Attitudes to gender roles: Modelling repeated ordered categorical data subject to dropout

Objectives // We analyze dropout and response from 17 waves of the British Household Panel Study over the period 1991 to 2007. The dependent variables examined are attitudinal and take the form of ordinal responses to a series of questions concerning gender roles. The data consist of repeated measurements over time from a set of individuals and is therefore a longitudinal panel subject to dropout. In order to take account of the repeated measures aspect of the data, we use random effects models to capture the associated unobserved heterogeneity. // Method/Models //
Univariate models for dropout and single attitudinal responses are fitted using binary logit and ordered logit random effects respectively. We also model jointly both dropout and response processes, and pairs of responses, as bivariate random effects models. Such joint models are not routinely available in standard software and this has led to further development of the software package SABRE. The model specifications themselves consist of both individual and household characteristics plus, for dropout, variables measuring the interviewers' level of access to the respondents in previous waves. // Results and Conclusions // We find that gender and age are amongst the most important determinants of gender roles, and that attitudes within society as a whole became generally more egalitarian over the period of study. The collection of attitudinal outcomes gives a largely consistent set of estimates and the correlations between the random effects in the joint response models confirm the traditionalist/egalitarian orientation of respondents towards gender roles.