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

Background The 12-item General Health Questionnaire (GHQ-12) is used routinely as a unidimensional measure of psychological morbidity. Many factor-analytic studies have reported that the GHQ-12 has two or three dimensions, threatening its validity. It is possible that these 'dimensions' are the result of the wording of the GHQ-12, namely its division into positively phrased (PP) and negatively phrased (NP) statements about mood states. Such 'method effects' introduce response bias which should be taken into account when deriving and interpreting factors. Methods GHQ-12 data were obtained from the 2004 cohort of the Health Survey for England (N = 3705). Following exploratory factor analysis (EFA), the goodness of fit indices of one, two and three factor models were compared with those of a unidimensional model specifying response bias on the NP items, using structural equation modelling (SEM). The hypotheses were (1) the variance of the responses would be significantly higher for NP items than for PP items because of response bias, and (2) that the modelling of response bias would provide the best fit for the data. Results Consistent with previous reports, EFA suggested a two-factor solution dividing the items into NP and PP items. The variance of responses to the NP items was substantially and significantly higher than for the PP items. The model incorporating response bias was the best fit for the data on all indices (RMSEA = 0.068, 90%CL = 0.064, 0.073). Analysis of the frequency of responses suggests that the response bias derives from the ambiguity of the response options for the absence of negative mood states. Conclusion The data are consistent with the GHQ-12 being a unidimensional scale with a substantial degree of response bias for the negatively phrased items. Studies that report the GHQ-12 as multidimensional without taking this response bias into account risk interpreting the artefactual factor structure as denoting 'real' constructs, committing the methodological error of reification. Although the GHQ-12 seems unidimensional as intended, the presence of such a large response bias should be taken into account in the analysis of GHQ-12 data