Analyzing consumers’ Profile of Mood States (POMS) data using the proportional odds model (POM) for clustered or repeated observations and R package ‘repolr’

Abstract

Consumers’ affective states associated with product development are hot topics and an important area of consumer research. The Profile of Mood States 2nd EditionTM – Adult (POMS 2-A™) with 65 items and 5 negative scales can be used to assess adult consumers’ affect. Consumer POMS data collected in the US for the NUTRILITE™ brand multivitamin multimineral supplement product are analyzed using an advanced statistical model and the R package ‘repolr’. This proportional odds model (POM) fitted with generalized estimating equations (GEE) is more appropriate for repeated ordinal categorical data than conventional ANOVA. Analysis of the data collected before product testing show that there are significant user group effects for each of the 7 POMS scales and significant gender and age group effects for some scales. Product users, females and age group 45–55 show larger scores for positive scales and smaller scores for the negative scales than non-users, males and other age groups. Analysis of the data collected after product testing shows that there are significant product effects for some scales of POMS (Depression-Dejection, Anger-Hostility, and Total Mood Disturbance). The estimated effect parameters in the model, i.e., log odds ratio values, can be transformed into Thurstonian discriminal distance δ or d-prime and other sensory measures, offering all the advantages of Thurstonian modeling and ease of interpretation of the results. Further analysis shows that although the results using conventional ANOVA are consistent with the results using the advanced POM model, the POM model for the POMS data is more valid and powerful than conventional ANOVA.