On Suitable Model Selection Approach for Analysis of Mental Health Service use Data

By : Dulal K. Bhaumik , Subhash Aryal , Scott Hur

Page No: 212-223

Abstract
Various methods are available in the literature for analysis of multi-level longitudinal service utilization and expenditure data. We discuss implementation of several methods to analyze service utilization and cost data. Hierarchical longitudinal service utilization data was analyzed using a log-gamma model. Mixed-effects models with random intercept and trend are the most appropriate method for analyzing data from hierarchical longitudinal studies and they provide the best fitting models. But these models also suffer from non-convergence problem, especially in the presence of random-effects at multiple levels. Researchers generally perform log transformation and use normal-based method to analyze service utilization data due to ease of use. Even though they are difficult to implement, generalized linear models is a more appropriate choice for analyzing service utilization data.

Authors :
Dulal K. Bhaumik
Department of Psychiatry, University of Illinois at Chicago, USA.

Subhash Aryal
Department of Biostatistics and Epidemiology UNT Health Science Center Fort Worth, TX, USA.

Scott Hur
Department of Psychiatry, University of Illinois at Chicago, USA.

DOI: DOI-https://doi.org/10.32381/IAPQRT.2019.43.02.6