CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: A family of models for multivariate rating scale data accounting for response styles Authors:  Sabrina Giordano - University of Calabria (Italy) [presenting]
Abstract: Rating scales have been widely used to investigate attitudes and opinions in sociological and psychological contexts. However, observed ratings may not represent the true opinion. In fact, it is generally observed that unaware respondents may use only a few of the given options irrespective of their real opinion. Someone has a strong tendency to mark the endpoints, others take shelter in the middle category, someone else responds with agreement(disagreement) regardless of item content, optimists (pessimists) may prefer the positive (negative) side of the scale. These behaviours are typically considered as response styles. A recently proposed family of models for multivariate rating scale data distinguishes aware from unaware responses. It directly and explicitly models the response styles by specifying the distribution of the unaware responses. We present an alternative approach defining the unaware distribution as a transformation of the distribution of the aware responses. To this aim, we use a known adjacent categories model accounting for response styles. Moreover, we aim also at investigating the influence of covariates on response styles. Competing models are compared by the Vuong test for misspecified non-nested models as the distributions that model response styles are only an approximation of the true mechanism generating unaware responses.