Platt L. Mobility and missing data: what difference does non-response make to observed patterns of intergenerational class mobility by ethnic group? Working Paper 2005-10 of the Institute of Economic and Social Research, University of Essex, Colchester. 2005.
While there is an extensive literature on intergenerational class mobility in Britain there is less work on the impact of these different forms of missing data on the observed transitions between parent’s class and child’s class. It may be particularly pertinent to consider the impact of missing data when evaluating ethnic group differences in intergenerational mobility, since unemployment is known to differ systematically with ethnicity, as do survey and census response. Research into intergenerational class transitions as they differ by ethnic group in Britain is the subject of only a small number of studies, and there has not been any detailed examination of the impact of missing data in studying ethnic group differences in intergenerational social mobility. The contribution of this paper is to highlight the issue of missing data in studying social mobility, to outline its potential implications, and to offer some preliminary analysis of the impact of missing data on our understanding of intergenerational social mobility as it varies by ethnic group.
The paper examines three forms of missing data that we might expect to effect the measurement of ethnic group differences in intergenerational social mobility. First, where measured origins or destinations are not accommodated by class categories – e.g. the unemployed; second, those whose origins are measured but who do not survive in the study sample, because of non-response or emigration; and third, those for whom information on variables of interest that are, or might be, associated with achieved social class is missing.
These three forms of missing data are explored by (a) examining the impact of operating different assumptions in relation to non-class outcomes, particularly unemployment but also ‘other’ outcomes; (b) investigating non-response at the point of measurement of destinations, by exploring and comparing the characteristics of those for whom we know the reason for non-response (emigration, recorded through embarkations data) and those for whom we do not know the cause of their non-response. This section of the paper goes on to look at whether there are selection effects in observed patterns of origins and destinations. In the following section, the paper examines (c) whether the treatment of missing values on variables included in models of emigration, attrition and social class outcomes affect the results obtained. Here just two of the potential treatments of missing data, listwise deletion and using dummy variables for missing values are contrasted.
The paper concludes that, despite big differences between ethnic groups in levels of non-class outcomes, emigration and non-response, the impact of these on results as far as it was investigated here was modest. Nevertheless, missing data remains an issue potentially highly pertinent for ethnically differentiated mobility research, and thus, it is argued, being selfconscious about how missing data are treated is important in itself.