Setting criterion thresholds for estimating prevalence: what is being validated?

Abstract

Much of the debate over how best to estimate the prevalence of problem gambling in the general population is driven by a number of misconceptions, misinterpretations, and questionable, sometimes erroneous assumptions. Among the latter is the failure to understand that what is being validated is not the test but the interpretation of test scores for a specific purpose. In addition there has been a lack of attention to defining the clinical and/or epidemiologic relevance of case definitions in terms of severity and other clinical attributes, a misunderstanding of how test values are interpreted when criterion thresholds or cut-off points are selected, and a failure to replicate the validation of criterion thresholds for defining cases of problem gambling. It is argued further that the distinction between dichotomy and continuum is a false choice, and any emphasis on overestimation is misdirected. Alternative methods for evaluating tests and estimating prevalence are described and a pragmatic empirical approach to the interpretation of prevalence estimates is recommended.