Dan Hedlin

Abstract

It is common in surveys to avoid sampling from part of the population for practical or regulatory reasons. This is referred to as cutoff sampling. Although cutoff sampling is widely used in practice, for example in business surveys where small businesses are not usually sampled, there is little research on bias and estimation issues associated with cutoff sampling. Below the cutoff, the data source is usually only administrative data, above it there are both administrative and survey data on the variable of interest.

The aim is to estimate the finite population total, including units below the cut off, which are not sampled. Hence, we have to rely on a strong model. I shall present an approach where the model uncertainty is taken into account.