| Session Title: A Sampler of Applications of Multilevel Modeling to the Evaluation of Community Collaboration |
| Multipaper Session 519 to be held in Suwannee 17 on Friday, Nov 13, 10:55 AM to 11:40 AM |
| Sponsored by the Human Services Evaluation TIG |
| Chair(s): |
| Adam Darnell, EMSTAR Research Inc, adam_darnell@yahoo.com |
| Discussant(s): |
| James Emshoff, EMSTAR Research Inc, jemshoff@gsu.edu |
| Abstract: Georgia's Family Connection Partnership (GFCP) represents the largest network of community collaboratives in the nation, one serving each county of the state (n=157). GFCP collaboratives are associations of social service providers, civic organizations, community leaders, and private citizens cooperatively addressing challenges to their communities' well-being. In this session we describe results from two different applications of multilevel modeling to the study of Family Connection collaboratives. In the first, multilevel modeling was applied to data from the Collaborative Member Survey, a self-report measure of collaborative functioning completed by numerous respondents from each collaborative. Dimensions of collaborative functioning were examined using multilevel confirmatory factor analysis. In the second paper, multilevel modeling was applied to longitudinal data on child abuse measured at the county-level annually from 1994-2006. Multilevel modeling was used to examine change in county-level child abuse rates as an outcome of various indicators of community context and collaborative structure and function. |
| A Multilevel Confirmatory Factor Analysis of the Collaborative Member Survey |
| Jack Barile, EMSTAR Research Inc, jpbarile@hotmail.com |
| Scott Weaver, Georgia State University, srweaver@gsu.edu |
| Adam Darnell, EMSTAR Research Inc, adam_darnell@yahoo.com |
| Steve Erickson, EMSTAR Research Inc, ericksoneval@att.net |
| James Emshoff, EMSTAR Research Inc, jemshoff@gsu.edu |
| Gabriel P Kuperminc, Georgia State University, gkuperminc@gsu.edu |
| The collaborative member survey assesses dimensions of collaborative functioning including communication, planning, leadership, budgeting and family involvement. It is administered on a voluntary basis to multiple respondents from each Family Connection collaborative. A measurement model of collaborative functioning was tested using multilevel confirmatory factor analysis. This measurement structure was also tested for measurement invariance across three years of Collaborative Member Survey data from 2006 to 2008. Then, using the most recent data, we discuss effects on collaborative functioning of both within-collaborative predictors (i.e., characteristics of individual respondents) and between-collaborative predictors (e.g., collaborative age, meeting frequency, county SES and demographic composition). Implications from both measurement and structural modeling will be discussed, including dimensions of collaborative functioning and correlates of high collaborative functioning. |
| Identifying Effects of Community Collaboration on Child Abuse Using Latent Growth Modeling |
| Adam Darnell, EMSTAR Research Inc, adam_darnell@yahoo.com |
| Scott Weaver, Georgia State University, srweaver@gsu.edu |
| Jack Barile, Georgia State University, jpbarle@hotmail.com |
| Gabriel P Kuperminc, Georgia State University, gkuperminc@gsu.edu |
| Steve Erickson, EMSTAR Research Inc, ericksoneval@att.net |
| James Emshoff, EMSTAR Research Inc, jemshoff@gsu.edu |
| Child abuse is one of the most commonly targeted outcomes among GFCP collaboratives. Rates of substantiated cases of child abuse for each Georgia county were measured annually from 1994 to 2006. Longitudinal change in abuse rates was modeled using latent growth modeling. Between-county differences in change in abuse rates were examined for association with the introduction into the county of a collaborative directly targeting abuse, controlling for community context (e.g., SES, population size, demographic composition). Several different approaches to testing the effect of collaboration were used, including a sequential process growth model relating change in abuse rates to change in a collaborative's propensity to target abuse in the period prior to measured abuse rates. We also report results from a cross-lagged autoregressive model of the relationship between collaboration and abuse rates. Discussion will address the strengths and weaknesses of each approach in terms of causal inference and practical challenges. |