The structure of the General Health Questionnaire (GHQ-12): two meta-analytic factor analyses

ABSTRACT

The General Health Questionnaire (GHQ-12) is a popular measure of psychological distress. Despite its widespread use, an ongoing controversy pertains to its internal structure. Although the GHQ-12 was originally constructed to capture a unitary construct, empirical studies identified different factor structures. Therefore, this study examined the dimensionality of the GHQ-12 in two independent meta-analyses. The first meta-analysis used summary data published in 38 primary studies (total N = 76,473). Meta-analytic exploratory factor analyses identified two factors formed by negatively and positively worded items. The second meta-analysis included individual responses of 410,640 participants from 84 independent samples. Meta-analytic confirmatory factor analyses corroborated the two-dimensional structure of the GHQ-12. However, bifactor modelling showed that most of the variance was explained by a general factor. Therefore, subscale scores reflected rather limited unique variance. Overall, the two meta-analyses demonstrated that the GHQ-12 is essentially unidimensional. It is not recommended to use and interpret subscale scores because they primarily reflect general mental health rather than distinct constructs.

KEYWORDS:

• Mental health
• distress
• factor analysis
• meta-analysis
• wording effects

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Timo Gnambs http://orcid.org/0000-0002-6984-1276

Notes

1 Random-effects meta-analyses using a small number of samples can result in unstable estimates of between-studies variances (Borenstein, Hedges, Higgins, & Rothstein, 2010). Accordingly, for some subgroup analyses respective random-effects model did not converge and did not give meaningful heterogeneity estimates for several pooled correlations. Therefore, subgroup analyses pertaining to correlation matrices as effects sizes as well as different language versions were based on a fixed-effects model, whereas all other analyses adopted a random-effects model.