The Dimensionality of the 12-Item General Health Questionnaire (GHQ-12): Comparisons of Factor Structures and Invariance Across Samples and Time

• Main Authors: Sigurd W. Hystad, Bjørn Helge Johnsen
• Format: Article
• Language: English
• Published: Frontiers Media S.A. 2020-06-01
• Series: Frontiers in Psychology
• Subjects: General Health Questionnaire (GHQ-12); confirmatory factor analysis (CFA); bifactor models; wording effects; factor structure; measurement invariance
• Online Access: https://www.frontiersin.org/article/10.3389/fpsyg.2020.01300/full

Because of its brevity, the 12-item General Health Questionnaire (GHQ-12) has become one of the most popular and used measure for detecting psychological distress. Originally intended as a unidimensional measure, the majority of subsequent factor-analytic studies have failed to support GHQ-12 as a unitary construct and have instead proposed a plethora of multidimensional structures. In this study, we further examined the factor structure in two different military samples, one consisting of crewmembers from four different frigates deployed in anti-piracy operations and Standing NATO Maritime Group deployments (N = 591) and one consisting of crewmember from three different minehunters/sweepers serving in Standing NATO Mine Counter-Measures Group deployments (N = 196). Results from confirmatory factor analyses (CFA) performed in the first sample supported a bifactor model, consisting of a general factor representing communality among all items and two specific factors reflecting common variance due to wording effects (negatively and positively phrased items). A multi-group CFA further confirmed this structure to be invariant across our second sample. Structural equation modeling also showed that the general factor was strongly associated with symptoms of insomnia and mental health, whereas the specific factors were either non-significantly or considerably weaker associated with the criterion variables. Overall, our results are congruent with the notion that the multidimensionality demonstrated in many previous investigations is most likely an expression of method-specific variance caused by item wording. The explained unique variance associated with these specific factors was further relatively small. Ignoring the multidimensionality and treating GHQ-12 as a unitary construct will therefore most likely introduce minimal bias to most practical applications.