Longitudinal validation of the GHQ-12 and PHQ-2 in Chilean social housing populations in urban regeneration contexts

Brief questionnaires have been applied in poor urban populations to monitor their mental health. Mental health outcomes encompass diverse symptoms that differ in cause and functional impact; however, most studies combine these symptoms as if they represented a single construct. Longitudinal validation of health instruments requires understanding the psychometric properties and causal structures between measurement points to distinguish measurement error from true change before interpreting intervention effectiveness. However, the longitudinal psychometric properties of the General Health Questionnaire (GHQ-12) and Patient Health Questionnaire (PHQ-2) in Latin American social housing populations remain unknown. This study assessed the dimensionality and longitudinal psychometric properties of these instruments in Chilean populations targeted by urban regeneration interventions. To assess the longitudinal psychometric properties of the General Health Questionnaire (GHQ-12) and Patient Health Questionnaire (PHQ-2) in poor Chilean urban populations undergoing urban regeneration interventions. We analyzed two waves of mental health data (6-month intervals) from 955 residents of social housing neighborhoods in Santiago and Viña del Mar, Chile. We evaluated item-level and scale-level statistics, confirmatory factor analysis, construct validity, test-retest reliability, and longitudinal measurement invariance. Following recent advances in causal inference methodology, we examined measurement error structures via directed acyclic graph principles to understand the causal implications of psychometric findings. Both instruments demonstrated good psychometric properties and construct validity. The three-factor GHQ-12 structure (dysphoria, social dysfunction, loss of confidence) showed an optimal fit (comparative fit index = 0.996, Tucker‒Lewis index = 0.992, root mean square error of approximation = 0.025). Dysphoria exhibited the strongest correlation with the PHQ-2 (r = 0.65) and highest temporal stability (0.60), whereas social dysfunction showed the lowest stability (0.48), suggesting differential sensitivity to environmental interventions. Evidence of a response shift emerged: while configural and metric invariance held across time, scalar invariance was violated (change in comparative fit index = 0.105), indicating systematic changes in item thresholds rather than true mental health changes. The GHQ-12 and PHQ-2 are reliable instruments for longitudinal mental health assessment in poor urban populations. However, scalar non-invariance suggests that residents recalibrate their mental health standards while urban regeneration begins, which has important implications for interpreting intervention effects. Future studies should incorporate measurement invariance testing and latent variable approaches when evaluating complex environmental interventions.