Grappling with Grouping III Social Network Analysis David Henry University of Illinois at Chicago Allison Dymnicki American Institutes for Research Advancing Health Practice and Policy through Collaborative Research Acknowledgments Families and Communities Research Group Patrick Tolan Deborah Gorman-Smith Michael Schoeny Social Network and N Normative ti IInfluence fl P Projects j t Fern Chertok Daneen Deptula Allison Dymnicki y Jane Jegerski Christopher Keys Kimberly Kobus Jennifer Watling Neal Zachary Neal Michael Schoeny Advancing Health Practice and Policy through Collaborative Research Acknowledgments • This work was supported by grants from the Centers for Disease Control and Prevention and the National Institute of Justice. The content of this presentation is solely p y the responsibility p y of the authors and does not necessarily represent the official views of the funders. • For more information information, visit www.ihrp.uic.edu. www ihrp uic edu Advancing Health Practice and Policy through Collaborative Research Grappling with Grouping I. Cluster Analysis II. Clustering methods for binary variables III. Social Network Analysis Central theme: Clustering approximates uniqueness in the same way that a sample mean approximates a population. Advancing Health Practice and Policy through Collaborative Research N Network kA Analysis l i • Widely used in community psychology research • 28 studies since 2000 in jjust two jjournals ((AJCP,, JCP)) – Search terms: “Network Analysis” • Similar search using the term “cluster analysis” returned 14 studies. Advancing Health Practice and Policy through Collaborative Research Community y Studies Employing p y g Network Analysis y Study Variables Type of Analysis Swindle et al., 2000 Positive and negative social transactions in networks of HIV+ persons Rating scales Hirsch et al., 2002 Differences in strength of ties by race Ego network Ying, 2002 Social network composition of Taiwanese graduate students Rating scales Langhout, 2003 Rating scales Zea et al., 2004 A single case study using ego networks Violence among female gang members increases before pregnancy and decreases afterward Criticism practical support were significant predictors of Criticism, practical support were significant predictors of mental health for battered women. Differences between nearly homeless and housed women on beliefs about netweok members as housing resources Target‐specific factors were related to the probability of disclosure. Chia, 2006 Sociometric nominations in a work organization Sociometric Knowlton & Latkin, 2007 Ego networks Ego network Dominguez & Maya‐Lariego, 2008 Ego network support characteristics Ego network Pernice Duca 2008 Pernice‐Duca, 2008 Social Support in clubhouse mental health programs Social Support in clubhouse mental health programs Rating scales Rating scales Fleisher & Krienert, 2004 Levendosky et al., 2004 Toohey et al., 2004 Ego network Qualitative Interviews Rating scales Rating scales Advancing Health Practice and Policy through Collaborative Research Community y Studies Employing p y g Network Analysis y Study Variables Type of Analysis y Toro et al., 2008 Social Support in homeless adults ‐ ego networks Campo et al., 2009 Latkin et al., 2009 Convergent and discriminant validity with other measures Rating scales Network drug use contributed to perceptions of neighborhood disorder. Ego network Neal, 2009 Density, centrality, and relational aggression Informant Trotter & Allen, 2009 Ego networks, qualitative analysis Ego network Crowe, 2010 Personal and Organizational Community Networks Sociometric L Lugue et al., 2010 t l 2010 Haines et al., 2011 Community Cancer Network C it C N t k S i Sociometric ti Social Support buffered ecological risk effects on psychological distress Rating scales Network characteristics of an interdisciplinary collaboration based on multiple types of relationships Sociometric Neal et al (2011) Neal et al. (2011) Cohesion vs structural similarity in teacher advice networks Cohesion vs. structural similarity in teacher advice networks Prelow et al., 2010 Ego network Sociometric Advancing Health Practice and Policy through Collaborative Research Variety V i t off Network N t kM Methods th d in Community Psychology Studies Type Number Sociometric nominations 5 Informant-based Methods 1 Ego-networks 7 Rating Scales 8 Qualitative 1 Advancing Health Practice and Policy through Collaborative Research Outline • Overview of methods – Sociometrics – Informants – Ego-networks Ego networks – Dynamic • For each (-1) – Theory/method/measures – Software – Strengths and limitations Advancing Health Practice and Policy through Collaborative Research Network Analysis with Sociometrics • Data source: Relationships – “Who are your friends?” (Kobus & Henry, 2009) – “What a o organizations ga a o s do you be belong o g to?” o (C (Crowe, o e, 2010) • Analysis: Matrix and Graph Representations A B C D E F G A 0 1 0 1 0 0 0 B 1 0 0 1 0 0 0 C 0 0 0 0 1 1 0 D 1 1 0 0 0 0 0 E 0 0 1 0 0 1 0 F 1 0 0 0 0 0 0 G 0 0 0 0 0 1 0 B E F C A D G Advancing Health Practice Policy through CollaborativeResearch Research Advancing Health Practice andand Policy Through Collaborative Sociometrics: Network Measures DENSITY X g (g ( g 1) 12 / 42 0.28 where X = relationships (ties) = 12 g = network size (# of potential relationships) = 42 A B C D E F G Σ A 0 1 0 1 0 0 0 2 B 1 0 0 1 0 0 0 2 C 0 0 0 0 1 1 0 2 D 1 1 0 0 0 0 0 2 E 0 0 1 0 0 1 0 2 F 1 0 0 0 0 0 0 1 G 0 0 0 0 0 1 0 1 Σ 3 2 1 2 1 3 0 12 B E F C A D G Advancing Health Practice Policy through CollaborativeResearch Research Advancing Health Practice andand Policy Through Collaborative Sociometrics: Networks Measures 2 2 g M L 2 ( 1 ) L2 2 0.77 2 2 L( g 1) L L2 • Mutuality Index M = number of mutual relationships = 5 g = network size = 7 L = sum off the th outdegree td off the th total t t l network t k = 12 L2 = sum of squares of the outdegree of the total network =22 A B C D E F G Σ A 0 1 0 1 0 0 0 2 B 1 0 0 1 0 0 0 2 C 0 0 0 0 1 1 0 2 D 1 1 0 0 0 0 0 2 E 0 0 1 0 0 1 0 2 F 1 0 0 0 0 0 0 1 G 0 0 0 0 0 1 0 1 Σ 3 2 1 2 1 3 0 12 B E F C A D G Advancing Health Practice and Policy through Collaborative Research Sociometrics: Network Measures • Boundary Density (Hirsch, 1980) Tactual BD 0.083 Tpossible Tactual = number of actual ties across subgroups = 2 Tpossible = number of possible ties across subgroups = 24 A B C D E F G Σ A 0 1 0 1 0 0 0 2 B 1 0 0 1 0 0 0 2 C 0 0 0 0 1 1 0 2 D 1 1 0 0 0 0 0 2 E 0 0 1 0 0 1 0 2 F 1 0 0 0 0 0 0 1 G 0 0 0 0 0 1 0 1 Σ 3 2 1 2 1 3 0 12 B E F C A D G Advancing Health Practice and Policy through Collaborative Research Sociometrics: Measures of Individuals D B ij Mean Geodesic Distance j ij j oor D B ji j ij j where D = Distance and B = Reachability In: 1.33 for F, 2.0 for D and 2.16 for G A B C D E F G Σ A 0 1 0 1 0 0 0 2 B 1 0 0 1 0 0 0 2 C 0 0 0 0 1 1 0 2 D 1 1 0 0 0 0 0 2 E 0 0 1 0 0 1 0 2 F 1 0 0 0 0 0 0 1 G 0 0 0 0 0 1 0 1 Σ 3 2 1 2 1 3 0 12 B E F C A D G Advancing Health Practice and Policy through Collaborative Research Sociometrics: Measures of Individuals B E • Position – Member – Liaison – Isolate F C A D G • Liaisons > Members or Isolates on tobacco and alcohol use (2 studies) • Members and Isolates more influenced by peer substance use than Liaisons. Advancing Health Practice and Policy through Collaborative Research Sociometrics: Statistical Models Example D d h Dyads have a ttendency d tto b become ttriads: i d “birds of a feather? or “friends of friends?” We can model the likelihood of triad closure, but the chance models are complex Random graph models and double permutation tests provide alternatives suitable for predicting network structures or individual ties. Advancing Health Practice and Policy through Collaborative Research N Network kA Analysis l i with i h Sociometrics S i i • Strengths – Unbiased assessment of social influence – Patterns of diffusion and communication – Rich measurement and theory • Limitations Li it ti – Missing nominators compromise accuracy – Costly assessment – Complex coding and analysis – Requires q bounded social space p Advancing Health Practice and Policy through Collaborative Research Network Analysis with Sociometrics Software • Stand alone Programs Stand-alone – UCINET (http://www.analytictech.com/ucinet/) – Krackplot (Freeware – visualization software) (http://www.andrew.cmu.edu/user/krack/krackplot.sh http://www andrew cmu edu/user/krack/krackplot sh tml) • R (http://www.r-project.org/) – iGraph – sna • Excel – NodeXL: N d XL Freeware F http://nodexl.codeplex.com/ htt // d l d l / Advancing Health Practice and Policy through Collaborative Research Network Analysis from Informants • Data Source: “Who hangs out together?” Informants Informant 1 Port Port Compilation Kit Tunner 1 0 Kit 0 Port Kit 1 Tunner .5 .5 5 Informant 2 Port Kit 1 1 1 • Variations – Cognitive Social Structures (Krakhardt, 1987) – Social Cognitive Mapping (Cairns et al., 1985) Advancing Health Practice and Policy through Collaborative Research Informants: Examples • Cairns, Leung, Buchannan, and Cairns (1995) used social cognitive mapping to study the fluidity, reliability, and interrelations of social networks of 4th and 7th graders over a 3-week period. • Neal (2009) used Cognitive Social Structures to study the influence of centrality and density on relational aggression in a sample of 3rd through 8th grade children. Advancing Health Practice and Policy through Collaborative Research Informants: Software • Cognitive Social Structures R ij R i, j – consensus aggregation k across ac oss k informant o a matrices a ces can be done in Excel or R. – See Krackhardt (1987) for specific instructions. • Social Cognitive Mapping – Contact Man-Chi Leung, Ph.D. at UNC (manchi.leung@unc.edu ) for a copy of the SCM 4.0 program p g and manual. Advancing Health Practice and Policy through Collaborative Research Network Analysis from Informants • Strengths: – Provides valid estimates of network ties with comparatively few informants. – Missing data does not decrease accuracy – Economical to administer • Limitations – Difficult to assess directed relations – Requires bounded social space (e.g., classrooms, g ) schools,, organizations) Advancing Health Practice and Policy through Collaborative Research Ego Network Analysis • Theory: Best for unbounded networks where saturation is not possible • Data Source: – Prompts for different social functions – Demographics, relationships, frequency – Behavior of network members • Analysis: – Network Net ork size, si e densit density, diversity, di ersit boundary bo ndar densit density, heterogeneity, position, behavior of network members Advancing Health Practice and Policy through Collaborative Research Ego Networks: Measures • Heterogeneity n Ak 2 e HeterogeneityiA 1 1 n where A = a categorical attribute (e.g., gender, race) Ak = number of individuals with the attribute e = number of individuals with valid data on A n = total number of traits of A in the ego network Advancing Health Practice and Policy through Collaborative Research Ego g Networks: Examples p • Dominguez & Maya-Lariego, 2008 – Ego networks of host individuals and immigrants in the U.S. and Spain – Host individuals had lower centralityy than did immigrants according to multiple measures. • T Tolan, l G Gorman-Smith, S ith & H Henry, 2003 – Ego network assessments of delinquent involvement in adolescent males – Network violence predicted future individual violence. Advancing Health Practice and Policy through Collaborative Research E Ego Networks N k • Strengths – Does not require assessment of entire network – Can provide social network and social support i f information ti – Does not require bounded social space. • Limitations – Possible bias in the direction of the individual’s behavior – Ego is central by definition, so meaning of position and centrality are problematic. Advancing Health Practice and Policy through Collaborative Research E Ego Networks: N k S Software f • Like informant-based network data, ego networks populate matrices and graphs of the type we have been discussing. • Ego network data can be visualized in Krackplot and other programs and analyzed using any software program you would use to analyze sociometric data. • Because ego networks tend to be smaller than networks derived from sociometric studies, analyses can often be conducted by hand or using Excel or SPSS. Advancing Health Practice and Policy through Collaborative Research Dynamic Social Network Analysis • Theory – Social relationships are dynamic – Most SNA is static – Static analysis may miss important characteristics of the social world. • Examples – Is “liaison” a position or a transition state? – Changes g in p parent g groups p over the course of intervention. Advancing Health Practice and Policy through Collaborative Research Group p 212: Pre Advancing Health Practice and Policy through Collaborative Research Group p 212: Session 4 Advancing Health Practice and Policy through Collaborative Research Group p 212: Session 9 Advancing Health Practice and Policy through Collaborative Research Group p 212: Session 14 Advancing Health Practice and Policy through Collaborative Research SAFE-E Group 212 3 25 2.5 2 1.5 1 0.5 0 1 0.9 08 0.8 0.7 0.6 0.5 04 0.4 0.3 0.2 0.1 0 # of contacts Density Advancing Health Practice and Policy through Collaborative Research Dynamic SNA • Methods – Berger-Wolf method • α (Persistence) • β (turnover) • γ (membership) – Software • tnet package in R does analysis of time-stamped ties - http://toreopsahl.com/tnet/ p p • DNA (Discourse Network Analyzer) http://www.philipleifeld.de/ Advancing Health Practice and Policy through Collaborative Research Summary Saturation Possible? Absentees or nonparticipants? Multiple Time Points N Y N Y N Y Sociometrics - + + - + - Informants - + + + + - Ego-Networks + ? + - + - Dynamic y SNA ? ? ? ? - + Advancing Health Practice and Policy through Collaborative Research