Bivariate transition model for analysing ordinal and nominal categorical responses: an application to the Labour Force Survey data

In many panel studies, bivariate ordinal--nominal responses are measured and the aim is to investigate the effects of explanatory variables on these responses. A regression analysis for these types of data must allow for the correlation among responses of the same individual. To analyse such ordinal--nominal responses using a proper weighting approach, an ordinal--nominal bivariate transition model is proposed and maximum likelihood is used to find the parameter estimates. We propose a method in which the likelihood function can be partitioned to make possible the use of existing software. The approach is applied to the Labour Force Survey data in Iran, where the ordinal response, at the first period, is the duration of unemployment for unemployed people and the nominal response, in the second period, is economic activity status of these individuals. The interest is to find the reasons for staying unemployed or moving to another status of economic activity.