Help Wanted: The Politicization of Job Loss in High-Unemployment Contexts Matthew B. Incantalupo1 Joint Degree Program in Politics and Social Policy Princeton University mincanta@princeton.edu This is a work in progress. Please do not cite or share without permission. Abstract: When Americans are laid off from work, who should be mainly responsible for helping them? I develop a new theory, Hardship-in-Context, which argues that citizens form political perceptions about their personal problems using relevant contextual information. Unemployed Americans perceive their hardship as individualized in low-unemployment contexts and as politicized in high-unemployment contexts. I examine several surveys over a period of time that includes both high and low unemployment and find support for this claim. When and where the unemployment rate is low, jobless Americans believe they are mainly responsible for helping themselves. When and where it is high, jobless Americans believe that government and employers are responsible for helping people who are laid off from work. Employed Americans do not display the same pattern of individualization in low-unemployment contexts and politicization in high unemployment contexts. I support these results using evidence from a recent longitudinal survey as well. Overall, I find strong evidence that the interaction of personal experiences and the broader context in which they occur helps to shape political attitudes. This paper is based on chapters 1, 2 and 3 of my dissertation project, “Help Wanted: Unemployment and the Politicization of Personal Economic Hardship.” I would like to thank Amy Lerman, Christopher Achen, Martin Gilens, Benjamin Bishin, Scott Abramson, Sarah Brayne, Kevin Collins, and Steve Rogers for helpful feedback and support. All remaining errors and omissions are my own. 1 1 When, in a city of 100,000, only one man is unemployed, that is his personal trouble, and for its relief we properly look to the character of the man, his skills, and his immediate opportunities. But when in a nation of 50 million employees, 15 million men are unemployed, that is an issue, and we may not hope to find its solution within the range of opportunities open to any one individual. The very structure of opportunities has collapsed. Both the correct statement of the problem and the range of possible solutions require us to consider the economic and political institutions of the society, and not merely the personal situation and character of a scatter of individuals. (Mills 1959, 9) In this paper, I advance and test the claim that individuals in high-unemployment contexts politicize the personal hardship of unemployment and perceive it as a socially-centered political problem.2 I argue that the effects of personal experience with job loss on how Americans perceive unemployment vary as a function of the broader economic context, specifically labor market conditions. This implies that out-of-work citizens view unemployment as a personal problem in low-unemployment contexts and as a political problem in highunemployment contexts. When unemployment becomes politicized, jobless Americans shift their foci of expectations away from themselves and look to government for solutions and assistance (Brody and Sniderman 1977). I test this implication using a series of surveys merged with statelevel unemployment data. To exploit as much variation in contextual factors as possible, I examine multiple samples that utilize the same survey instrument over an eight-year period that spans both good and bad economic times to understand how individuals’ perceptions of unemployment vary as a function of the broader context in which they experience this form of economic hardship. I also support these findings using a recent panel survey of Americans who lost their jobs in the Great Recession. I find support for my theory; unemployed Americans in low-unemployment contexts are likely to report that unemployed workers are responsible for helping themselves while 2 High-unemployment contexts refer to circumstances or places when and where the unemployment rate is high. Alternatively, I define low-unemployment contexts to refer to circumstances or places when and where the unemployment rate is low and near “full” employment. 2 unemployed individuals in high-unemployment contexts believe that government and employers should help individuals who lose their jobs. Furthermore, unemployed individuals in high unemployment contexts are unlikely to believe that the unemployed should be mainly responsible for helping themselves. The politicization and deindividualization of unemployment as a function of local economic context is unique to jobless Americans; employed respondents do not exhibit this pattern of responses to the same extent as out-of-work respondents. This is strong evidence that unemployment is best understood in context, particularly with respect to how it can shape political beliefs and ultimately contribute to mobilization and political participation (Incantalupo 2012a). Background In general, Americans do not expect government to help them deal with their personal problems and hardships. “Citizens are likely to think the government has some responsibility to help only if the type of problem of most concern to them is manifestly beyond the capacity of any one person to deal with all by himself. Otherwise, they are overwhelmingly likely to insist that it is up to them to cope with the problem on their own” (Brody and Sniderman 1977, 340). This rugged individualism and the belief that individuals are personally responsible for their economic successes and failures shape Americans’ attitudes towards economic and social policy (Free and Cantril 1968; Huber and Form 1973; Kluegal and Smith 1986). Americans rely primarily on themselves to cope with the personal hardships that they face and expect government to take on socially-located problems that are perceived as not amenable to individual influence, such as monetary inflation or natural disasters (Sniderman and Brody 1977). Building off of this personal-social dichotomy, we can consider a continuum that ranges from fully self-located to fully socially-located on which Americans place their personal 3 concerns (Brody and Sniderman 1977). Socially-located concerns become politicized and affect political attitudes as well as influence participation and voting. Self-located concerns do not link up to political behavior, and may in fact inhibit participating in politics or paying attention to current events because they are distracting. For example, job loss and financial strain place significant psychological (Schlozman and Verba 1979), familial (Feather 1989; Schlozman and Verba 1979), and even health burdens (Warr 1987) on those who experience them, which make Americans less likely to think about or participate in politics. Writing about unemployment, Rosenstone concludes, “When a person experiences economic adversity his scarce resources are spend on holding body and soul together – surviving – not on remote concerns like politics” (1982, 26). I begin with this personal-social continuum in developing a new theory of how Americans politicize their personal hardships. However, existing theories argue that citizens believe government has a responsibility to help if the type of problem that most concerns them is one that they cannot manage on their own (Brody and Sniderman 1977; Sniderman and Brody 1977). This implies that citizens locate problems along the personal-social continuum based on their content, which is problematic because it requires too much from citizens, most of whom spend little time thinking about politics and more time worrying about their problems. Americans do not possess a great deal of political information (Delli Carpini and Keeter 1996 and may lack the necessary schemas to connect what is troubling them to politics (Lodge and Hamill 1986). Citizens often punish incumbents for things politicians cannot control or influence so long as they can connect their problems to government through a “folk story” (Achen and Bartels 2002). They can also be happy with government despite poor performance based on recent good 4 economic performance (Bartels 2008) or even more banal events like sports victories (Healy, Malhotra, and Mo 2010). In contrast, I argue that individuals locate their personal problems along this continuum based on the broader context in which they experience these hardships. By incorporating contextual information into the process of addressing and coping with their personal problems, Americans determine how widespread these problems are, the extent to which they are amenable to individual influence, and the degree to which they should shift their foci of expectations away from themselves and towards government. In this way, two citizens who are experiencing the same hardship can reach two different conclusions about whether their problems should be addressed by individual or government action if they experience them in disparate social and economic contexts. Considering the role of context in determining how Americans respond to personal hardship allows us to reconcile the fact that problems that seem very personal by their content sometimes lead to political action. Having or caring for someone with a serious illness would be characterized as an individually-centered problem at face value (Brody and Sniderman 1977; Sniderman and Brody 1977). However, state and local health departments receive over 1,000 requests to investigate suspected cancer clusters each year (Thun and Sinks 2004). Actual cancer clusters are very rare, and most suspected cancer clusters are simply unfortunate coincidences reported by cancer patients or their loved ones who believe that they have noticed an unusual pattern of cancer incidences (Trumbo 2000; Robinson 2002). But, in the proper context, a 5 hardship that usually would not be politicized, a cancer diagnosis, contributes to political action that in some cases can continue for years.3 The same can be said for numerous other hardships. A foreclosure sale on a home is a self-located concern, but when hundreds of homes in a single county are facing foreclosure, citizens will perceive their own foreclosure as part of a socially-located problem. Organizations like the Foreclosure Working Group, an offshoot of Occupy Greensboro, boast hundreds of members experiencing foreclosure who are now engaging in political action and protests against major banks. As of June 2012, Guilford County, which contains Greensboro, had the fourthhighest foreclosure rate in North Carolina, and borders the county with the second-highest rate in the state.4 And just as we observe “Occupy Homes” groups springing up in response to high foreclosure rates, we observe “99ers” groups forming to advocate for extended unemployment benefits to jobless Americans who have exhausted their unemployment compensation after a maximum of 99 weeks. The proliferation of jobless advocacy groups following the most recent economic recession indicates that unemployment can meaningfully affect political behavior, at least if unemployment is sufficiently high and widespread. This is in contrast to Schlozman and Verba (1979), which finds that the unemployed are largely disorganized, in part due to a lack of class consciousness in the United States, but does not consider how the effects of unemployment may vary depending on the social and economic contexts in which Americans experience it. Economic context seems to affect attitudinal measures as well. The belief that hard work pays 3 A confirmed cancer cluster at Marine Corps Base Camp Lejeune has been the subject of over a decade of investigation, recently concluding in federal legislation to provide medical care for dozens of military personnel and their family members who have developed cancer from contaminated drinking water. Ordonez, Franco, and Barbara Barrett, (McClatchy), "Obama Signs Law Giving Health Care To Lejeune TaintedWater Victims", Raleigh News & Observer, 7 August 2012 4 These data are available at http://www.npr.org/templates/story/story.php?storyId=111494514 and come from www.realtytrac.com, which bills itself as “the most trusted source of foreclosure information.” 6 off, an important pillar of American political culture, increases following periods of economic growth and falls during recessions (Pew Research Center 2012). More locally, mass layoffs positively associate with voter turnout at the county level (Healy 2009; Margalit 2011); In the Canadian case, neighborhood-level unemployment negatively associates with support for free trade (Cutler 2007). Thus, there exists ample evidence in the existing literature that Americans’ attitudes related to work and unemployment can be shaped by broader economic conditions. I argue that personal experience with unemployment is perceived as a self-located problem that is best remedied through individual action in low-unemployment contexts. When relatively few Americans are unemployed, the hardship of unemployment is individualized and jobless Americans are less likely to believe that the government should be responsible for helping them.5 When and where the economy is struggling and the unemployment rate is high, unemployed Americans are more likely to perceive unemployment as a social problem and believe that government is responsible for helping them deal with it. I define individuals’ perceptions of unemployment as politicized to the extent that they believe that government bears responsibility for helping people who are unemployed. Citizens’ perceptions of unemployment are individualized to the extent that they believe that the unemployed are primarily responsible for helping themselves. I expect gainfully employed Americans to be more likely to perceive unemployment as individualized than as politicized in high and low-unemployment contexts. I expect unemployed Americans’ attitudes about unemployment to associate with unemployment context in the following manner: in lowunemployment contexts, unemployed Americans should individualize the hardships associated 5 That is not to say that the unemployed do not collect unemployment benefits from the government. Unemployment benefits, temporary relief for laid-off workers, are generally less maligned than more controversial safety net programs such Food Stamps (Gilens 1999). 7 with being out of work. In high-unemployment contexts, unemployed Americans should politicize this experience and believe that government should step in and help. Cross-Sectional Evidence I make use of multiple editions of the Work Trends Poll, a survey of labor force participants administered by the John J. Heldrich Center for Workforce Development at Rutgers University. The advantages of using this poll are its focus on members of the labor force and emphasis on public attitudes related to work, employers, the government, and economic issues. The polls feature attitudinal measures related to work not included on surveys more familiar to political scientists, such as the American National Election Studies. Unfortunately, relying upon these polls has its disadvantages as well. The polls do not contain many measures that typically accompany attitudes about the government and the economy, such as a measure of political ideology. Still, I believe these are the best available data to test if perceptions of who is responsible for helping the unemployed are shaped through the interaction between personal experience with unemployment and the broader context in which unemployment takes place. I use the following survey question from the Work Trends Poll series to measure the politicization of job loss: “When people are laid off from work, who should be mainly responsible for helping them? Is it government, employers, or workers themselves?” This is a very strong measure of the politicization of unemployment for several reasons. First, it asks a specific question about people who lose their jobs without adding additional information about deservingness or the circumstances under which they are laid off from work. This is a significant contrast from a measure such as asking respondents to agree or disagree with a statement like “It is the responsibility of government to take care of people who cannot take care of themselves,” which removes any ambiguity about individual efficacy and agency in solving a personal 8 hardship. Additionally, it forces respondents to choose who they feel are mainly responsible for helping individuals who lose their jobs, as opposed to asking them the extent to which they believe government, employers, or workers themselves should help the unemployed as individual measures. Given Americans’ complicated beliefs about individualism and government’s role in the economy, it would not be surprising to find a large proportion of Americans believes that both government and individuals are responsible for dealing with the problem of unemployment simultaneously if asked about their roles separately. This question wording allows for a very straightforward test that unemployment becomes a politicized problem when experienced in a high-unemployment context. Unemployed individuals in highunemployment contexts should be likely to believe that government is mainly responsible for helping the jobless.6 This question is repeated using the same wording in six cross-sectional Work Trends Polls, spanning from June 2003 through July 2011.7 I present some basic information about the polls used in this analysis in Table 1. [Table 1 here] Table 1 serves as a useful summary of the sample sizes for each Work Trends Poll, as well as a summary of the economic context during each poll’s fielding. Perhaps the most glaring detail contained in Table 1 is the fact that the national unemployment rate increased by four points in just one year from May 2008 to May 2009. For each poll in this analysis, I also indicate the 6 In some administrations of the Work Trends Poll, respondents can voluntarily offer more than one choice (usually coded as “a combination”) or say that all three are equally responsible for helping people who are laid off. I remove these individuals from my analysis, as I am unable to determine who they believe should help the unemployed. 7 I include a discussion of some earlier Work Trends Polls with alternative question wordings in Appendix A to this paper. While the evidence in Appendix A cannot be treated as conclusive, it suggests that question wording does significantly affect who Americans believe should be responsible for helping the unemployed. One alternative question wording mentions workers who are laid off “through no fault of their own.” In this wording, I observe an increase in the belief that employers should be responsible for helping the unemployed. Another alternative wording mentions workers who are laid off in times of economic downturn. Here, I observe an increase in the belief that government should be responsible for helping the unemployed. 9 highest and lowest state unemployment rates. Particularly in the July 2010 poll, we observe considerable variation in unemployment at the state level. As a simple first step, I plot respondents’ answers to my dependent variable, which asks who should be responsible for helping people who are laid off from work. I disaggregate the responses by employment status and poll to allow us to see if my expectations hold up under a very simple and descriptive analysis and present the results in Fig. 1. For each poll, I perform a simple chi-squared test to see if the pattern of responses is different between employed and unemployed respondents. [Fig. 1 here] This is good early evidence in favor of Hardship-in-Context. Over the course of the 6 polls, we observe unemployed respondents becoming more likely to say that government should be responsible for helping the unemployed and considerably less likely to say that the unemployed should help themselves. We observe a similar pattern but to a lesser extent among the employed. Even in July 2010, with the economy continuing to suffer under very high unemployment, employed respondents’ modal response is that people who are laid off from work should be responsible for helping themselves. In 4 of the 6 polls (June 2003, May 2008, May 2009, and July 2010), the chi-squared test is statistically significant at the 90% level or better, indicating that the pattern of responses we observe between employed and unemployed respondents is unexpected if employment status has no effect on this outcome and the responses we observe are due to chance alone. Now that we see that this pattern holds in the most basic sense, we can incorporate additional rigor into this analysis. 10 Explaining Politicization I operationalize personal experience with unemployment using respondents’ stated employment statuses. As Work Trends Polls sample only those in the labor force, I code employment status using a binary variable that indicates whether a respondent is unemployed (out of work and currently looking for work) or employed. I include several demographic controls in my analysis as well, including an indicator variable for female respondents, who tend to be more favorable towards economic redistribution and social welfare policies than men are (Gilens 1999). I include another binary indicator for married respondents, control for respondents’ age in years, and include a factor variable for a respondents’ race or ethnicity. I indicate if respondents identify as black, Hispanic or Latino, or some other racial or ethnic group, with white respondents set aside as an excluded category. I control for educational attainment using a 4-point scale (no high school diploma, high school diploma, some college, college degree or more). This is especially important because of the disparity in unemployment between Americans with low educational attainment and Americans with high educational attainment. Members of the labor force who lack college or high school degrees are significantly more likely to be unemployed than workers who hold degrees.8 While controlling for educational attainment by no means erases our concerns over selection bias into who experiences unemployment, it does help to adjust for one of the demographic differences between the employed and unemployed. 8 For more information, see the graph of unemployment rates by education attainment in Appendix A to this paper. Generally speaking, the ratio of unemployed Americans with college degrees to unemployed Americans without college degrees is fairly stable at the national level. From 1992, when unemployment rates by educational attainment are first available, to 2012, the ratio of the unemployment rate for people without college degrees to the unemployment rate for people with college degrees ranges between a minimum of 1.63 in December, 2001 and a maximum of 2.56 in August, 2006. The average ratio over roughly 20 years of monthly observations is 2.05. 11 I also control for respondents’ household income. Because income is measured differently across Work Trends Polls, I am forced to use a somewhat imprecise measure of it in my analysis. Respondents are coded on a three-point scale, corresponding to low (less than $30,000 per year), medium ($30,000 – $75,000 per year), and high (more than $75,000 per year). I include a separate category for respondents who refuse to report their household incomes.9 Economic context Local contextual information is an important determinant of public opinion (Cutler 2007). I expect unemployed Americans in high-unemployment contexts to perceive job loss as a socially-centered problem that requires government help and to be unlikely to perceive it as a personal problem that should be fixed through individual action. I use a very straightforward measure to operationalize local economic context: the seasonally-adjusted unemployment rate in a respondent’s state of residence at the time when he or she is interviewed as part of a Work Trends Poll.10 This metric reflects the economic circumstances in which a jobless individual must cope with the problem of his or her unemployment and seek out a remedy for his or her personal hardship. I summarize this measure of economic context for each year in my sample using boxand-whisker plots in Figure 2. Since the purpose of this graph is to describe local economic 9 I present a summary of the basic demographic information about the respondents in my sample in the Appendix A to this paper. 10 All unemployment rates are from the Bureau of Labor Statistics. It is not possible to use a more local measure of unemployment, such as county or metro area-level unemployment. Work Trends Polls do not record these geographical measures. Furthermore, unemployment rates at these very local levels are measured with more error than at the state level. I have also considered using alternative measures of economic context. The Bureau of Labor Statistics estimates and publishes alternative measures of the unemployment rate, including one that counts “discouraged” workers and people who report working part time for economic reasons among the unemployed. This measure is known as the U6 unemployment rate and is regarded by some as the “true” measure of unemployment, because it includes additional individuals who are experiencing economic hardship. However, it does not contribute additional information beyond what is contained in the more commonly-cited U3 unemployment rate. Monthly observations of the U3 and U6 at the national level from January 1992 through the July 2013 have a correlation coefficient of .995. Using the U6 instead of the U3 does not change any of the results in this paper in a meaningful way. Additionally, using mass layoffs data from the BLS in place of the U3 does not change any of the results in this paper. 12 context during each administration of the Work Trends Poll that I use, my unit of analysis in Fig. 2 is the state. Therefore, each boxplot is generated using 51 observations, one for each state and the District of Columbia.11 [Fig. 2 here] The important takeaway from the information presented in Fig. 2 is that we observe considerable variation in economic context both across states at a given point in time and over time as the economy experiences growth and recession. A central contribution of this project is to incorporate this variation into the study of how unemployment affects political behavior. Both the information presented in Fig. 2 (and the right half of Table 1) make it abundantly clear that some states, such as North Dakota and South Dakota, can enjoy relatively low levels of unemployment while other states experience a devastating shortage of jobs, like we observe in Nevada in 2010. This additional information would be obscured and unused if we look only at the national unemployment rate. Before I explain the statistical model I will use to test if this variation in economic context across both time and space affects the relationship between employment status and attitudes about who should help people who lose their jobs, I first need to discuss how I incorporate party identification into this framework. Party Identification The Work Trends Polls do not provide ideal measures of party identification. In five of the six polls, respondents are asked if they identify as Democrats, Republicans, or Independents. Unfortunately, the polls do not measure the strength of these partisan attachments. Even more unfortunately, respondents who identify as Independents or express no partisan attachment are not asked whether they lean closer to one of the two major political parties following the first 11 Since Work Trends Polls use nationally-representative samples, the distribution of economic context at the individual level in the sample I use for my analysis is roughly the same but slightly more cluttered, so I choose to present the graph that is easier for the reader to comprehend. 13 question about partisanship. For these polls, I simply code whether respondents identify as Democrats or Republicans and leave Independents and those without a party preference as an excluded category. Most problematically, the July 2010 Work Trends Poll does not contain a measure of party identification. This means that I cannot simply include it in any analysis in which I would like to control for the effects of partisanship on citizens’ beliefs about who should help the unemployed or would just have to ignore the effects of partisanship. Since Americans’ political views are heavily influenced by their partisan identifications (Campbell, Converse, Miller, and Stokes 1960), I believe that this would introduce damaging omitted variable bias into my analyses. To avoid ignoring the July 2010 poll or ignoring the effects of partisanship, I employ two solutions. First, I use multiple imputation to estimate partisan identification for July 2010 respondents using the measured partisanship of respondents in other Work Trends Polls, and demographic measures that bridge across surveys. Because the July 2010 poll neglects to ask about party identification, I can treat party identification as missing at random, as respondents did not choose to skip or refuse to answer the question. I perform multiple imputation twenty times, which allows me to incorporate a great deal of uncertainty into this measure of party identification. In this way, it is a very conservative solution to my missing data problem. In order to validate my use of imputation and to avoid relying entirely upon simulated data in my analysis, I employ an additional solution to this missing data problem. I proxy for party identification by using a question on the June 2010 poll that I believe reflects respondents’ partisan loyalties without directly asking about partisanship. The poll asks “Who do you trust to do a better job handling the economy?” Answer choices are “President Obama,” “Republicans in 14 Congress,” and a third category I created by combining the “both” and “neither” responses. Respondents were coded as Democrats if they chose President Obama, Republicans if they chose Republicans in Congress, and Independents if they chose both or neither. While this is not a perfect measure of partisanship, it is very plausible to assert that this question will be heavily influenced by partisan attitudes.12 I report the cross-tabulations of this proxy measure by employment status in Table 2. [Table 2 here.] Two important things jump out from the data presented in Table 2. First, the chi-squared statistic is highly significant, indicating that employed and unemployed respondents significantly differ in who they trust to handle the economy. Unemployed respondents are considerably more likely than employed respondents to trust Pres. Obama and only one in eight unemployed respondents trust Republicans in Congress more to handle the economy. Second, more than half of all respondents, both employed and unemployed, answer both or neither to this economic trust question. This almost certainly overstates the true number of Independents in the sample. Individuals who offer one of the partisan responses are more likely to be extreme partisans than weak partisans, which means that this measure of party identification is probably overdetermined by the strength of a respondents’ true partisan attachments. This is not very problematic, since I also have an imputed measure of partisanship. Taken together, the “true” partisanship of the sample is probably somewhere in between the imputed and proxy measures. Still, I prefer to use a pair of imperfect measures of partisanship in order to avoid having to either disregard observations in my sample or disregard the effects of party on attitudes about who should help the unemployed. 12 I validated this measure as part of a survey conducted in June 2013 for another project by asking the proxy question alongside the standard party identification questions. In that sample, the proxy measure associates strongly with partisan self-identification. The details of this validation are contained in Appendix C to this paper. 15 Cross-sectional Model Because my outcome variable of interest -- a respondent’s belief about who is responsible for helping people who are laid off from work -- is categorical, I estimate a multinomial logit model.13 This is the most appropriate model for evaluating the predictions of Hardship-inContext, because it will allow me to estimate the probability that a respondent will offer each of the three responses and then see if these estimated probabilities vary as a function of employment status and local unemployment context.14 I estimate these probabilities as functions of several factors: The first is employment status, measured using an indicator variable for if the respondent is unemployed. Since Work Trends Polls only include individuals in the labor force in their samples, the excluded category for this variable is composed entirely of people who are employed. Therefore, the coefficient attached to this indicator variable is the effect of being unemployed on the probability that a respondent chooses one option over the baseline option. Second, I include economic context, measured using the unemployment rate (U3) in a respondent’s state at the time of his or her interview. Third, I include the interaction between personal experience with unemployment and unemployment context. This allows for separate effects of unemployment context for employed and unemployed respondents. Hardship-in-Context implies that the coefficient attached to the interaction between unemployed status and the local unemployment rate, should be positive for the probability that respondents choose the categorical outcome that government should help jobless Americans. Fourth, I include several demographic controls (age, race, gender, marital 13 The choices of the government, employers, and workers themselves represent three discrete options that do not logically fall along an ordered scale. 14 Recall that I exclude respondents who volunteer responses that a combination or none of the three choices should help the unemployed from this analysis. 16 status, education, and income). Fifth, I include one of two measures of party identification.15 The first is imputed using demographic characteristics that bridge across Work Trends Polls, and the second is the proxy measure that I outlined above. Finally, I include fixed effects for each individual poll and for each individual state. The poll effects help to account for unmeasured factors unique to the period of time in which each Work Trends Poll conducted its interviews, such as national economic conditions. The state fixed effects help to control for state-level factors that might affect responses, such as statewide ideology or the administration of unemployment benefits by state governments. I present the results of this analysis in Table 3 in the next section. Cross-sectional Results The results to the multinomial logit analysis are presented in Table 3. I estimate the model once using my imputed measure of party identification and a second time using my proxy measure for party identification. For both estimations, the first column of coefficients represents the change in the odds of the dependent variable in favor of the choice of “government” over the baseline choice of “workers themselves” associated with a one-unit change in each explanatory variable. The second column represents the change in favor of “employers” over “workers themselves” and the third column represents the change in favor of “government” over “employers.”16 [Table 3 here] In the first run of the model (with party identification imputed), only two of my key explanatory variables attain an accepted level of statistical significance. The first is the effect of 15 Recall that I use party identification as measured for every poll except for the July 2010 poll, which is the only poll that did not include a question about party identification. 16 To consider one of these choices in reverse, like the change in favor of “employers” over “government,” for example, simply multiply the coefficients by -1. 17 being unemployed on the probability of choosing “employers” over “workers,” which is negative and significant at the 90% level. The other is the coefficient on the interaction between personal experience with unemployment and state economic context. This coefficient is positive and significant at the 95% level. Before addressing the substantive significance of these results, I will first look to see if they are robust to switching to my alternative measure of party identification, the proxy measure. The right-hand side of Table 3 presents the estimates for an identical model but using the proxy measure of partisanship. Importantly, the coefficients in this model are fairly similar to the coefficients for the model that uses multiply-imputed party identification. It is important to note that the coefficients attached to party identification in this model are on average larger and estimated with less uncertainty than the coefficients in the imputed model. This is likely due to attenuation bias in the first run of the model owing to the presence of measurement error in multiple imputations of party affiliation, and possibly due to the fact that the proxy measure of party identification picks up on more intense partisan attitudes. In sum, these two measures of partisanship seem to be free of major issues, and switching from one to the other does not drastically change our estimates. That said, we do see more of the coefficients attached to our variables of interest attain statistical significance in the model that uses the proxy measure of party identification for July 2010 respondents. The interaction of personal unemployment with the state unemployment rate is positive and significant at the 90% level for the choice of “government” over a baseline of “workers themselves.” As in the previous run of the model, we see the coefficients for unemployed status and unemployment context are negative and significant and positive and significant, respectively, for the choice of “employers” over a baseline category of “workers 18 themselves.” Finally, we see that personal experience with unemployment has a positive and weakly significant effect for the choice of “government” over a baseline of “workers themselves.” These effects persist in the presence of demographic controls, controls for partisanship, and state and poll fixed effects. Multinomial logit coefficients are fairly ungainly, so I present a few graphs to better explain the results contained in Table 3. Figure 3 displays the predicted probability that a respondent will say that government should be responsible for helping people who are laid off from work. I plot this separately for employed and unemployed respondents across the observed values of local economic context, measured as the seasonally-adjusted unemployment rate in a respondent’s state at the time of his or her interview.17 [Fig. 3 here] While there is considerable overlap in the confidence intervals surrounding the two lines, it is still possible to see that the predicted probability of choosing “government” when asked who should help people who are laid off from work increases with the unemployment rate among unemployed respondents and slightly decreases with the unemployment rate among gainfully employed respondents.18 Thus, we have evidence that suggests that employed and unemployed citizens react to economic context in different ways. While I do not graph these quantities, I compute the marginal effects of unemployment on the probability of choosing government and find that it has a significant and positive effects ranging from about .051 to .066 between unemployment rates of 5 and 10 percent. When and where the unemployment rate is high, unemployed respondents are more likely than employed respondents are to believe that government should step in and help the jobless. 17 All other variables are held at their means. I use the multinomial logit model with the proxy measure of party identification to generate the figures in this section. 18 19 I plot the predicted probability that employed and unemployed respondents choose “employers” as a function of unemployment context in Fig. 4. [Fig. 4 here] In this instance, we see strong support for Hardship-in-Context. Employed respondents are virtually nonresponsive to the unemployment rate, barely more likely to respond that employers should help the unemployed in high-unemployment contexts than they are in low-unemployment contexts. On the other hand, we see strong contextual effects among the unemployed. In lowunemployment contexts, unemployed respondents are less likely than employed respondents to offer “employers” as a response, with a predicted probability below 0.2. This relationship is reversed in high-unemployment contexts, in which unemployed respondents are more likely than employed respondents to believe that employers have a responsibility to help people who are laid off. This is entirely due to an increase in the predicted probability of choosing employers among the unemployed, which approaches 0.4 in high-unemployment contexts. Additionally, the marginal effect of unemployment on the predicted probability of saying employers are responsible to help the unemployed (not graphed) is negative and significant in lowunemployment contexts and positive and significant in high-unemployment contexts. I plot the predicted probability of saying that laid off workers are responsible for helping themselves as a function of employment status and unemployment context in Fig. 5. [Fig. 5 here] The pattern of results visible in Fig. 5 is a reversal of what we observe in Fig. 4. In this instance, unemployed respondents are slightly more likely than employed respondents to say that workers themselves are responsible for helping themselves in low-unemployment contexts but are significantly less likely to offer the same response in high-unemployment contexts. When and 20 where the unemployment rate is high, employed respondents are more likely than jobless respondents to believe that unemployed Americans need to provide their own assistance. Again, we observe no sensitivity to economic context among employed Americans, but we do see that unemployed Americans’ responses are meaningfully influenced by economic context. Figs. 3-5 present results that support Hardship-in-Context. The experience of unemployment is individualized and self-located in low-unemployment contexts, approximately at the level of full employment. As outlined and hypothesized by Hardship-in-Context, individuals who lose their jobs under moderate to high levels of unemployment perceive their hardship as deindividualized and politicized and look to government to provide them with some assistance. While about half of unemployed respondents consider endorse an individual-based remedy to unemployment in low-unemployment contexts, this proportion falls to below one-fifth under the blisteringly high unemployment states like Nevada experienced in the aftermath of the 2007 economic recession. Longitudinal Evidence Having found support for Hardship-in-Context using cross-sectional data, I turn to the Work Trends Panel survey, a 4-wave longitudinal study of individuals who reported losing their jobs between August 2008 and August 2009. This means that everyone in the sample experienced unemployment during the Great Recession. This is important because it reduces the possibility that any effects that we observe are due to selection into who is employed and who is unemployed at a given point in time. The entire sample for this panel is drawn from a common pool of individuals who have experienced unemployment during the same period of nationwide economic contraction. 21 Some of the respondents remained unemployed while other respondents found new jobs or left the labor force by the start of wave 1 in August 2009. As before, I consider only those respondents who are either employed or unemployed and looking for work, so I exclude individuals who are not in the labor force from my analysis.19 I summarize the number of employed and unemployed respondents across all 4 waves in Table 4. [Table 4 here] A high proportion of the sample, nearly seven in ten respondents, is unemployed at wave 1, administered in August 2009. This proportion declines to roughly one in three respondents by wave 4, administered two years later. Of course, some of this is due to panel attrition, as nearly half of the sample has left either the survey or the labor force by wave 4. Table B1 in the appendix to this paper reports the transition probabilities from one employment state to another across the four survey waves. Once again, the dependent variable is measured using the question “When people are laid off from work, who should be mainly responsible for helping them?” As before, the choices are “government,” “employers,” or “workers themselves.” I report the percentage of the sample that gave each response in Table 4, and graph these percentages by employment status in Fig. 6. Just as in Fig. 1, I also report chi-squared statistics. [Fig. 6 here] Most interestingly, the proportion of unemployed respondents who believe that employers should be responsible for helping people who are laid off increases over time, nearly approaching the proportion of jobless respondents who believe that government should be responsible for helping by wave 4. While the pattern of responses for employed respondents 19 I only exclude them while they are not in the labor force. If respondents reenter the labor force, I include them in the analysis for the survey waves during which they are in the labor force. 22 stays fairly similar across survey waves, we observe some shifting among the unemployed. Table B2 in the appendix reports transition properties on this dependent variable measure across the 4 survey waves. Longitudinal Model Omitted variable bias is one potential concern that might arise from the cross-sectional analysis presented in the previous section. It is possible that factors that we do not observe, such as the circumstances under which an individual loses his or her job, or his or her attachment to that job before losing it, might affect the extent to which an unemployed American views his or her job loss as a personal or a social problem. Fortunately, I can utilize the longitudinal design of the Work Trends Panel to reduce potential concerns about omitted variable bias. I do this by employing an individual fixed-effects model, which accounts for factors that are unique to respondents in my sample that do not change over time. Therefore, I am able to estimate beliefs about who should be responsible for helping the unemployed as a function of measures that change over time, namely employment status and unemployment context while controlling for any individual-level time-invariant characteristics, observed or unobserved. Unfortunately, this means I must abandon using a multinomial logit model, as implementing such a model with individual fixed-effects is currently not possible but is under development (Pforr 2011). I recode my dependent variable of interest into three binary variables that capture whether a respondent chose “government,” “employers,” “or workers themselves.”20 I run three separate fixed-effects regressions for each binary choice. Although my dependent variable is binary, I use OLS regression as it does not require me to discard observations that do not change across the 4 survey waves, which is necessary in a fixed-effects or conditional logit For example, I code respondents who say “government” as 1 for my “government” variable and code respondents who say “employers” or “workers themselves” as 0. I repeat this for separate variables for “employers” and for “workers themselves.” 20 23 model. My estimates are substantively unchanged if I switch to a fixed-effects logit model, so I choose to present the results of an OLS regression with individual fixed effects. I also include an indicator variable for each survey wave, which accounts for any wave-specific variation such as national economic conditions or political events such as electoral campaigns. The results for this fixed-effects model are presented in Table 5. Longitudinal results [Table 5 here.] In the fixed-effects model, which accounts for all individual-level variation that remains constant over time, we see support for Hardship-in-Context in that the belief that workers should be responsible for helping themselves decreases among the unemployed as we move from a relatively low-unemployment context to a high-unemployment context. While we do not see any significant results for the government or employers outcomes, this negative result is still noteworthy, because I have confirmed at least one of my findings from the cross-sectional analysis using a very strong test in the form of a fixed effects model on a unique survey sample that all experienced job loss during the same period of economic turmoil. In sum, we see evidence that Americans who experience the personal hardship of unemployment are more likely to believe that they should help themselves in low-unemployment contexts and less likely to believe that they should help themselves in high-unemployment contexts. In high-unemployment contexts, the jobless believe that government and employers should step in to help the unemployed. Using both cross-sectional and longitudinal data, I find evidence that unemployed Americans shift their foci of expectations away from themselves and towards government as a function of the context in which they experience job loss. In the next section, I discuss these findings in more detail. 24 Discussion and Conclusion In this paper, I provide a substantial amount of evidence that the effects of unemployment on attitudes about who should help the unemployed vary as a function of the severity of unemployment, measured locally at the state level. One potential counterargument to this paper and Hardship-in-Context more broadly is the claim that individuals who experience job loss in low-unemployment contexts are systematically different and therefore incomparable to individuals who lose their jobs during severe economic recessions. I do as much as I can to control for demographics and still observe significant differences between the unemployed and employed as a function of local unemployment rates. While I cannot fully rule it out in my crosssectional sample, I do not observe clear evidence that cohort selection into who is unemployed is driving the pattern of results we observe.21 Most importantly, my use of panel data and implementation of a fixed effects model allow me to identify changes in attitudes as a function of changes in employment status, unemployment context, and the interaction between the two while controlling for any time-invariant characteristics that are unique to the individual respondents. This includes demographics and any personal characteristics, psychological orientations, beliefs, or habits, insofar as they do not change over time. The sample design of the Work Trends Panel, composed entirely of individuals who recently lost their jobs, and the estimation techniques I employ help to minimize potentially confounding selection effects. 21 I also do not think we observe this pattern of results because of demographic imbalances in who is unemployed in low versus high-unemployment contexts. I report demographic characteristics across surveys in Appendix A to this paper and while they do vary at times, this variation is not remarkable and not likely to be producing the pattern of results we observe. I also include a graph of unemployment rates by educational attainment in the appendix to this paper in order to make the point that the composition of the unemployed stays roughly the same with respect to education over time. 25 Implications for additional research The main finding of this paper, that unemployment is viewed as a politicized problem and not as an individualized problem in high-unemployment contexts, provides an important link to other research in how personal experience with unemployment affects political attitudes and behaviors as a function of unemployment context. This finding joins a growing literature on the effects of unemployment on political attitudes. Paolino (2010) finds that long-term unemployment undermines support for individualist values in Americans. Additionally, new research finds that the economic “shock” of job loss increases support for unspecified social welfare spending (Margalit 2013). This effect is significantly larger among Republicans than it is among Democrats. My paper is similar to these recent studies in that it makes use of a longitudinal survey of the unemployed in order to reach conclusions about how job loss meaningfully affects public attitudes. However, my paper is novel in that it provides a new theory to explain the divergent effects of job loss on political attitudes, incorporates contextual factors into our understanding of how personal experiences shape political beliefs, and examines who Americans believe is responsible for helping the unemployed, not simply how much help Americans believe the unemployed should receive. It is also an example of how we can better understand how economic hardship influences political behavior through purposeful sampling and repeated surveys of understudied groups of Americans. In a separate paper (Incantalupo 2012a), I find that unemployment affects voter turnout as a function of broader economic conditions. Job loss and unemployment are a demobilizing influence on voter turnout in low-unemployment contexts and a mobilizing influence on turnout in high-unemployment contexts. This finding seems especially plausible in light of what we now know about individualized and politicized responses to job loss as a function of local 26 unemployment context. Taken together, these findings have important implications for research in political participation and economic voting. Despite a body of research that emphasizes “sociotropic” factors in structuring political behavior to a greater extent than “pocketbook” factors (Kinder and Kiewiet 1982; Mutz 1992), I find that individual and contextual factors can interact to affect public attitudes. In another paper (Incantalupo 2012b), I examine whether local unemployment context affects blame and attribution following job loss. I find that unemployed Americans are less warm towards immigrants (documented and undocumented), more likely to blame immigrants for their own job loss, and more likely to accuse them of taking jobs away from native-born Americans in high-unemployment contexts than they are when the economy is doing relatively well. This is another promising application of Hardship-in-Context and has implications for research into attitudes towards immigrants (Hainmueller and Hiscox 2010), as well as theories of group conflict and attribution theory in social psychology. In summary, Hardship-in-Context is a rich theory of how Americans perceive and react to the very serious personal hardship of job loss and provides a framework by which we can understand other hardships. Finally, the United States remains in a prolonged period of high unemployment. Understanding how unemployment (and other economic hardships that can become politicized in context, such as foreclosures) affects political behavior sheds additional light on a phenomenon that affects millions of citizens and could contribute to better policies for providing assistance or relief to out-of-work Americans. For example, state unemployment offices could tailor the kinds of unemployment benefits they give based on the broader state of the economy, which affects how unemployed Americans view their circumstances. States, aware that job loss in lowunemployment contexts is typically viewed as an unemployed person’s problem to fix on his or 27 her own, can divert resources to job training programs and other policies to help people who believe they are mainly responsible for helping themselves. During periods of high unemployment, governments could provide more direct relief in the forms of increased unemployment benefits or create government jobs as a solution to high unemployment that constituents perceive as a socially-centered problem that government should address. Elected officials who ignore the very probable reality that millions of unemployed Americans view their personal hardships as part of a much broader politicized problem that government is mainly responsible to remedy do so at their own risk, particularly since job loss in high-unemployment contexts has a mobilizing effect on an individual’s probability of turning out to vote (Incantalupo 2012a). In this paper, I set out to examine whether citizens politicize the personal hardship of unemployment. The answer, of course, is that it depends. Americans perceive unemployment as a self-located problem or as a socially-located problem depending on the context in which it is experienced. The analyses I have presented in this paper demonstrate that there exists a strong relationship among personal experience, social and economic context, and political beliefs that has heretofore not been discussed and therefore constitutes a valuable contribution to our understanding of the oftentimes complex linkages between the personal, the contextual, and the political. 28 References Achen, Christopher H. and Larry M. Bartels. 2002. “Blind Retrospection: Electoral Responses to Droughts, Flu, and Shark Attacks." Paper presented at the Annual Meeting of the American Political Science Association, Boston, MA. Bartels, Larry M. 2008. Unequal Democracy: The Political Economy of the New Gilded Age. Princeton: Princeton University Press. Brody, Richard A. and Paul M. Sniderman. 1977. "From Life Space to Polling Place." British Journal of Political Science 7: 337-360. Campbell, Angus, Philip E. Converse, Waren E. Miller, and Donald E. Stokes. 1960. The American Voter. New York: John Wiley & Sons, Inc. Cutler, Fred. 2007. “Context and Attitude Formation: Social Interaction, Default Information, or Local Interests?” Political Geography 26: 575-600. Delli Carpini, Michael X., and Scott Keeter. 1996. What Americans Know About Politics and Why It Matters. New Haven: Yale University Press. Feather, Norman T. 1989. The Psychological Impact of Unemployment. New York: SpringerVerlag. Free, Lloyd A., and Hadley Cantril. 1968. The Political Beliefs of Americans. New York: Simon & Schuster. Gilens, Martin. 1999. Why Americans Hate Welfare: Race, Media, and the Politics of AntiPoverty Policy. Chicago: University of Chicago Press. Gordus, Jeanne Prial, Paul Jarley, and Louis A. Ferman. 1981. Plant Closings and Economic Dislocation. Kalamazoo: W.E. Upjohn Institute for Employment Research. Hainmueller, Jens, and Michael J. Hiscox. 2010. “Attitudes toward Highly Skilled and Low Skilled Immigration: Evidence from a Survey Experiment.” American Political Science Review 104: 61-84. Healy, Andrew J. 2009. “Individual Unemployment, Layoffs, and Voting in U.S. Presidential Elections.” Working Paper. Available from: http://130.203.133.150/viewdoc/summary?doi=10.1.1.144.7276 29 Healy, Andrew J., Neil Malhotra, and Cecilia Hyunjung Mo. 2010. “Irrelevant Events Affect Voters' Evaluations of Government Performance." Proceedings of the National Academy of Sciences 107: 12804-12809. Huber, Joan., and William H. Form. 1973. Income and Ideology: An Analysis of the American Political Formula. New York: Free Press. Incantalupo, Matthew B. 2012a. “The Effects of Unemployment on Voter Turnout in U.S. National Elections.” Working paper. Princeton University. Incantalupo, Matthew B. 2012b. “Blame and Attribution Following Job Loss” Working Paper. Princeton University. Kinder, Donald R. and D. Roderick Kiewiet. 1981. “Sociotropic Politics: The American Case.” British Journal of Political Science 11:129-161. Kluegal, James R. and Elliot R. Smith. 1986. Beliefs about Inequality: Americans’ Views of What Is and What Ought to Be. New York: Aldine de Gruyter. Lodge, Milton and Ruth Hamill. 1986. “A Partisan Schema for Political Information Processing.” American Journal of Political Science 80: 505-520. Margalit, Yotam. 2011. “Costly Jobs: Trade-related Layoffs, Government Compensation, and Voting in U.S. Presidential Elections.” American Political Science Review 105: 166-188. Margalit, Yotam. 2013. “Explaining Social Policy Preferences: Evidence from the Great Recession.” American Political Science Review 107: 80-103. Forthcoming. McBride, Bill. 2012. “Employment Graphs: Construction Employment, Unemployment by Education.” Calculated Risk. Updated 10 September. http://www.calculatedriskblog.com/2012/09/employment-graphs-construction.html (Sept. 10, 2012). Mills, C. Wright. 1959. The Sociological Imagination. New York: Oxford University Press. Mutz, Diana C. 1992. “Mass Media and the Depoliticization of Personal Experience.” American Journal of Political Science 36: 483-508. Paolono, Philip. 2011. “Long-Term Unemployment and Individualist Values.” Presented to the 2011 Annual Meeting of the Midwest Political Science Association. Chicago, IL. March 31-April 3. 30 Pew Research Center. 2012. “The Lost Decade of the Middle Class.” 22 August. Available from: http://www.pewsocialtrends.org/2012/08/22/the-lost-decade-of-the-middle-class/ (1 September 2012). Pforr, Klaus. 2011. “Implementation of a Multinomial Logit Model with Fixed Effects” Presented at the Ninth German Stata Users Group Meeting. Bamberg. Germany. Robinson D. 2002. “Cancer Clusters: Findings vs. Feelings.” MedGenMed 4: 16. Rosenstone, Steven J. 1982. “Economic Adversity and Voter Turnout.” American Journal of Political Science 26: 25-46. Schlozman, Kay Lehman, and Sidney Verba. 1979. Injury to Insult: Unemployment, Class, and Political Response. Cambridge: Harvard University Press. Sniderman, Paul M. and Richard A. Brody. 1977. "Coping: The Ethic of Self-reliance." American Journal of Political Science 21: 501-521. Thun, Michael. J. and Thomas Sinks. 2004. “Understanding Cancer Clusters.” CA: A Cancer Journal for Clinicians 54: 273–280. Trumbo, Craig W. 2000. “Public Requests for Cancer Cluster Investigations: A Survey of State Health Departments.” American Journal of Public Health 90: 1300-1302. Warr, Peter. 1987. Work, Unemployment, and Mental Health. New York: Oxford University Press. 31 Tables and Figures Poll Jun. 2003 N (employed) 758 N (unemployed) 86 National U3 6.3% Low U3 High U3 3.6% 8.6% (ND, SD) (OR) Feb. 2004 678 66 5.6% 3.4% 7.6% (ND) (AK, OR) May 2008 389 50 5.4% 2.9% 7.7% (SD, WY) (MI) May 2009 446 72 9.4% 4.3% 13.5% (ND) (MI) Jul. 2010 322 259 9.5% 3.9% 14.9% (ND) (NV) Jul. 2011 183 144 9.1% 3.5% 13.5% (NE) (NV) Table 1: Summary of cross-sectional Work Trends Polls (2003-2011). “U3” refers to the seasonally-adjusted unemployment rate as measured by the Bureau of Labor Statistics. 32 Employed Unemployed Total Democratic Proxy 21.5% 36.7% 28.3% (President Obama) Republican Proxy 21.8% 12.35% 54.1% (Republicans in Congress) Independent Proxy 56.7% 51.0% 17.6% (Both/Neither) χ2 (2) = 19.84; p < .01 Table 2: Proxy measure of Party Identification (July 2010 Work Trends Poll) by employment status. Survey question is “Who do you trust to do a better job handling the economy?” Percentages are given by column and may not add up to 100% due to rounding. 33 Choice Base Category Unemployed Unemployment Rate (U3) Unemployed*U3 Female Married Black Latino Other Age Income Income (Ref.) Education Democrat (imputed) Republican (imputed) Government Workers -.106 (.432) .001 (.061) .064 (.049) .205*** (.094) -.086 (.112) .535*** (.170) .384** (.172) .136 (.196) -.011*** (.004) -.242*** (.077) -.568 (.246) -.190*** (.055) .549*** (.130) -.380*** (.142) Employers Workers -.802* (.473) .023 (.064) .117** (.054) .482*** (.096) -.190* (.109) .377** (.175) .196 (.189) .192 (.197) -.004 (.004) -.191** (.077) -.792*** (.273) -.063 (.056) .406*** (.133) -.263* (.142) Government Employers .697 (.471) -.023 (.069) -.054 (.053) -.275*** (.105) .104 (.119) .159 (.165) .188 (.186) -.057 (.213) -.007 (.004) -.052 (.084) .224 (.298) -.126** (.059) .143 (.144) -.118 (.167) Democrat (proxy) Republican (proxy) Cons. Other vars. (not shown) N -.463 (.701) -2.01*** (.670) State and Poll 3183 1.54** (.785) Government Workers -.359 (.435) -.005 (.061) .086* (.050) .204** (.096) -.075 (.113) .479*** (.174) .363** (.175) .103 (.206) -.012*** (.004) -.231*** (.077) -.636** (.260) -.200 (.056) Employers Workers -1.12** (.481) .022 (.065) .148*** (.055) .494*** (.097) -.185* (.111) .406** (.176) .171 (.088) .210 (.200) -.003 (.004) -.179** (.076) -.681** (.278) -.059 (.056) Government Employers .764* (.486) -.028 (.071) -.062 (.054) -.290*** (.106) .110 (.121) .073 (.166) .192 (.190) -.107 (.221) -.008** (.004) -.052 (.085) .045 (.310) -.140 (.059) .727*** (.116) -.470*** (.131) -.975 (.817) .380*** (.119) -.429*** (.126) -2.05*** (.671) State and Poll 3091 .347*** (.124) -.042 (.150) 1.07 (.880) Table 3: Results for combined Work Trends Polls with partisan identification (2003-2011). Cell entries are multinomial logit coefficients with robust standard errors in parentheses. *p < .10; *p < .05; ***p < .01 34 Wave (Date) Employment Status 1 (Aug. 2009) 2 (Mar. 2010) 3 (Nov. 2010) 4 (Aug. 2011) Employed Unemployed Responsible for helping the unemployed 31.0% 69.0% 47.1% 52.83 58.2% 41.81% 64.60% 35.40% Government 41.1% 43.2% 40.6% 36.6% Employers 21.0% 23.1% 32.5% 28.1% Workers Themselves 37.9% 33.7% 27.0% 35.3% 1111 808 636 544 N 9.6% 9.8% 9.8% 9.1% National U3 Table 4: Summary of Work Trends Panel (2009-2011). “U3” refers to the seasonally-adjusted unemployment rate as measured by the Bureau of Labor Statistics. 35 Government Employers Workers Themselves -.034 -.161 .195** (.097) (.113) (.098) -.007 .020 -.013 Unemployment Rate (U3) (.017) (.017) (.016) .004 .018 -.021** Unemployed*U3 (.010) (.011) (.010) .462*** .007 .531*** Cons. (.173) (.181) (.164) Wave Wave Wave Other vars. (not shown) 1163 1163 1163 N 3099 3099 3099 N*T 2.7 2.7 2.7 T (avg.) Table 5: Longitudinal results for Work Trends Panel. Cell entries are fixed-effects OLS coefficients with robust standard errors in parentheses. The dependent variable is noted at the top of each column and if a binary variable created from “When people are laid off from work, who should be mainly responsible for helping them?” *p < .10; **p < .05; ***p < .01 Y=1 Unemployed 36 Jun. 2003 Feb. 2004 2 (2) =4.65* May 2008 2 (2) =2.15 =7.98** 0 20 40 60 2 (2) Gov. Emp. Self Gov. Emp. Self Gov. Emp. Self May 2009 Emp. Self Gov. Emp. Self 2 (2) =9.76*** Gov. Emp. Self Emp. Self Jul. 2011 2 (2) =34.15*** =4.01 0 20 40 60 2 (2) Gov. Jul. 2010 Gov. Emp. Self Employed Gov. Emp. Self Unemployed Gov. Emp. Self Employed Gov. Emp. Self Unemployed Gov. Emp. Self Employed Gov. Unemployed Fig. 1: “When people are laid off from work, who should be mainly responsible for helping them?” Graphs by Work Trends Poll and employment status. “Gov.” refers to “government,” “Emp.” refers to “employers,” and “Self” refers to “workers themselves.” Combined responses (volunteered) are not included. *p < .10; *p < .05; ***p < .01 37 15 10 5 0 Jun. 2003 Feb. 2004 May 2008 May 2009 Jul. 2010 Jul. 2011 Fig. 2: The distribution of the seasonally-adjusted unemployment rate (U3) by state (plus the District of Columbia) for each administration of the Work Trends Poll I use in my crosssectional analysis. 38 .5 .4 .3 .2 .1 3 4 5 6 7 8 9 10 State Unemployment Rate Employed 11 12 13 14 15 Unemployed Fig. 3: Predicted probability of answering “government” to “When people are laid off from work, who should be mainly responsible for helping them?” by employment status and unemployment context. The coefficients used to generate this graph are presented in in Table 3 (right-hand side). 39 .6 .5 .4 .3 .2 .1 3 4 5 6 7 8 9 10 State Unemployment Rate Employed 11 12 13 14 15 Unemployed Fig. 4: Predicted probability of answering “employers” to “When people are laid off from work, who should be mainly responsible for helping them?” by employment status and unemployment context. The coefficients used to generate this graph are presented in in Table 3 (right-hand side). 40 .8 .6 .4 .2 0 3 4 5 6 7 8 9 10 State Unemployment Rate Employed 11 12 13 14 15 Unemployed Fig. 5: Predicted probability of answering “workers themselves” to “When people are laid off from work, who should be mainly responsible for helping them?” by employment status and unemployment context. The coefficients used to generate this graph are presented in in Table 3 (right-hand side). 41 Wave 1 (Aug. 2009) 2 Wave 2 (Mar. 2010) 2 = 5.87* (2) = 2.76 0 10 20 30 40 50 (2) Gov. Emp. Self Gov. Emp. Self Gov. Emp. Wave 3 (Nov. 2010) 2 Gov. Emp. Self Emp. Self Wave 4 (Aug. 2011) 2 = 6.81** (2) = 12.29*** 0 10 20 30 40 50 (2) Self Gov. Emp. Employed Self Gov. Emp. Unemployed Self Gov. Emp. Employed Self Gov. Unemployed Fig. 6: When people are laid off from work, who should be mainly responsible for helping them?” Graphs by Work Trends Panel wave and employment status. “Gov.” refers to “government,” “Emp.” refers to “employers,” and “Self” refers to “workers themselves.” *p < .10; *p < .05; ***p < .01 42 Appendix A: Supplementary information (Cross-sectional) Question Wording Poll N emp. N unemp. Aug. 1998 778 52 When workers get laid off from their jobs through no fault of their own, who do you think should be primarily responsible for providing financial support for them until they find another job? (No fault wording) Aug. 1999 798 55 Oct. 2001 617 62 In times of economic downturn, many companies experience large layoffs. Who should be primarily responsible for providing services to workers who have been laid off from their job? (Economic downturn wording) Jun. 2003 758 86 When people are laid off from work, who should be mainly responsible for helping them? (Laid off wording) Feb. 2004 678 66 May 2008 389 50 May 2009 446 72 Jul. 2010 322 259 Jul. 2011 183 144 Table A1: Work Trends Poll question wordings. The shaded cells (laid off wording) are used in the cross-sectional analysis in the main paper. Earlier administrations of the Work Trends Poll used different question wordings. I plot the responses to these polls by employment status in Fig. A1 43 Aug. 1998 2 Aug. 1999 2 = 4.77* (2) Oct. 2001 2 =1.60 (2) =1.70 0 20 40 60 (2) Gov. Emp. Self Gov. Emp. Self Gov. Emp. Self Jun. 2003 2 Emp. Self Gov. Emp. Self Feb. 2004 2 =4.65* (2) Gov. Emp. Self Gov. Emp. Self Gov. Emp. Self May 2008 2 =2.15 (2) =7.98** 0 20 40 60 (2) Gov. Gov. Emp. Self Gov. Emp. Self Gov. Emp. Self May 2009 2 Emp. Self Gov. Emp. Self Jul. 2010 2 =9.76*** (2) Jul. 2011 2 =34.15*** (2) =4.01 0 20 40 60 (2) Gov. Gov. Emp. Self Employed Gov. Emp. Self Unemployed No fault wording Gov. Emp. Self Employed Gov. Emp. Self Unemployed Economic downturn wording Gov. Emp. Self Employed Unemployed Laid off wording Fig. A1: Work Trends Polls with alternative question wordings. Notice that the modal response is employers in the “no fault wording” and government in the “economic downturn wording.” Also the pattern of responses across employment groups is unremarkable. Because these question wordings prime particular types of unemployment, I exclude them from my analysis and use only those with the “laid off” wording. While I cannot claim that these potential question wording effects are fully robust since these data are from repeated cross-sections and not from a controlled, randomized survey experiment, the data suggest that beliefs about who should help the unemployed might be affected by how job loss is framed. 44 Fig. A2: Unemployment Rate by Level of Education (McBride 2012). The educational composition of the unemployed stays roughly the same over time with some minor perturbations. Periods of economic recession are indicated by a blue background. 45 Feb. 2004 May 2008 0 20 40 60 80 Jun. 2003 Male Female Male Female Male Female Female Male Female Jul. 2010 Male Female Male Female Jul. 2011 0 20 40 60 80 May 2009 Male Male Female Employed Male Female Unemployed Male Female Employed Male Female Unemployed 46 Male Female Employed Unemployed Fig. A3: Gender by Employment Status and Work Trends Poll. Feb. 2004 May 2008 0 20 40 60 Jun. 2003 Ref. Low Mid High Ref. Low Mid High Ref. Low Mid High Jul. 2010 Ref. Low Mid High Ref. Low Mid High Jul. 2011 0 20 40 60 May 2009 Ref. Low Mid High Ref. Low Mid High Ref. Low Mid High Ref. Low Mid High Ref. Low Mid High Ref. Low Mid High Ref. Low Mid High Fig A4: Income Group by Employment Status and Work Trends Poll. Individuals who refused to indicate their household incomes are noted using the “Ref.” category. The “Low” income category consists of indivduals with household incomes below $30,000 per year. The “Mid” income category consists of indivduals with household incomes between $30,000 and $75,000 per year. The “High” income category consists of individuals with household incomes greater than $75,000 per year. 47 Feb. 2004 May 2008 0 10 20 30 40 50 Jun. 2003 No HS HS Some Col. Col. No HS HS Some Col. Col. No HS HS Some Col. Col. No HS HS Some Col. Col. No HS HS Some Col. Col. Jul. 2010 No HS HS Some Col. Col. Jul. 2011 0 10 20 30 40 50 May 2009 No HS HS Some Col. Col. Employed No HS HS Some Col. Col. Unemployed No HS HS Some Col. Col. Employed No HS HS Some Col. Col. Unemployed No HS HS Some Col. Col. Employed No HS HS Some Col. Col. Unemployed Fig. A5: Educational Attainment by Employment Status and Work Trends Poll. The education attainment categories noted in the graph are no high school diploma, high school diploma or an equivalent certificate, some college (no degree), and a bachelor’s degree and any advanced degrees. 48 Feb. 2004 May 2008 0 20 40 60 80 Jun. 2003 White Black Latino Other White Black Latino Other White Black Latino Other White Black Latino Other Jul. 2010 White Black Latino Other Jul. 2011 0 20 40 60 80 May 2009 White Black Latino Other White Black Latino Other White Black Latino Other White Black Latino Other White Black Latino Other White Black Latino Other White Black Latino Other Employed Unemployed Employed Unemployed Employed Unemployed Fig. A6: Race/Ethnicity by Employment Status and Work Trends Poll. The groups used in this analysis are white, black, Latino, and an additional category for any other race/ethnic groups. 49 Feb. 2004 May 2008 0 20 40 60 Jun. 2003 Dem. Ind. Rep. Dem. Ind. Rep. Dem. Ind. Rep. Dem. Ind. Rep. Jul. 2010 Dem. Ind. Rep. Jul. 2011 0 20 40 60 May 2009 Dem. Ind. Rep. Dem. Ind. Rep. Dem. Ind. Rep. Dem. Ind. Rep. Dem. Ind. Rep. Dem. Ind. Rep. Dem. Ind. Rep. Employed Unemployed Employed Unemployed Employed Unemployed Fig. A7: Party Identification by Employment Status and Work Trends Poll. Note that the July 2010 Work Trends Poll did not ask party identification, so I use a proxy measure for this poll. 50 Appendix B: Supplementary Information (Longitudinal) *p < .10; *p < .05; ***p < .01 Employed (Wave 2) Unemployed (Wave 2) 79.3% 20.5% 32.3% 67.7% Employed (Wave 3) Unemployed (Wave 3) 86.8% 13.2% Employed (Wave 2) 35.2% 64.8% Unemployed (Wave 2) Employed (Wave 4) Unemployed (Wave 4) 89.1% 11.0% Employed (Wave 3) 26.0% 74.0% Unemployed (Wave 3) Table B1: Employment status transition probabilities across Work Trends Panel Waves. Percentages are given by row. Respondents who leave the labor force or panel are not included in this table. Employed (Wave 1) Unemployed (Wave 1) 51 Government (Wave 1) Employers (Wave 1) Workers Themselves (Wave 1) Government (Wave 2) 72.2% 31.8% 19.7% Employers (Wave 2) 17.1% 44.9% 17.4% Workers Themselves (Wave 2) 10.7% 23.3% 62.9% Government (Wave 2) Employers (Wave 2) Workers Themselves (Wave 2) Government (Wave 3) 64.9% 27.4% 16.6% Employers (Wave 3) 25.6% 58.1% 24.8% Workers Themselves (Wave 3) 9.5% 14.5% 58.6% Government (Wave 3) Employers (Wave 3) Workers Themselves (Wave 3) Government (Wave 4) 66.3% 24.4% 11.7% Employers (Wave 4) 18.5% 47.5% 13.3% Workers Themselves (Wave 4) 15.2% 28.1% 75% Table B2: Dependent variable transition probabilities across Work Trends Panel Waves. Percentages are given by row. Respondents who leave the labor force or panel are not included in this table. 52 Appendix C: Validation of Proxy Measure for Partisan Self-Identification In June 2013, I conducted an online survey as part of another project and included the proxy measure of partisanship that I use in place of the standard measure of party identification for the June 2010 Work Trends Poll. The survey used Amazon Mechanical Turk as a subject recruitment platform. Table C1 below reports the distribution of partisan self-identification, recorded in the same manner that it was recorded on the June 2010 Work Trends Poll, by respondents’ answers to the proxy measure of partisanship. Partisan Self-identification Democrat Independent Republican Democratic Proxy 92.6% 5.43% 1.96% (President Obama) Independent Proxy 48.0% 28.2% 23.8% (Both/Neither) Republican Proxy 87.2% 7.47% 5.34% (Republicans in Congress) 59.0% 17.6% 23.4% Total 2 χ (4) = 1200; p < .001 Table C1: Validation of proxy measure for party identification. I report the proportions of partisan-self-identification for respondents by on their answer to the proxy measure of partisanship, “Who do you trust to do a better job handling the economy?” Answer choices are “President Obama,” “Republicans in Congress,” and a middle category I created by combining the “both” and “neither” responses. Percents are reported by row and may not add up to 100 due to rounding. The pattern of responses indicates that this economic trust question is a reasonable proxy measure for partisan self-identification in this sample, which is composed of 2,466 adult Amazon Mechanical Turk users who reported being either employed or unemployed. While I cannot replicate this analysis for the June 2010 Work Trends Poll, and cannot assume that the relationship between the two measure holds perfectly across both time and survey sample, I have provided this as a robustness check on the proxy measure of party identification that I employ in my analysis. 53