Purpose: Psychometric models and statistical techniques are cornerstones of research into latent structures of specific psychopathology and general mental health. We discuss “pivot points” for future research efforts from a psychometric epidemiology perspective, emphasising sampling and selection processes of both indicators that guide data collection as well as samples that are confronted with them.
Method: First, we discuss how a theoretical model of psychopathology determines which empirical indicators (questions, diagnoses, etc.) and modelling methods are appropriate to test its implications. Second, we deal with how different research designs introduce different (co-)variances between indicators, potentially leading to a different understanding of latent structures. Third, we discuss widening the range of statistical models available within the “psychometrics class”: the inclusion of categorical approaches can help to enlighten the debate on the structure of psychopathology and agreement on a minimal set of models might lead to greater convergence between studies. Fourth, we deal with aspects of methodology that introduce spurious (co-)variance in latent structure analysis (response styles, clustered data) and differential item functioning to gather more detailed information and to guard against over-generalisation of results, which renders assessments unfair.
Conclusions: Building on established insights, future research efforts should be more explicit about their theoretical understanding of psychopathology and how the analysis of a given indicator–respondent set informs this theoretical model. A coherent treatment of theoretical assumptions, indicators, and samples holds the key to building a comprehensive account of the latent structures of different types of psychopathology and mental health in general.
- Latent class analysis