Correcting Selectivity of Non-Probability Samples by Means of Sample Matching
Roel C. A. Huijskens · Utrecht University Repository (Utrecht University) · 2020
The increase in non-response and costs of employing probability-based surveys has motivated national statistical offices and other statistical agencies to investigate new potential sources of data for statistical inferences. However, many promising new sources of data are not obtained based on a predefined sampling design (non-probability samples), and as a consequence, selectivity correction methods are needed to obtain unbiased estimates. This thesis investigates the performance of sample matching as a bias correction method for non-probability samples through a simulation study, both using simulated and real data. The benefits of using this method over more traditional bias correction methods are the ease of use, little reliance on complex modeling assumptions, and the fact that no information on population totals is required. Further, we investigated what the effects were of sample sizes, matching methods, and sampling methods on the performance of sample matching. Results showed that sample matching is a promising bias correction method and that using a simple random sample provides the best results in terms of bias and variance.