluated more closely and given feedback (Gensch, 1987; Hiubl & Trifts, 2000; Kovacs & Sharkey, 2014). In examining the role of status in evaluative outcomes, extant research has commonly focused on one of these two stages, thus failing to uncover the degree to which discrimination persists across stages (Kovacs & Sharkey, 2014). Status has been found to play an important role in both stages of the evaluation process. In line with hypothesis 1, in the selection stage, status provides a signal of quality helping to clarify any underlying uncertainty (Goode, 1979; Podolny, 1994, 2005; Simcoe & Waguespack, 2011). As the evaluation process moves from selection stage to the rating and feedback stage we would expect uncertainty and the reliance on and effect of status to attenuate, however, there is recent evidence to the contrary (Kovacs & Sharkey, 2014). Extant research focused on the effect of stratification on important economic outcomes has found discrimination at both stages of the evaluation process. Hiring audit studies have found evidence that applicant's names, and other signals suggesting the gender or race of a candidate on a resume, lead to biased outcomes in the resume screening process (Bertrand & Mullainathan, 2004; Goldin & Rouse, 2000; Neumark et al., 1996; Pager et al., 2009). For example, resumes suggesting that the applicant was a white male were considerably more likely to be called for an interview than those with names likely to be for female or minority applicants. Other research, focusing on the evaluation and feedback stage of the evaluation process, has also found bias (E. J. Castilla, 2008). Despite being hired into the firm and receiving equal performance ratings, for example, Castilla (2008) shows that pay continues to lag for women and racial minorities. While double standards theory does not differentiate between these two stages, we propose that the female penalty will be moderated in the second stage. To the extent that status-based discrimination is contributing to a preference for male candidates in the first stage, those with the strongest biases against women, or preferences for men, would likely exclude women from the consideration set in the first stage. As a result, the evaluators who are most apt to be assessing female candidates in the second stage are those with either weaker, or non-existent, gender biases. Further, uncertainty decreases in the second stage as more information becomes available and 67 evaluators have increased interactions with candidates. As a result, we would expect that evaluations are more likely to be based on the available objective criteria. For these reasons, evaluators will be less likely to employ double standards, or to base assessments on gender, when evaluating candidates in the second stage. Hypothesis 2: Evaluators will use more objective criteria in the second stage of evaluation, such that investment recommendations submitted by women will receive less of a penalty in the ratings or feedback received. 3.2.4 Search Costs Whereas statistical discrimination is invoked to address a problem of limited information, excessive information poses a different problem impacting the conditions under which status-based discrimination will be most prevalent. The set of actors or objects being considered during the selection stage of the evaluation process varies in size. As it increases, it is reasonable to expect greater uncertainty in the selection stage process and a greater reliance on the use of signals to understand quality (Podolny, 1994, 2005; Spence, 1973). For example, in the labor market, the applicants' level of education, graduating institution, and grade point average may be used to narrow the set to a more reasonable number. At the firm level, Stuart and colleagues (1999) find that, in the biotechnology industry, affiliations between new firms and established partners helps signal quality and are most important in times of uncertainty. Further, other work has found that entrepreneurial firms will even incur a cost in order to obtain these high-status affiliation signals (Hsu, 2004), and that under conditions of decreased search and uncertainty these status signals do not impact important outcomes (Simcoe & Waguespack, 2011). A similar logic likely applies with respect to the conditions under which double standards will be activated. To the extent that gender is viewed as a status characteristic and used as a means for sifting through large amounts of information, we would expect the prevalence of double standards to be a function of the number of candidates being evaluated. Specifically, if evaluators are selecting from a large number of candidates with similar performance, it is more likely that they will use gender as a means of simplifying the evaluative process, thus resulting in a male advantage. If 68 instead the observed bias is the result of systematic distaste for women, then we would not expect search costs to affect the selection stage of the evaluation process. Hypothesis 3: There will be a greater female penalty in viewership during periods where there is a high-volume of investment recommendations, such that recommendations posted by women will be less likely to be viewed in higher-volume than in lower-volume periods. 3.3 3.3.1 DATA AND METHODS Empirical Context We study the Real Investor's Club (RIC, a pseudonym), a private online platform that brings together buyside (e.g., hedge fund, mutual fund) investment professionals with the goal of sharing relevant investment recommendations with other investment professionals. The main goal of the platform is for investment professionals to share profitable buy or sell (i.e., short selling) recommendations for stocks connected to firms. Once an investment recommendation is submitted, it is added to the repository of all existing recommendations on RIC and is then among the set that may be selected for viewing and subsequently for rating and feedback. In this context, a given investment professional may simultaneously act as an author by writing an investment recommendation to be viewed by others, and as an evaluator by selecting and assessing the recommendations posted by others. Importantly, interviews with members of RIC reveal that most investment professionals using the platform do not know the other investment professionals on the platform prior to joining. Thus, individual investment professionals do not have prior knowledge about others on this platform that would potentially influence their evaluations of recommendations. In submitting an investment recommendation, the author posits a buy or sell recommendation for a security connected to a firm (e.g., common stock) along with an in-depth justification for this position. The analysis provides the price target, or the price the author proposes that the security will reach in the future, and the investment horizon, or the length of time the author expects it will take for this price target to be reached (e.g., eight months). The detailed justifications average over 1,000 words. And, the content of these write-ups vary across recommendations, but broadly they include a justification for the price target, a summary of 69 supporting information gleaned from company reports (e.g., quarterly reports), a discussion of macroeconomic trends, and a discussion of the valuation technique used (e.g., discounted cash flow). This setting is particularly well-suited for testing our hypotheses for several reasons. First, as detailed below, investment recommendations are evaluated by other investment professionals in two stages: the first stage of selection and the second stage of rating and feedback. These multiple stages of evaluation provide an opportunity to unpack whether and how biases in evaluation persist. Second and third, also discussed in further detail below, performance information is visible to evaluators and provides an unbiased signal of the author's quality. To effectively and accurately test how double standards contribute to gender inequality in evaluations it is necessary to isolate the role of gender above and beyond the presence of other, more relevant, indicators of quality. Finally, this setting allows us to isolate the action of individual investment professionals. Here, a market-based performance measure is directly attributed to each investment recommendation and each investment recommendation is directly linked to individual investment professionals. Given that funds are commonly managed by a team and sometimes multiple managers, the majority of extant research has been unable to attribute performance of investment professionals directly to a single manager or analyst. As a result, studies have either focused on fund-level outcomes to make claims about individual performance or relied on individual-level data for non-investment professionals. Together these features of our setting enable us to develop a more nuanced understanding of the effects of gender in evaluations by isolating the role of gender in evaluations from other possibly confounding factors and processes. Process of evaluation and the evaluating audience. When the investment professionals are not working on submitting a new recommendation, or acting as authors, they are consumers of the recommendations posted by others thus serving as evaluators in this setting. Investment recommendations are evaluated across two stages. The first, or selection, stage is where the evaluator opts whether or not to 'click' and view a given investment recommendation (i.e., give it attention). The second, or the rating and feedback, stage provides the evaluator an opportunity to give the investment recommendation a numerical rating and leave comments or questions. This 70 rating and feedback stage is conditional on the first stage, such that for an evaluator to assess an investment recommendation in this stage she must have first decided to view the idea in the selection stage. Investment recommendations are listed in reverse chronological order with the most recently posted recommendation listed first. Evaluators are presented with a minimal amount of information in the first stage of evaluation. Prior to selecting a recommendation for view (i.e., the first stage of evaluation), the evaluator is able to see the following information about the investment recommendation: the firm name for the stock being recommended, the recommendation (e.g., long versus short), and the return since the recommendation was submitted; and the following information about the investment professional who submitted the recommendation: their name and their employer's name. In order to access additional information about a given recommendation the evaluator must click the recommendation, thus including the recommendation in the subset surviving the selection stage of evaluation. To facilitate the search for recommendations, evaluators have the ability to screen recommendations along several objective criterion, such as industry, firm size, or expected return. When interviewed, most investment professionals claimed that their goal was to find high quality investment recommendations that could help them improve performance. While a few of the investment professionals interviewed stated that they keep their search to certain industries or firms, the majority claimed to not restrict their search to a specific screening protocol. However, these investment professionals also stated that they were less likely to take action (buy or sell) a security based on an investment recommendation outside of their expertise. While the salience of gender in the finance industry is unmistakable, interviews with investment professionals in this context make it unclear whether evaluators incorporate the author's gender in assessments of investment recommendations. Most investment professionals stated that they inferred a recommendation author's gender from the name listed on the recommendation. As one investment professional stated, "When I see a 71 name like Mary it sticks out because there are so few [women] in this industry." Most investment professionals argued that gender did not enter into their consideration when choosing a recommendation to view. The decision to select, or view, a particular recommendation was posited to be based solely on the potential of providing information that if followed would lead to the generation of profits. As one industry insider stated, "I am in the business of making money, I don't care who you are, as long as you can help me in that goal." However, another insider had a different viewpoint about the industry, "Whoever says there is no bias against females in this industry is bullshitting you." In the second stage, the evaluator is presented with the all of the facts and information included in the investment recommendation. In this rating and feedback stage of the evaluation, the evaluator can choose to rate the idea along two dimensions-the quality of the analysis and the return-as well as provide comments on the recommendation. RIC's intended purpose of the quality of analysis rating is to measure whether the investment professional was able to get their point across in a rigorous manner. In other words, this rating is intended to assess whether the position was substantiated by relevant and well thought out analysis. Importantly for our study, the quality of analysis rating is was referred to by one interviewee as the "good analyst rating" demonstrating the intimate link between a recommendation and its author. In interviews, investment professionals, both on and off RIC, frequently discussed the importance of a well presented analysis in both supporting an investment recommendation and indicating the investment professional's skill. The return rating is even more straightforward, as it measures the evaluator's belief that the recommendation (e.g., buy versus sell) and the magnitude (e.g., price target) is correct and likely to be achieved. Comments are free-form entries that allow evaluators to leave feedback or reactions related to the recommendation. Performance. To effectively and accurately test the role of double standards in these evaluations, it is necessary for performance to be visible, standard, and accorded via an unbiased mechanism. First, since the reliance on ascriptive characteristics is not completely surprising in the absence of other easily observable indicators of quality, to test whether double standards are being 72 applied in evaluations it is necessary that the evaluator is able to clearly and easily observe relevant performance information. To the extent that performance is as easy to view directly as gender, basing evaluations on gender suggests an evaluator bias that cannot be attributed to a lack of information (e.g., statistical or status-based accounts). Second, this visible performance measure must be standardized, thus conveying information about quality that is interpreted similarly by evaluators. In other words, for the performance metric to indicate quality what constitutes a good versus a bad performance must be agreed upon by the vast majority of the actors in the setting. Third, to assess gender differences in evaluative outcomes the performance metric must itself not be biased. For performance to be accorded via an unbiased mechanism any two actors taking the same action must receive the same performance outcome. An underlying bias in the attribution of performance would confound the results, such that what appears to be a double standard may actually be the result of bias in properly assessing performance. While previous research has leveraged lab-based experimental designs that largely meet these conditions (Foschi, 1996; Foschi & Valenzuela, 2008, 2012), the construction of performance in these studies is neither natural nor definitively unbiased. Hence, as aforementioned, variation in the degree to and direction in which these performance metrics have been unbiased in past studies may account for the heterogeneity of results. Even though subjects are commonly told that performance is assessed by an unbiased third party, this may not be enough to convince an evaluator. Further, the lack of a competitive market in these studies dilutes the meaning and importance of performance (Altonji & Blank, 1999; Arrow, 1971; Gary Stanley Becker, 1957; Charles & Guryan, 2008). The investment industry, and RIC in particular, provides a well-suited setting for this analysis given the visible, standard, and unbiased nature of performance within the industry. Performance being visible means that investment professionals can observe how a given recommendation is performing before deciding whether to view the recommendation. Performance in the investment industry is measured as the rate of return on a given investment (realized return, Rt). It is calculated simply as, R = (pt+-pt)+dt , where Pt+i is the price of the security at the time of sale, dt accounts for any distributions during the investment period (e.g., dividends), and Pt is the 73 price of the security at the time of its purchase. Lastly, few processes are as unbiased as a security's performance which is accorded by the stock market, making this a great strength of our setting. If any actor, regardless of ascriptive characteristics, takes a certain position in the market her performance would be identical to any other actor that took the same position at that same time. During the time period under study, the difference of the average investment recommendation's return, above the S&P 500, between men and women is only about 0.30 percentage points, with women outperforming men.8 However, this difference is not statistically significant (p = 0.97). Low accountability and motivation for a lack of bias in selection. recommendation is posted, it enters a repository of all recommendations Once a allowing investment professionals to sort through recommendations using broad classifications before choosing to 'click' on the recommendation to view its entirety. After an investment professional chooses to view a recommendation they have the option to anonymously rate the recommendation and/or comment and ask questions regarding details of the recommendation. The author of a recommendation cannot track how many other investment professionals viewed their recommendation. Because neither viewers nor raters are visible to the investment professional being evaluated, evaluators in this setting are not accountable for their evaluations to the author or to a broader audience of other evaluators. Generally, accountability serves to reduce the propensity for evaluators to base their decisions on bias (Salancik & Pfeffer, 1978; P. E. Tetlock, 1992), but the low-level of accountability in this setting makes it probable that evaluators will act in accordance with their preferences. Furthermore, investment professionals are motivated to make accurate assessments of others, especially in selecting which investment recommendations to view. Given that evaluators in this setting are time constrained and are hoping to gain knowledge about potential investment opportunities that will be lucrative, we will assume that it is rational for these evaluators to select the best investment recommendations. This return calculation assumes that the stock was held for approximately the time stated in the investment recommendation. Financial market data were only available until the end of 2013, therefore, the return was calculated until the end of the investment horizon or until the end of 2013. 8 74 3.3.2 Sample This study will analyze unique and proprietary data collected from RIC on the investment professionals using this platform, the investment recommendations posted, and the evaluations conducted by these investment professionals. The data span the period from 2009 to 2013 and are supplemented by external financial databases. Only recommendations pertaining to firms listed on a U.S. exchange (e.g., NASDAQ and NYSE) and covering common stock-as opposed to debt or options-will be analyzed in this study. Using these boundaries, this study focuses on the 3,003 recommendations submitted by 1,345 investment professionals. On average, each investment professional posted 2.23 recommendations during this 5 year window. For analysis, detailed data were collected on the characteristics of both the investment professionals and the recommendations. These measures will be used to isolate the effect of gender at both the selection and rating and feedback stages of evaluation. Further, to rule out a possible alternative explanation-that females may be acting in a manner systematically different from their male counterparts, as frequently highlighted in studies of gender (Baldiga, 2013; Bromiley & Curley, 1992; Ding, Murray, & Stuart, 2006; Eckel & Grossman, 2008; Fernandez-Mateo, 2009; Halaby, 2003; Miller & Hoffmann, 1995)measures of risk aversion were also collected and analyzed. 3.3.3 Dependent Variables: The Evaluation Process Selection stage and rating and feedback stage. For the first stage of the evaluation process, our analysis will focus on the level of attention received by a given investment recommendation in the selection stage of evaluation. This will be measured by the outcome variable views (logged), which is operationalized as the number of times a recommendation is 'clicked on', or selected. For the second stage of the evaluation process, our analysis will utilize three measures (all logged), write-up rating, return rating, and total comments. Each rating is measured on a 1-10 scale, with 10 being the highest rating. The write-up rating assesses the quality of the write-up, whereas the return rating assess the likelihood that the recommendation will reach its price target. Total comments represents the total number of comments a given investment recommendation receives from the time the recommendation was posted through the end of the study window. It is important 75 to note that in order for an evaluator to provide a rating or a comment on a recommendation, it is necessary for the evaluator to have selected to view the recommendation. It is this fact about our empirical context that allows for identifying the role of gender at each stage of evaluation. 3.3.4 Independent Variable: Gender At both stages of evaluation, the gender of the investment recommendation's author is not directly noted. Evaluators are only able to infer gender from the author's name', which is visible in both the selection and rating stage. Inferring individual-level ascriptive information from a name is not unique to this setting (Bertrand & Mullainathan, 2004; Neumark et al., 1996; Pager et al., 2009). We used the IBM InfoSphere Global Name Management Tool to score each investment professional's name in terms of likelihood to be female. This tool takes as its input an individual's first name and then compares that name to its database of 750 million names from around the world. Names are then scored in terms of how likely the name is female on a scale from 0 to 99, with a score of 0 signifying almost zero probability that the name is traditionally female. Figure 1 provides a histogram of the names on RIC. As expected, the great majority of names are associated with the male gender (approximately, 86.45% percent have a score less than six), and only about 4.43 percent of the names are more likely to be female than male (i.e., the name received a score of 50 or higher). The effect of this variable will demonstrate the degree to which evaluators favor investment recommendations from male investment professionals in the selection and ratings stages of evaluation in this setting. [INSERT FIGURE 1] 3.3.5 Control Variables 9 Members of RIC are not required to provide their gender, and only approximately 48 percent do, with no difference between men and women. Others can find this information, however, as discussed in the results section of this paper. Our results indicate that evaluators are unlikely to be looking within RIC for this information as we find that our results hold even when limiting the analysis to self-identified male investment professionals. 76 Both investment professional- and recommendation-level characteristics will be included as controls. Importantly, in our models, we attempt to control for the variables that are visible to an evaluator when assessing an investment recommendations at a given stage of the evaluation process as well as other variables that could affect the evaluation process. At the investment professionallevel, these include measures for education, the rank of both undergraduate and graduate institution; location; and the number of investment recommendations posted before the focal recommendation. At the recommendation-level, we control for recommendation type (i.e., long or short), firm information, and include both year and industry fixed effects. For undergraduate education, the 2013 US News College Ranking (US News and Report, 2013) was used to match investment professional undergraduate institutions to its respective ranking. For graduate education, the 2013 US News MBA Ranking (US News and Report, 2013) was used to match investment professional graduate institutions to its ranking and 2013 Financial Times Global MBA Ranking (Financial Times, 2013) was used for non-US business schools. To deal with missing rankings, investment professionals were grouped into four categories: top ranked (under)graduate institution (for a ranking of 1-10), mid ranked (under)graduate institution (for a ranking of 11-50-reference category), bottom ranked (under)graduate institution (for a ranking of 51-100), and unranked (under)graduate institution. Additionally, we created the variable no graduate school to account for those who did not attend graduate school and coded it one for these individuals. 10 Since all investment professionals in this setting have at a minimum an undergraduate degree, using these rankings together with the indicator of no graduate school serves as a proxy for education level. To control for location we use two dichotomous variables: major city and non-US location. Major city represents all cities where two percent or more of investment professionals are located, and is coded one for all investment professionals from these cities. For example, some of these cities include: Boston, Chicago, New York City, and San Francisco. Additionally, the variable Results are robust to changing the ranking cutoff. Additionally, there may be a concern of investment professionals attending a school previously when it was ranked differently, however, an analysis of the ranking showed that while there are a lot of changes, very few schools moved among these defined buckets. 10 77 non-US location is coded one for all investment professionals located in a city outside of the United States. At the recommendation-level, short is the dichotomous variable that takes the value of one if the investment recommendation recommends short selling, and zero if instead it is recommending taking a long position (i.e., buy and hold). For investment horizon we use a dichotomous variable, short investment horizon, that takes the value of one if the investment professional is recommending holding onto the stock for less than a year and zero otherwise. We also control for the investment style of the recommendation: growth, value (reference category), event-based, or other; firm size (i.e., price per share * shares outstanding), expected return (price target-priceat reco m endation , logged); and week-one performance. Here, week-one performance is measured as the return of the focal investment recommendation (Rt,i) in the first week after being posted, less how the market (S&P 500) performed during this same period (Rtm). In other words, the degree to which the investment recommendation outperformed the market benchmark during this period. Our decision to use one week performance is based on the fact that recommendations receive more than 50% of views within the first five calendar days of being posted, with a long right tail in the distribution. Table 1 provides summary statistics for each of the variables used for the analyses. This table splits the data into two panels: Panel A includes summary statistics for the investment recommendations where the investment professional's female name score is less than or equal to 10, and Panel B includes summary statistics for the investment recommendations where the investment professional's female name score is greater than 10. We find that the most consistently unbalanced variables between these two groups is related to education. Evidence suggests that those more likely to be classified as female attended higher ranked institutions. Given the few number of females in the investment industry, it is not surprising that there would be a higher bar for entry into the profession. Therefore, to the extent that women differ from men in terms of key independent variables in this setting, we find evidence that women possess better indicators of quality than do men in terms of education. 78 [INSERT TABLE 1] 3.3.6 Empirical Model We begin our empirical analysis by evaluating the effect of an investment professional's gender, inferred from their name, on a given recommendation's viewership. Specifically, we estimate the following ordinary least squares (OLS) regression: Y = fl 1 Female Name Score + yXi + At + 6j + Ej, where Y is the views (logged) and i indexes the recommendation; Xi is a vector of recommendationand investment professional-level controls; Ai is a vector of year dummies that control for time trends; and 6i represents 24 Global Industry Classification Standard (GICS) industry group subcodes, and an indicator for a missing GICS group subcode. Robust standard errors are clustered at the investment professional-level given the possibility that viewership may be correlated within investment professional. For the second stage of the evaluation process, Y represents total comments, write-up rating, or return rating (all logged). Ideally, names would be randomly assigned to recommendations and then evaluated by the audience, however, this is not the case. Our ability to closely mimic the data generation process of inferring gender from a name and the ability to control for most of the information that investment professionals claim to use, and have available to them through the platform, increases our confidence in measuring the effect of gender on our outcomes variables. 3.4 3.4.1 RESULTS Is there a Gender effect? To understand whether there are double standards in the first stage of evaluation, we examine whether investment recommendations posted by authors with a name more likely to be associated with a female receive fewer views. We use the number of views (logged) that a given investment professional's recommendation receives in these analyses. Regardless of the model specification used (Table 2), we find evidence that there is a negative effect of approximately -0.0014 79 for every point increase in female name score (p < 0.01). For example, in approximately the middle of the scale, an investment recommendation posted by an individual named Bowen, which is scored a 60, receives approximately 8.40 percent fewer views than a recommendation posted by the unquestionably male name of Matthew, scored a 0. On the far end, the penalty increases: a recommendation posted by an individual named Mary (scored a 99) receives approximately 14.17 percent fewer views than a recommendation submitted by Matthew. Therefore, we find support for hypothesis 1. There is evidence that in the selection stage of the evaluation process evaluators use a stricter standard and investment recommendations submitted by investment professionals likely to be female are significantly less likely to be selected, or to receive attention. From Table 2, it is evident that this is a competitive market. Evaluators are attracted to investment recommendations that have a positive return in the short-term indicated by the positive coefficient of week-one performance (p < 0.05). Additionally, and in line with conversations with investment professionals, risk is rewarded. Riskiness can be measured a few different ways, such as recommending a short selling position, an aggressive price target, and shorter investment horizons. The number of views is higher for recommendations that more aggressive price targets (expected return, logged) and for short selling recommendations. Investment recommendations that for a short selling positions receive almost 25 percent more views than a recommendation for a buy position (p < 0.001). Similarly, investment recommendations with more aggressive price targets are significantly more likely to be viewed in the selection stage. While these models control for this risk taking behavior, one may still worry that men and women are acting systematically different in terms of risk taking behavior. Specifically, the audience may be using gender as a proxy for an individual's risk appetite, or the audience may be sorting on risk-loving characteristics, thereby excluding females. In other words, to the extent that male and female investment professionals systematically differ in their propensity to be risk taking, observed gender differences in attention may be the result of the audiences' preference for higher risk, and not for men per se, thus overstating the female penalty. 80 [INSERT TABLE 2] Next, we focus on second stage of the evaluation process, the rating and feedback stage. Here, evaluators give rating scores and comments to those recommendations that they selected to view. Overall, we find support for hypothesis 2. Conditional on the recommendation being viewed, women do not statistically differ from men in terms of number of comments received (Table 3). Therefore, once an investment professional chooses to view a recommendation, the gender of the investment professional does not seem to play any role in likelihood that the recommendation receives a comment. We then estimate the role of gender for the evaluator's rating of the recommendation's write-up quality (logged; Table 4, Models 3A and 3B) and return rating (logged; Table 4, Models 4A and 4B) to determine if women receive lower ratings than do men, conditional on receiving a rating. In terms of ratings, there is no evidence to suggest that female name score effects the write-up quality rating or the return rating that a given investment recommendation receives. Since there is a small number of observations with a female name score greater than zero we may still be interested in the coefficient's magnitude. However, the economic significance of these outcome variables is small, and even positive in one case (the number of comments). [INSERT TABLE 3] [INSERT TABLE 4] In summation, we find support for both hypothesis 1 and hypothesis 2. There is strong evidence that women receive less attention in this setting: women receive fewer views than men when controlling for various investment professional- and recommendation-level characteristics. However, we do not find evidence that this bias persists in subsequent evaluation. Therefore, given that an investment professional's female name score does not affect the ratings or comments a recommendation receives, we only find evidence of a double standard in the first stage of the evaluation process. Of course, we cannot rule out the fact that all of the biased evaluators are selected out of the second stage of the evaluation process. However, given the fact that few of the variables included in the analysis helped predict variation in the number of comments or the ratings 81 received, it is more likely that the second stage of the evaluation process depends on the actual investment recommendation and not ascriptive characteristics. These results bring an important question to the forefront: Is this bias in the selection stage due to a mechanism or simply bias against women? 3.4.2 Ruling out the alternative that women are doing things differently? While not explicitly tested, one common interpretation of recent findings contradicting double standards theory is that the status value of gender is declining (Foschi & Valenzuela, 2008, 2012). And indeed, a number of recent studies support this perspective suggesting that the status value of gender is decreasing and gender is becoming a less salient characteristic (Jasso & Jr., 1999; Okamoto & Smith-Lovin, 2001). There is, however, some evidence suggesting an alternative mechanism that may account for these conflicting results-that the presence of other desirable characteristics, more commonly demonstrated by men, drive what appear to be double standards. Specifically, evaluators may show preference not for men, but for certain characteristics that are perceived to be indicators of future performance (Foschi & Valenzuela, 2008). To the extent that men are more likely to possess these desirable characteristics, preferences for these characteristics may be interpreted as double standards disadvantaging women. But, instead there may be a universal standard and both men and women possessing the characteristic will be evaluated as more competent than those who do not possess the characteristic. Since in the financial market, the ideal worker is male (Acker, 1990; Williams, 1999) characteristics typically associated with men are likely to be most valued. Over time, members of a given occupation come to share an image of the ideal worker that includes the characteristics and behaviors of people who performed the work successfully in the past or are expected to perform it successfully in the future (Gorman, 2005). This image serves as an unofficial norm, becoming the standard against which candidates are evaluated and that specifies hiring and promotion criteria (Heilman, 1983). In order to attribute observed differences to the activation of double standards it is necessary to rule out the possibility that these results are not driven by preferences for other characteristics, especially those commonly associated with gender. 82 Specifically, research in economics, management, and sociology commonly discuss the link between gender and risk-aversion with many studies finding evidence that women display greater levels of risk aversion in various contexts (Baldiga, 2013; Bromiley & Curley, 1992; Ding et al., 2006; Eckel & Grossman, 2008; Fernandez-Mateo, 2009; Halaby, 2003; Miller & Hoffmann, 1995). Halaby (2003) finds that men prefer high-risk high-return career paths. Similarly, Fernandez-Mateo (2009) finds that in contract employment women are less likely to change clients. In a recent experimental study, Baldiga (2013) administered exam questions to male and female students. When asked to answer these questions, participants had similar completion rates. However, when participants were notified that there would be a penalty for wrong answers, women were four to six percent more likely to skip a question. In the finance industry, the link between women and risk-aversion is particularly salient (Barber & Odean, 2001; Jianakoplos & Bernasek, 1998; Niessen & Ruenzi, 2007; Sunden & Surette, 1998) (Bajtelsmit & VanDerhei, 1997; Beckmann & Menkhoff, 2008). In studies of lay investors, women have been found to allocate their investments more conservatively (Sunden & Surette, 1998) and trade less often than men (Barber & Odean, 2001). A recent survey of research and press reports discussing women in fund management, completed by the National Council for Research on Women, concluded that women investment professionals take a measured and long-term approach to risk [for Research on Women, 2009]. Similarly, Beckmann and Menkhoff (2008), as well as Niessen and Ruenzi (2007), find that women fund managers are more risk averse in their investing. Accounting for risk-aversion is critical for assessments of double standards theory focusing on gender because if the action being evaluated has underlying risk, especially since there is evidence that risk is rewarded in this industry, failing to control for measures of risk will lead to confounded results. Further, even if the actor exhibits superior performance as a result of being risk-averse, she may still receive a penalty when being evaluated by an audience that does not act in a similar fashion, nor values this approach. Given that risk aversion has been found to be correlated with gender in general settings (Baldiga, 2013; Bromiley & Curley, 1992; Ding et al., 2006; Eckel & Grossman, 2008; Fernandez- 83 Mateo, 2009; Halaby, 2003; Miller & Hoffmann, 1995) as well as financial markets (Barber & Odean, 2001; Beckmann & Menkhoff, 2008; Jianakoplos & Bernasek, 1998; Niessen & Ruenzi, 2007; Sunden & Surette, 1998), this is a plausible alternative explanation for the observed female penalty in views, or the first stage of evaluation. If on average female investment professionals are being more riskaverse in their recommendations than the majority of their audience, a preference for riskier recommendations and not gender may explain the female penalty in views. This is especially true in this setting since the audience can observe indicators of risk aversion before deciding to view an investment recommendation. Additionally, given the evidence that risk is preferred in this setting, a penalty associated with risk-aversion is particularly plausible in this context. Historically it has been difficult to unpack the role of gender and risk aversion given that these two characteristics are commonly highly correlated. Inherently, measuring risk at the individual-level is empirically difficult: "Reliable demographic data on individual risk attitudes is virtually nonexistent" (Friedman, 1994). This has led most studies to rely on self-reported measures of risk-aversion. An issue, however, is that they could be positively biased, as women have been shown to have greater concerns about violating gender stereotypical behaviors (Deaux & Major, 1987; Heilman, 1980). Therefore, it is plausible that women may self-report their risk-attitudes and preferences in a way that aligns with what is expected instead of how they actually behave. This issue is also salient in the investment management industry. An additional empirical roadblock in this literature stems from the use of fund-level measures of risk. Specifically, even though most funds have one portfolio manager the choice to add or remove securities to the fund often resembles more of a democracy than a dictatorship. This is important because a fund-level measure may confound the risk-attitudes of the team with those of the individual female manager. While these results may speak to gender differences in management style, they may not be informative about the individual's actual risk preferences. Given that women have been found to be less confident, especially in male dominated fields (Beyer, 1990; Beyer & Bowden, 1997), it is also plausible that female investment professionals are more affected by the team consensus. Therefore, a true measure of risk would focus on the decision solely of the individual. 84 Using our unique setting, we are able to address this challenge to better isolate the role of gender in evaluation above and beyond differences in risk-taking. We will address these empirical difficulties in two ways, first, by using investment professional-recommendation-level measures of risk aversion: recommendation of short selling, short investment horizon, and expected return (logged). The use of multiple measures of risk aversion addresses the concern that any given proxy of riskaversion may capture other effects. Though no single measure is enough to categorize an actor as risk-averse, evaluating all of these measures together attenuates this concern. Short selling is the practice of selling borrowed shares of a security with the promise of purchasing and returning these shares to the lender-along with any distributions, such as a dividend-at a later time. This strategy, which is predicated on the opinion that the security in question is over-valued, is in direct contrast to taking a long position, which is based on an expectation that the security is undervalued. Investment professionals will be compared in their likelihood to recommend a short-selling position. A second measure of risk-aversion is based on the use of short investment horizon. Historically, shorter investment horizons are seen as riskier, as this usually indicates that the recommendation is exploiting an anticipated market change or condition as opposed to a long-term view (Barber & Odean, 2001). Therefore, we will examine whether there are gender differences in investment professionals likelihood of recommendation being for an investment horizon of less than one year. Our third measure of risk-aversion is based on the expected return of a given investment. In other words, how aggressive was the price target chosen by the investment professional. The goal of any investment recommendation is to hit the stated price target. Therefore, investment professionals are more apt to meet this goal, thus assuming less risk, when minimizing the gap between the security's price on the day of recommendation and the price target. For example, if an investor believes a stock currently priced at $15.00 per share will increase, they will be more likely to be correct if they state the price target at $17.50 as opposed to a more aggressive target of $21.50. While this also can be viewed as a confidence measure, we are using it to measure the investment 85 professional's risk appetite. A risk-averse investment professional would be more likely to recommend a smaller gap, in other words, have smaller rates of expected return. Using these measures of risk aversion we only find an association between the likelihood of a shorter time frame and female name score (p < 0.05). Interestingly, this association is opposite of what we would expect if women were more risk averse: more female names tend to recommend investment horizons that are shorter (i.e., more risky). Thus, if anything, there is weak evidence that women may be slightly more risk-loving in this setting. All other measures provide no evidence that there is a link between risk aversion and female name scores (Table 5). While there is still a valid concern that no single measure is enough to categorize a group as risk-averse, the lack of consistent evidence across all three of these measures suggests that, in this setting, men and women do not systematically differ in terms of risk-aversion. Therefore, while risk is favored in this context, this alternative explanation does not seem to be a driver of our observation that evaluators prefer recommendations by male authors. This result allows us to rule out endogeneity concerns related to gender and actions closely related to gender. [INSERT TABLE 5] 3.4.3 Status and Increased Search Costs To test if gender was treated as a status characteristic, hypothesis 3 posited that any differences should be more pronounced when searching becomes more difficult (Podolny, 1994, 2005; Simcoe & Waguespack, 2011). Using a year-long subset of microdata of each investment professional's viewing habits, we find that a recommendation gets more than 50 percent of its viewership within the first five days of being posted--with the distribution having a long right-tail. We constructed a variable, traffic, which summed the total number of recommendations in the fiveday window, before and after the recommendation was posted. The distribution of this variable is similarly concentrated. Given this distribution we constructed two dichotomous variables, low-traffic and high-traffic, which represent the bottom and top deciles of traffic, respectively. As we would expect, a given recommendation receives less total views (about 11.82 percent less) when posted 86 during a high traffic window and more total views when posted during a low-traffic window relative to recommendations posted during an average traffic window (Table 6). To test hypothesis 3, or the effect of female name score in times of increased search costs, we interacted female name score with both the low-traffic and high-traffic variables. First, we intuitively find that there is a negative relationship between increased traffic and viewership, and a positive relationship between decreased traffic and viewership. Then, using the interaction, we find that as female name score increases viewership receives a larger penalty during a high-traffic period, and no penalty in times of low search costs. This provides evidence that gender is used as a sorting mechanism during times of increased search costs, which is consistent with the idea that status indicators are most salient under these conditions and supports hypothesis 3. [INSERT TABLE 6 3.4.4 Unobserved gender differences There may still be concerns that women are acting in a different manner that cannot be easily measured. In an ideal examination we would combine features of laboratory studies with those of a competitive market setting by randomly assigning names to these investment recommendations. To address this concern, we limited our sample to those investment professionals who self-report as male. The rationale being that an analysis on this subset allows us to control for unobserved heterogeneity related to gender. Additionally, if our results hold it suggests that evaluators are not conducting their own due diligence using industry records and the Internet. When the sample is limited to this subset of self-reported males, our results are robust. We find the same relationship between the author's female name score and the number of views a recommendation receives (Table 7). In other words, investment recommendations posted by men with more ambiguous or female names (e.g., Bowen) received fewer views than did those posted by men with more male names (e.g., Matthew) 87 This check helps rule out the concern that women on this platform may be acting differently in a hard to measure way. Descriptively (Table 2) men and women are largely matched in terms of observed characteristics, but this is insufficient for ruling out that an unobserved gender difference, though unlikely given the robustness of our models, is driving our female penalty in views. If our effect were due to unobserved gender differences, we would not expect to see this female penalty present in this subsample. 3.4.5 Robustness Checks While a strength of the IBM InfoSphere Global Name Management Tool is the fact that it uses a vast and diverse database for its naming algorithm, there may be concern that the evaluators are not familiar with the likely gender of more unfamiliar names. To alleviate this concern, we used the frequently occurring first name data from the 1990 public use U.S. Census [U.S. Census Bureau, 1995]. Approximately 80.45 percent of the investment professionals first names were matched to these census data. We then restricted the analysis to investment recommendations that were submitted by an investment professional with a first name that was on this list. Upon rerunning the main analyses, we find that our results are robust (Table Al). 3.5 DISCUSSION Evaluation processes are a cornerstone of most market and firm settings, and their outcomes have significant economic implications. It is often the case, due to constraints, that only a small subset of a given population can be selected for further evaluation. The inherent uncertainty in this process has led to an influential literature that stresses the importance of signals and categorization in this process (Podolny, 2005; Spence, 1973; Zuckerman, 1999), and highlights how these processes are unscientific at times. Given the unequivocal role and implications of these evaluative processes, a burgeoning literature has focused on uncovering whether these processes contribute to systematic disadvantage for some groups (Bertrand & Mullainathan, 2004; Foschi, 1989, 1996, 2009; Foschi et al., 1995; Foschi & Valenzuela, 2012; Pager et al., 2009; Reskin, 1988). This research has shown evidence of bias in multiple stages of the evaluation process, but is unable to test the persistence of 88 bias for those that make it to the second stage of evaluation, or hold the evaluator constant. Additionally, it has been difficult to identify the underlying mechanisms that drive reliance on ascriptive characteristics in evaluations. Our study identifies the role of gender for evaluations in a setting where evaluators are motivated to identify the most qualified candidate and where more objective signals of quality are available. In the first stage of the evaluation process, we find evidence of an economically substantial attention penalty, where women are as much as 14 percent less likely to be selected. However, in the second stage of evaluation we do not find evidence of a gender-based penalty. Therefore, it seems that bias occurs only in the first stage of the evaluation process and does not play a role in subsequent, more detailed assessments. It is then important to try to understand the mechanisms that contribute to this first stage bias. Consistent with viewing gender as a status characteristic (Berger, 1977; Correll & Ridgeway, 2003), we find that gender is most salient as a sorting mechanism in times of high search costs. Importantly, we are able to rule out the popular alternative explanation that women are more risk averse than males. Overall, this study provides more direct evidence that status-based mechanisms of discrimination, and not simply taste-based preferences, play an important role in perpetuating gender inequality. Our ability to examine the role of gendered names among a subset of confirmed males in the study population is particularly revealing. The fact that men with a gender ambiguous or more female name receive fewer views than do men with more clearly masculine names provides compelling evidence of a female penalty in the first stage of evaluation. This approach provides clean identification of the effect of gender as a persistent signal even when more relevant indicators of quality are present and cheaply observable. Despite the pervasiveness of gender in the first stage of evaluation, our results highlight the importance of taking performance or actual quality into account when estimating the impact of gender. A common criticism of gender inequality research is a failure to adequately account for an 89 unbiased measure of quality or performance. As our results indicate, performance plays a key role in evaluator's decisions of whether to view a given recommendation. Failing to capture performance, therefore, could lead to a gross overestimation of gender effect. Future research aiming to identify the extent to which women are disadvantaged in evaluations must carefully account for relevant indicators of quality or performance. It is only through a meticulous comparison of men and women that we can uncover the mechanisms serving to perpetuate gender inequality and subsequently take steps to redress these inequities. Our findings also suggest that there is not a systematic bias against women in our setting. The central role of search costs coupled with a lack of evidence of a second stage penalty suggest that gender is being used as a means for sorting through information. In other words, evaluators are not simply exercising a universal preference for men, but rather using gender as a status signal to facilitate in the identification of quality. Thus we argue that status-based discrimination, or double standards, are evoked to resolve uncertainty related to a problem of excessive information. While statistical discrimination is employed under conditions of limited information, our results indicate that ascriptive characteristics may also be used under conditions where evaluators need to resolve uncertainty resulting from too much information. In this way, relying on gender may not indicate a bias against women, but rather a mechanism for simplify large amounts of information. The role of gender in this setting may be particularly salient given the low concentration of women in the investment industry (Beckman & Phillips, 2005). Based on Kanter's Men and Women of the Corporation, among investment professionals women are tokens-defined as a member of minority representation (less than 15 percent) in a setting that also has a dominant group (1977). Balancing the gender composition within the industry is difficult. From the perspective of the supply side, the women in this setting seem to be among those with highest quality education credentials, therefore, there seems to be a dearth of women from lower ranked schools possibly suggesting a higher bar for entry. Further, pinpointing demand side mechanisms is difficult. For example, another possibility, as found in law firms by Beckman and Philips (2005), is that the gender composition of exchange partners may be a major driver of the industry's gender composition. 90 Regardless of the factor, it is important to face this issue. Recently, there has been an attempt by certain state governments and institutions to increase the number of women in the investment industry. For example, a report by the Rothstein Kass Institute highlighted state-level initiatives to increase the involvement of women in the investment management industry. Connecticut and Maryland have allocated over $60 million combined to support funds that are led by women [Institute, 2012: 8]. Further, knowledge sharing platforms, such as the setting highlighted in this research, are able to give equal access to investment professionals and may be crucial in helping attenuate this bias. This increased exposure to women should help lessen the reliance on ascriptive characteristics by making gender less salient. Women investment professionals that were interviewed credited platforms, such as RIC, with allowing women to "take down the traditional barriers" and have a voice among their peers. Future research could benefit from identifying whether the evaluative processes contributing to gender inequality persist in non-male typed settings. To achieve this goal it is necessary to examine whether gender factors into evaluations in settings like the one we study, where there is objective performance information and multiple stages of evaluation. 91 3.6 FIGURES AND TABLES Figure 1. Distribution of Female Name Score 20 40 80 Female Name Score 80 92 10o Table 1: Summary Statistics of Key Variables Panel A: Female Name Score of less than or equal to 10 Obs. Mean Std. Dev. Min. Max. Dependent Variables (all logged) Views Write-up Rating Return Rating Total Comments Count 2,647 2,541 988 2,647 5.122 1.995 2.027 0.879 0.611 0.245 0.315 0.876 2.944 0.693 1.099 0.000 7.466 2.398 2.398 4.19 Control XVariables-idea-Level Recommendation: Short Week-One Performance Expected Return (logged) Idea Type: Growth Idea Type: Event-based Idea Type: Other Short Investnment Horizon Market Cap. 2,647 2,647 2,647 2,647 2,647 2,647 2,647 2,647 0.159 0.006 0.569 0.198 0.156 0.100 0.393 13.824 0.366 0.072 0.415 0.399 0.363 0.301 0.488 1.918 0.000 -0,333 0.010 0.000 0.000 0.000 0.000 8.917 1.000 0.837 4.894 1.000 1.000 1.000 1.000 20.204 2,647 2,647 2,647 2.357 .736 0.050 3.496 0.441 0.218 0.000 0.000 0.000 31.000 L000 2,647 2,647 2,647 2,647 2,647 2.647 2,647 0,238 0.116 0.322 0.272 0.029 0.037 0.564 0.426 0.320 0.467 0.445 0.169 0.190 0.496 0.000 0.000 0 0.000 0.000 0.000 0.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Control Variables User-Level Number of Past Ideas Location: Major City Location: Non-US Education (Ref: Mid-Rank) Undergraduate Rank: Top Undergraduate Rank. Bottom Undergraduate Rank: Unranked Graduate Rank: Top Graduate Rank: Bottom Graduate Rank: Unranked Graduate Rank: No Grad. 1.000 Panel B: Female Name Score of greater than 10 Obs. Meatn Std. Dev. Mirt. Max. Dependent Variables (all logged) Views Write-up Rating Return Rating Total Comments Count 356 333 142 356 5.034 1.991 1.967 0.910 0.650 .268 0.383 0.875 0.000 0.693 1.099 0.000 7.343 2.398 2.398 4.585 Control \variables-Idea-Level Recommendation: Short Week-One Performance Expected Return (logged) Idea Type: Growth Idea Type: Event-based Idea Type: Other Short Investment Horizon Market Cap. 356 356 356 356 356 356 356 356 0.177 0.004 0.537 0.250 0.169 0.104 0.497 14.068 .382 0.066 0.316 0.434 0.375 0.306 0.501 1.856 0.000 -0.450 0.010 0.000 0.000 0.000 0.000 9,275 1.000 0.267 2.152 1.000 1.000 1.000 1.000 19.433 356 356 356 2.618 0.817 0.062 4.116 0.387 0.241 0.000 0.000 0.000 22.000 1.000 1.000 356 356 356 356 356 356 356 0.275 0.065 0.261 0.351 0.008 0.051 0.503 0.447 0.246 0.440 0.478 0.092 0.219 0.501 0.000 0.000 0.000 0.000 0.000 0.000 0.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Control Variables~-User-Level Number of Past Ideas Location: Major City Location: Non-US Education (Ref: Mid-Rank) Undergraduate Rank: Top Undergraduate Rank: Bottom Undergraduate Rank: Unranked Graduate Rank: Top Graduate Rank: Bottom Graduate Rank: Unranked Graduate Rank: No Grad. 93 Table 2: OLS Regression of Views (logged) on Female Name Score. Unit of analysis is user-idea. Model 1 k 0.0014 * Recommendation: Short Expected Return (logged) Idea Type: Growth Idea Type: Event-based Market -0.1082 (00283) 0.1660 (0.0350) *** Horizon C(ap. * *** (0.0493) 0.01.92 (0.0239) -0.0265 *** (0.0069) Week-One Performance Number of Past 0.1832 (0.0322) -0.1117 (0.0280) 0.0496 Idea Type: Other Short Investment 0.2511 (0.0424) 0 1935 (0.0330) Ideas Location: Major City Location: Non-US 1.ndergraduate Rank: Top 0.1596 (0.0346) 0.049 1 (0.0490) 0.0 125 (0.0239) -0.0225 (0.0066) 0.3132 (0.1394) 0.0112 (0.0052) 0.1111 (0.0344) 0.0376 (0.0549) 0.1103 (0.0366) 0.1133 Undergradiate Hank: Bottom (0.0501) 0.0934 indergraduate Rank: Unranked (0.0339) -,0.0001 (0.0555) -0.1005 (0.0906) Graduate Rank: Top Graduiate [lank: Bottom Graduiate [Rank: Unranked 0.0962 Graduate Rank: No Grad. (0.0686) 0.0262 (0.0529) Constant M-Square Adj. Observations 5.2362 ** (0.0005) 0.2438 (0.0412) *** 5.0073 (0.1148) (0. 126 9) 0.251 0.266 3,003 3,003 Note: Models contain year and industry fixed eflects. Robust Standard errors clustered at the user-level are in parentheses. Significance Levels: + p ! 0.10, * p 5 0.05, ** p S 0.01, *** p S 0.001. 94 * 0.0013 (0.0005) ** * Female Name Score Model 113 ** Table 3: OLS Regression of Comments Count (logged) on Female Name Score. Unit of analysis is user-idea. Model 2 Idea Type: Growth Idea Type: Event-based Idea Type: Other Short Investment Horizon Market Cap. Week-One Performance Number of + -0.0171 (0.0267) -0.0188 (0.0071) -0.1847 (0.1863) Past Ideas -0.0029 (0.0053) -0.0844 (0.0324) 0.0458 (0.0570) -0,0308 (0.0338) 0.0602 (0.0464) 0.0471 (0.0355) 0.0046 (0.0500) 0.1605 (0.0745) 0.0483 (0.0746) -0.0108 (0.0467) ** -2.9828 (0.2008) 0.408 3,003 Location: Major City Location: Non-US Undergraduate Rank: Top Undergraduate Rank: Bott nm ludergraduate Raik: LUranked Graduate Rank: Top Graduate Rank: Bottom Graduate Rank: Unranked Graduate Rank: No Grad. Constant 0.0007 (0.0007) 0.8545 (0.0322) 0.0376 (0.0443) 0.0569 (0.0362) -0.0030 (0.0322) 0.0645 (0.0374) 0.0684 (0.0535) ---0.0137 (0.0265) -0.0210 (0.0071) *.. + Recommendation: Short Expected Return (logged) 0.0005 (0,0007) 0.8432 (0.0315) 0.0332 (0.0447) 0.0673 (0.0357) -0.0080 (0.0323) 0.0604 (0.0379) 0.0717 (0.0542) -2.9.988 * (0.1998) -Square Adj. 0.403 Observations 3,003 Note: Models contain year and industry fixed eflects. Robust Standard errors clustered at the user-level are in p4 rentheses. * p 5 0.001. Significance Levels: + p 0.10, 1 p < 0.05, ** P < 0.01. 95 * Views (logged) Model 2B * Fenale Name Score Table 4: OLS Regression of Write-up Quality (logged) and Return Ratings (logged) on Female Name Score. Unit of analysis is user-idea. Write-up Model 3A Female Name Score 0.0001 (0.0003) 0.0268 (0.0172) 0.0321 Recommendation: Short Expected Return (logged) (0.0118) Idea Type: Growth -0.0670 (0.0127) 0.0354 Idea Type: Event-based (0.0140) Idea Type: Other Short Investment Market 0.0482 (0.0199) -0.0062 (0.0101) -00165 (0.0026) Horizon Cap. Week-One Performance Number of Pasti deas Location: Major City Location: Non-US Undergraduate Rank: 'op Undergraduate Rank: Bottorn Return Model 3B -0.0002 (0.0003) 0.0232 (0.0169) 0.0283 (0.0110) -0.0685 (0.0127) 0.0334 (0.0138) 0.0504 (0.0199) -0.0074 (0.0101) -0.0152 (0.0025) 0.2132 (0.0598) 0.0041 (0.0016) 0.0186 (0.0148) 0.0067 (0-0265) 0.0288 (0.0144) 0.0241 -* Model 4A -0.0007 (0.0006) -0.0282 (0.0363) -0.0102 (0.0318) -0.0999 (0.0273) -0.0007 (0.0006) -0.0310 (0.0353) -0.0222 (0.0320) -0.1021 (0.0277) 0.0204 (0.0261) 0.0325 (0.0416) -0.0280 (0.0205) -0.0184 (0.0057) 0.3946 (0.1387) 0.0224 0.0267 (0.0258) (0.0416) -0.0212 (0.0206) -0.0193 (0.0056) 0.0028 (0.0024) 0,0245 (0.0236) -0.0336 (0.0473) 0.0347 (0.0266) 0.0385 (0.0338) 0,0375 (0.0).') Undergraduate Rank: Unranked Graduate Rank: Top GradIate Rank: Bottom Graduate flank: Unranked Graduate Rank: No Grad. Constant 2.1838 (0.0419) It-Square Ad 0. 4 Observations 2,874 Note: Models contain year and industry hxed effects. Robust Standard errors clustered at the user-level are in parentheses. Significan ce Levels: -+ p 5 0.10, * p 5 0.05, ** p < 0.01, ** p . 0.001. Table 5: 0.0099 (0.01,42) 0.0097 (0.0219) -0.0515 (0.0378) -0.0424 (0.0296) 0.0068 (0.0203) 2.1307 (0.0453) 0.095 2,874 (0.0255) 0.0109 (0.0363) -0.1001 (0.0733) -0.0545 (0.0563) -0.0183 (0.0344) 2.4053 (0.1743) 0.060 1,130 *12.3549 OLS Regression Evaluating Risk-Aversion. Unit of analysis is user-idea. Short Ren. Short Horizon Expected Return (log) Fenale Namei Score 0.0003 (0.68) 0.0015 (2.12) * -. 0004135 (-1.17) Note: Models contain user- and idea-level controls, excluding other risk aversion measures, as well as year and industry fixed effects. Robust Standard errors clustered at the user-level. T-statistic in parentheses. Significance Levels: + p K 0.10, * p K 0.05, ** p < 0.01, *** p K 0.001. 96 Model 4B (0.1770) 0.og 1,130 Table 6: OLS Regression of Vi ews (logged) on Female Name Score in times of increased traffic. U nit of analysis is user-idea. Model 5A Low Traffic (0.0005) -0.1182 (0.0392) 0.0847 (0.0289) * ** Female Name Score X High Traffic -0.0033 (0.0013) -0.0006 Female Name Score X Low Traffic + -0.0010 ** * High Traffic -0.0015 (0,0005) -0.1385 (0.0376) 0.0812 (0.0276) ** * Female Name Score Model 5B Short Investment Horizon Market Cap. Location: Majoy City Location: Non-US Undergraduate Rank: Top Undergraduate Rank: Bottom Undergraduate Rank: Unranked Graduate Rank: Top Graduate Rank: Bottom Graduate Rank: Unranked * * * 0.1026 ** (0.0371) 0.1232 (0.0540) 0.0874 (0.0341) -0.0143 (0.0597) -0.1141 (0.0904) -0.1194 * * (0.0730) * * -0.1134 (0.0281) 0.1590 (0.0346) 0.0385 (0.0482) 0.0089 (0.0237) -0.0236 (0.0067) 0.1192 (0.0351) 0.0400 (0.0590) * Idea Type: Other (0.0321) * * Idea Type: Event-based -0.1143 (0.0281) 0.1595 (0.0346) 0.0410 (0.0483) 0.0081 (0.0237) -.0.0235 (0.0067) 0.1212 (0.0351) 0.0374 (0.0592) 0.1046 (0.0371) 0.1227 (0.0540) 0.0869 (0.0341) -0.0141 (0.0598) -0.1133 (0.0902) -0.1177 ** * (0.0321) Idea Type: Growth * * * 0.2509 (0.0409) 0.3287 (0.1402) 0.1882 * Expected Return (logged) * (0.0731) Graduate Rank: No Grad. 0.0140 (0.0568) 5.0443 * Constant (0.1318) 0.269 R-Square Adj. Observations 3,003 Note: Models contain year and industry fixed effects. Robust Standard errors are in parentheses. Significance Levels: + p < 0.10, * p < 0.05, ** p < 0.01, * 97 0.0146 (0.0568) 5.0404 (0.1319) 0.269 3,003 p < 0.001. * Week-One Performance 0.2486 (0.0410) 0.3306 (0.1398) 0.1885 * (0.0015) Recommendation: Short Table 7: OLS Regression of Views (logged) on Female Name Score Sample only those users who are confirmed males. Unit of analysis is user-idea. Model 6A Recommendation: Short * -.0.0021 (0.0009) 0.2173 * Female Name Score 0.1313 (0.0409) -0.1586 (0.0388) 0.1424 (0.0457) 0.0035 (0.0774) Idea Type: Event-based idea Type: Other Short Investment Horizon 0.0333 (0.0372) Number of Past Ideas 0.0116 (0.0049) -0.0260 (0.0094) Market Cap. Location: Major City Location: Non-US Undergraduate Rank: Top Undergraduate Rank: Bottom Undergraduate Rank: Unranked Graduate Rank: Top Graduate Rank: Bottom ** ** 0.1424 ** (0.0489) 0.1120 (0.0723) 0.0812 (0.0581) 0.0667 (0.0666) 0.0611 (0.0477) -0.0549 (0.0814) -0.2142 + Idea Type: Growth * Expected Return (logged) * 0.4889 (0.1941) * Week-One Performance * (0.0603) -0.1556 (0.0947) Graduate Rank: No Grad. -0.0595 (0.0803) Constant 5.3032 (0.1790) R-Square Adj. 0.239 Observations 1,405 Note: Models contain year and industry fixed effects. * (0.1241) Graduate Rank: Unranked Robust Standard errors are in parentheses. Significance Levels: + p < 0.10, * p K 0.05, ** p K 0.01, * 98 p K 0.001. Table Al: OLS Regression of Evaluation Process (all logged) on Female Name Score-Census Unit of analysis is user-idea. Week-One Performance Expected Return (logged) Views (logged) Model A2 -0.0012 (0.0005) 0.2627 (0.0495) 0.2131 (0.1579) 0.2045 (0.0412) * * 0.0744 (0.0383) 0.8505 (0.0359) -0.0013 (0.0060) 0.0043 (0.0358) 0.0823 (0.0424) 0.1240 (0.0632) Number of Past Ideas Idea Type: Growth Id Type: Event-based Idea Type: Other Short Investment Horizon Market Cap. Location: Major City Location: Non-US Undergraduate Rank: Top Undergraduate Rank: Bottom Undergraduate Rank: Unranked Graduate Rank: Top Graduate Rank: Bottom Graduate Rank: Unranked Graduate Rank: No Grad. Constant R-Scquare Adj. Observations -0.1220 (0.0308) 0.1682 (0.0383) 0.0430 (0.0564) 0.0322 (0.0270) -0.0247 (0.0074) 0.1122 (0.0377) 0.0685 (0.0757) 0.1079 (0.0406) 0.1157 (0.0570) 0.1015 (0.0378) * * ** * * -0.0001 (0.0004) 0.0222 (0.0190) 0.2087 (0.0687) 0.0382 (0.0116) Return Rating Model A4 -0.0008 (0.0007) -0.0600 (0.0411) 0.3449 (0.1530) -0.0214 (0.0444) + + + (0.0139) 0.0295 (0.0153) 0.0617 (0.0212) -0.0097 (0.0114) -0.0120 (0.0028) 0.0159 (0.0156) 0.0089 (0.0331) 0.0329 (0.0155) 0.0216 (0.0208) -0.0003 (0.0159) 0.0137 (0.0245) 0.0304 * (0.0429) -0.0488 (0.0341) 0.0140 (0.0223) 2.0746 (0.0500) 0.090 2,311 0.0031 (0.0031) 0.0963 -0.0684 (0.0296) -0.0269* (0.0079) -0.0905 * (0.0357) 0.0724 (0.0739) -0.0141 (0.0365) (0.0503) 0.0632 (0.0392) -0.0126 (0.0550) 0.0983 (0.0848) 0.1101 (0.0906) -0.0084 (0.0517) -2.8876 (0.2248) 0.415 2,416 Rating Model A3 0.0036 (0.0018) 0.0650 Note: Models contain year and mdustry hxed effects. Robust Standard errors clustered at the user-level are in parentheses. Significance Levels: + p < 0.10, * p < 0.05, ** p < 0.01, *** p K 0.001. 99 + -0.0108 -0.0305 (0.0624) 0.1627 (0.1017) -0.1585 (0.0835) -0.0104 (0.0598) 5.0562 (0.1453) 0.274 2,416 0.0006 (0.0007) 0.0298 (0.0526) 0.3079 (0.2162) WVrite-up + (0.0311) 0.0224 (0.0310) 0.0608 (0.0510) -0.0367 (0.0234) -0.0171 (0.0066) 0.0147 (0.0270) -0.0850 (0.0671) 0.0571 (0.0306) 0.0295 (0.0377) 0.0348 (0.0303) -0.0007 (0.0407) 0.0382 (0.0952) -0.0231 (0.0796) -0.0171 (0.0379) 2.3460 (0.1943) 0.058 885 * Recommendation: Short Comments * Feinale Naine Score Views Model Al matched first names. 4 FORMALIZATION REVISITED: CONSIDERING THE EFFECTS OF MANAGER GENDER AND DISCRETION ON CLOSING THE WAGE GAP 4.1 INTRODUCTION Despite fifty years of political mobilization and dedicated public policy, the gender wage gap remains one of the most persistent forms of workplace inequality (Cha & Weeden, 2014; Cotter, DeFiore, Hermsen, Kowalewski, & Vanneman, 1997; England, 1992; Huffman & Velasco, 1997; Reskin, 2000). Comparisons of earnings for male and female workers in the United States consistently reveal that women earn substantially less than similarly qualified men working in the same organizational positions. In an effort to uncover the sources of this form of inequality, a growing body of research has focused on the role of organizational practices and pay systems (Baron, Davis-Blake, & Bielby, 1986; Briscoe & Kellogg, 2011; E. Castilla, 2008; Castilla, 2011; Castilla & Benard, 2010; Fernandez & Fernandez-Mateo, 2006; Kalev, Kelly, & Dobbin, 2006; Reskin, 2003). In particular, the degree to which organizational systems are formalized," or & governed by clear written rules and procedures (Anderson & Tomaskovic-Devey, 1995; Huffman Velasco, 1997; Weber, 1958), has been shown to have direct implications for the level of pay inequality among employees. Most studies on formalization and pay inequality demonstrate that compensation systems that are highly formalized, with unambiguous rules, reduce sex-based bias by limiting decision & makers' use of discretion and subjectivity when evaluating employees (Baron et al., 1986; Elvira Graham, 2002; Pfeffer & Cohen, 1984; Reskin, 2000; Sutton, Dobbin, Meyer, & Scott, 1994). These studies argue that when managers have discretion they are subject to decision-making biases, which lead them to favor male employees when allocating resources (Reskin, 2000; Ridgeway, 1997). Thus, formalized organizational pay practices that constrain managerial discretion are posited to minimize 1 See Elvira and Graham (2002) for a detailed discussion of pay formalization. 100 gender disparities because they limit the opportunities for managerial biases to factor into employee compensation decisions (Elvira & Graham, 2002; Reskin, 2000). and pay equity into question (E. Castilla, 2008; Dencker, 2008; Kalev et al., 2006; Konrad & Conversely, a handful of recent studies call this positive relationship between formalization Linnehan, 1995). Some of this research has found that, when compared to less formalized practices that allow managerial discretion, formalized personnel practices are associated with higher levels of gender inequality (Dencker, 2008). For example, Dencker (2008) posits that less formalized personnel practices may empower organizational decision makers to redress inequality, particularly when organizations face pressures for gender equity, in a way that highly formalized pay systems do not. Thus, recent evidence suggests that the discretion offered by less formalized work practices may lead to gender equitable outcomes under some conditions. longstanding research finding that organizational Furthermore, this is consistent with bureaucracies and formalized routines, more broadly, exacerbate gender and racial inequality in the workplace (Acker, 1990; Kanter, 1977). In this study, I advance our understanding of the relationship between formalization of pay and gender pay inequality by shifting the debate to focus on uncovering the conditions under which less formalized personnel practices reduce pay inequality. In particular, I consider the gender of the decision maker to determine whether male and female managers differ in how they use the discretion afforded by less formalized pay systems. While there may be a number of conditions under which less formalized practices serve to redress inequality, manager gender is of central importance as it is not clear that both male and female decision makers share pro-male biases. For example, ingroup biases have been posited to lead male managers to allocate resources, such as wages, in ways that benefit other men and disadvantage women (Gary S Becker, 1957; Bielby, 2000; Bridges & Nelson, 1989; Cotter et al., 1997; Ely, 1995; Nelson & Bridges, 1999; Reskin, 2000). Thus, as a result of a preference for the ingroup, female managers may favor members of their gender ingroup, namely female employees. In support of this view, while not directly considering pay formalization, research in the organizational demography tradition suggests that a greater proportion of women in management is associated with less gender wage inequality among non-managerial 101 employees Hultin & Szulkin, 1999; Joshi, Hui Liao, & Jackson, 2006; Kulis, 1997; Pfeffer, Davis-Blake, & (Baron, Mittman, & Newman, 1991; Cohen & Huffman, 2007; Huffman, Cohen, & Pearlman, 2010; Daniel J. Julius, 1995). I demonstrate that the effect of pay formalization on gender pay inequality varies by manager gender. Consistent with most extant research, I do not find inequality in terms of highly formalized components of pay. Both male and female managers display similar equitable practices. On the other hand, I demonstrate variation in pay equity under less formalized components of pay. In particular, I find less inequality in base salary, which is far less formalized in this setting, for employees reporting to female, as compared to male, managers. However, female managers do not ubiquitously use discretion to close the gender pay gap. I find that they only reduce pay inequality among employees in female-typed positions in the lowest organizational ranks. As such, my findings provide conditional support for the perspective that less formalized personnel practices may result in gender pay equity for employees reporting to female, but not male, managers. Thus, failing to account for manager gender may lead to erroneous, or oversimplified, claims that less formalized organizational policies perpetuate workplace inequality in all cases. Since allowing managers discretion also provides firms with the opportunity to enhance performance (Gomez-Mejia & Balkin, 1992), understanding when less formalized practices lead to equitable outcomes for men and women is critical for designing pay systems that simultaneously benefit the firm and its employees. My findings suggest that less formalized pay systems may be effective for achieving gender pay equity only in settings where women are present in management and oversee employees in female-typed, lower-status positions. Undoubtedly, a primary explanation for our limited understanding of when less formalized pay practices reduce gender pay inequality is the challenge of collecting appropriate data. To identify whether male and female managers impact employee pay differently when given discretion via less formalized pay systems is particularly difficult. It requires comparing relative wages for male and female employees reporting directly to male versus female managers in a setting where there are components of pay that vary in the degree of formalization. Using longitudinal data from a large 102 retail financial services firm in which I was able to link employees to their managers, this study addresses this challenge. Specifically, I analyze whether the relationship between formalization of pay systems and within-job gender pay inequality varies by manager gender. By simultaneously considering formalization of pay policies and manager gender, this study identifies both the conditions under which, and how, less formalized reward systems reduce gender pay inequality within organizations. I proceed as follows. I begin with a discussion of current theory on formalization and gender inequality. Drawing on social identity theory and status characteristics theory, I then develop hypotheses about whether, and when, male and female managers may use the discretion afforded by less formalized organizational pay policies differently. Next, I introduce the research setting and analytical method used to test whether manager gender explains the conditions under which less formalized practices reduce gender pay inequality among employees. Then, I present results which demonstrate that under certain conditions there is evidence of less gender inequality in terms of less formalized components of pay for employees reporting to female, relative to male, managers. I conclude with a discussion of the theoretical and managerial implications of my findings for our understanding of gender pay inequality, and workplace inequality more broadly, as well as the importance of considering characteristics of the manager in inquiries regarding workplace inequality. 4.2 4.2.1 THEORY AND HYPOTHESES Formalization of Pay and Gender Inequality Organizational resources, including wages, are allocated on the basis of the organization's reward structure (E. Castilla, 2008; Gibbons, 1998). While generally managers possess some discretion over employee pay within this structure, organizational practices and processes govern the degree of discretion available to managers (Elliott & Smith, 2004; Hultin & Szulkin, 1999, 2003). Specifically, organizations can elect pay systems that vary in degree of formalization (Dobbin, Sutton, Meyer, & Scott, 1993; Osterman, 1987), or the extent to which written rules and procedures govern pay decisions (Anderson & Tomaskovic-Devey, 1995; Huffman & Velasco, 1997; Weber, 1958). 103 Even within a firm the degree of pay formalization may vary, with some components of pay being more formalized than others (Elvira & Graham, 2002; Gomez-Mejia & Balkin, 1992). By establishing explicit practices guiding managerial decisions related to pay, highly formalized pay systems constrain the discretion managers have over subordinate pay decisions relative to less formalized systems (Bielby, 2000; Reskin & McBrier, 2000). To illustrate how manager discretion is constrained, consider commission pay for sales employees where the commission an employee receives is a fixed percentage of that individual's sales. As a result of the unambiguous rule governing commission pay decisions, managers do not have any discretion over final commission pay in this highly formalized pay system. Because they allow for minimal managerial discretion, highly formalized pay systems are commonly thought to be effective in reducing workplace gender inequality (Bielby, 2000; Reskin, 2003). Generally, organizational decision makers are posited to perpetuate inequality by, either consciously or unconsciously, favoring male employees when allocating resources (Reskin, 2000; Ridgeway, 1997). Whereas managers are unlikely to make personnel decisions based solely on merit when left to their own discretion (Castilla, 2011; Perry et al., 1994; Reskin, 2003), formalized personnel practices are posited to lead to more equitable outcomes by removing this managerial discretion from personnel decisions (Baron & Bielby, 1980; Pfeffer, 1977; Reskin, 2000). In fact, opportunity legislation and of the human resources profession since the late 1960s (Anderson & formalized personnel practices have been a central tenet in the development of equal employment Tomaskovic-Devey, 1995; Dobbin & Kelly, 2007; Huffman & Velasco, 1997; Kalev et al., 2006; Reskin & McBrier, 2000). Furthermore, a considerable body of empirical research supports this positive relationship & between formalization of organizational reward systems and gender equitable outcomes (Anderson & Tomaskovic-Devey, 1995; Baron, Hannan, Hsu, & Kogak, 2007; Elvira & Graham, 2002; Pfeffer Cohen, 1984; Reskin & McBrier, 2000; Sutton et al., 1994). For example, Elvira and Graham (2002) compare three types of pay, varying in the degree of formalization, for male and female employees 104 within a financial services firm. While not taking manager gender into consideration, they find that with the most formalized type of pay there is less gender pay inequality on average. In an examination of hiring outcomes, Baron and colleagues (2007) find that high-tech firms with a full time HR employee, which they argue signals that more objective employment practices are in place, hired more female scientists. Similarly, using firm-level data to examine the effects of organizations' employment practices on sex-based ascription in access to managerial jobs, Reskin and McBrier (2000) find that firms with more formalized personnel practices had a greater proportion of women in management positions. While the preponderance of empirical evidence supports the argument that more formalized pay systems serve to reduce gender pay inequality, some studies call this result into question (Acker, 1990; E. Castilla, 2008; Castilla, 2011; Castilla & Benard, 2010; Dencker, 2008; Huffman & Velasco, 1997; Kanter, 1977; Konrad & Linnehan, 1995). For example, Dencker (2008) finds a higher level of gender inequality in promotion under highly formalized personnel practices, arguing that formalized practices prevent organizational decision makers from responding to gender equity pressures. Relatedly, other research suggests that less formalized practices, which offer more managerial discretion, may actually serve to reduce inequality even in the absence of explicit goals for equity. In an experimental design, Castilla and Benard (2010) found that presenting subjects with language emphasizing managerial discretion was associated with a bias in favor of women and conjectured that the presence of discretion "may create the perception of the existence of bias and may therefore motivate individual attempts to correct it" (Castilla & Benard, 2010: 567). These mixed results raise an important question: Under what conditions might less formalized pay systems serve to reduce, rather than perpetuate, gender pay inequality? It is only by advancing this line of inquiry that we can uncover the effect of different organizational reward systems on gender pay inequality. By focusing on whether less formalized organizational reward systems perpetuate inequality, an implicit assumption of existing research on formalization and gender inequality has been that organizational decision makers will use the discretion afforded by these policies uniformly. Theories of ingroup preference and status, however, shed doubt on this 105 assumption. Therefore, it is imperative to extend this debate by including the characteristics of the organizational decision makers, such as their gender, in our inquiries. While I did not directly measure managerial intentions or motivations, I draw on social identity theory and status characteristics theory to argue that when given discretion through less formalized pay systems male and female managers may differ in their propensity to disadvantage women, or to favor men. 4.2.2 Considering Manager Gender and Differences in Gender Biases using Social Identity Theory Social identity theory provides a basis for understanding why male and female managers' pay decisions are likely to impact gender pay inequality among non-managerial employees differently. According to this theoretical perspective, individuals use visible demographic attributes, such as gender, to classify themselves and others (Ashforth & Mael, 1989; Tajfel & Turner, 1979). Using these categorizations, demographically similar others are classified as members of the ingroup, whereas those who are demographically dissimilar are seen as members of the outgroup. This categorization process is posited to be largely automatic, occurring "independently of decision makers' group interests or their conscious desire to favor or harm others" (Reskin, 2000: 321). Nonetheless, it leads individuals to favor ingroup members, and in some cases disfavor outgroup members, in evaluations and reward allocations (Bielby, 2000; Hewstone, Rubin, & Willis, 2002; Reskin, 2000). There are at least two reasons to expect that male and female managers will engage, either consciously or not, in a gender-based in- versus out-group categorization process, leading to a preference for their gender ingroup, with implications for employee pay. First, gender is a salient characteristic in the workplace as gender differences among individuals are clearly visible (Ridgeway, 1991). As a result, employees are more likely to engage in gender-based categorization and tend to associate with, and to privilege, other same-gender employees (Elliott & Smith, 2004; Gorman, 2005; McPherson, Smith-Lovin, & Cook, 2001). Second, while the propensity to make distinctions between the in- and outgroup is universal, it is most pronounced for actors occupying positions of power (Goodwin, Operario, & Fiske, 1998) as these individuals often aim to maintain the status quo (Jost 106 & Banaji, 1994). Given their situational power within the organization, male and female managers may be more predisposed to engage in this categorization process. Thus, just as the discretion afforded by less formalized organizational reward systems may lead male managers to advantage their male subordinates, female managers may act in favor of their female subordinates. Consistent with this notion that managers are apt to favor members of their gender ingroup, gender pay inequality has been attributed to the historical predominance of men in managerial positions because men allocate resources, such as wages, in ways that benefit other men and disadvantage women (Gary S Becker, 1957; Bielby, 2000; Bridges & Nelson, 1989; Cotter et al., 1997; Ely, 1995; Nelson & Bridges, 1999; Reskin, 2000). As the representation of women in management in the United States has been rising, 2 a body of research in the organizational demography tradition argues that a greater proportion of women in management is associated with greater gender pay equity among non-managerial employees (Baron et al., 1991; Cohen & Huffman, 2007; Huffman et al., 2010; Hultin & Szulkin, 1999; Joshi et al., 2006; Kulis, 1997; Pfeffer et al., 1995). Since these studies do not account for the degree of formalization characterizing pay systems in the settings under study, it is difficult to infer that these observed associations are the result of female manager impact. To the extent that these pay systems were highly formalized, for example, female managers may not even have had the opportunity, even if they wanted, to redress gender pay inequality among employees. By simultaneously considering formalization of pay and manager gender, this study more accurately isolates whether female managers serve to reduce gender pay inequality when given discretion through less formalized pay systems. Given the extensive evidence showing the behavioral effects of ingroup and outgroup membership categorization, I predict that the relationship between formalization of pay systems and employee pay will vary by manager gender such that: Hypothesis la: For less formalized components of pay, female employees will earn more when reporting to a female, as compared to a male, manager. 12 Data supporting this trend obtained from a U.S. Bureau of Labor Statistics report entitled "Women in the Labor Force: A Databook." The report was issued in May 2014 and accessed in March 2015 via www.bls.gov. 107 Hypothesis 1b: For less formalized components of pay, male employees will earn more when reporting to a male, as compared to a female, manager. 4.2.3 Explaining Variation in Manager Preferences Using Status Characteristics Theory Despite the general tendency for individuals to prefer people like themselves, there is evidence that individuals may act ambivalently toward the ingroup, or even advantage members of the outgroup, in some cases. Thus, this difference in employee pay for those employees reporting to female versus male managers may not be ubiquitous. Specifically, I argue it varies based on the status and gender-typing of the job of the target, or employee, being assessed. Status characteristics theory is useful for understanding why this is the case. According to this perspective, status characteristics emerge when greater social esteem and competence are accorded to members possessing one category of the characteristic (e.g. male, white) than another (i.e. female, black) (Berger et al., 1977). Because these status beliefs are widely shared, it is posited that both members of the high- and low-status groups accept that the high-status group is perceived as generally more competent than the low-status group. For members of high status groups, these consensual status beliefs intensify the ingroup preference, "[b]ut these same status beliefs blunt the ingroup bias of lower status group members who are torn between sticking with their own or favoring those from highstatus groups" (Ridgeway, 2014: 79). As such, these status beliefs "transform competing ingroup favoritisms into a consensual belief that, whether individuals like it or not, people in one group are accorded more social respect than are those in another" (Ridgeway & Correll, 2006: 433). Consequently, members of lower-status demographic groups are generally less apt to favor ingroup members than are members of high-status groups (Bettencourt, Charlton, Dorr, & Hume, 2001; Ely, 1994, 1995; Ibarra, 1992; Ridgeway & Correll, 2006). Generally in the U.S., gender is one such status distinction where men are more highly valued, and generally perceived as more competent, than are women (Berger et al., 108 1977; Correll & Ridgeway, 2003). Since this gender status belief is widely-held and institutionalized, there may be social costs associated with challenging this status ordering, leading people to largely comply even if they do not personally endorse that men are higher status (Ridgeway & Correll, 2004, 2006). As a result of this gender-based status difference, ingroup preferences and preferences for the higher status gender align for men. On the other hand, for women these preferences are at odds and potentially pose a tension. Thus, while male and female managers may generally favor their same gender employees, as members of the lower-status gender, female managers may not always exercise a preference for their female employees. Extant research suggests that gender status beliefs are strongest, and result in the greatest male advantage, when actors performing tasks that are stereotypically masculine, or higher-status, are being evaluated (Ridgeway, Ridgeway & Smith-Lovin, 1999). 1993, 1997; Ridgeway & Correll, 2004; Jobs become labeled as masculine when there is a predominance of men employed in the job, such that over time, even in sex-segregated settings, both managers and occupants of the job reinforce this gender stereotypic labeling (Ridgeway, 1997). Unsurprisingly, jobs typically higher in the organizational hierarchy are typed male as the traits seen as necessary for success in these higher-status jobs are more & commonly associated with men (Eagly, 1987; Heilman, Block, & Martell, 1995; Rudman Glick, 1999). On the other hand, jobs lower in the organizational hierarchy are typically considered female and perceived to be lower-status (Baron & Newman, 1990; Ibarra, 1993; Kilbourne, England, & Beron, 1994). For example, Glick (1991) found a negative relationship between prestige and the belief that occupations are female-typed, or require feminine personality traits (see also England, Herbert, Kilbourne, Reid, & Megdal, 1994). Research examining the prevalence of pro-male biases suggests that female managers, like their male counterparts, are most apt to privilege members of the higher status gender (i.e. male employees) when evaluating employees in higher-status, male-typed positions. For example, in lab-based studies, whereas both men and women exhibit a preference for men in male-typed tasks, only male subjects have been found to favor men in gender neutral tasks 109 (Foschi et al., 1994). Similarly, Moss-Racusin and colleagues (2012) found that both male and female faculty members favored male student applicants over identical female applicants in the traditionally male-typed fields of biology, chemistry and physics. Furthermore, this tendency for women to hold pro-male biases when evaluating others in higher-status, maletyped, as compared to lower-status, female-typed, tasks may be most pronounced among women in higher levels of the organizational hierarchy. Some studies have found that women in managerial positions are more likely to distance themselves from other women (Chattopadhyay, Tluchowska, & George, 2004; Derks, Laar, Ellemers, & Groot, 2011; Duguid, Loyd, & Tolbert, 2012) and initiatives targeted at advantaging the position of women (Ng & Chiu, 2001). Following this logic, female managers are most likely to share the same preferences as male managers and, as a result, not favor their ingroup (i.e. female employees) when assessing employees in male-typed, high-status positions. A direct implication of the research outlined here is that, to the extent female managers do favor their ingroup, they are most likely to do so when assessing employees in lower-status, female-typed organizational positions. While generally it is less clear whether female employees will be favored in femaletyped jobs, with research suggesting at most a modest female advantage (Ridgeway, 1993; Ridgeway & Correll, 2004; Swim & Sanna, 1996), these studies have not considered manager gender. Since female managers are less apt to hold pro-male biases when assessing others in female-typed, lower-status organizational positions, I propose that female managers will favor their gender ingroup in these positions, such that: Hypothesis 2a: For less formalized components of pay, female employees in the lower-level, female-typed organizationalpositions will earn more when reporting to a female, as compared to a male, manager. Hypothesis 2b: For less formalized components of pay, male employees in the lower-level, female-typed organizationalpositions will earn less when reporting to a female, as compared to a male, manager. 4.3 DATA AND METHODS 110 4.3.1 Research Setting I examine whether the relationship between formalization of pay systems and gender pay inequality varies by manager gender using data from multiple retail branches of a globally diversified financial services firm (FinServ). I leveraged the richness of data that results from focusing on a single firm in order to accurately test the hypotheses outlined herein (Fernandez, Castilla, & Moore, 2000; Sorensen, 2000). FinServ provided the annual personnel databases for all employees in its U.S. operations for the 41-month period from January 1996 through May 1999 (the study period). These databases contained demographic characteristics, including gender, race, and age, as well as employment specific characteristics, including date of hire, pay, and performance. These data are particularly well-suited for identifying whether less-formalized pay systems lead to different levels of pay inequality based on whether employees report to a female versus a male manager for at least four reasons. First, employees within FinServ's branch system were eligible to receive two distinct forms of compensation, base and bonus pay, with the latter being far more formalized." Examining whether male and female managers pay decisions differ in terms of more formalized pay, where managers do not have discretion, provides a way to rule out an alternative explanation. Specifically, observing a difference in more formalized pay by manager gender would suggest that an unobserved factor, and not the manager, was driving the observed difference. Second, the same manager makes the pay decisions for each of their employees in terms of both types of pay, allowing for a within-manager analysis of pay for male and female employees reporting to male versus female managers. By comparing pay decisions made by the same group of evaluators, this setting allows me to reduce concerns that observed differences, or a lack of a difference, in employee pay by manager gender is driven by unobserved differences in the types of managers making each pay decision. A detailed discussion of these two components of pay, particularly focusing on how these differ in degree of formalization, is presented in the section below titled Compensation Structure and Formalization. 13 111 Third, personnel databases included complete workgroups with organizational codes linking individual managers to each non-managerial employee reporting to them. To the extent that managers have the discretion to affect employee outcomes, they are most apt to have an impact on those employees reporting to them directly (c.f. Hultin & Szulkin, 2003). Therefore, these data present a rare opportunity for a proximate examination of the relationship between relative pay for male and female employees, for both more and less formalized types of pay, and the gender of their direct manager. Fourth, by examining differences in pay outcomes for employees reporting to female versus male managers within FinServ's retail branches, this study compares structurally equivalent subunits (i.e. branches), allowing for a commensurate comparison of the effects of male versus female managers on employee pay. Each manager in this setting oversees a single comparable branch that employs individuals in each of the five non-managerial positions. These managers operate under the same organizational context and pay guidelines, which is particularly important given the focus of this study. By comparing branches within the same organization, this analysis exploits variation in manager gender across these subunits while holding potentially confounding organizational attributes constant. My setting allows me to disentangle whether the gender of an employee's manager has implications for her pay from other organization- or branch-level alternative explanations, by isolating manager gender while holding constant organization-level policies and controlling for attributes that vary across branches. 4.3.2 Participants FinServ's retail branches are client facing offices that offer a wide range of financial products and services to retail financial services customers. Among others, these services and products include mutual fund accounts, mortgages and margin lending accounts. These retail branches employ individuals in six unique organizational positions: five non-managerial and one managerial, as depicted in Figure 1. Each branch employs an average of approximately nine non-managerial workers all reporting to a single branch manager. 112 Using human resources records, I constructed a longitudinal database for all full-time nonmanagerial employees and managers employed in FinServ's 120 retail branches across eight U.S. states during the four-year study period. Since employees enter and exit the branches at different points, this database of full-time employees is unbalanced, with each employee being present in the retail branches data between one and four years. During the study period, employees may be present for all years of the study, be hired into the branch system after the beginning of the study, leave the branch system before the end of the study window, or both enter and exit during the study window. For example, an employee present for the entire duration of my study will have a total of four person-year observations. On average, non-managerial branch employees are present in these data for two and a half years. The analyses included a total of 897 unique full-time non-managerial employees, for a total of 2,139 person-years for non-managerial employees, and 156 full-time branch managers, for a total of 359 person-years for branch managers. 4.3.3 Compensation Structure and Formalization As aforementioned, FinServ's compensation structure included base salary and bonus pay for all branch employees, which together constituted total compensation for employees in the branch system. In order to understand the compensation structure and the relative degree of formalization for each of the two types of pay, I leveraged various organizational documents, primarily FinServ's employee handbook and materials from a presentation on the compensation structure targeted to branch managers. These materials provided explicit evidence of the degree of formalization characterizing these two components of pay (cf. Elvira and Graham 2001). Base salary comprises the greatest proportion of total compensation, as described in Table 1, for employees across all branch positions. Generally, highly formalized pay systems are characterized by unambiguous rules and procedures (Anderson & Tomaskovic-Devey, 1995; Huffman & Velasco, 1997; Weber, 1958), which was not the case for base salary at this firm. Organizational documents prescribed the approved base salary range for every level of the job hierarchy and provided explicit guidelines that employees should be compensated within the bounds of their pay grade. Within these wide pay ranges branch managers had discretion over setting employee base salary. The upper 113 bound of pay ranges were deliberately written as "up to" the prescribed dollar amount in an effort to "leave room for flexibility", according to the employee handbook. Furthermore, for any given position there were up to four possible pay grades with a spread between the 50th percentile of the lowest and highest pay grade within a position of up to 20,000 dollars. For example, for lowest-level organizational position of teller, the pay ranged from just under $18,000 to nearly $31,000. Managers were not given explicit criteria for assigning employees to a particular pay grade within the job category or for establishing employee pay within a given grade. Furthermore, in a presentation on the compensation structure, FinServ communicated to branch managers the organizational goal to "empower managers to make compensation decisions" and to "institute flexible versus mechanical compensation". Consistent with this perspective, the employee handbook indicates that the "branch manager owns compensation decisions." Thus, base salary was not highly formalized as characterized by the loosely defined guidelines around these pay decisions. In contrast, bonus pay at FinServ was more formalized allowing managers less discretion, as there were clear organizational guidelines dictating how managers would determine bonus pay for individual employees. Bonus pay was allocated based on "one consistent, cohesive plan," according to the employee handbook. Generally, bonus pay tends to be variable and not guaranteed, often being tied to financial performance of the organization or unit (Gerhart & Milkovich, 1990). In line with this argument, bonus pay was based on a combination of employee performance and performance of the branch. One might expect that a high level of managerial discretion would enter at the point of performance evaluations, but because performance ratings were limited to three categories - below par, at par, and above par - and based on objective criteria there was minimal discretion at this stage." While managers were not mandated to allocate a certain percentage of FinServ, branch managers evaluate all employees along six predetermined performance dimensions, which are the same for all employees in the same branch position. Managers assess whether the target employee is below par, at par, or above par for each dimension and based on these individual scores assigns a single overall performance rating of below par, at par, or above par. Each branch position comes with same performance goals, used to create a scorecard for each branch position with which managers assess employees. While the manager has the ability to reallocate the position goals, changes affect the criteria used to assess all employees within a position. Thus, these criteria used for evaluation are not negotiated with individual employees. The branch goal is the sum of all the position goals. 1At 114 employees to each of the three performance categories, over 60 percent of employees received a rating of at par. Consistent with the notion that formalization limits variability in decisions across managers (Finkelstein & Hambrick, 1990) , there was no difference in the distribution of male and female employees to the three performance categories by manager gender. Furthermore, bonus pay decisions were monitored by both the senior managers overseeing branch managers and the nonmanagerial employees receiving the bonus pay. Specifically, all bonuses were approved centrally by the business unit manager and the formula used to calculate bonus pay was available to employees. Another key measure for the degree of formalization of pay is the amount of documentation devoted to defining a component of pay (Elvira & Graham, 2002; Gomez-Mejia & Balkin, 1992). For FinServ, the organizational documents devoted far more attention to bonus pay relative to base salary. This illustrates that at FinServ bonus pay was far more formalized than was base salary, such that managers had more discretion over base salary pay decisions. 4.3.4 Measures Annual Pay. The main dependent variables in this study are the two components of annual non-managerial employee pay, which, as described above, vary in degree of formalization. Formalized pay was measured using annual bonus pay, which was highly formalized allowing minimal managerial discretion. Less formalized pay was measured as annual base salary, which allowed managers greater discretion over compensation decisions. As is commonly the case for employee compensation, both formalized pay and less formalized pay were log-normally distributed, and therefore, I transformed each of the measures of pay to achieve a distribution that more closely resembled the normal distribution. Organizational positions. Figure 1 depicts the hierarchy of positions, as represented in FinServ documents, as well as the key responsibilities of each role. FinServ refers to the position of branch manager as a "manage others" position, given that a key set of responsibilities is related to personnel management and managing the branch team. The five non-managerial branch positions follow a clear hierarchy and there is a structure of cumulative skills and knowledge (see Figure 1). 115 Incumbents of each position were expected to be expert in particular skills associated with their position as well as the skills of each lower level position. For example, representatives are focused on unsecured products and sales and service, but they are also responsible for the tasks of tellers, namely basic products and transactions. [INSERT FIGURE 1] At FinServ, organizational positions vary in their gender-typing and status, making these positions particularly well-suited for testing hypotheses 2a and 2b. First, as depicted in Figure 1, the organization categorizes each of the five non-managerial jobs as either a clerical or a sales position, with all positions except for the lowest-level position of teller being labeled as sales. As is evident in the job descriptions (Figure 1), the position of teller is the only branch position without a sales component as this position deals exclusively with customer service tasks. The degree to which positions contribute to sales increases along the branch hierarchy. Generally, sales jobs are gendertyped as masculine, whereas clerical occupations, including bank teller, are largely perceived to be female-typed (e.g. Glick, 1991). Consistent with this gender-typing of jobs argument, women at FinServ tend to be concentrated in the lower-level organizational positions involving minimal sales related tasks (Figure 2). Thus, lower-level positions, particularly the position of teller, are femaletyped. [INSERT TABLE 1] Second, as with most organizations, lower-level positions at FinServ were not only femaletyped, but also lower-status. Consistent with the branch hierarchy depicted in Figure 1, Table 1 reveals that both the mean annual base salary and bonus pay increase steadily along the organizational hierarchy, with the lowest paid position being teller and the highest non-managerial position being relationship manager. As depicted in the last column of Table 1, the proportion of total pay that is bonus pay also generally increases along the hierarchy, with bonus pay ranging from approximately 1.5 to 7.5 percent of total annual pay. Taken together, these distinctions between 116 occupational positions, particularly with jobs lower in the organizational hierarchy being female" typed and lower-status, further validate the use of the branch jobs to test hypotheses 2a and 2b. Organizational positions was measured using dummy variables for each of the 5 nonmanagerial branch positions - teller (0 = "not teller," 1 = "teller"), representative (0 = "not representative," 1 = "representative"), officer (0 = "not officer," 1 = "officer"), executive (0 = "not executive," 1 = "executive"), and relationship manager (0 = "not relationship manager," 1 = "relationship manager"). To facilitate interpretation, the lowest position in the organizational hierarchy, teller, was omitted from all models and serves as the reference category. Gender variables. The main explanatory variables are employee female and manager female to capture the gender of the focal employee and the employee's manager, respectively (0 = "male," 1 = "female"). The main focus of this study is to determine whether the relative wages for male versus female employees differs by the manager's gender. Therefore, the interaction between manager female and employee female is used to capture this difference. Furthermore, to test hypotheses 2a and 2b, some models also include the three-way interactions between employee female, manager female, and each organizational position to assess whether any observed difference in employee pay by manager gender varies by organizational position. Control variables. To reduce potential confounding effects, I controlled for several variables at the individual-, manager-, and branch-level that are commonly thought to correlate with employee pay. At the non-managerial employee-level, personnel databases were used to determine length of organizational tenure, age, marital status, ethnicity, and performance for all non- managerial employees. I control for tenure and age to account for differences in human capital. Employee age and employee tenure are measured in years for each year from 1996 to 1999 based on The choice to focus on jobs, as opposed to the broader categorization of clerical versus sales, is deliberate as it allows for within-job wage comparisons where men and women are performing the same tasks. It is well established that a substantial portion of gender wage inequality is due to the differential sorting of men and women into different jobs (e.g. Peterson & Morgan, 1995). For example, in this case, because women are disproportionately employed in the lower-level positions, lumping all sales jobs together would lead to an overestimation of the wage gap (i.e. wages for women would be downward weighted towards the lower-level position of representative, whereas wages for men would be weighted toward the higher-level positions). 15 117 the employee's date of birth and date of hire, respectively. To account for potential differences in pay linked to ethnicity, I include non-white to represent the employee's ethnicity (0 "white," 1 = "non- white"). Married identifies whether an employee is married in a given year (0 = "married") married and is included because extant research has found that "not married," 1 = women are disadvantaged relative to their male counterparts (e.g. Correll et al., 2007; Killewald & Gough, 2013). In addition to employees' demographic characteristics, 16 the models also include measures of promotion and performance, both of which are likely associated with employee pay. Promoted represents whether an employee received a promotion during the study period (0 = "not promoted," 1 = "promoted"), which could indicate a new position or an increase in pay grade. Performance was assigned by an employee's manager and is based on a one to three scale as explained above. The performance scores capture how well an employee meets expectations and represent the following: one indicates "does not meet expectations", two indicates "meets expectations", and three indicates "exceeds expectations". Since performance was missing at random for approximately 155, or 17 percent of, non-managerial employees, models including performance are based on the remaining 742 non-managerial employees. At the manager-level, personnel databases were used to determine branch manager's length of organizational tenure and age. These controls were included to capture potential differences in manager compensation decisions for more versus less experienced managers. For example, less experienced employees may seek legitimacy and, as a result, be more reluctant to exercise discretion. Manager age and manager tenure, each measured in years for each year from 1996 to 1999, were calculated based on the manager's date of birth and date of hire, respectively. A third set of controls is related to branch-level attributes. Larger, more lucrative branches may be of higher status, indicating that managers at larger branches may differ in some substantive Educational information (i.e. highest degree) is only populated for 39 percent of employees included in the study population. This information is provided by employees on a voluntarily basis. Since this information is not missing at random, education is not included in the main analyses. However, for robustness, including education does not significantly change the main results. Results of this analysis for tellers are included in the appendix, Table Al. 118 way from, or have more resources than, managers of smaller, less profitable branches. Therefore, all models control for branch size, coded as the number of non-managerial employees for each year from 1996 to 1999. As a proxy for branch performance and market share, data on branch revenues were collected from the Federal Deposit Insurance Corporation (FDIC).1 Branch revenues were log- normally distributed so this variable was transformed to achieve a distribution that more closely resembles the normal distribution. The gender composition of the work group may also have implications for gender pay inequality (c.f. Joshi et al., 2006). Therefore, using the count of male and female employees in a branch in a given year, percent female was calculated for each branch. Lastly, all models include state fixed effects for the state where the branch is located to account for any regional differences in wages. 4.3.5 Analytical Strategy I estimated various cross-sectional time-series linear models using the method of generalized estimating equations (GEE) 8 to examine when and how gender pay inequality differs based on manager gender. I used multivariate analyses with the logarithm of pre-tax less formalized pay (i.e. base salary) and formalized pay (i.e. bonus pay) as the dependent variables to examine whether relative pay for male and female employees differs by manager gender. Because employees in this setting were nested within branches, individuals within branches may have more in common than individuals across branches. This indicates that employees within branches may not be independent, violating a key assumption of ordinary least squares (OLS), therefore OLS may result in biased estimates of standard errors. Furthermore, the data I analyzed were pooled, cross-sectional time series (yearly) data. A common approach for analyzing data that are structured in this way is to use fixed-effects models, which would capture within-individual and within-branch, over time variation. Given my research question, I am inherently interested in the gender of non-managerial employees These data were collected based on a useful suggestion from an anonymous reviewer. 18 Results are largely consistent when estimating models using less structure on the variance-covariance matrix (i.e. 1 OLS). 119 and the gender of branch manager, two attributes that are time-invariant.1 9 Therefore, a fixed-effects model is not suitable for this study as both employee gender and branch manager gender would be dropped from such models. As aforementioned, the data that I analyzed are longitudinal and include repeat observations for individuals over time. Therefore, I ran diagnostics to test two key assumptions of these models, namely that errors are homoscedastic and not auto-correlated across time periods. As expected, errors in these data are heteroscedastistic, as the variance of the errors varies across individual nonmanagerial employees. Additionally, and as expected, the Woolridge test for autocorrelation in panel data reveals that there exists first-order auto-correlation, where the errors at time t are correlated with the errors at time t-1. Thus, longitudinal GEE models accounting for both heteroscedasticity and first-order autocorrelation were used to estimate the effect of manager gender on gender-based wage inequality. 4.4 4.4.1 RESULTS Descriptive Statistics [INSERT TABLE 2] Table 2 provides overall means, standard deviations, and correlations for all variables used in this study. In terms of distribution of employees across branch positions, Table 3 shows that over 75 percent of all employees occupy one of the three lower-level positions of teller, representative or officer. Consistent with the notion that women are underrepresented in higher-level organizational positions (Albrecht, Bj6rklund, & Vroman, 2003; Morrison, White, Velsor, & Leadership, 1987; Powell & Butterfield, 1994), Figure 2 reveals that the proportion of women in the lower-level 19 Theoretically, branch manager gender may vary in cases where a branch is managed by a female manager in one year and a male manager in a subsequent year (or vice versa), for example. During this time period, however, these switches in manager gender were very rare (n = 50). Only 3 branches, for a total of 50 individual non-managerial employees, experienced such an event. Furthermore, the direction of this switch in branch manager gender was not consistent: 1 branch went from having a female to a male manager and the other 2 branches went from a male to a female manager. Therefore, branch fixed effects would not allow for estimating parameters for the effect of manager gender for most cases in this study. 120 positions of teller and representative is considerably higher than in higher-level positions. However, this observed pattern is independent of manager gender. Therefore, while I am not able to draw any conclusions regarding the impact of female relative to male managers on this distribution of women across levels of the organizational hierarchy, 0 the gender composition of jobs is the same for both male and female managers. [INSERT FIGURE 2] [INSERT TABLE 3] The branch managers comprise approximately 13 percent of all full-time employees in the retail branches. Each branch is headed by a single manager with sole personnel responsibilities to oversee all branch employees and operations. As presented in Table 1, the average annual base salary for managers was approximately $65,000 and exactly half of all managers were female. While Finserv was committed to generally fostering diversity and providing equal employment opportunities, this equal representation of women in management is not the result of explicit gender equity initiatives. Given that the focus of this study is to determine whether the degree of gender inequality among employees differs based on the gender of their manager, it is important to identify if male and female managers in this setting differ in any substantive ways other than gender. I found that male and female managers differed in pay with female managers earning slightly less than male managers on average (p < 0.05), despite being approximately two years older (p < 0.05) and having nearly three more years of experience (p < 0.01) on average. Given the aim of this study, it is necessary to not only compare male and female managers, but also whether male and female managers oversee comparable branches and individuals. While female managers run branches that are similar in terms of number of employees and gender 20 While this pattern is evident, this is simply a descriptive pattern as these are post-hire data which do not provide insight into the mechanisms leading to this apparent glass ceiling (for a similar discussion of the limitations of posthire data see Fernandez & Weinberg, 1997). For example, not knowing the proportion of women that are in the consideration set for higher-level positions (i.e. relationship manager) makes it equally plausible that 1) women are simply not applying to higher level jobs (as has been shown by Barbulescu & Bidwell, 2013, for example) or 2) managers are showing a preference in selection for male applicants for higher level jobs. In order to isolate the mechanisms at play in creating these observed patterns it is necessary to examine pre-hire data. 121 composition, female managers head branches with significantly lower branch deposits on average (p < 0.01). This is consistent with the perspective that even within the same organizational positions, women are the lower status group (Ridgeway, 1991, 1997) managing less powerful, non-growing organizational units (Stover, 1994) than their male counterparts. To be sure that these differences are not accounting for any observed relationship between employee pay and manager gender, all models control for manager age, manager tenure, and branch deposits. To compare characteristics of the employees that male and female managers oversee, Table 4 presents key descriptive statistics for all variables of interest for employees by both employee and branch manager gender. To attribute differences in the degree of gender inequality in less formalized components of pay to the gender of the manager, it is necessary that male and female managers be overseeing employees that are similar on key dimensions. Therefore, the goal of this comparison is to determine whether the employees reporting to male and female managers differ in any substantive way so that these attributes can be included as controls in the regression models. On average, female non-managerial employees reporting to female managers do not differ significantly along any dimension from those reporting to male managers. On the other hand, male non-managerial employees reporting to female managers earned lower annual base salary and bonus pay (p < 0.05), are slightly older (approx. 2.5 years, p < 0.05), and are more apt to be married (p < 0.05), on average than male employees reporting to male managers. Given the overall similarities between employees reporting to male and female managers, any observed difference in employee pay based on manager gender is unlikely to be driven by characteristics of the employees. But to be sure this is the case, all models control for each available employee characteristic. [INSERT TABLE 4] 4.4.2 Overall Effect of Formalization on Gender Pay Inequality To highlight the shortcomings of previous studies failing to simultaneously account for formalization of pay and manager gender, the first two sets of models in Table 5 present the average gender difference in pay for male and female non-managerial employees across organizational 122 positions by type of pay. These models account for formalization of pay by separately predicting less formalized pay (models la and 2a) and formalized pay (models lb and 2b), but do not factor in manager gender. Consistent with most existing research on formalization and pay inequality, I find an overall gender wage gap for less formalized pay, but not formalized pay. While model la predicting less formalized pay reveals that on average across organizational positions women earn a base salary that is 21 percent (p < 0.001) less than that of similar male counterparts, model lb indicates that there is no gender difference among employees in formalized pay, or bonus pay. As model 2a and 2b reveal, this finding is robust to the inclusion of performance, with evidence of an even larger gender disparity in formalized pay (model 2a) between male and female employees with equal performance ratings. Based on these results, one would conclude that less formalized pay systems ubiquitously lead to unequal pay for men and women. While not entirely inaccurate, I will show that claims about the relationship between formalization and gender pay inequality from this oversimplified analysis lead to incomplete conclusions. In the ensuing discussion, I demonstrate the importance of simultaneously considering formalization of pay, manager gender, and organizational positions to better understand the conditions under which less formalized pay systems may instead lead to equitable pay outcomes. [INSERT TABLE 5] 4.4.3 Formalization and Male versus Female Manager Impact on Pay To test the hypotheses set forth in this study, the remaining analyses examine whether the relationship between formalization and employee pay varies by manager gender or the organizational position of the non-managerial employee. Models 3a and 3b in Table 5 take manager gender into account. By including manager gender in the models predicting less formalized and formalized pay, I am able to determine whether male and female managers use the discretion afforded by less formalized pay systems differently, particularly in a way that impacts gender pay inequality (Hia, 1b). Then, the models in Table 6 examine this relationship by the organizational position of the non- 123 managerial employees. An examination by organizational position establishes whether the impact of female managers on employee pay varies across the organizational hierarchy, or the gender-typing of the position (H2a, 2b). Main Effects of Manager Gender on Employee Pay. Since both male and female managers are apt to favor their same-gender employees, hypothesis la and lb proposed that in terms of less formalized pay, where managers have greater discretion over pay outcomes, employees would earn more when reporting to a same-, rather than opposite-, gender manager. Consistent with hypothesis 1b, model 3a in Table 5 shows that, on average, male employees reporting to male managers earn approximately five percent (p < 0.01) higher base salaries than do male employees reporting to female managers. On the other hand, the non-significant coefficient of the interaction between manager female and female in Model 3a reveals that there is no difference in female employees' base salaries based on their manager's gender. Thus, model 3a in Table 5 provides support for hypothesis 1b, but not hypothesis la. To highlight that female managers only impact less formalized components of pay, Model 3b in Table 5 compares employees' formalized pay based on whether they report to a male versus female manager. As expected, given that formalization restricts managerial discretion, there is no difference in formalized pay (i.e. bonus pay) for men or women irrespective of whether they report to a sameversus opposite-gender manager. This is evidenced by the non-significant coefficients of each of the manager gender variables (main effect and two-way interaction). Female Manager Impact on Pay by Organizational Position. Before assessing whether female managers' impact on employee pay varies by organizational position as predicted by hypotheses 2a and 2b, the first two sets of models in Table 6 examine the degree of gender pay inequality within each non-managerial branch position. In terms of less formalized pay, Model la shows that female non-managerial employees earned lower base salaries across all occupational positions. As shown by the coefficient of female and the interactions between female and each of the organizational positions, women earned between four and 13 percent less than did similar men in the 124 same branch position, conditional on all covariates. Similarly, Model 2a reveals that while performance is significantly related to base salary, controlling for performance does not account for gender differences in pay. In fact, for the positions of teller, officer, and relationship manager, the magnitude of the gender pay differences are magnified after accounting for performance. In contrast to the observed gender differences in less formalized pay, models lb and 2b show that male and female employees earn comparable bonus pay in each organizational position. In sum, these results provide evidence that there is overall within-job gender pay inequality in terms of less formalized pay across all organizational positions, but not in terms of formalized pay. To test hypotheses 2a and 2b, the last two models in Table 6 (models 3a and 3b) are fully interacted models estimating the effect of manager gender on less formalized pay and formalized pay, respectively. To determine whether the observed impact of female managers on gender pay inequality varies by organizational level, it is necessary to compare the relative pay, in terms of both formalized and less formalized components, received by male and female employees in the same jobs, reporting to female versus male managers. These effects are represented by the coefficients of manager gender (main effect, two-way interactions, and three way interactions). [INSERT TABLE 6] Both hypothesis 2a and 2b proposed that the effect of female managers on less formalized components of employee pay would be limited to employees in lower-status, female-typed organizational positions. Because the position of teller is the reference category, the negative and significant coefficient of the main effect of manager female and the positive and significant coefficient of the interaction between manager female and female in model 3a provides support for hypothesis 2a and 2b. Specifically, among tellers reporting to female managers women earn base salaries approximately six percent (p < 0.05) higher, and men earn base salaries over four percent (p < 0.05) lower, than do female and male tellers reporting to male managers. An examination of the other coefficients including manager female (two-way and three-way interactions) reveals that there 125 is no significant difference in employee pay for those reporting to a female, relative to a male, manager in any of the other organizational positions. Model 3b reveals that female managers only compensate employees differently than male managers in terms of less formalized pay systems, as evidenced by the non-significant coefficients of all variables including manager gender (main effect, two-way, and three-way interactions). Because managerial discretion is constrained, there is no difference in formalized pay (i.e. bonus pay) for men or women in any branch position based on whether they report to a same- versus opposite-gender manager. As a result of the impact of female managers on less formalized pay, male and female tellers reporting to female managers receive nearly equal base salaries whereas there is a seven and one half percent (p < 0.01) wage differential among tellers reporting to male managers. Therefore, in support of hypothesis 2a and 2b, these results show that when female managers are afforded discretion via less formalized pay systems, they close the gender pay gap, but only among employees in lowerstatus, female-typed organizational positions. The fact that the impact of female managers is limited to tellers helps to explain why hypothesis la was not supported. Based on models aggregating across organizational positions (Table 5, model 3a), there was no evidence of female managers paying female employees higher base salaries than do male managers. Since female managers do not pay female employees more than do male managers in the other four organizational positions, it is not surprising that the average effect of manager female was not significant in this pooled model. To fully elucidate the relationship between manager gender and less formalized components of employee pay, it was necessary also to account for the organizational positions employees occupy. In summary, these analyses provide conditional support for the notion that male and female managers impact employee pay differently when given discretion over pay decision via less formalized pay systems. Whereas models not accounting for manager gender and organizational position suggested that less formalized pay systems universally lead to inequitable pay outcomes, the more 126 nuanced approach taken in this study suggests that there are boundary conditions to this claim. Female managers used the discretion afforded by less formalized pay systems to pay employees in the lower-level organizational ranks more equally than did male managers. Together these results highlight the importance of considering both manager gender and organizational positions in examinations of the relationship between pay formalization and gender inequality. 4.5 DIsCUSSION Gender differences in earnings are a persistent feature of the U.S. labor market, despite the increased organizational and policy initiatives aimed at eradicating this form of inequality. Scholars have largely posited that formalized pay systems that reduce managerial discretion through unambiguous written rules serve to reduce gender pay inequality (e.g. Elvira & Graham, 2002; Reskin, 2000), though some recent studies call this into question (e.g. E. Castilla, 2008; Dencker, 2008). Drawing on social identity theory and status characteristics theory, this study developed and tested hypotheses that the impact of pay formalization for gender pay inequality varies based on the gender of the organizational decision maker and the organizational position the manager oversees. Using unique personnel data from 120 branches of a large retail financial services firm, I demonstrate that manager gender and the organizational position of the employees being assessed do in fact matter. Specifically, failing to account for these may lead to erroneous, or oversimplified, claims about the relationship between pay formalization and gender pay inequality. By analyzing the degree of gender pay inequality across two components of pay, namely base salary, which is less formalized, and bonus pay, which is highly formalized, I am able to first show that male and female managers impact gender pay inequality differently. Consistent with existing research, generally formalization was associated with more equitable pay in this setting, as evidenced by the equal bonuses awarded to male and female employees irrespective of manager gender. However, contrary to what most of the literature would expect, I find evidence of greater gender equity in terms of less formalized pay systems for employees reporting to female managers. Specifically, I find that female managers use the discretion afforded to them by less formalized pay systems to pay male and female 127 employees more equitably than do male managers. This effect is limited to employees in the lowest organizational ranks providing conditional support for the perspective that less formalized personnel practices may result in gender pay equity for employees reporting to female, but not male, managers. 4.5.1 Importance of Considering Characteristics of Manager and Formalization These findings have several implications for research on the formalization of rewards systems and gender inequality. First, this study demonstrates the importance of simultaneously considering characteristics of the decision maker and formalization of organizational reward systems. By taking manager gender into account, this study helps us move beyond our understanding of whether formalized pay systems are an effective way to redress workplace inequality, to identifying when this is most likely to be the case. In terms of gender inequality, most existing research has led to conclusions that less formalized pay systems, which allow managers more discretion, lead managers to privilege male employees and devalue female employees (e.g. Elvira & Graham, 2002; Reskin, 2000; Ridgeway, 1997). Implicit in this theory is that all organizational decision makers will use the discretion afforded by less formalized pay systems uniformly. However, my study finds that, unlike male managers, when female managers have discretion they advantage their female employees, thus reducing gender pay inequality. Therefore, this study suggests that the common association between formalization of pay systems and gender inequality is at least partly due to the historical overrepresentation of men in decision-making positions. As such, this study offers a corrective to existing research on formalization of pay and gender inequality by showing that it is necessary to consider manager gender. Relatedly, my study directly builds on recent research closely examining the relationship between formalization and inequality by highlighting an additional condition under which less formalized practices redress inequality. For example, Dencker (2008) offers a provocative finding that when organizations are facing pressures for gender equity, less formalized practices empower managers to react to these pressures and reduce inequality in promotion of higher-level employees. By introducing manager gender, my study offers a possible extension to Dencker (2008). In his study, women comprised a nontrivial proportion of managers, ranging from nearly 36 to over 57 128 percent across functions and departments. Thus, it is plausible that female managers, in particular, are contributing to the female advantage observed during the period where practices were less formalized, further highlighting the importance of accounting for manager gender. My study also contributes to the literature on the impact of women in management for gender inequality by highlighting the importance of accounting for both manager gender and formalization of reward systems. Specifically, I demonstrate that failing to account for formalization, which dictates the level of discretion managers have over pay decisions, may lead to incorrect estimates and conclusions about whether having women in management leads to more equitable pay. For example, while not accounting for formalization, Penner and colleagues (2012) use withinorganization employment records for a grocery retailer operating under a collective bargaining agreement. " They find that there is no difference in gender wage inequality among supermarket workers based on the gender of the manager to whom they report, arguing that like their male counterparts female managers devalue work performed by female employees. However, since collective bargaining agreements are commonly associated with more formalized pay systems and less managerial discretion (c.f. Hirsch, 2012), it is plausible that their finding is at least partly due to female managers not having sufficient discretion over employee pay to have an impact in this setting. 4.5.2 Importance of Accounting for Context-specific Status Ordering Second, this study demonstrates that the impact of female managers on gender pay inequality varies by organizational position, or level of the hierarchy. As members of the lower status group, female managers face a tension between favoring members of their gender ingroup and favoring the higher-status outgroup members. Therefore, to understand when female managers are Penner et al. (2012) makes the claim that managers have control in their research setting; however the retail grocer they study has several distinct features that cast doubt on the appropriateness of this setting for addressing whether female managers impact wage inequality. First, as the original paper analyzing these data states, "all non-managerial employees were covered by collective bargaining agreements" in this setting (Ransom & Oaxaca, 2005: 222). Therefore, employee wages are likely based on union negotiations severely limiting managerial discretion. Furthermore, the within job wage ranges for both non-managerial and managerial employees were narrow, with standard deviations for hourly wages ranging from $0.00 to $1.13. Therefore, while the authors do not find that female managers reduce gender inequality, it is plausible that this finding may be attributed to particular features of the setting. For additional details on this setting, see Ransom and Oaxaca (2005). 129 most apt to favor their female employees it is necessary to consider conditions under which we expect the preference for the ingroup to dominate the preference for the high-status other. In line with status characteristics theory, which highlights the importance of taking the local status ordering into account (e.g. Ridgeway & Correll, 2006), this study considers how female managers may be more apt to reduce gender inequities for lower-status employees, or those in lower-level organizational positions. Relatedly, this study offers a refinement to the conclusions in organizational demography research examining whether the presence of women in management leads to lower levels of gender inequality on average by highlighting the importance of accounting for the jobs of employees being assessed. Extant studies have largely found that a greater proportion of women in management leads to more pay equity among non-managerial employees (e.g. Baron et al., 1991; Cohen & Huffman, 2007; Huffman et al., 2010; Hultin & Szulkin, 1999; Joshi et al., 2006; Kulis, 1997), concluding that female managers redress inequality. However, these studies are conducted at the organizational level analyses and aggregate employees across jobs and levels of the organizational hierarchy. As Table 5 shows, not accounting for organizational level at FinServ would similarly lead to an overly simplified conclusion. While not entirely inaccurate, concluding that female managers redress inequality is incomplete because once organizational level is taken into account it becomes evident that female managers only reduce inequality for employees in the lowest-level organizational position of teller. 4.5.3 Limitations and Future Research Because this study is based on data from a single organization, it is necessary to consider limitations to the generalizability of results. While it is not possible to definitively rule out, there is no reason to assume that the managerial practices of this firm are atypical of other large firms (c.f. Kalleberg, Knoke, Marsden, & Spaeth, 1996). This is especially unlikely given that the analysis is centered on the population of branches of a Fortune 500 retail bank. Furthermore, the validity of this critique would require that not only the organization, but also the compensation decisions of individual male and female managers within the organization, be atypical. Given that the branches studied are distributed across eight U.S. states and are managed by 156 different branch managers, 130 these branches are parallel organizations with variation in the key independent variable, namely the gender of the branch manager. Again, as with any case study, it is not possible to discount this explanation in absolute terms; yet, it is unlikely that the female managers in this setting impact gender wage inequality differently than would female managers in other settings. To bolster the generalizability of these findings, future research could replicate the empirical strategy used in this study across firms. A second limitation stems from constraints resultant from the records maintained by FinServ. Specifically, I was unable to control for employee education2 2 or job tenure. I address this by including age, as a proxy for experience, and organization tenure, as a proxy for job tenure. 23 There is also an opportunity for future research to expand on this study by considering other manager characteristics that may serve as bases for ingroup categorization in examinations of formalization and inequality. While it is plausible that in- versus outgroup categorizations along other status distinctions, such as race, lead to similar results as those observed in this study, assuming these processes operate the same way across different status characteristics may lead to incorrect conclusions (Joshi et al., 2006; Ridgeway, 2014). Relatedly, my finding that the impact of female managers on pay varied by organizational position suggests that future research should simultaneously consider characteristics of the manager and the context-specific status ordering. These approaches will yield a more complete understanding of the conditions under which providing managers with discretion via less formalized rewards systems may lead to more equitable outcomes. Another profitable direction for future research is to further unpack the mechanisms driving female managers to use the discretion afforded by less formalized pay systems to both pay female employees more and pay male employees less than do male managers. Existing theories of ingroup favoritism suggest that actors will advantage their ingroup. The most direct way that female Education was a self-reported variable not required of FinServ Employees. As a result this field was missing for nearly 70% of the employees in my study. Even more importantly for this study, men underreported educational information such that it is not missing at random. Despite this large amount of missing data, Table Al in the appendix predicting base salary for tellers (the only position for which there is a manager gender effect) includes education and reveals results identical to those in the main analyses. 23 During my study window, only 77 out 857 individuals, or 1 in approximately 11 people, were promoted. Therefore, in the absence of direct job tenure information, organization tenure provides a strong proxy for job tenure at FinServ. 22 131 managers could favor their ingroup is by paying of their female employees more than do male managers. A second way that female manager could elevate the position of women relative to men is by paying male employees less than do male managers. An observable difference in employee pay would result for employees reporting to female versus male managers if female managers engage in either, or some combination of both, of these compensation strategies. While I find that female managers do both, and importantly in a manner that leads to more equal pay for employees, these data do not allow me to isolate the mechanism driving this finding. Thus, to more fully unpack how female managers compensate employees differently from male managers, a profitable direction for future research may be to further isolate these mechanisms. 4.5.4 Implications for Organizations In an ideal scenario, organizational practices and rewards systems would be adopted uniformly by all managers and in a manner that resulted in equitable outcomes for all employees. While striving for this ideal should be the underlying objective of organizations seeking gender equity, in the absence of that, my study suggests we should pay attention to the organizational practices that are at a manager's disposal and how they may be used differently across managers. From a practical standpoint, my research warns against blindly assuming that either increasing formalization, or increasing the presence of women in management, will provide an effective means for reducing gender pay inequality. Increasing pay formalization may be more effective for reducing gender pay inequality in settings where women are absent from management or managers oversee employees in higher-status, male-typed jobs. Similarly, increasing the presence of women in management is most apt to lead to gender equitable outcomes in a setting where formalization of rewards systems is low, thus allowing female managers discretion, and female managers oversee employees in lower-level, female-typed positions. In order to develop appropriate strategies for reducing gender pay inequality, organizations must concurrently consider the potential role of both female managers and level of the employee they oversee. Similarly, since allowing managers discretion has been posited to provide firms with the opportunity to enhance performance (Gomez-Mejia & Balkin, 1992), my research points to when 132 allowing discretion may simultaneously yield benefits for the firm and its employees. While discretion is commonly perceived to exacerbate gender inequality, my findings suggest that this is not always the case. Rather than favoring women, the female managers in my setting used discretion to close the gender earnings gap among lower-level employees by paying these male and female employees more equally. Thus, these findings suggest that less formalized systems allowing managerial discretion are most apt to advantage the firm without exacerbating gender inequities when women are in management and overseeing lower-level employees. 133 4.6 FIGURES AND TABLES FIGURE 1 Branch Positions: Hierarchy and Description of Responsibilities Non-managerial Positions Pst Magra Rtlaionaship anager Ex ccniv. 4 bhi li ( flticer * itttpitsn . \Ivane. -'I busiw s I . os s m ani m-4ow iiii i I ot simet rv ItwfIaIitt14 t.* TelIed rts i Iiusn wo . stm11 ----- k .11c+ .1 3 Ii k Sales Clerital Posit ions Position FIGURE 2 Percent Female by Job for Non-Managerial Employees, by Manager Gender, Yearly Average Over 4 Years . . ..T *1~ I *1 T a FiIal( Manger a Male 20% 10% 1 %4 Telltr Ihepro;SI11ent ivI Officvr xive Relatgioship .Nmaager Branch Position 134 'Managcr TABLE 1 Mean Annual Base Salary and Bonus Pay by Position for All Retail Branch Employees, Yearly Average Over 4 Years Annual Pay (in dollars) Base Salary Bonus Pay Branch Position Mean s.d. Mean s.d. As Percentage of Total Annual Pay Teller 24,301.31 3,179.76 378.27 592.71 1.53% Representative 28,400.35 4,023.26 762.45 904.60 2.61% Officer 38,635.47 7,550.21 1,746.06 2,716.79 4.32% Account Executive 48,598.29 6,585.50 3,932.69 6,616.21 7.49% Relationship Manager 60,159.13 11,678.58 3,062.93 4,104.40 4.84% Branch Manager 65,342.91 14,002.73 4,446.74 5,099.57 6.37% 135 Variable 1. Annual base salary 2. Annual bonus pay 3. Bonus pay (as % of base pay) 4. Female (%) 5. Age 6. Organizational tenure 7. Non-white (%) 8. Married (%) 9. Performance 10. Teller (%) 11. Representative (%) 12. Officer (%) 13. 14. 15. 16. 17. 18. 19. Executive (%) Relationship Manager Manager female (%) Manager age Manager tenure Branch size Percent female (%) 20. Branch deposits (in 0 (%) 000s) TABLE 2 Descriptive Statistics and Correlations among Study Variables in Person-Period Data Seta Mean s.d. 1 2 3 4 5 6 7 8 9 10 11 12 35,821.81 12,256.55 1,493.75 2,913.68 0.27 0.03 0.05 0.94 0.15 0.67 0.47 -0.26 -0.11 -0.06 36.59 10.05 0.32 -0.03 -0.07 0.14 6.42 6.85 0.22 -0.03 -0.06 0.18 0.55 0.56 0.50 -0.18 -0.06 -0.04 0.06 -0.12 -0.08 0.44 0.50 0.14 -0.02 -0.03 0.02 0.14 -0.06 0.01 2.33 0.54 0.12 0.28 0.30 0.01 -0.08 -0.10 0.06 0.06 0.25 0.43 -0.56 -0.22 -0.22 0.11 -0.15 -0.13 0.14 0.02 -0.11 0.14 0.35 -0.25 -0.10 -0.06 0.12 -0.07 0.00 0.06 -0.06 -0.04 -0.24 0.46 0.50 0.21 0.08 0.13 0.00 0.13 0.13 -0.08 -0.07 0.08 -0.54 -0.38 0.06 0.23 0.26 0.21 0.16 -0.15 0.00 -0.07 -0.09 0.02 0.08 -0.14 -0.10 -0.23 0.08 0.28 0.60 0.16 0.07 -0.19 0.10 0.02 -0.08 0.17 0.00 -0.17 -0.12 -0.28 0.46 0.50 -0.06 -0.04 -0.02 0.04 0.06 0.03 -0.01 0.01 -0.02 -0.01 0.03 0.04 39.99 8.74 -0.01 -0.02 -0.03 0.08 0.06 0.06 0.01 0.03 -0.01 -0.04 0.06 0.02 2.32 6.28 0.00 -0.04 -0.05 0.05 0.07 0.18 -0.01 0.07 -0.01 0.01 0.02 -0.01 9.05 4.48 0.00 0.05 0.07 -0.07 -0.10 -0.10 0.06 0.06 0.05 -0.01 0.00 -0.02 0.67 0.21 -0.19 -0.08 -0.06 0.44 0.08 0.02 0.02 0.03 -0.08 0.06 0.08 0.01 87,230.02 68,923.26 0.03 0.01 0.00 -0.10 0.00 0.11 0.12 0.07 0.07 0.07 -0.03 -0.06 *n = 2,319, except for performance where n = 1,821. Coefficients with an absolute value equal to or larger than .04 are significant at p < .05. 13 14 15 16 17 18 19 -0.07 -0.03 0.01 0.01 -0.16 0.07 0.16 0.03 -0.08 0.10 -0.19 0.20 -0.13 0.19 -0.07 -0.04 0.11 0.19 -0.15 0.42 -0.23 -0.07 -0.03 -0.03 -0.04 0.06 -0.06 -0.04 TABLE 3 Distribution of Employees across Branch Positions, Yearly Average Over 4 Yearsab Organizational Position Percentage of Total Employees Teller 23.85 Representative 13.83 Officer 38.29 (276) (160) Account Executive (443) 4.24 Relationship Manager (49) 6.31 Branch Manager (73) 13.48 (156) a Parentheses indicate number of employees bThere are a total of 104 employees (including 77 non-managerial employees) who change postion within the branch during the years under study. Therefore, the combined total of employees (n=1,157, including managers) is greater than the number of unique employees (n = 1,053). 137 TABLE 4 Basic Descriptive Statistics for Non-managerial Employee-level Variables, by Employee Gender and Manager Gender, Yearly Average Over 4 Years Female Managers Male Managers Mean s.d. Mean s.d. Annual base salary 33,422.14 9,185.85 Annual bonus pay 1,281.57 2,290.27 1,270.22 2,367.02 10.53 Variables t-testa Female Employees Age 33,752.16 10,429.10 37.80 10.09 37.32 Tenure 7.45 6.50 7.11 6.96 Performance 2.33 52.64 2.33 53.89 Married (%) 42.86 49.52 46.09 49.88 Non-white (%) Promotion (%) 56.25 49.64 58.68 49.27 10.27 30.38 10.60 30.80 persons 29 9b 335 b person-years 672 755 Annual base salary 38,705.04 14,787.99 41,449.06 15,173.16 Annual bonus pay 1,534.42 2,362.82 2,224.90 4,546.41 Age 36.13 10.00 33.57 8.37 Tenure 4.87 6.97 4.57 6.57 Performance 2.28 0.54 2.35 54.40 Married (%) 48.18 50.05 38.88 48.81 Non-white (%) 52.15 50.00 51.10 50.05 Promotion (%) 10.89 10.02 30.07 persons 140 179b person-years 303 409 a * Male Employees indicates whether differences in values for male and female managers are statistically significant based on two-sided t-tests b There are a total of 56 employees who worked for both a male and a female manager during the study period (i.e. switchers). Therefore, the combined total of employees reporting to male and female managers is greater than the number of unique employees by 56 as these people are counted for years reporting to male managers and years reporting to female managers (n = 953). ** p 0.05 0.01 p 0.001 * p 138 TABLE 5 Generalized Estimation Equation Regression Models Predicting Log Annual Base Salary and Log Annual Bonus Pay of Non-Managerial Model la Variables (Base Salary) Female Model lb Less Formalized Pay Formalized Pay Model 2a Model 2b' Less Formalized Pay Formalized Pay (Bonus Pay) (Base Salary) -0.208*** 0.173 (0.020) (0.221) Model 3ac Model 3bc Less Formalized Pay Formalized Pay (Bonus Pay) (Base Salary) -0.223*** 0.114 -0.227*** 0.068 (0.022) (0.220) (0.024) -0.050** (0.281) 0.144 (0.017) 0.013 (0.324) 0.112 (0.021) (0.384) Manager Female Manager Female x Female (Bonus Pay) Individual -level controls Performance 0.018*** 1.343*** 0.017*** 1.340*** Weekly hours 0.007 0.097 (0.005) 0.008 (0.161) 0.013 (0.005) 0.008 (0.161) 0.014 (0.005) Employee age 0.010*** (0.099) -0.033** (0.006) 0.009*** (0.096) -0.039*** (0.006) 0.009*** (0.096) -0.040*** Employee tenure 0.015*** (0.012) 0.060*** (0.001) 0.015*** (0.011) 0.028 (0.001) 0.014*** (0.011) 0.027 Married (0.002) 0.074*** (0.018) 0.495* (0.002) 0.063** (0.017) 0.405* (0.002) 0.063** (0.017) 0.406* Non-White (0.019) -0.074*** (0.196) -0.206 (0.021) -0.071*** (0.194) -0.176 (0.021) -0.072*** (0.194) -0.172 (0.019) 0.027*** (0.196) Promoted 1.353*** (0.020) 0.026*** (0.193) 1.176*** (0.020) 0.025*** (0.193) 1.175*** (0.007) (0.348) (0.008) (0.351) (0.008) (0.350) (0.001) Manager -level controls Manager age Manager tenure -0.000 -0.010 -0.001 -0.013 0.000 -0.014 (0.000) 0.001 (0.011) -0.016 (0.000) 0.001 (0.011) -0.021 (0.001) 0.000 (0.011) -0.021 (0.001) (0.015) (0.001) (0.015) (0.001) (0.015) Branch -level controls Branch size Percent female (%) 0.001 0.066** 0.001 0.100*** 0.001 0.102*** (0.001) (0.024) -0.041 -0.969 (0.001) -0.049 (0.024) -1.145* (0.001) -0.055* (0.025) -1.171* (0.023) (0.514) (0.027) (0.519) (0.027) (0.520) Log of Branch deposits (in 000s) Constant Observations Number of GEID Time Periods Wald Chi2 DF -0.023 0.021 -0.016 -0.238 -0.019 -0.216 (0.017) 10.728*** (0.192) 9.348* (0.018) 10.699*** (0.193) 6.581 (0.018) 10.733*** (0.194) 6.387 (0.296) (4.655) (0.335) (4.563) (0.334) (4.578) 2,139 2,139 897 897 1,821 742 1,821 742 1,821 742 1,821 742 4 4 4 4 4 4 674.33*** 88.21*** 557.05*** 183.18*** 577.45*** 184.60*** 19 19 20 20 22 22 SAll models include controls for scheduled weekly hours, employee age, employee tenure, married, non-white, promoted, manager age, manager tenure, number of non-managerial employees in branch, percent female non-managerial employees in branch, branch deposits, and fixed effects for the state where the branch is located. The omitted category for non-white is "white'; for married is 'single'; for job title is 'teller"; for female is 'male"; and for manager female is "manager male". b Standard errors in parentheses c These models include a control for performance. Because performance is missing for 115 employees, these models are based on 742 employees. * p 0.05 ** p ** p 5 0.01 0.001 139 TABLE 6 Generalized Estimation Equation Regression Models Predicting Log Annual Base Salary and Log Annual Bonus Pay of Non-Managerial Employees by Organizational Positiona, Model la Variables Less Formalized Pay Formalized Pay (Base Salary) Female Model lb Model 2ac b Model 2b' Less Formalized Pay Formalized Pay (Bonus Pay) (Base Salary) -0.038* 0.588 (0.017) (0.413) (Base Salary) -0.040* 0.173 -0.075** -0.191 (0.020) (0.442) (0.024) (0.613) Manager Female x Female Officer Executive Relationship Manager Rep x Female Officer x Female Exec x Female Relation x Female Model 3bc (Bonus Pay) Manager Female Representative Model 3a' Less Formalized Pay Formalized Pay (Bonus Pay) -0.045* -0.254 (0.023) (0.723) 0.063* 0.792 (0.027) (0.828) 0.142*** 0.796 0.130*** 0.499 0.127*** -0.220 (0.017) (0.575) (0.018) (0.607) (0.024) (0.852) 0.341*** 2.017*** 0.338*** 1.395** 0.322*** 1.154 (0.016) (0.413) (0.017) (0.441) (0.022) (0.614) 0.549*** 1.245* 0.549*** 0.231 0.527*** 0.185 (0.025) (0.598) (0.027) (0.591) (0.032) (0.780) 0.731*** 1.974*** 0.725*** 1.558** 0.724*** 1.368 (0.026) (0.546) (0.028) (0.559) (0.035) (0.744) -0.048* 0.277 -0.043* 0.457 -0.013 1.363 (0.019) (0.648) (0.021) (0.678) (0.028) (0.950) -0.043* -0.383 -0.054** -0.106 -0.023 0.279 (0.018) (0.480) (0.020) (0.506) (0.026) (0.701) -0.050 1.517 -0.058 2.133* -0.058 2.143 (0.035) (0.860) (0.037) (0.836) (0.049) (1.108) -0.094** -0.725 -0.127** -0.003 -0.082 0.588 (0.035) (0.767) (0.040) (0.812) (0.049) (1.033) Rep x Mgr Female Officer x Mgr Female Exec x Mgr Female Relation x Mgr Female Rep x Female x Mgr Female Officer x Female x Mgr Female Exec x Female x Mgr Female Relation x Female x Mgr Female -0.004 1.474 (0.033) (1.193) 0.019 0.523 (0.026) (0.846) 0.023 0.007 (0.045) (1.159) -0.020 0.300 (0.044) (1.065) -0.049 -1.903 (0.038) (1.340) -0.054 -0.875 (0.033) (0.985) 0.003 0.116 (0.068) (1.662) -0.098 -1.463 (0.064) (1.684) Individual -level controls Performance Weekly hours Employee age Employee tenure Married Non-White Promoted 0.014** 1.244*** 0.013** 1.244*** (0.005) (0.160) (0.005) (0.160) 0.014*** 0.090 0.013** 0.026 0.013** 0.025 (0.004) (0.096) (0.004) (0.094) (0.004) (0.094) 0.006*** -0.045*** 0.006*** -0.048*** (0.011) 0.006*** (0.001) -0.047*** (0.001) (0.011) (0.001) (0.011) 0.007*** 0.046** 0.007*** 0.018 0.007*** 0.019 (0.001) (0.017) (0.001) (0.017) (0.001) (0.017) 0.027* 0.475* 0.024* 0.431* 0.025* 0.436* (0.010) (0.193) (0.011) (0.194) (0.011) (0.194) -0.029** -0.024 -0.024* -0.082 -0.025* -0.089 (0.010) (0.191) (0.011) (0.191) (0.011) (0.191) 0.026*** 1.373*** 0.026*** 1.197*** 0.026*** 1.192*** (0.007) (0.350) (0.007) (0.352) (0.007) (0.352) 140 TABLE 6 (continued) Generalized Estimation Equation Regression Models Predicting Log Annual Base Salary and Log Annual Bonus Pay of Non-Managerial Employees by Organizational Positiona, Model Variables la Model lb Less Formalized Pay Formalized Pay (Base Salary) (Bonus Pay) Model 2a b Model 2bc Less Formalized Pay Formalized Pay (Base Salary) (Bonus Pay) Model 3a Model 3bc Less Formalized Pay Formalized Pay (Base Salary) (Bonus Pay) Manager -level controls Manager age Manager tenure 0.000 -0.012 0.000 -0.016 0.001 -0.018 (0.000) (0.010) (0.000) (0.011) (0.000) (0.011) 0.000 -0.017 0.000 -0.021 0.000 -0.021 (0.001) (0.015) (0.001) (0.014) (0.001) (0.015) Branch -level controls Branch size Percent female (%) Log of Branch deposits (in Constant Observations Number of GEID Time Periods Wald Chi2 DF 000s) 0.000 0.057* -0.000 0.091*** -0.000 0.093*** (0.001) (0.023) (0.001) (0.024) (0.001) (0.024) -0.045* -0.835- -0.046* -0.937 -0.045* -0.947 (0.019) (0.501) (0.021) (0.511) (0.021) (0.513) -0.007 0.121 -0.005 -0.157 -0.006 -0.125 (0.009) (0.186) (0.010) (0.189) (0.010) (0.190) 10.613*** 6.904 10.552*** 5.713 10.582*** 5.613 (0.198) (4.520) (0.220) (4.486) (0.220) (4.510) 2,139 2,139 1,821 1,821 1,821 1,821 897 897 742 742 742 742 4 4 4 4 4 4 3899.55*** 167.48*** 3121.15*** 240.39*** 3142.89*** 245.87*** 27 27 28 28 38 38 a All models include controls for scheduled weekly hours, employee age, employee tenure, married, non-white, promoted, manager age, manager tenure, number of nonb Standard errors in parentheses c These models include a control for performance. Because performance is missing for 115 employees, these models are based on 742 employees. *p! 0.05 0.01 * p 0.001 141 Table Al Generalized Estimation Equation Regression Model Predicting Log Annual Base Salary for Tellers Model Variables a, b la Less Formalized Pay (Base Salary) Female -0.104* Manager Female -0.131** Manager Female x Female 0.236*** (0.05) (0.04) (0.06) Individual -level controls 0.004 Performance (0.01) 0.063 Bachelors Degree (0.04) Weekly hours -0.031* Employee age -0.001 (0.01) (0.00) Employee tenure 0.031*** Married 0.107** (0.01) (0.04) 0.03 Non-White (0.05) Manager -level controls Manager age -0.001 Manager tenure -0.006 (0.00) (0.00) Branch -level controls Branch size -0.003 Percent female (%) -0.024 (0.00) (0.08) 0.026 Log of Branch deposits (in 000s) (0.03) Constant 8.575*** (0.72) Observations 69 Number of GEID 33 4 Time Periods Wald Chi2 104.88*** DF 21 models include controls for scheduled performance, education, weekly hours, employee age, employee tenure, married, non-white, manager age, manager tenure, number of non-managerial employees in branch, percent female non-managerial employees in branch, branch deposits, and fixed a All effects for the state where the branch is located. The omitted category for non-white is "white"; for married is "single"; for job title is "teller"; for female is "male"; and for manager female is "manager male". b Standard errors in parentheses * * p 0.05 ** p 0.01 p 0.001 142 5 5.1 CONCLUSION CONTRIBUTIONS This dissertation enhances our understanding of the effect of gender in evaluative processes by carefully elucidating when and how assessments of similar men and women vary systematically. Whereas prior research has largely focused on documenting that gender is often taken into account in evaluations, this dissertation shifts our focus to identifying the conditions under which this is the case and explicating the evaluative mechanisms perpetuating these observed gender differences. In addition to the specific contributions of the independent research papers, as detailed in each of the preceding chapters, this dissertation also makes four broader contributions to research on gender inequality and evaluative processes. First, by more carefully detailing when and how gender affects evaluative processes, this dissertation extends existing theory stemming from lab-based studies in the status characteristics tradition. These studies have shown that even when evaluators have performance information they sometimes base assessments on gender (e.g., Foschi, 1989). Despite documenting the persistence of gender in evaluations, these studies have neither identified scope conditions for when we might expect this to be the case nor uncovered the mechanisms leading to the inclusion of gender in the presence of performance information. Thus, this dissertation helps to fill this gap. Second, this dissertation demonstrates evaluators may use gender not only to deal with a problem of missing performance information, but also as a means for solving other types of information problems. Specifically, as detailed in Chapter 2, I show that evaluators only exhibit a male preference when the success of their evaluations is based on approval from a third-party contact. Similarly, in Chapter 3, we find that evaluators are most apt to use gender as a sorting heuristic when faced with increased search costs. These findings are consistent with the general argument that the inclusion of gender in evaluations is a means for remedying information inefficiencies, though existing research has focused on the use of gender to address missing performance or quality information. These results provide evidence that evaluators also incorporate 143 gender as a means for solving other types of information problems, namely to sort through too much information and to satisfy the unknown preferences or expectations of others. Though this behavior can still be classified as biased, since I do not find evidence that men are objectively outperforming women in my settings, my results suggest that it is not only overt bias that leads to unequal evaluations of men and women, but more subtle processes are also often at play. Third, this dissertation helps advance a burgeoning body of literature showing how evaluators' assessments are influenced by their relevant audience. I find that the evaluator's propensity to incorporate gender is also a function of whether they are the end-user who is evaluating for their own consumption, or acting as a mediator who is accountable to an audience. As detailed in Chapter 2, when an evaluator moves from a position as the end user to a position as a mediator they are more apt to base their assessments on the gender of their network contacts. I argue that this shift in how gender affects evaluations is because in the latter case the evaluator's assessment incorporates what they anticipate to be the preferences or expectations of their own evaluating audience, in this case their third-party contacts. This notion that evaluators' decisions are influenced by a relevant audience, or third-parties, is not entirely novel. For example, economists attribute labor market inequality partly to customer discrimination, where managers disproportionately hire employees who they believe their customers prefer (e.g. Neumark, Bank, and Van Nort, 1996). In terms of gender specifically, expectations of customer or client preferences for men over women has been posited to contribute to a male advantage in hiring (Fernandez-Mateo and King, 2011) and in promotion decisions (Beckman and Phillips, 2005). However, these studies have been unable to account for personal gender beliefs, or preferences, making it unclear to what extent these findings are actually attributable to the presence of third-party actors. By comparing how the same evaluators incorporate gender in their assessments both when they rely on approval from a third-party and when they do not, this dissertation more precisely identifies that the effect of gender in evaluative processes is related to the presence of third-parties. Fourth, this dissertation highlights the importance of accounting for characteristics of the evaluator more generally, and gender in particular. Specifically, Chapter 4 provides some initial 144 evidence that male and female evaluators may differ in their propensity to factor a candidate's gender into their assessments and enact organizational processes differently. I find evidence of less gender pay inequality among lower-level employees reporting to a female, as opposed to a male, manager. Thus, this dissertation suggests that who the evaluator is matters, such that it may be incorrect to assume that gender affects evaluations uniformly across different categories of evaluators. 5.2 FUTURE RESEARCH There are several fruitful paths for future research in further addressing the unifying research question from these projects. While the preceding chapters include recommendations for specific research directions stemming from each study, here I focus on two broader research directions following from this dissertation and my plans to advance these. First, this dissertation points to the importance of taking the perspective of the evaluator to identify whether the propensity to exhibit gender bias in evaluations differs based on characteristics of evaluator. With the exception of Chapter 4, this dissertation centers on identifying the extent to which there are differences in the evaluations received by similar men and women. While inherently these gender differences emerge because evaluators incorporate gender in their assessments, to sharpen our understanding of the precise mechanisms driving these differences, future research should more directly examine how these processes differ across evaluators, or categories of evaluators. For example, the core findings in Chapter 2 - that anticipatory third-party biases result in female entrepreneurs receiving fewer third-party referrals when they are in male-dominated occupations - raises the question of whether this anticipatory bias is reinforced by both male and female referrers (e.g. the evaluators). Using data on the characteristics of the evaluator, I plan to examine whether the propensity to base decisions about resource sharing on gender varies by the resource-holder's own gender, the gender of their third-party, or the interaction of these. Second, though this dissertation suggests that receiving lower assessments leads to a female disadvantage, these studies are not able to causally link observed gender differences in evaluations to 145 substantive outcomes. As a result, the ultimate implications of gender differences in evaluations for gender inequality remain unclear. For example, the findings in Chapter 2 on referrals among entrepreneurs raise the following question: How much lower are revenues for female entrepreneurs as a result of receiving fewer third-party referrals? Similarly, the results in Chapter 3, that female investment professionals receive less attention, raises the following question: Does receiving less attention impact the propensity of women sharing recommendations in the future? The logical next step in uncovering the effect of gender in evaluative processes is to identify how observed gender differences in assessments translate to substantive differences in ultimate outcomes for men and women. To begin to address this gap, I plan to use data on revenue generated from referrals to assess the economic implications of receiving fewer third-party referrals for women. Linking gender differences in evaluations to inequality in tangible outcomes for men and women will help clarify the extent to which evaluative processes actually contribute to the persistence of gender inequality. Despite gains in the relative position of women, inequality persists. In order to devise solutions to remediate these inequities, it is necessary to understand the precise sources of these gender differences. By examining when and how evaluation processes, in particular, contribute to unequal outcomes for similar men and women, this dissertation begins to make headway in this direction and elucidates paths for deepening our understanding of this timely and important issue. 146 6 REFERENCES Acker, J. 1990. Hierarchies, Jobs, Bodies: A Theory of Gendered Organizations. Gender and Society, 4(2): 139-158. Adut, A. 2005. A Theory of Scandal: Victorians, Homosexuality, and the Fall of Oscar Wilde. American Journal of Sociology, 111(1): 213-248. Albrecht, J., Bj6rklund, A., & Vroman, S. 2003. Is There a Glass Ceiling in Sweden? Journal of Labor Economics, 21(1): 145-177. Alden. 2013. Returns at Hedge Funds Run by Women Beat the Industry, Report Says. DealBook. http://dealbook.nytimes.com/2013/01/10/returns-at-hedge-funds-run-by-women-beat-theindustry-report-says/. Aldrich, H., Elam, A., & Reese, P. 1997. Strong Ties, Weak Ties, and Strangers: Do Women Business Owners Differ from Men in Their Use of Networking to Obtain Assistance? In S. Birley & I. Macmillan (Eds.), Entrepreneurship in a Global Context 1-25. London; New York: Routledge. Aldrich, H., Reese, P. R., & Dubini, P. 1989. Women on the verge of a breakthrough:networking among entrepreneurs in the United States and Italy. Entrepreneurship & Regional Development, 1(4): 339-356. Altonji, J. G., & Blank, R. M. 1999. Chapter 48 Race and gender in the labor market. In Orley C. Ashenfelter and David Card (Ed.), Handbook of Labor Economics, vol. Volume 3, Part C: 3143-3259. Elsevier. Anderson, C., & Tomaskovic-Devey, D. 1995. Patriarchal Pressures - An Exploration Organizational Processes That Exacerbate and Erode Gender Earnings Inequality. of Work and Occupations, 22(3): 328-356. Arrow, K. 1971. Some Models of Racial Discrimination in the Labor Market. http://www.rand.org/pubs/research-memoranda/RM6253.html. Ashforth, B. E., & Mael, F. 1989. Social Identity Theory and the Organization. Academy of Management Review, 14(1): 20-39. Babcock, L., & Laschever, S. 2003. Women Don't Ask: Negotiation and the Gender Divide. Princeton, N.J: Princeton University Press. Bajtelsmit, V., & VanDerhei, J. 1997. Risk Aversion and Pension Investment Choices. In M. S. Gordon, 0. S. Mitchell, & M. M. Twinney (Eds.), Positioning Pensions for the TwentyFirst Century. University of Pennsylvania Press. 147 Baldiga, K. 2013. Gender Differences in Willingness to Guess. Management Science, 131105054351001. Barber, B. M., & Odean, T. 2001. Boys Will be Boys: Gender, Overconfidence, and Common Stock Investment. The Quarterly Journal of Economics, 116(1): 261-292. Barbulescu, R., & Bidwell, M. 2013. Do Women Choose Different Jobs from Men? Mechanisms of Application Segregation in the Market for Managerial Workers. Organization Science, 24(3): 737-756. Baron, J. N., & Bielby, W. T. 1980. Bringing the Firms Back in: Stratification, Segmentation, and the Organization of Work. American Sociological Review, 45(5): 737-765. Baron, J. N., Davis-Blake, A., & Bielby, W. T. 1986. The Structure of Opportunity: How Promotion Ladders Vary Within and Among Organizations. Administrative Science Quarterly, 31(2): 248-273. Baron, J. N., Hannan, M. T., Hsu, G., & Kogak, 0. 2007. In the Company of Women Gender Inequality and the Logic of Bureaucracy in Start-Up Firms. Work and Occupations, 34(1): 35-66. Baron, J. N., Mittman, B. S., & Newman, A. E. 1991. Targets of Opportunity: Organizational and Environmental Determinants of Gender Integration within the California Civil Service, 19791985. American Journal of Sociology, 96(6): 1362-1401. Baron, J. N., & Newman, A. E. 1990. For What It's Worth: Organizations, Occupations, and the Value of Work Done by Women and Nonwhites. American Sociological Review, 55(2): 155-175. Becker, G. S. 1957. The Economics of Discrimination. Chicago: University of Chicago Press. Beckman, C. M., & Phillips, D. J. 2005. Interorganizational Determinants of Promotion: Client Leadership and the Attainment of Women Attorneys. American Sociological Review, 70(4): 678-701. Beckmann, D., & Menkhoff, L. 2008. Will Women Be Women? Analyzing the Gender Difference among Financial Experts. Hannover Economic Papers (HEP) no. dp-391, Leibniz Universitdt Hannover, Wirtschaftswissenschaftliche Fakultdt, https://ideas.repec.org/p/han/dpaper/dp-391.html. Berger, J. 1977. Status characteristics and social interaction: an expectation-states approach. Elsevier Scientific Pub. Co. Berger, J., Fisek, H. F., Norman, R. Z., & Zelditch, M. 1977. Status characteristics and social interaction: an expectation-states approach. Elsevier Scientific Pub. Co. 148 Berger, J., Rosenholtz, S. J., & Zelditch, M., Jr. 1980. Status Organizing Processes. Annual Review of Sociology, 6: 479-508. Bernard, H. R., Kilworth, P. D., & Sailer, L. 1981. Summary of research on informant accuracy in network data and the reverse small world problem. Connections, 4(2): 11 - 25. Bertrand, M., & Mullainathan, S. 2004. Are Emily and Greg More Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination. The American Economic Review, 94(4): 991-1013. Bettencourt, B. A., Charlton, K., Dorr, N., & Hume, D. L. 2001. Status differences and in-group bias: A meta-analytic examination of the effects of status stability, status legitimacy, and group permeability. Psychological Bulletin, 127(4): 520-542. Beyer, S. 1990. Gender differences in the accuracy of self-evaluations of performance. Journal of Personality and Social Psychology, 59(5): 960-970. Beyer, S., & Bowden, E. M. 1997. Gender Differences in Seff-Perceptions: Convergent Evidence from Three Measures of Accuracy and Bias. Personality and Social Psychology Bulletin, 23(2): 157-172. Bielby, W. T. 2000. Minimizing Workplace Gender and Racial Bias. Contemporary Sociology, 29(1): 120-129. Bielby, W. T., & Baron, J. N. 1986. Men and Women at Work: Sex Segregation and Statistical Discrimination. American Journal of Sociology, 91(4): 759-799. Bird, S. R., & Sapp, S. G. 2004. Understanding the Gender Gap in Small Business Success: Urban and Rural Comparisons. Gender and Society, 18(1): 5-28. Booth, A., & Leigh, A. 2010. Do employers discriminate by gender? A field experiment in femaledominated occupations. Economics Letters, 107(2): 236-238. Braddock, J. H., & McPartland, J. M. 1987. How Minorities Continue to Be Excluded from Equal Employment Opportunities: Research on Labor Market and Institutional Barriers. Journal of Social Issues, 43(1): 5-39. Brass, D. J. 1985. Men's and Women's Networks: A Study of Interaction Patterns and Influence in an Organization. The Academy of Management Journal, 28(2): 327-343. Bridges, W. P., & Nelson, R. L. 1989. Markets in Hierarchies: Organizational and Market Influences on Gender Inequality in a State Pay System. The American Journal of Sociology, 95(3): 616-658. 149 Briscoe, F., & Kellogg, K. C. 2011. The Initial Assignment Effect Local Employer Practices and Positive Career Outcomes for Work-Family Program Users. American Sociological Review, 76(2): 291-319. Bromiley, P., & Curley, S. P. 1992. Individual differences in risk taking. Risk-taking behavior 87132. Oxford, England: John Wiley & Sons. Burt, R. 1995. Structural Holes: The Social Structure of Competition. Massachusetts: Harvard University Press. Cambrdige, Burt, R. 2007. Secondhand Brokerage: Evidence on the Importance of Local Structure for Managers, Bankers, and Analysts. Academy of Management Journal, 50(1): 119-148. Buttner, E. H., & Rosen, B. 1988. Bank loan officers' perceptions of the characteristics of men, women, and successful entrepreneurs. Journal of Business Venturing, 3(3): 249-258. Campbell, K. E. 1988. Gender Differences in Job-Related Networks. Work and Occupations, 15(2): 179-200. Castilla, E. 2008. Gender, Race, and Meritocracy in Organizational Careers. The American Journal of Sociology, 113(6): 1479-1526. Castilla, E. J. 2008. Gender, Race, and Meritocracy in Organizational Careers. American Journal of Sociology, 113(6): 1479-1526. - Castilla, E. J. 2011. Bringing Managers Back In. American Sociological Review, 76(5): 667 694. Castilla, E. J., & Benard, S. 2010. The Paradox of Meritocracy in Organizations. Administrative Science Quarterly, 55(4): 543-576. Cejka, M., & Eagly, A. 1999. Gender-stereotypic images of occupations correspond to the sex segregation of employment. Personality and Social Psychology Bulletin, 25(4): 413- 423. Charles, K. K., & Guryan, J. 2008. Prejudice and Wages: An Empirical Assessment of Becker's The Economics of Discrimination. Journal of Political Economy, 116(5): 773-809. Chattopadhyay, P., Tluchowska, M., & George, E. 2004. Identifying the Ingroup: A Closer Look at the Influence of Demographic Dissimilarity on Employee Social Identity. The Academy of Management Review, 29(2): 180-202. Cha, Y., & Weeden, K. A. 2014. Overwork and the Slow Convergence in the Gender Gap in Wages. American Sociological Review, 79(3): 457-484. 150 Clark, C. R., Clark, S., & Polborn, M. K. 2006. Coordination and status influence. Rationality and Society, 18(3): 367-391. Cohen, P. N., & Huffman, M. L. 2007. Working for the Woman? Female Managers and the Gender Wage Gap. American Sociological Review, 72(5): 681-704. Coleman, J. S. 1988. Social Capital in the Creation of Human Capital. The American Journal of Sociology, 94: S95-S120. Correll, S. J., & Benard, S. 2006. Biased estimators? Comparing status and statistical theories of gender discrimination. Advances in Group Processes, 23: 89-116. Correll, S. J., Benard, S., & Paik, I. 2007. Getting a Job: Is There a Motherhood Penalty? The American Journal of Sociology, 112(5): 1297-1338. Correll, S., & Ridgeway, C. 2003. Expectation States Theory. Handbook of social psychology 29 - 51. New York: Kluwer Academic/Plenum. Cotter, D. A., DeFiore, J., Hermsen, J. M., Kowalewski, B. M., & Vanneman, R. 1997. All Women Benefit: The Macro-Level Effect of Occupational Integration on Gender Earnings Equality. American Sociological Review, 62(5): 714-734. Davison, H. K., & Burke, M. J. 2000. Sex Discrimination in Simulated Employment Contexts: A Meta-analytic Investigation. Journal of Vocational Behavior, 56(2): 225-248. Deaux, K., & Major, B. 1987. Putting gender into context: An interactive model of gender-related behavior. Psychological Review, 94(3): 369-389. Dencker, J. C. 2008. Corporate Restructuring and Sex Differences in Managerial Promotion. American Sociological Review, 73(3): 455-476. Derks, B., Laar, C. V., Ellemers, N., & Groot, K. de. 2011. Gender-Bias Primes Elicit Queen-Bee Responses Among Senior Policewomen. Psychological Science, 22(10): 1243-1249. Ding, W. W., Murray, F., & Stuart, T. E. 2006. Gender Differences in Patenting in the Academic Life Sciences. Science, 313(5787): 665-667. Dobbin, F., & Kelly, E. 2007. How to stop harassment: Professional construction of legal compliance in organizations. American Journal of Sociology, 112(4): 1203-1243. Dobbin, F., Sutton, J., Meyer, J., & Scott, W. 1993. Equal-opportunity Law and the Construction of Internal Labor Markets. American Journal of Sociology, 99(2): 396-427. Drentea, P. 1998. Consequences of Women's Formal and Informal Job Search Methods for Employment in Female-Dominated Jobs. Gender and Society, 12(3): 321-338. 151 Duguid, M. M., Loyd, D. L., & Tolbert, P. S. 2012. The Impact of Categorical Status, Numeric Representation, and Work Group Prestige on Preference for Demographically Similar Others: A Value Threat Approach. Organization Science, 23(2): 386-401. Eagly, A. H. 1987. Sex differences in social behavior: a social-role interpretation. Psychology Press. Eagly, A. H. 2004. Prejudice: Toward a More Inclusive Understanding. In A. H. Eagly, R. M. Baron, & V. L. Hamilton (Eds.), The social psychology of group identity and social conflict: Theory, application, and practice- 45-64. Washington, DC, US: American Psychological Association. Eagly, A. H., & Karau, S. J. 2002. Role congruity theory of prejudice toward female leaders. Psychological Review, 109(3): 573-598. Eckel, C. C., & Grossman, P. J. 2008. Men, Women and Risk Aversion: Experimental Evidence- 1061-1073. Handbook of Experimental Economics Results, Elsevier. Elliott, J. R., & Smith, R. A. 2004. Race, Gender, and Workplace Power. American Sociological Review, 69(3): 365-386. Elvira, M. M., & Graham, M. E. 2002. Not Just a Formality: Pay System Formalization and SexRelated Earnings Effects. Organization Science, 13(6): 601-617. Ely, R. J. 1994. The Effects of Organizational Demographics and Social Identity on Relationships among Professional Women. Administrative Science Quarterly, 39(2): 203-238. Ely, R. J. 1995. The Power in Demography: Women's Social Constructions of Gender Identity at Work. The Academy of Management Journal, 38(3): 589-634. Emerson, R. M. 1983. Holistic Effects in Social Control Decision-Making. Law & Society Review, 17(3): 425-455. Emirbayer, M., & Goodwin, J. 1994. Network Analysis, Culture, and the Problem of Agency. American Journal of Sociology, 99(6): 1411-1454. England, P. 1992. Comparable Worth: Theories and Evidence. Aldine Transaction. England, P., Herbert, M. S., Kilbourne, B. S., Reid, L. L., & Megdal, L. M. 1994. The Gendered Valuation of Occupations and Skills: Earnings in 1980 Census Occupations. Social Forces, 73(1): 65-100. Fernandez-Mateo, I. 2009. Cumulative Gender Disadvantage in Contract Employment. American Journal of Sociology, 114(4): 871-923. 152 Fernandez-Mateo, I., & King, Z. 2011. Anticipatory Sorting and Gender Segregation in Temporary Employment. Management Science, 57(6): 989-1008. Fernandez, R., Castilla, E., & Moore, P. 2000. Social capital at work: Networks and employment at a phone center. American Journal of Sociology, 105(5): 1288-1356. Fernandez, R. M., & Fernandez-Mateo, 1. 2006. Networks, Sociological Review, 71(1): 42-71. Race, and Hiring. American Fernandez, R. M., & Sosa, M. L. 2005. Gendering the Job: Networks and Recruitment at a Call Center. The American Journal of Sociology, 111(3): 859-904. Fernandez, R. M., & Weinberg, N. 1997. Sifting and Sorting: Personal Contacts and Hiring in a Retail Bank. American Sociological Review, 62(6): 883-902. Financial Times. 2013. Global MBA Ranking. Technical report, London, U.K.: Financial Times Ltd. Finkelstein, S., & Hambrick, D. C. 1990. Top-Management-Team Tenure and Organizational Outcomes: The Moderating Role of Managerial Discretion. Administrative Science Quarterly, 35(3): 484-503. Foschi, M. 1989. Status Characteristics, Standards, and Attribution. In J. Berger, M. Zelditch, & B. Anderson (Eds.), Sociological theories in progress: New Formulations. 58 - 72. Sage Publications. Foschi, M. 1996. Double standards in the evaluation of men and women. Social Psychology Quarterly, 59(3): 237-254. Foschi, M. 2000. Double Standards for Competence: Theory and Research. Annual Review of Sociology, 26(1): 21-42. Foschi, M. 2009. Gender, performance level, and competence standards in task groups. Social Science Research, 38(2): 447-457. Foschi, M., & Foddy, M. 1988. Standards, performances, and the formation of self-other expectations. In M. Webster & M. Foschi (Eds.), Status generalization: New theory and research: 248-260. Stanford University Press: Stanford University Press. Foschi, M., Lai, L., & Sigerson, K. 1994. Gender and Double Standards in the Assessment of Job Applicants. Social Psychology Quarterly, 57(4): 326-339. Foschi, M., & Lapointe, V. 2002. On Conditional Hypotheses and Gender as a Status Characteristic. Social Psychology Quarterly, 65(2): 146-162. Foschi, M., Sigerson, K., & Lembesis, M. 1995. Assessing Job Applicants - the Relative Effects of Gender, Academic Record, and Decision Type. Small Group Research, 26(3): 328-352. 153 Foschi, M., & Valenzuela, J. 2008. Selecting job applicants: Effects from gender, self-presentation, and decision type. Social Science Research, 37(3): 1022-1038. Foschi, M., & Valenzuela, J. 2012. Who is the better applicant? Effects from gender, academic record, and type of decision. Social Science Research, 41(4): 949-964. Friedman, D. 1994. Experimental Methods: A Primer for Economists. Cambridge University Press. Gensch, D. H. 1987. A Two-Stage Disaggregate Attribute Choice Model. Marketing Science, 6(3): 223-239. Gerhart, B., & Milkovich, G. T. 1990. Organizational Differences in Managerial Compensation and Financial Performance. Academy of Management Journal, 33(4): 663-691. Gibbons, R. 1998. Incentives in Organizations. The Journal of Economic Perspectives, 12(4): 115-132. Glick, P. 1991. Trait-based and sex-based discrimination in occupational prestige, occupational salary, and hiring. Sex Roles, 25(5-6): 351-378. Goldin, C., & Rouse, C. 2000. Orchestrating Impartiality: The Impact of "Blind" Auditions on Female Musicians. The American Economic Review, 90(4): 715-741. Gomez-Mejia, L. R., & Balkin, D. B. 1992. Compensation, Organizational Strategy, and Firm Performance. South-Western Publishing Company, College Division. Goode, W. J. 1979. The Celebration of Heroes: Prestige As a Social Control System. University of California Press. Goodwin, S. A., Operario, D., & Fiske, S. T. 1998. Situational Power and Interpersonal Dominance Facilitate Bias and Inequality. Journal of Social Issues, 54(4): 677-698. Gorman, E. H. 2005. Gender Stereotypes, Same-Gender Preferences, and Organizational Variation in the Hiring of Women: Evidence from Law Firms. American Sociological Review, 70(4): 702-728. Granovetter, M. S. 1995. Getting a job: a study of contacts and careers. University of Chicago Press. Gulati, R., & Srivastava, S. n.d. Bringing Agency Back into Network Research: Constrained Agency and Network Action. In D. Brass, G. Labianca, A. Mehra, D. Halgin, & S. Borgatti (Eds.), Research in the Sociology of Organizations- 73 - 93. Emerald Group Publishing Ltd. 154 Gupta, V. K., Turban, D. B., Wasti, S. A., & Sikdar, A. 2009. The Role of Gender Stereotypes in Perceptions of Entrepreneurs and Intentions to Become an Entrepreneur. Entrepreneurship Theory and Practice, 33(2): 397-417. Halaby, C. N. 2003. Where Job Values Come from: Family and Schooling Background, Cognitive Ability, and Gender. American Sociological Review, 68(2): 251-278. Hallen, B. L. 2008. The Causes and Consequences of the Initial Network Positions of New Organizations: From Whom Do Entrepreneurs Receive Investments? Administrative Science Quarterly, 53(4): 685-718. Hanson, S., & Pratt, G. 1991. Job search and the occupational segregation of women. Annals of the Association of American Geographers, 81(2): 229. Hiubl, G., & Trifts, V. 2000. Consumer Decision Making in Online Shopping Environments: The Effects of Interactive Decision Aids. Marketing Science, 19(1): 4-21. Heilman, M. E. 1980. The impact of situational factors on personnel decisions concerning women: Varying the sex composition of the applicant pool. Organizational Behavior and Human Performance, 26(3): 386-395. Heilman, M. E. 1983. Sex bias in work settings: The Lack of Fit model. Research in Organizational Behavior, 5: 269-298. Heilman, M. E., Block, C. J., & Martell, R. F. 1995. Sex stereotypes: Do they influence perceptions of managers? Journal of Social Behavior and Personality, 10(6): 237-252. Hewstone, M., Rubin, M., & Willis, H. 2002. Intergroup Bias. Annual Review of Psychology, 53(1): 575-604. Hinz, R. P., D. D. McCarthy, and J. A. Turner. 1997. Are Women Conservative Investors? Gender Differences in Participant Directed Pension Investments. Michael S. Gordon, Olivia S. Mitchell, and Marc M. Twinney, Eds.) Positioning Pensions for the Twenty-first Century, 91-103. University of Pennsylvania Press. Hirsch, B. T. 2012. Unions, dynamism, and economic performance. (C. L. Estlund & M. L. Wachter, Eds.)Research Handbook on the Economics of Labor and Employment Law, 107- 145. Hsu, D. H. 2004. What Do Entrepreneurs Pay for Venture Capital Affiliation? The Journal of Finance, 59(4): 1805-1844. Huffman, M. L., Cohen, P. N., & Pearlman, J. 2010. Engendering Change: Organizational Dynamics and Workplace Gender Desegregation, 1975-2005. Administrative Science Quarterly, 55(2): 255-277. 155 Huffman, M. L., & Velasco, S. C. 1997. When More is Less Sex Composition, Organizations, and Earnings in U.S. Firms. Work and Occupations, 24(2): 214-244. Hultin, M., & Szulkin, R. 1999. Wages and Unequal Access to Organizational Power: An Empirical Test of Gender Discrimination. Administrative Science Quarterly, 44(3): 453-472. Hultin, M., & Szulkin, R. 2003. Mechanisms of Inequality. Unequal Access to Organizational Power and the Gender Wage Gap. European Sociological Review, 19(2): 143-159. Hundley, G. 2001. Why Women Earn Less Than Men in Self-Employment. Journal of Labor Research, 22(4): 817-829. Hurlbert, J. S., Haines, V. A., & Beggs, J. J. 2000. Core Networks and Tie Activation: What Kinds of Routine Networks Allocate Resources in Nonroutine Situations? American Sociological Review, 65(4): 598-618. Ibarra, H. 1992. Homophily and Differential Returns: Sex Differences in Network Structure and Access in an Advertising Firm. Administrative Science Quarterly, 37(3): 422-447. Ibarra, H. 1993. Personal Networks of Women and Minorities in Management: A Conceptual Framework. The Academy of Management Review, 18(1): 56-87. Ibarra, H. 1997. Paving an Alternative Route: Gender Differences in Managerial Networks. Social Psychology Quarterly, 60(1): 91-102. Ibarra, H., Kilduff, M., & Tsai, W. 2005. Zooming in and out: Connecting Individuals and Collectivities at the Frontiers of Organizational Network Research. Organization Science, 16(4): 359-371. Jasso, G., & Jr., M. W. 1999. Assessing the Gender Gap in Just Earnings and Its Underlying Mechanisms. Social Psychology Quarterly, 62(4): 367-380. Jensen, M. 2006. Should We Stay or Should We Go? Accountability, Status Anxiety, and Client Defections. Administrative Science Quarterly, 51(1): 97-128. Jianakoplos, N. A., & Bernasek, A. 1998. Are Women More Risk Averse? Economic Inquiry, 36(4): 620-630. Joshi, A., Hui Liao, & Jackson, S. E. 2006. Cross-Level Effects of Workplace Diversity on Sales Performance and Pay. Academy of Management Journal, 49(3): 459-481. Jost, J. T., & Banaji, M. R. 1994. The role of stereotyping in system-justification and the production of false consciousness. British Journal of Social Psychology, 33(1): 1-27. 156 Kalev, A., Kelly, E., & Dobbin, F. 2006. Best Practices or Best Guesses? Assessing the Efficacy of Corporate Affirmative Action and Diversity Policies. American Sociological Review, 71(4): 589-617. Kalleberg, A. L., Knoke, D., Marsden, D. P. V., & Spaeth, J. L. 1996. Organizations in America: Analysing Their Structures and Human Resource Practices. Sage Publications, Inc. Kanter. 1977. Men and Women of the Corporatn. Basic Books. Katz, J., & Williams, P. 1997. Gender, self-employment and weak-tie networking through formal organizations. Entrepreneurship & Regional Development, 9(3): 183-198. Kilbourne, B., England, P., & Beron, K. 1994. Effects of Individual, Occupational, and Industrial Characteristics on Earnings: Intersections of Race and Gender. Social Forces, 72(4): 11491176. Killewald, A., & Gough, M. 2013. Does Specialization Explain Marriage Penalties and Premiums? American Sociological Review, 78(3): 477-502. Kleinbaum, A. M. 2012. Organizational Misfits and the Origins of Brokerage in Intrafirm Networks. Administrative Science Quarterly, 57(3): 407-452. Kleinbaum, A. M., Stuart, T. E., & Tushman, M. L. 2013. Discretion Within Constraint: Homophily and Structure in a Formal Organization. Organization Science, 24(5): 1316-1336. Kmec, J. A. 2008. The Process of Sex Segregation in a Gender-Typed Field: The Case of Male Nurses. Sociological Perspectives, 51(2): 259-279. Konrad, A. M., & Linnehan, F. 1995. Formalized HRM Structures: Coordinating Equal Employment Opportunity or Concealing Organizational Practices? The Academy of Management Journal, 38(3): 787-820. Kovacs, B., & Sharkey, A. J. 2014. The Paradox of Publicity How Awards Can Negatively Affect the Evaluation of Quality. Administrative Science Quarterly, 59(1): 1-33. Kulis, S. 1997. Gender Segregation among College and University Employees. Sociology of Education, 70(2): 151-173. Lerner, J. S., & Tetlock, P. E. 1999. Accounting for the effects of accountability. Psychological Bulletin, 125(2): 255-275. Lincoln, J. R., & Miller, J. 1979. Work and Friendship Ties in Organizations: A Comparative Analysis of Relation Networks. Administrative Science Quarterly, 24(2): 181-199. Lin, N. 2001. Social capital: a theory of social structure and action. Cambridge University Press. 157 Lorber, J. 1984. Women physicians: careers, status, and power. New York: Tavistock Publications. Loscocco, K. A., Robinson, J., Hall, R. H., & Allen, J. K. 1991. Gender and Small Business Success: An Inquiry into Women's Relative Disadvantage. Social Forces, 70(1): 65-85. Marin, A. 2012. Don't mention it: Why people don't share job information, when they do, and why it matters. Social Networks, 34(2): 181-192. Marsden, P. V. 1987. Core Discussion Networks of Americans. American Sociological Review, 52(1): 122-131. McDonald, S. 2011. What's in the "old boys" network? Accessing social capital in gendered and racialized networks. Social Networks, 33(4): 317-330. McGuire, G. M. 2000. Gender, Race, Ethnicity, and Networks The Factors Affecting the Status of Employees' Network Members. Work and Occupations, 27(4): 501-524. McGuire, G. M. 2002. Gender, Race, and the Shadow Structure: A Study of Informal Networks and Inequality in a Work Organization. Gender and Society, 16(3): 303-322. McPherson, M., Smith-Lovin, L., & Cook, J. M. 2001. Birds of a Feather: Homophily in Social Networks. Annual Review of Sociology, 27: 415-444. Mencken, F. C., & Winfield, I. 2000. Job search and sex segregation: does sex of social contact matter? Sex Roles, 42(9/10): 847-864. Merrett, C. D., & Gruidl, J. J. 2000. Small Business Ownership in Illinois: The Effect of Gender and Location on Entrepreneurial Success. Professional Geographer, 52(3): 425. Miller, A. S., & Hoffmann, J. P. 1995. Risk and Religion: An Explanation of Gender Differences in Religiosity. Journal for the Scientific Study of Religion, 34(1): 63-75. Moore, D. P., & Buttner, E. H. 1997. Women Entrepreneurs: Moving Beyond the Glass Ceiling (First Edition). SAGE Publications, Inc. Moore, G. 1990. Structural Determinants of Men's and Women's Personal Networks. American Sociological Review, 55(5): 726-735. Morrison, A. M., White, R. P., Velsor, E. V., & Leadership, C. for C. 1987. Breaking the Glass Ceiling: Can Women Reach the Top of America's Largest Corporations?. Addison Wesley Longman Publishing Co. Moss-Racusin, C. A., Dovidio, J. F., Brescoll, V. L., Graham, M. J., & Handelsman, J. 2012. Science faculty's subtle gender biases favor male students. Proceedings of the National Academy of Sciences of the United States of America, 109(41): 16474-16479. 158 Muchnik, L., Aral, S., & Taylor, S. J. 2013. Social Influence Bias: A Randomized Experiment. Science, 341(6146): 647-651. National Council for Research on Women. 2007. Women in Fund Management. June 3 2009: http://www.ncrw.org/reports-publications/women-fund-management-report. Accessed: August 17, 2014. Nelson, R. L., & Bridges, W. P. 1999. Legalizing Gender Inequality: Courts, Markets, and Unequal Pay for Women in America. Cambridge [England]: Cambridge University Press. Neumark, D., Bank, R. J., & Nort, K. D. V. 1996. Sex Discrimination in Restaurant Hiring: An Audit Study. The Quarterly Journal of Economics, 111(3): 915-941. Ng, C. W., & Chiu, W. C. K. 2001. Managing equal opportunities for women: sorting the friends from the foes. Human Resource Management Journal, 11(1): 75-88. Niessen, A., & Ruenzi, S. 2007. Sex matters: Gender differences in a professional setting. no. 06-01, CFR Working Paper, http://www.econstor.eu/handle/10419/57738. Okamoto, D. G., & Smith-Lovin, L. 2001. Changing the Subject: Gender, Status, and the Dynamics of Topic Change. American Sociological Review, 66(6): 852-873. Osterman, P. 1987. Choice of Employment Systems in Internal Labor Markets. Industrial Relations, 26(1): 46-67. Pager, D., Bonikowski, B., & Western, B. 2009. Discrimination in a Low-Wage Labor Market: A Field Experiment. American Sociological Review, 74(5): 777-799. Penner, A. M., Toro-Tulla, H. J., & Huffman, M. L. 2012. Do Women Managers Ameliorate Gender Differences in Wages? Evidence from a Large Grocery Retailer. Sociological Perspectives, 55(2): 365-381. Perry, E. L., Alison Davis-Blake, & Kulik, C. T. 1994. Explaining Gender-Based Selection Decisions: A Synthesis of Contextual and Cognitive Approaches. The Academy of Management Review, 19(4): 786-820. Petersen, T., & Morgan, L. A. 1995. Separate and Unequal: Occupation-Establishment Sex Segregation and the Gender Wage Gap. The American Journal of Sociology, 101(2): 329-365. Pfeffer, J. 1977. Toward an Examination of Stratification in Organizations. Administrative Science Quarterly, 22(4): 553-567. Pfeffer, J., & Cohen, Y. 1984. Determinants of Internal Labor Administrative Science Quarterly, 29(4): 550-572. 159 Markets in Organizations. Pfeffer, J., Davis-Blake, A., & Daniel J. Julius. 1995. AA Officer Salaries and Managerial Diversity: Efficiency Wages or Status? Industrial Relations, 34(1): 73-94. Phelps, E. S. 1972. The Statistical Theory of Racism and Sexism. The American Economic Review, 62(4): 659-661. Pinnington, A. H., & Sandberg, J. 2013. Lawyers' Professional Careers: Increasing Women's Inclusion in the Partnership of Law Firms. Gender, Work & Organization, 20(6): 616- 631. Podolny, J. M. 1993. A Status-Based Model of Market Competition. The American Journal of Sociology, 98(4): 829-872. Podolny, J. M. 1994. Market Uncertainty and the Social Character of Economic Exchange. Administrative Science Quarterly, 39(3): 458-483. Podolny, J. M. 2005. Status Signals: A Sociological Study of Market Competition. Princeton University Press, http://books.google.com/books?id=7-W8vgADb5YC. Podolny, J. M., & Baron, J. N. 1997. Resources and Relationships: Social Networks and Mobility in the Workplace. American Sociological Review, 62(5): 673-693. Powell, G. N., & Butterfield, D. A. 1994. Investigating the "Glass Ceiling" Phenomenon: An Empirical Study of Actual Promotions to Top Management. The Academy of Management Journal, 37(1): 68-86. Quintane, E., & Kleinbaum, A. M. 2011. Matter Over Mind? E-mail Data and the Measurement of Social Networks. Connections, 31(1): 20-41. & Ransom, M., & Oaxaca, R. L. 2005. Intrafirm Mobility and Sex Differences in Pay. Industrial Labor Relations Review, 58(2): 219-237. Renzulli, L. A., Aldrich, H., & Moody, J. 2000. Family Matters: Gender, Networks, and Entrepreneurial Outcomes. Social Forces, 79(2): 523-546. Reskin, B. 2003. Including mechanisms in our models of ascriptive inequality. American Sociological Review, 68(1): 1-21. Reskin, B. F. 1988. Bringing the Men Back in: Sex Differentiation and the Devaluation of Women's Work. Gender and Society, 2(1): 58-81. Reskin, B. F. 2000. The Proximate Causes of Employment Discrimination. Contemporary Sociology, 29(2): 319-328. Reskin, B. F., & McBrier, D. B. 2000. Why Not Ascription? Organizations' Employment of Male and Female Managers. American Sociological Review, 65(2): 210-233. 160 Reskin, B. F. . R., Patricia A., & Roos, P. A. 1990. Job Queues, Gender Queues: Explaining Women's Inroads into Male Occupations. Temple University Press. Riach, P. A., & Rich, J. 2006. An Experimental Investigation of Sexual Discrimination in Hiring in the English Labor Market. in Economic Analysis & Policy, 5(2). http://www.degruyter.com.libproxy.mit.edu/view/j/bejeap.2006.5.2/bejeap.2006.5.2.1416/be Advances jeap.2006.5.2.1416.xml?format=INT. Ridgeway, C. 1991. The Social Construction of Status Value: Gender and Other Nominal Characteristics. Social Forces, 70(2): 367-386. Ridgeway, C. 1993. Gender, Status, and the Social Psychology of Expectations. In P. England (Ed.), Theory on gender/feminism on theory 175-1198. Transaction Publishers. Ridgeway, C. L. 1997. Interaction and the Conservation of Gender Inequality: Considering Employment. American Sociological Review, 62(2): 218-235. Ridgeway, C. L. 2011. .f-amed by Gender: How Gender Inequality Persists in the Modern World. Oxford University Press. Ridgeway, C. L. 2014. Why Status Matters for Inequality. American Sociological Review, 79(1): 1-16. Ridgeway, C. L., & Correll, S. J. 2000. Limiting Inequality through Interaction: The End(s) of Gender. Contemporary Sociology, 29(1): 110-120. Ridgeway, C. L., & Correll, S. J. 2004. Unpacking the Gender System: A Theoretical Perspective on Gender Beliefs and Social Relations. Gender and Society, 18(4): 510-531. Ridgeway, C. L., & Correll, S. J. 2006. Consensus and the Creation of Status Beliefs. Social Forces, 85(1): 431-453. Ridgeway, C. L., & Smith-Lovin, L. 1999. The Gender System and Interaction. Annual Review of Sociology, 25: 191-216. Robb, A., & Coleman, S. 2009. Characteristics of New Firms: A Comparison by Gender. SSRN Scholarly Paper no. ID 1352601, Rochester, NY: Social Science Research Network, http://papers.ssrn.com/abstract=1352601. Rosenfeld, R. A. 2002. What Do We Learn about Difference from the Scholarship on Gender? Social Forces, 81(1): 1-24. Rubineau, B., & Fernandez, R. M. 2014. How do Labor Market Networks Work? In R. Scott & S. Kosslyn (Eds.), Emerging 1ends in the Social and Behavioral Sciences. Hoboken, NJ: John Wiley and Sons. 161 Rudman, L. A., & Glick, P. 1999. Feminized management and backlash toward agentic women: The hidden costs to women of a kinder, gentler image of middle managers. Journal of Personality and Social Psychology, 77(5): 1004-1010. Ruef, M., Aldrich, H. E., & Carter, N. M. 2003. The Structure of Founding Teams: Homophily, Strong Ties, and Isolation among U.S. Entrepreneurs. American Sociological Review, 68(2): 195-222. Salancik, G. R., & Pfeffer, J. 1978. A Social Information Processing Approach to Job Attitudes and Task Design. Administrative Science Quarterly, 23(2): 224-253. Salganik, M. J., Dodds, P. S., & Watts, D. J. 2006. Experimental Study of Inequality and Unpredictability in an Artificial Cultural Market. Science, 311(5762): 854-856. Simcoe, T. S., & Waguespack, D. M. 2011. Status, Quality, and Attention: What's in a (Missing) Name? Management Science, 57(2): 274-290. Smith, E. B., Menon, T., & Thompson, L. 2012. Status Differences in the Cognitive Activation of Social Networks. Organization Science, 23(1): 67-82. Smith, S. S. 2000. Mobilizing Social Resources: Race, Ethnic, and Gender Differences in Social Capital and Persisting Wage Inequalities. Sociological Quarterly, 41(4): 509-537. Smith, S. S. 2005. "Don't put my name on it": Social Capital Activation and Job-Finding Assistance among the Black Urban Poor. American Journal of Sociology, 111(1): 1-57. Sorensen, J. B. 2000. The Longitudinal Effects of Group Tenure Composition on Turnover. American Sociological Review, 65(2): 298-310. Sorensen, J. B., & Sharkey, A. J. 2014. Entrepreneurship as a Mobility Process. American Sociological Review, 79(2): 328-349. Spence, M. 1973. Job Market Signaling. The Quarterly Journal of Economics, 87(3): 355-374. Stewart, A. 1990. The Bigman Metaphor for Entrepreneurship: A "Library Tale" with Morals on Alternatives for Further Research. Organization Science, 1(2): 143-159. Stover, D. L. 1994. The Horizontal Distribution of Female Managers within Organizations. Work and Occupations, 21(4): 385-402. Stuart, T. E., Hoang, H., & Hybels, R. C. 1999. Interorganizational Endorsements and the Performance of Entrepreneurial Ventures. Administrative Science Quarterly, 44(2): 315349. Stuart, T. E., & Sorenson, 0. 2007. Strategic networks and entrepreneurial ventures. Strategic Entrepreneurship Journal, 1(3-4): 211-227. 162 Sunden, A. E., & Surette, B. J. 1998. Gender differences in the allocation of assets in retirement savings plans. American Economic Review, 88(2): 207-211. Sutton, J. R., Dobbin, F., Meyer, J. W., & Scott, W. R. 1994. The Legalization of the Workplace. American Journal of Sociology, 99(4): 944-971. Swim, J. K., & Sanna, L. J. 1996. He's skilled, she's lucky: A meta-analysis of observers' attributions for women's and men's successes and failures. Personality and Social Psychology Bulletin, 22(5): 507-519. Tajfel, H., & Turner, J. C. 1979. An integrative theory of intergroup conflict. In S. Worchel & W. Austin (Eds.), The social psychology of intergroup relations, vol. 33: 74. Monterey, CA: Brooks/Cole. Tetlock, P. 1992. The impact of accountability on judgment and choice: Toward a social contingency model. In M. P. Zanna (Ed.), Advances in Experimental Social Psychology, Volume 25 (1 edition): 331 - 376. San Diego; London: Emerald publishing. Tetlock, P. E. 1992. The Impact of Accountability on Judgment and Choice: Toward A Social Contingency Model. In Mark P. Zanna (Ed.), Advances in Experimental Social Psychology, vol. Volume 25: 331-376. Academic Press. Th6baud, S. 2010. Gender and Entrepreneurship as a Career Choice: Do Self-assessments of Ability Matter? Social Psychology Quarterly, 73(3): 288-304. Turco, C. J. 2010. Cultural Foundations of Tokenism: Evidence from the Leveraged Buyout Industry. American Sociological Review, 75(6): 894-913. U.S. Census Bureau. 1995, Genealogy Data. https://www.census.gov/genealogy/www/ data/1990surnames/names-files.html. Accessed: 2014-08-17. U.S. News and World Report. 2013a. Best Business Schools. Technical report, Washington, D.C.: U.S. News and World Report. U.S. News and World Report. 2013b. National University Rankings. Technical report, Washington, D.C.: U.S. News and World Report. Valian, V. 1999. Why So Slow? The Advancement of Women. The MIT Press. Wagner, D. G., & Berger, J. 1993. Status characteristics theory: The growth of a program. In J. Berger, M. Zelditch, & Jr. (Eds.), Theoretical research programs: Studies in the growth of theory 23-63. Stanford University Press: Stanford University Press. Weber, M. 1958. From Max Weber: Essays in Sociology. (H. H. Gerth & C. W. Mills, Eds.). Princeton, N.J.: Oxford University Press. 163 Webster, M., Jr., & Hysom, S. J. 1998. Creating Status Characteristics. American Sociological Review, 63(3): 351-378. Williams, C. L. 1992. The Glass Escalator: Hidden Advantages for Men in the "Female" Professions. Social Problems, 39(3): 253-267. Williams, J. 1999. Unbending Gender: Why Family and Work Conflict and What To Do About It. Oxford University Press. Zeng, Z., & Xie, Y. 2008. A Preference-Opportunity-Choice Framework with Applications to Intergroup Friendship. American Journal of Sociology, 114(3): 615-648. Zuckerman, E. W. 1999. The Categorical Imperative: Securities Analysts and the Illegitimacy Discount. American Journal of Sociology, 104(5): 1398-1438. Zuckerman, E. W. 2004. Structural Incoherence Sociological Review, 69(3): 405-432. and Stock Market Activity. American Zuckerman, E. W., Kim, T.-Y., Ukanwa, K., & James von Rittmann. 2003. Robust Identities or Nonentities? Typecasting in the Feature-Film Labor Market. The American Journal of Sociology, 108(5): 1018-1074. 164