Tuesday 14th July      Wednesday 15th July      Thursday 16th July      Friday 17th July     

Tuesday 14th July, 16:00 - 17:30 Room: HT-102

Weighting issues in complex cross-sectional and longitudinal surveys 2
Convenor	Ms Nicole Watson (University of Melbourne )
Coordinator 1	Dr Olena Kaminska (University of Essex)

Session Details

A range of issues arise when constructing weights for surveys with complex designs. Use of mixed modes or multiple frames in cross-sectional, longitudinal or cross-national surveys may result in uncertainty in selection probabilities, fieldwork outcomes or response propensities that make it difficult to construct appropriate weights. Further, longitudinal surveys tend to have greater uncertainty over time (as interviews are no longer attempted with some people). Growing complexity of design for multi-purpose surveys calls for the development of weighting methods to reflect them.

How should weights be best constructed in the presence of uncertainty about inclusion probabilities or fieldwork outcomes? How should the response process be best modeled in cross-national surveys where countries differ in quality and type of sampling frame and other auxiliary data? For surveys with multiple frames, how do we best construct weights that combine samples from multiple sources that may have partial overlap in the presence of uncertainty about membership? In constructing weights for longitudinal samples, we need to consider how populations are defined over time, how to treat deaths and other out-of-scopes, how best to adjust for attrition. Further, in household-based longitudinal surveys we need to determine how to best incorporate new sample members arising from changes in the household structure.

This session seeks to bring together survey methodologists involved in constructing weights for complex surveys (both longitudinal and cross-sectional) to explore the approaches taken. Papers submitted to this session might include comparisons of alternative methods, analysis of the impact of a particular component of the weights, or suggestions for new methods.

Paper Details

1. Dual-Frame Telephone Surveys: Adjusting for Device-Specific Nonresponse
Mr Matthias Sand (GESIS - Leibniz Institute for the Social Sciences)

To avoid errors due to undercoverage in CATI-Surveys, the dual-frame approach that employs independently drawn samples from both the landline and the mobile frame might be an appropriate solution for nationwide surveys. A specific source of error that can arise within this sampling approach might be determined by individual usage-patterns for specific means of communication. Therefore this paper will introduce a weighting approach that takes such frame-dependent causes of nonresponse into account by subdividing the survey population into different categories based on their usage-patterns and employing a composite design weighting approach.

2. A simulation study to evaluate the weighting estimators used for dual-frame telephone surveys
Professor Michele Haynes (The University of Queensland)
Mr Shane Dinsdale (The University of Queensland)
Dr Bernard Baffour (The University of Queensland)

Dual-frame telephone surveys have become increasingly important in Australia with the proportion of adults owning mobile phones and living in households without a landline telephone increasing exponentially from 5% in 2005 to 27% in 2014.This paper reviews the dual-frame estimators available for estimating population quantities using combined data from overlapping telephone sampling frames. Selected estimators are assessed empirically in the Australian context, where data on differential non-response by sampling frame is unknown. Additionally, bias in each of the estimators is evaluated under different patterns of non-response for simulated data with known telephone frame population totals.

3. Constructing cross-sectional weights for the German Panel of Household Finances
Dr Panagiota Tzamourani (Deutsche Bundesbank)
Dr Tobias Schmidt (Deutsche Bundesbank)

The paper describes the construction of cross-sectional weights for the second wave of the Panel of Household Finances survey, the German part of the euro area Household Finance and Consumption Survey. The approach is the ‘base weight’ approach (eg Verma, Betti and Ghellini, 2006). It thus entails the construction of person weights, from which household weights are derived. The procedure includes adjustments of the initial weights of the panel members and their affiliated members, adjustments of the design weights of the refreshment sample, non-response adjustments and calibration of the merged sample.The procedure is compared with alternative approaches.

4. Partial MAR
Mrs Olena Kaminska (ISER, University of Essex)

The concept of MCAR, MAR and NMAR is useful theoretically, but in practice it is rare to observe a situation with pure MCAR (where nonrespondents are exactly the same as respondents on a particular estimate) or MAR (where nonresponse can be perfectly explained). The most common situation in practice is what we call ‘partial MAR’, where some, but not the whole, nonresponse process can be explained. The presentation will introduce the concept of partial MAR, provide theoretical background for its concept and showcase its usefulness in developing nonresponse weights using a few empirical examples from the UK Household Longitudinal Survey.