Introduction Born in Bradford is a large multi-ethnic birth cohort drawn from the North of England. As background to comparing mental health across our sample we conducted a confirmatory factor analysis (CFA) and invariance testing for each of the four-factor seven-item subscales of the GHQ-28 for a bi-ethnic subpopulation.

Methods We examined data from White (N=2011) and Pakistani-origin (N=1516) women who responded to the GHQ-28 in English. We fitted a single factor model to data from each subscale to the White subpopulation and tested for invariance against the Pakistani women's data. Where configural invariance was indicated for a proposed factor we followed a model generation strategy for each subpopulation and tested for invariance on this reduced item set.

Results In general, the models gave poor fit. Fit was acceptable when we reduced the number of items from seven to four for the Somatic, Anxiety/Insomnia and Severe Depression subscales. The unmodified Social Dysfunction subscale fitted both groups adequately. After correcting for differential item functioning (Item 4, Felt ill) we found Pakistani women were more likely to have worse scores on the reduced Somatic subscale (d=0.10, p<0.001). They also had worse scores on the unmodified Social Dysfunction (d=0.03, p<0.001) and reduced Severe Depression subscales (d=0.14 p<0.001). Severe fit difficulties were observed for the Anxiety/Insomnia subscale.

Conclusion The unmodified GHQ -28 subscales did not fit a CFA model well and modifications were needed to ensure measurement equivalence. Pakistani women in this preliminary dataset fared significantly worse on all measureable subscales than White women.