Abstract
Psychological distress often onsets during adolescence, necessitating an accurate understanding of its development. Assessing change in distress is based on the seldom examined premise of longitudinal measurement invariance (MI). Thus, we used three waves of data from Next Steps, a representative cohort of young people in the UK (N = 13,539) to examine MI of the General Health Questionnaire-12 (GHQ-12). We examined MI across time and gender from ages 15 to 25 in four competing latent models: (a) a single-factor model, (b) a three-factor correlated model, (c) a bifactor model of "general distress" and two orthogonal specific factors capturing positive and negative wording, and (d) a single-factor model including error covariances of negatively phrased items. We also tested acceptability of assumptions underlying sum score models. For all factor models, residual MI was confirmed from ages 15 to 25 years and across gender. The bifactor model had the best fit. While sum score model fit was not unequivocally acceptable, most mean differences across time and gender were equivalent across sum scores and latent difference scores. Thus, GHQ-12 sum scores may be used to assess change in psychological distress in young people. However, latent scores appear more accurate, and model fit can be improved by accounting for item wording.