Title: A new approach for analyzing response style data
Authors: Yu-Wei Chang - Department of Statistics, Feng Chia University (Taiwan) [presenting]
Abstract: In survey questionnaires studies, latent factors underlying Likert-type data are often the quantities of interest. Many empirical studies have shown that response style is an issue in survey questionnaires. For example, people from East Asia are more conservative than people from Latin America, and the former are less likely to answer ``very agree'' or ``very disagree'' in survey questionnaires, given that they have the same latent factors. Multiple-group factor analysis (M-FA) models are one of the common practice for taking the group difference into account so that we could have better estimates for the latent factors. However, the Likert-type data are treated as continuous in M-FA models. To better accommodate the ordinal categorical feature of the data, we suggest using Multiple-group categorical confirmatory factor analysis (MC-FA) models for the response style modeling. As other multiple-group analysis, we need one or some anchor item(s) for a fair comparison between groups. More specifically, before fitting a multiple-group model, we have to find one or some item(s) which functions the same between groups, and then parameters of the item(s) are restricted to be the same between groups. We provide a strategy for the anchor selection while MC-FA models are applied to the response style data. A justification of the procedure will also be given. Finally, the proposed procedure is applied to a real data set for illustration.