Getting the Gig Picture in Community Science: Methods That Capture Context

1. Getting the Gig Picture in Community Science: Methods That Capture Context Douglas A. Luke American Journal of Community Psychology, 2005

2. Introduction • CP should put context front and center as one of the core values of community psychology • Call for contextual methods instead of taking context into account just by theories, models and frameworks. • Methodologically , need to be in control of the quantitative methods, not vise versa.

3. Introduction • The Goal of this paper: provide several useful analytic methods to capture community context.

4. The Review of Statistical Practice in Community Science

5. Figure 1: Statistics usage

6. The point : if community scientists are seriously interested in understanding how the physical, social , organizational, cultural, economic and political context shape human behavior and health, it better try some other new innovative method. • The argument of paper is more philosophical and political than technical.

7. Contextualism and Community Science • Effective interventions based on community science can and should be aimed at the extra-individual level-------contextualism • Some empirical work conducted from • Fewer collected data from extra-individual levels or analyze the data using appropriate methods for multiple levels. • Contextual thinking Vs. Contextual methods

8. Multilevel Modeling • Hierarchical linear modeling, • Mixed-effects modeling • Growth-curve modeling

9. A Detailed Example • Tobacco control policy study by Luke & Krauss, 2004 • Typical single-level regression model would be : the percentage of time that a member of congress voted in a pro-tobacco industry direction based on his or her political part and the amount of money received from a tobacco industry political action committee

10. There is cluster effects within independent variables. • Congress member from same party but different states would vote differently • State-level characteristics may influence voting behavior apart from the individual party and money received. • Particular the size of the tobacco economy in a state, operationalized as size of tobacco harvest, is a type of contextual variable that might influence congressional voting • Not only individual level , but state level.

11. A multilevel model

12. A detailed example ( contd)

13. GIS • Definition • 3 illustrative studies • Wieczorek and Hanson (1997): DWI study • Tobacco control policy study • Goldstein (2003) tobacco free school policies study

14. Wieczoreck and Hanson (1997) • Examine the patterns of driving-while-intoxicated DWI to reveal the geographic context of DWI • GIS map shows that contour plots which show that DWI are not distributed randomly around the metropolitan area. • This type of analysis leads to policies and interventions that can be aimed more precisely , leading to lower costs and hopefully enhanced effectiveness

15. Luke, Esmundo, & Bloom, 2000 • Use GIS to collect the billboards in St Louis and analyze the location patterns of tobacco advertising. • Billboards are coded by the type of image found on the billboard, combined with census data. • Billboards with African American images on them tended to be concentrated in neighborhoods with higher proportions of African American residents. • Support an argument that the tobacco industry was targeting African Americans for their products

16. Goldstein (2003) • Identify important factors influencing the successful adoption of tobacco-free policies in 14 North Carolina school districts • GIS analyses revealed that all school districts passing policy were located in countries with relatively little tobacco production, although key informants suggested that the local tobacco economy had little direct influence on policy adoption.

17. Network Analysis • Social network analysis is a broad set of methods for the systematic study of social structure ( Degenne& Forse,2004) • Network tools are based on relational data rather than only attributes are analyzed • Relations can be any type of connection between actors, like friendship, recognition, money exchange, kinship, information exchange and respect, etc.

18. Network analysis • Originated from social support study • Perceived social support Vs. relationship or structural aspects of social support • Strokes’ study (1983) • Each participant identify up to 20 people who were important in their lives and with whom they had monthly contact, and drew lines between connecting each pair of people • Measure not only the source of support and satisfaction with support, but size and density of density. • Ego-centric network, determined by the perspective of simple person.

19. Ego-centric Network

20. Complete Network

21. Sample studies of Complete network • Tausig (1987): for analyze of a community mental health service system in a New York county • Using the presence or absence of interactions between all 45 mental health service agencies to identify different types of cracks in the service system, including missing links and conflicted links

22. Krauss, Mueller & Luke, (2004) • Evaluation study of 10 state tobacco control programs • View a state tobacco control program not as one activity directed by one lead agency but rather a set of interrelated activities coordinated by an organized network of tobacco control agencies and organizations. • Help to understand the structure of these complex state tobacco control programs, to identify control program to identify other state characteristics related to program structure, and to determine

23. From these network analysis, individual agencies vary in their centrality in the communication structure. • This centrality is measured using Freeman’s betweenness index

24. Network Sample Study

25. Centrality of network • Freeman’s index is defined as : for any individual actor i, the actor’s centrality is the proportion of times that actor I is included in the geodesics of actors J and K, summing across all j,k pairs. ( Geodesics is the shortest network path between any two actors.) • So the betweenness centrality measures how often an individual actor I is involved in the communication between other pairs of actors in the network • High-network centralization scores indicate hierachical communications structures, whereas low-centralization scores indicate flat communication structures.

26. To test how States financial and political climates influence the structure of state tobacco control networks • States that had a strong positive climate tended to have communication networks that were relatively hierarchical with the lead agency always being the most central actor in the network (ITPC). • States that had challenging climates tended to show a different pattern

28. Interpretation: state networks adapt to their environments. • When times are good, a lead agency takes a central role in guiding activities and distributing financial resources. • When times are not good, there is less money flowing into the program , and other non-lead agencies step up and become more actively involved

29. Cluster Analysis • Most methods deal with variables, but there is also tools tell how cases are related to each other: cluster analysis, and multidimensional scaling, network analysis • Useful tool for uncovering and describing the context pattern, especially the diversity and heterogeneity within the data • Cluster analysis works by grouping cases ( often people) based on their similarities and dissimilarities and is an exploratory technique that can to used to reveal unknown heterogeneity. • In traditional modeling method, heterogeneity is handled by using covariates to control or adjust for these individual differences, but controlling heterogeneity is a way to decontextualize the data

30. Sample Studies • Kuhn and Culhane (1998) study of types of public shelter usage • They used cluster analysis on administrative data on public shelter use by single adults in New York and Philadelphia and identified three clear types of homeless persons in the basis of patterns of shelter usage • Transitional homeless: once and short • Episodically homeless: frequently • Chronically homeless: less frequently, longer

31. Significance of the study • Not only counter the stereotypes of homelessness( transitional group made up of 80% of shelter use, chronic group only 10%), but influenced subsequent social policy at the federal level • Help us to understand the interplay between poverty , housing and social services.

32. Zapert , Snow and Tebes study • Zapert, Snow, and Tebes (2002): use cluster analysis to identify subgroups of teenagers on the basis of their patterns of drug use from 6th-10th grade. • Identified six distinctive clusters of substance users • Implication for intervention: interventions can be focused more specifically on the right type of high-risk adolescent at the right time given these types of data

33. A sample study

34. Discussion • Using wider array of methods are more likely to be able to explore and understand the complex data patterns of interest to community scientists. • Tools should be consistent with the values and what to be studied------ some real problems in community psychology.

35. Comments • Implication for suicide study • Analysis of context influencing the suicide would facilitate to the understanding of suicide • We should keeping learning about the method to advance our research skill • Research method should be chosed suitable for the research purpose. • Nested model is one kind of multilevel methods

36. Comments • Implication for training • The phenomena that some training program can only affect the knowledge of trainees instead of their behavior can be regarded as the function of context • The research focusing on the context can also contribute to the effectiveness of training