It is common practice for surveys to collect income data using income bands especially if income information is not the main focus of the survey. This is done for various reasons including reducing respondent burden. Longitudinal studies have a wealth of information on the same individual over time. It is often attractive to use income data beyond the tabulations of income bands with other categorical variables. The studying of income poverty or income dynamics; where percentile or quintile distribution of income is of interest, are examples of situations where continuous income is preferred to banded income. At each wave of all the four waves of the Millennium Cohort Study (MCS), banded family income data was collected. In this paper we first investigate longitudinal predictors of income item non response, and we use the predictors to impute missing data due to item non response, and to predict continuous income conditional on the observed bands. We use interval regression and multiple imputations for the imputation and prediction of continuous income. Secondly, for the purpose of studying child income poverty and family income dynamics, we compare the imputed income distributions to the commonly used band midpoint distribution.
Presented by:
Sos Ketende, Centre for Longitudinal Studies, Institute of Education, University of London
Date & time:
November 17, 2010 1:00 pm - November 17, 2010 2:00 pm
Internal seminars home