Comparative analysis of data reduction techniques for questionnaire validation using self-reported driver behaviors.

Area of Science:

• Psychometrics
• Data Analysis
• Transportation Psychology

Background:

• Factor Analysis (FA) and Principal Component Analysis (PCA) are commonly applied to ordinal data, despite being designed for metric measures.
• Questionnaire validation, particularly with Likert Scale data, requires appropriate statistical techniques.

Purpose of the Study:

• To compare the effectiveness of FA, PCA, and Categorical Principal Component Analysis (CATPCA) for validating the Driver Behavior Questionnaire (DBQ) using Likert Scale data.
• To assess which technique best explains variance and represents the original data structure.

Main Methods:

• An e-survey collected self-reported behaviors from 1,354 Brazilian and 348 Portuguese drivers using 20 DBQ items on a five-point Likert Scale.
• Exploratory data reduction techniques including FA, PCA, and CATPCA were applied to the collected data.

Main Results:

• All techniques confirmed the established three-factor structure of the DBQ (errors, ordinary violations, aggressive violations).
• CATPCA explained more variance (11% more than FA, 2% more than PCA) and showed significant differences in individual factor scores.
• While one item shifted dimensions in CATPCA, structural interpretability was maintained.

Conclusions:

• CATPCA is more advantageous for analyzing Likert Scale data due to its ability to better represent original data and account for data constraints.
• The findings suggest CATPCA is a more suitable method for DBQ validation, with potential applications in crash prediction models using derived factor scores.