Methods for addressing selection bias in non-probability samples via data integration
Conference
65th ISI World Statistics Congress
Format: IPS Abstract - WSC 2025
Keywords: data integration, nonprobability sample
Abstract
Statistical agencies are exploring new data sources for producing more timely and detailed statistics, while reducing costs and respondent burden. These data sources may include administrative records, big data such as bank transactions, and non-probability surveys such as online web panels. Selection bias and measurement error are two issues that may be present in these types of data. These errors may be corrected using available auxiliary information relating to the population of interest, such as from a census or reference probability sample.
In this presentation, a potpourri of work undertaken in the ABS to address selection bias in non-probability samples will be shared. The data scenario we are interested in is one where we have a non-probability sample collecting data for an item of interest as well as additional auxiliary variables. An existing probability sample can be made available for the auxiliary information, and we assume it is possible to identify the overlap between the two samples. Within this context, we have looked at a new method for estimating participation probabilities for the non-probability sample comparing the efficiency of this approach compared with several alternatives.