B0930
Title: Challenging the assumption of consistent responding behavior over time through a Markov switching model
Authors: Roberto Colombi - University of Bergamo (Italy)
Sabrina Giordano - University of Calabria (Italy) [presenting]
Abstract: When assessing attitudes or perceptions using a Likert scale, respondents often tend to behave according to a response style, by selecting the midpoint or extremes or the agreement or disagreement sides of the scale, regardless of the content. These response behaviors can introduce bias in estimates and misleading results. Recognizing the importance of response styles, the aim is to account for their time evolution, challenging the assumption of consistent answering behavior over time. To achieve this, a Markov switching model, driven by a bivariate latent Markov chain, is proposed for longitudinal rating data. At each time occasion, respondents are assumed to answer either based on a response style or by appropriately using the rating scale to accurately represent their own feelings. The model incorporates one binary variable to account for two answering regimes (RS or not) and another latent variable that captures dynamics and respondent-specific unobserved heterogeneity. For the observable ordinal variables, their distribution is specified under the two RS regimes. Under the RS regime, a highly flexible two-parameter distribution is used to accommodate all the possible response styles. In contrast, under the noRS regime, a stereotype logit model is employed. Finally, an application of the model is provided for real data concerning respondent perceptions.
