Distinguishing Persistent Versus Episodic Clusters of At-Risk Respondents on the Problem Gambling Severity Index.

Abstract

The Problem Gambling Severity Index (PGSI) is a popular tool for assessing past-year problems related to gambling. Multiple categorization schemes have been proposed, with scores 3-7 variously interpreted as reflecting a 'moderate' degree of problems. Crucially, it is possible to land in this Moderate-risk category by reporting one or two persistent problems, or up to seven problems that occur more sporadically. Given that DSM-V gambling disorder may occur either persistently or episodically, this confounding of problems' occurrence and their frequency necessitates the development of a method for delineating the PGSI's Moderate-risk category. We propose a variance clustering approach for understanding Moderate-risk cases on the PGSI. Using 3,868 Moderate-risk cases from an existing database of 18,494 Canadian online gamblers, we use K-means clustering to identify distinct subgroups within the variances of collected PGSI surveys. We find that three clusters (which correspond to lower [61.83%], higher [8.85%], and intermediate [29.32%] variance cases) are not equal in size, and are separated at cutoffs equal to 0.40 and 0.81. These clusters differ in terms of the number of PGSI items endorsed, and multiple dimensions of participants' sociodemographic background. These variance boundaries, and the case clusters they separate, are easy to compute and offer useful context that further informs summed survey scores falling in the Moderate-risk category of the PGSI. Additional applications, and avenues for further research are discussed.