| Session Title: Retention in a Longitudinal Outcome Study: Modeling Techniques and Practical Implications |
| Multipaper Session 333 to be held in Room 103 in the Convention Center on Thursday, Nov 6, 1:40 PM to 3:10 PM |
| Sponsored by the Quantitative Methods: Theory and Design TIG |
| Chair(s): |
| Robert Stephens, Macro International Inc, robert.stephens@macrointernational.com |
| Abstract: This multipaper session will explore the determinants of retention and patterns of participation in the national evaluation of the Comprehensive Community Mental Health Services for Children and Their Families Program. The data were collected as part of the CMHS system of care initiative in communities initially funded between 2002 and 2004 by the Center for Mental Health Services. In the three papers, we use data from the longitudinal outcome component of this evaluation, as well as site level information on communities participating in the evaluation, to investigate the effect of child, caregiver and site level characteristics on retention. Each of the three papers explores a different methodology to model retention. Specifically, we use latent class analysis (LCA), the sequential response model, panel data techniques, and multilevel modeling. The session will have both practical and methodological relevance, as participants will learn about determinants of retention, as well as techniques for studying retention. |
| Modeling Retention Over Time |
| Yisong Geng, Macro International Inc, yisong.geng@macrointernational.com |
| Megan Brooks, Macro International Inc, megan.a.m.brooks@macrointernational.com |
| In this analysis, two methods are used to model participation in the longitudinal outcome study. In both models, the outcomes are binary indicators of participation at each timeframe. In the first, the outcome of retention is modeled as a series of sequential decisions of whether or not to participate in the next longitudinal outcome study interview. Because an individual does not decide to be 'retained' in a study at one point, but rather makes a series of decisions at each interview of whether or not to continue to participate, this sequential model might capture these dynamics more accurately than other models. In the second model, panel data analysis techniques are used to focus on the impact of past participation on future participation. While retention is not modeled sequentially here, the decision to participate at each timeframe is considered separately, while taking into consideration participation at each prior timeframe. |
| A Latent Class Analysis of Patterns of Respondent Participation in a Longitudinal Outcome Study |
| Ye Xu, Macro International Inc, ye.xu@macrointernational.com |
| Robert Stephens, Macro International Inc, robert.stephens@macrointernational.com |
| This presentation will explore the patterns of respondents' participation in the national evaluation of the Comprehensive Community Mental Health Services for Children and Their Families Program. Through this study we hope to develop a classification system of the longitudinal outcome study participants who were heterogeneous in their participation in follow up data collection, and develop a set of key characteristic variables that predict these patterns of participation in the longitudinal study. We will present the utility of latent class analysis for accomplishing this type of classification. Latent class analysis (LCA) allows one to examine shared characteristics across groups of respondents with different distributions on several indicators at a point in time (Muthon, 2001) For this presentation retention is defined through a series of dichotomous variables that represent participation at each of the follow-up waves of data collection. |
| Determinants of Retention in a Longitudinal Study using a Multilevel Modeling Approach |
| Tesfayi Gebreselassie, Macro International Inc, tesfayi.gebreselassi@macrointernational.com |
| In any longitudinal study, participant loss during follow-up can potentially bias the results of analysis because of differences between those who dropped-out and those who continue to participate. In this presentation, we use data from the longitudinal outcome component of the national evaluation of the Comprehensive Community Mental Health Services for Children and Their Families program to implement system of care funded through the Center for Mental Health Services, as well as site level information on communities participating in the evaluation, to investigate child, caregiver and site level characteristics that predict retention in the longitudinal outcome study. |