Advances in mental health research are highly dependent on the quality of data analytic strategies available to investigators. The purpose of the proposed project is to extend an important class of statistical models, namely random-effects regression models (RRMs), to better accommodate types of data collected in mental health services research. RRMs are especially useful for analyzing data from two types of nested designs: longitudinal (observations nested within subjects) and/or clustered (subjects nested within clusters) designs, both of which are quite common in mental health services research. Thus far, random-effects procedures have been developed for continuous, dichotomous, and ordinal outcome variables. However, important types of services utilization data are not optimally or appropriately modeled using these existing methods. For example, the outcome variable """"""""type of service use"""""""" (e.g., outpatient psychotherapy, inpatient hospitalization, community mental health center, etc.) would best be treated as a nominal outcome variable, while """"""""frequency of service use"""""""" or """"""""number of hospitalizations"""""""" are counts. While these variables often represent the primary outcomes in services research studies, current methodology is insufficient for appropriately modeling these outcomes when they arise from either a longitudinal or clustered design. Also, even when appropriate statistical methods and software are available for analysis, researchers under-utilize such methodology due to the complexity involved in performing such analyses. This project will address these issues by focusing on development, implementation, and accessibility. The development phase will focus on generalizing current statistical methodology of RRM for nominal and count-type variables. In terms of implementation, public domain computer programs will be produced that are capable of modeling nominal and count-type response data arising from either a longitudinal or clustered dataset. To make these programs accessible, a user-friendly Windows-based interface will be implemented and an accompanying primer will be developed for each program. Thus, the goal of this proposal is to develop and provide further generalization of the RRM data analytic strategy for handling many of the challenges encountered in analyzing mental health services research data from longitudinal and clustered designs.
| Hedeker, Donald; Mermelstein, Robin J; Demirtas, Hakan (2007) Analysis of binary outcomes with missing data: missing = smoking, last observation carried forward, and a little multiple imputation. Addiction 102:1564-73 |
| Raman, Rema; Hedeker, Donald (2005) A mixed-effects regression model for three-level ordinal response data. Stat Med 24:3331-45 |
| Hedeker, Donald (2003) A mixed-effects multinomial logistic regression model. Stat Med 22:1433-46 |
| Hedeker, D; Siddiqui, O; Hu, F B (2000) Random-effects regression analysis of correlated grouped-time survival data. Stat Methods Med Res 9:161-79 |
| Hedeker, D; Mermelstein, R J; Weeks, K A (1999) The thresholds of change model: an approach to analyzing stages of change data. Ann Behav Med 21:61-70 |
