CESS 2018

All countries that deliver Labour Force Survey data to Eurostat make use of a rotational pattern that results in annual overlaps of data, and with the exception of Germany, also overlaps of quarterly data. Rotational patterns are generally used in sample surveys to reduce the variances of estimators. For the EU-Labour Force Survey (LFS), this longitudinal feature can be exploited to produce highly demanded estimates of labour market transitions between the status employed, unemployed and economically inactive. When defining the longitudinal sample, Eurostat has to rely on the available overlapping data; this means, that information on individuals who drop out of the survey before completing all interviews is lost. The individuals dropping out may have moved away, it may have been impossible to contact them, they may have refused to answer, or they may have died. Eurostat does not have information on the reason why individuals drop out,nor the information that would be necessary to fill or estimate the missing labour market data for those individuals. If those individuals that drop out are on average no different from those that stay in the sample, this attrition will only lead to less precise estimates. If, however, these individuals differ with regard to their labour market status from the remaining sample, estimates based on this sample would not be representative. In short, transition statistics derived by Eurostat using LFS data may possibly suffer from attrition bias. In this paper, simple binary regression is used to determine whether attrition bias may be a problem, and simple simulations are used to estimate the potential size effect differential attrition might have on the estimates.