Integrating Probability and Nonprobability Samples for Survey Inference
Presenter: Arkadiusz Wisniowski - University of Manchester
Authors: Arkadiusz Wisniowski, Joseph W. Sakshaug, Diego Andres Perez Ruiz, and Annelies G. Blom
Survey data collection costs have risen to a point where many survey researchers and polling companies are abandoning large, expensive probability-based samples in favour of less expensive nonprobability samples. The empirical literature suggests this strategy may be unwise for many reasons, amongst them probability samples tend to outperform nonprobability samples on accuracy when assessed against population benchmarks. However, nonprobability samples are often preferred due to convenience and cost effectiveness. Instead of forgoing probability sampling entirely, we propose a method of combining both probability and nonprobability samples in a way that exploits their strengths to overcome their weaknesses within a Bayesian inferential framework. By using simulated data, we evaluate supplementing inference based on small probability samples with prior distributions derived from nonprobability data. We demonstrate that informative priors based on nonprobability data can lead to reductions in variances and mean-squared errors of the linear model coefficients. The method is also illustrated with actual probability and nonprobability survey data. A discussion of these findings, their implications for survey practice, and possible research extensions are provided in conclusion.
Techniques for randomising protest survey samples to facilitate comparison with nationally representative samples.
Presenter: Clare Saunders - University of Exeter
Authors: Clare Saunders and Natalie Shlomo
In this paper, we compare two techniques for balancing non-random protest survey samples with random European Social Survey (ESS) samples of protesters. These techniques will become increasingly important in the study of protesters because protesters are a ‘rare’ sample. The power for analysis is very low if we rely on random samples of protesters alone. Therefore, it is important to find ways in which to randomise data from other data sources to provide rich insights into protesters. The two techniques compared are proportional weighting and sample matching. Proportional weighting involves post-stratification bench-marking based on UK census data. Sample matching is an approach modified from quasi-experimental designs. Propensity score matching is used to balance non-random samples. We show that that the sample matching approach to induce randomization of the non-random sample of protesters through the integrations with the ESS has the highest standard of replication of the distribution of variables in a random sample of protesters. We are able to correct for under-selection of centrists and over-selection of left-wingers, especially among the stalwart protesters.
5:00 Opening by Chair: Nick Moon
5:10 Presentation I
5:55 Presentation II
Attendance is free and open to all, whether fellows of the RSS or not, but pre-registration is required.