The factor structure of the twelve item General Health Questionnaire (GHQ-12): the result of negative phrasing?

Global self-esteem based on M. Rosenberg's (1965) scale is typically treated as a unidimensional scale. However, factor analyses suggest separate factors associated with positively and negatively worded items, and there is an ongoing debate about the substantive meaningfulness of this distinction. Confirmatory factor analysis (CFA) was used to evaluate alternative 1- and 2-factor models and to test hypotheses about how the factors vary with reading ability and age. Responses based on the National Longitudinal Study of 1988 (S.J. Ingles et al., 1992) reflected a relatively unidimensional factor and method effects associated with negatively worded items. Such effects are common in rating scale responses, and this CFA approach may be useful in evaluating whether factors associated with positively and negatively worded items are substantively meaningful or artifactors.

Different versions of the General Health Questionnaire (GHQ), including the GHQ-12 and GHQ-28 have been subjected to factor analysis in a variety of countries. The World Health Organization study of psychological disorders in general health care offered the opportunity to investigate the factor structure of both GHQ versions in 15 different centres. The factor structures of the GHQ-12 and GHQ-28 extracted by principal component analysis were compared in participating centres. The GHQ-12 was completed by 26,120 patients and 5,273 patients completed the GHQ-28. The factor structure of the GHQ-28 found in Manchester in this study was compared with that found in the earlier study in 1979. For the GHQ-12, substantial factor variation between centres was found. After rotation, two factors expressing depression and social dysfunction could be identified. For the GHQ-28, factor variance was less. In general, the original C (social dysfunction) and D (depression) scales of the GHQ-28 were more stable than the A (somatic symptoms) and B (anxiety) scales. Multiple cross-loadings occurred in both versions of the GHQ suggesting correlation of the extracted factors. In Manchester, the factor structure of the GHQ had changed since its development. Validity as a case detector was not affected by factor variance. These findings confirm that despite factor variation for the GHQ-12, two domains, depression and social dysfunction, appear across the 15 centres. In the scaled GHQ-28, two of the scales were remarkably robust between the centres. The cross-correlation between the other two subscales, probably reflects the strength of the relationship between anxiety and somatic symptoms existing in different locations.