Comparison of statistical analysis methods for object case best-worst scaling.

Area of Science:

• Health Technology Assessment (HTA)
• Health Economics
• Decision Analysis

Background:

• Object case best-worst scaling (BWS) is a valuable tool for preference elicitation.
• Various statistical methods exist for analyzing BWS data, each with potential strengths and weaknesses.
• Understanding the most appropriate analysis method is crucial for accurate interpretation of BWS results.

Purpose of the Study:

• To compare common statistical analysis methods for object case BWS.
• To evaluate the advantages and limitations of count analysis, multinomial logit, mixed logit, latent class analysis, and hierarchical Bayes estimation.
• To assess the impact of different methods on ranking results and identifying respondent heterogeneity.

Main Methods:

• Analysis of BWS data using five distinct statistical methods: count analysis, multinomial logit, mixed logit, latent class analysis, and hierarchical Bayes estimation.
• Comparison of ranking outcomes across the employed methods.
• Application of a case study involving 136 policymakers and HTA experts assessing barriers to HTA usage.

Main Results:

• All five statistical methods produced broadly similar rankings, especially for the most and least important factors.
• Latent class analysis successfully identified five distinct respondent segments.
• Mixed logit models indicated significant preference heterogeneity among respondents for most factors.

Conclusions:

• Standard statistical methods for object case BWS provide consistent factor rankings.
• Count analysis serves as a reliable and straightforward initial approach for preference elicitation.
• Latent class and mixed logit models offer enhanced insights by revealing underlying segments and individual preference variations.