Detection of Problem Gambler Subgroups Using Recursive Partitioning

Abstract

The multivariate socio-demographic risk factors for problem gambling have been well documented. While this body of research is valuable in determining risk factors aggregated across various populations, the majority of studies tend not to specifically identify particular subgroups of problem gamblers based on the interaction between variables. The identification of problem gambling subgroups offers the potential for improved harm-reduction initiatives in particular geographic contexts. We introduce an analytical approach termed recursive partitioning, commonly used in the health sciences but infrequently employed in gambling research, to identify specific gambler subgroups based on the interaction of a range of predictor variables. Recursive partitioning creates groups of cases (e.g. gamblers) with similar outcomes by repeatedly splitting each group into smaller and more homogenous subgroups. We employ it to define problem gambler subgroups within a diverse population context (i.e. northern Australia) and compare the results with a multivariate analysis of the same dataset using a generalized linear regression model. We assess the advantages and disadvantages of each approach, and argue that recursive partitioning is an easily-interpretable approach that may be useful both in identifying problem gambling subgroups and in developing targeted harm-minimisation strategies.