Booster samples of marginal groups vs separate focussed studies

Abstract

Executive summary

Recent longitudinal study designs have had samples which are sufficiently large to support a range of subgroup analyses. But for very specialised subpopulations, the sample sizes are too small. The design can be adjusted to over‐represent these cases, but this has an impact on general analysis, particularly by increasing variability in the weights. It seems methodologically preferable to use standalone studies in these cases, though these would be vulnerable to funding changes.

Small studies should be harmonised as far as possible with the main panel/cohort, to facilitate combined analysis. How to combine samples potentially sampled in quite different ways to make the best use of the data is an open question needing further research.

Boost samples could be employed regularly in the main studies to replace sample losses from attrition and to increase coverage of subpopulations of particular interest. How to manage analysis in such a dynamic system also needs to be addressed, but it is not so far removed from a rotating panel design.
Coordinated sampling might help to make fieldwork procedures more efficient.

A small‐scale comparison of probability and non‐probability sampling approaches for marginal groups in a UK context would be valuable; it should contain at least two waves to investigate the impact on sample retention.

More information

Identifiers

ORCID for Paul A. Smith: orcid.org/0000-0001-5337-2746

Catalogue record

Export record