Construct validity and internal consistency of three factor structures and two scoring methods of the 12-item General Health Questionnaire

Abstract

INTRODUCTION: The analysis of the psychometric properties of an instrument is important for the control of measurement bias. OBJECTIVE: To compare the construct validity and internal consistency of three factorial structures of the General Health Questionnaire-12. MATERIALS AND METHODS: We conducted a validation study in a nested sample of 483 dentistry students from Cartagena, Colombia. The instrument was applied along with questions about personal factors. The construct validity of the following structures was evaluated: one-dimensional, one-dimensional with correction of response bias, and the two and three-factor structure, and these scoring systems: Likert scale (0-1-2-3) and dichotomous (0-0-1-1) scoring method. Validity was determined by confirmatory factorial analysis, and the internal consistency with Cronbach's alpha coefficient for the Likert scale and the Kuder-Richardson Formula 20 for the dichotomous scoring using the Stata™, v. 13.2 (Stata Corp, USA) and Mplus™, v.7.11 (Muthen and Muthen, USA) software. RESULTS: Adjustment indexes revealed that the best model was the one-dimensional one with correction of response bias based on a dichotomous scoring (degrees of freedom=36; chi square=52.432; root mean square error of approximation=0.03; 90% CI: 0.008-0.048; comparative fit index=0.982; Tucker-Lewis index=0.966). The internal consistency was 0.70. CONCLUSIONS: The adjustment of the model in this study allowed for the control of measurement bias and guaranteed external validity of results when using the General Health Questionnaire-12 to evaluate mental conditions in this population. The psychometric properties of the scales should be critically evaluated before results analysis.