Weighting Strategy for Understanding Society
Weighting Strategy for Understanding Society
Key takeaways
Bibliography: Lynn, P., Kaminska, O., 2010. Weighting Strategy for Understanding Society. Institute for Social and Economic Research.
Authors:: Peter Lynn, Olena Kaminska
Collections:: UCL UKHLS Dump
First-page: -1
This paper outlines the strategy for the development and provision of analysis weights for Understanding Society. The strategy is placed within the context of a number of practical and statistical issues that arise when a survey has such a complex design involving multiple waves, samples, instruments and analysis bases. A fundamental issue considered is for which data subsets weights should be produced, where a subset may be defined by a combination of waves, instruments and samples. Secondary issues concern the auxiliary data for non-response adjustment and post-stratification, the modelling methods, and the subsequent weight calculation methods. There are also important considerations relating to documentation, access and user support.
content: "@lynnWeightingStrategyUnderstanding2010" -file:@lynnWeightingStrategyUnderstanding2010
Reading notes
Imported on 2024-06-26 11:23
⭐ Important
- & When analysing survey data, the analyst must decide what “weight” to give each unit (e.g. person or household) in the sample. A simple solution is to give equal weight to each sample unit, but this is not always appropriate. Certain types of units may be under-represented, in which case they should be given a larger weight, in order to make the (weighted) sample representative of the population. (p. -1)
- & he definition of unit respondents is clearly crucial. If the definition is too broad, then many units with weights will in fact be excluded from the analysis. The weighting is then clearly suboptimal. But if the definition is too narrow, then cases that could have been used in estimation will be excluded as they do not have weights. There is a trade-off to be made. (p. 1)
- & Weights aim to reduce bias caused by under-coverage, sampling or non-response. There is a strong user demand for weights. Much of the information required for weight calculation is not available to users. (p. 6)
- & Design weights are needed in order to deal with differences in selection probabilities. (p. 7)
- & Design weights should be adjusted to reflect different propensities of selected households to participate. (p. 7)
- & We note that the BHPS sample does not represent immigrants to the UK since 1991, so such immigrants in the UKHLS-GPS sample would need to be identified and up-weighted to compensate. (p. 9)
- & Ex-BHPS sample: This sample already has wave 18 BHPS longitudinal weights, which are post-stratified design weights adjusted for successive waves of non-response. This weight could form the base weight for weighting from wave 2 onwards or we could return to the wave 1 (1991) base weight and use a new adjustment strategy. (p. 13)