Abstract

In recent years, non-probability data, i.e. data not collected by means of a known and well-designed sampling mechanism such as administrative data and big data, are more and more used for producing official statistics. In some cases official statistics are based on non-probability data solely, in other cases official statistics are based on a combination of non-probability data and traditional sample survey data. When official statistics are (partly) based on non-probability data, assessment of the quality of estimates for parameters of interest is usually much more complicated than when such estimates are based on sample survey data only. For instance, administrative data and big data are sometimes based on a selective part of the population, which may lead to selection bias in estimates for parameters of interest. Also, when estimates are based on several datasets – either non-probability datasets or survey samples – in which the same target variable is measured, measurement error in this variable may become apparent and needs to be assessed. In this talk we will discuss measuring the quality of estimates (partly) based on non-probability data. We will focus on assessing selection bias in practical situations and on assessing the effect of measurement error on the quality of estimates.