Statistical modelling of vignette data in psychology

Area of Science:

• Psychology and Social Sciences
• Quantitative Research Methods

Background:

• Vignette methods are prevalent for scenario-based data collection in social sciences.
• Quantitative data from vignette studies pose statistical modeling challenges due to design intricacies.

Purpose of the Study:

• To review challenges in statistical modeling for vignette studies.
• To recommend efficient design and data allocation strategies for vignette research.
• To propose multilevel models for handling complex vignette study designs.

Main Methods:

• Distinguishing between factorial survey experiments and experimental vignette studies based on vignette allocation.
• Identifying potential issues like fractional or unbalanced designs and unmodeled heterogeneity.
• Applying multilevel models to nested and crossed designs, including unbalanced and fractional data.

Main Results:

• Factorial survey experiments result in nested designs (single random factor).
• Experimental vignette studies can lead to crossed random factor designs.
• Unmodeled heterogeneity and unbalanced designs necessitate advanced statistical approaches.

Conclusions:

• Multilevel models provide a unified framework for analyzing diverse vignette study designs.
• Proper design and statistical modeling are crucial for valid inferences in vignette research.
• The proposed approach enhances the analysis of complex vignette data, as demonstrated with a bullying and harassment study.