Proportional Odds Model ( POM ) for Analyzing Rate‐All‐That‐Apply ( RATA ) Data: Detecting Sample Effects

Rate‐all‐that‐apply (RATA) data, as an extension of check‐all‐that‐apply (CATA) data, are being used more frequently in sensory and consumer research. How to analyze RATA data is of theoretical and practical importance. This paper proposes and advocates the proportional odds model (POM) or the proportional odds mixed model (POMM) as a standard first‐selected model for the analysis of RATA data. Using a food‐related case study, it is shown that POM or POMM are more appropriate and useful statistical models than the conventional linear model (LM) or linear mixed model (LMM) for RATA data. Log of odds‐ratio in POM or POMM is a more meaningful index to measure sample effects and sensory differences between samples for RATA data than the difference of rating means in LM or LMM. Generalized linear models (GLM) and generalized linear mixed models (GLMM) with binomial distribution are also discussed for when the proportional odds assumption fails.