Analysing conjoint data with OLS and PLS regression: a case study with wine

Abstract

This paper presents a case study with wine where two statistical methods for the analysis of rating-based conjoint analysis data were applied. Traditionally, ordinary least squares (OLS) regression is used to estimate the relative importance of the experimental factors and the part-worth utilities of factor levels. Partial least squares (PLS) regression, which is a popular tool in sensory and consumer science, can also be used for the analysis of interval-level conjoint data. Using conjoint analysis, purchase intentions for Californian red and white wine were obtained from a convenience sample of young US adults (n ≈ 250). OLS and PLS regression uncovered the same systematic patterns in the data: negative utility associated with more expensive wine, and positive utility associated with famous wine regions. While OLS regression provided more accessible top-line results, an advantage of PLS regression was the graphical format of results. This provided easy insight to individual differences in the importance attached to the factors driving purchase intention. OLS and PLS regression can complement each other in the analysis of interval-level conjoint data. Dual analysis can help to ensure that the right insights are drawn from the study and communicated to internal/external clients. It may also facilitate communication within project teams.