Gender attitudes: Modelling bivariate repeated ordinal categorical data

Data collected as part of large national longitudinal surveys such as the British Household Panel Study (BHPS) routinely include multiple Likert items which comprise ordered categories. For example, the BHPS includes a range of Likert items which ask respondents about their attitudes towards gender roles. Three such items are:(a) Both the husband and wife should contribute to the household income; (b) A husband’s job is to earn money; a wife’s job is to look after the home and family; (c) Children need a father to be as closely involved in their upbringing as the mother.
These items comprise the five response categories: ‗Strongly Agree‘, ‗Agree‘, ‗Neither Agree Nor Disagree‘, ‗Disagree‘ and ‗Strongly Disagree‘. This talk will explore the extent to which these three items are associated with each other. We will achieve this by adopting a correlated random effects approach. We start by fitting separate models to the three responses. The cumulative logit or proportional odds model will be used to model each ordered response. Residual heterogeneity will be incorporated by using a random effect in all three models. These random effects will be assumed to be independent of each other. We extend this analysis by fitting a trivariate response model with correlated random effects, thereby allowing us to estimate three additional association parameters, and to test whether there is significant association between each possible pair of responses. This trivariate model has been implemented within the software package SABRE (http://www.sabre.lancs.ac.uk), undertaken as part of the ESRC‘s National Centre for Research Methods [Phase 2].