Understanding the population structure of the GHQ-12: evidence for multidimensionality using Bayesian and Exploratory Structural Equation Modelling from a large-scale UK population survey
AbstractMental health and its complexity, measurement and social determinants are increasingly important avenues of research for social scientists. Quantitative social science commonly investigates mental health as captured by population screening metrics. One of the most common of these metrics is the 12-Item General Health Questionnaire (GHQ-12). Despite its canonical use as an outcome of interest in social science, the traditional use of the summed scores of summed questionnaires carries empirical and substantive assumptions which are often not fully considered or justified in the research. We outline the implications of these assumptions and the restrictions imposed by traditional modelling techniques and advocate for a more nuanced approach to population mental health inference. We use novel Exploratory Structural Equation Modelling (ESEM) on a large, representative UK sample taken from the first wave of the Understanding Society Survey, totalling 40,452 respondents. We use this to exemplify the potential of traditional, restrictive assumptions to bias conclusions and policy recommendations. ESEM analysis identifies a 4-factor structure for the GHQ-12, including a newly proposed “Emotional Coping” dimension. This structure is then tested against leading proposed factor structures from the literature and is demonstrated to perform better across all metrics, under both Maximum Likelihood and Bayesian estimation. Moreover, the proposed factors are more substantively dissimilar than those retrieved from previous literature. The results highlight the inferential limitations of using simple summed scores for mental health measurement. Use of the highlighted methods in combination with population studies offers quantitative social scientists the opportunity to explore predictors and patterns of underlying processes of population mental health outcomes, explicitly addressing the complexity and measurement error inherent to mental health analysis.
