CMStatistics 2017: Start Registration
View Submission - CMStatistics
B0528
Title: Modelling multiple Likert items through an adjacent categories ordinal paired comparison model Authors:  Brian Francis - Lancaster University (United Kingdom) [presenting]
Regina Dittrich - WU Vienna (Austria)
Abstract: The modelling of multiple Likert items measured on the same Likert scale can be challenging. We review existing methodologies, which can be classified into two approaches - joint modelling of the absolute Likert ratings, and modelling of the relative differences between Likert Items. One approach for modelling relative differences has been suggested, which led to a paired comparison model with three outcomes for each pair of Likert items the first item preferred, the second item preferred and no preference. Such a model can be fitted as a Poisson log-linear model, with each item placed on a ``worth'' (0,1) scale; in addition, covariates can be included. That model is extended to allow for a distance metric between the pairs of items. This leads to an adjacent categories ordinal paired comparison model with ties. This model makes better use of the underlying Likert rating scale, and we compare the two approaches through worth plots. Numerical issues and implementation in R are both discussed, together with typical restrictions on the size of the problem that can be dealt with. Future work will be discussed.