On testing factorial invariance: A reply to J.C.F. de Winter

Since its publication, the Driver Behavior Questionnaire (DBQ) has been used for comparing subgroups of respondents on the constructs formed through factor analyzing the questionnaire items. However, not enough attention has been paid to ascertaining that the instrument actually measures the same constructs in the same way in all respondent groups. I recently published an article (Mattsson, 2012) that aimed to do this for the Finnish 28-item version of the DBQ using the stage-wise factorial invariance approach in the Exploratory Structural Equation Modeling (ESEM) context. de Winter (2013) commented on the publication, arguing that the results were artifacts due to measurement error that too many factors were extracted and that too strict criteria for invariance were applied. In this contribution, I reply to each criticism and suggest methodological approaches for ensuring the measurement invariance of self-report instruments such as the DBQ.     

Measurement invariance of the Driver Behavior Questionnaire across samples of young drivers from Finland and Ireland

This article investigates the factor structure of the 27-item Driver Behavior Questionnaire (DBQ) in two samples of young drivers (18–25 years of age); one from Finland and the other from Ireland. We compare the two-, three-, and four-factor solutions using Confirmatory Factor Analysis (CFA) and show that the four-factor model (with the latent variables rule violations, aggressive violations, slips and lapses) fits the data from the two countries best. Next, we compare the fit of this model across samples by the means of a measurement invariance analysis in the CFA framework. The analysis shows that the four-factor model fails to fit both samples equally well. This is mainly because the socially-oriented latent variables (rule violations and aggressive violations) are different in nature in the two samples. The cognitively-oriented latent variables (slips and lapses) are, however, similar across countries and the mean values of slips can be compared using latent variable models. However, the common practice of calculating sum scores to represent the four latent DBQ variables and comparing them across subgroups of respondents is unfounded, at least when comparing young respondents from Finland and Ireland.     

Age,gender, mileage and the DBQ: The validity of the Driver Behavior Questionnaire in different driver groups

The Driver Behavior Questionnaire (DBQ) is one of the most widely used instruments for measuring self-reported driving behaviors. Despite the popularity of the DBQ, the applicability of the DBQ in different driver groups has remained mostly unexamined. The present study measured aberrant driving behavior using the original DBQ (Reason, J.T., Manstead, A., Stradling, S.G., Baxter, J., Campbell, K., 1990. Errors and violations on the road – a real distinction. Ergonomics, 33 (10/11), 1315–1332) to test the factorial validity and reliability of the instrument across different subgroups of Danish drivers. The survey was conducted among 11,004 Danish driving license holders of whom 2250 male and 2190 female drivers completed the questionnaire containing background variables and the DBQ. Exploratory and confirmatory factor analysis showed that the original three-factor solution, a four-factor solution and a two-factor solution had acceptable fit when using the whole sample. However, fit indices of these solutions varied across subgroups. The presents study illustrates that both the original DBQ and a Danish four-factor DBQ structure is relatively stable across subgroups, indicating factorial validity and reliability of the DBQ. However, as the Danish DBQ structure has an overall better fit, the present study highlights the importance of performing an explorative analysis when applying the DBQ in order to assess the problem areas within a driving population.     

Using Structural Equation Modeling (SEM) for the Study of Impurity Profiles of Drug Substances

The impurity profile is one of the most important quality characteristics of a drug substance. Although it is always desirable to determine the chemical structure of all impurities forming the impurity profile, unfortunately this is not always economically and technically feasible because of the extremely low concentrations at which some impurities may be found in the drug substance. Therefore, alternative approaches to the chemical analysis are needed for trying to determine the origin of the unidentified impurities.

In a previous study conducted by our group, based on exploratory (principal component and hierarchical cluster) analysis, we were able to suggest a hypothesis for explaining the origin of the unidentified impurities of a drug substance. However, there was still a concern that alternative hypotheses might explain the same phenomenon equally well. This article explores the application of recent developments in structural equation modeling for the systematic generation and selection of hypotheses (models) worthy of being confirmed by chemical research.