Combining the power of social scientific longitudinal surveys with biological information provides the potential to bring new insights about the complex interaction of the biological and social contexts in our daily life. Whilst using such data is becoming more popular the issue of missing data has the potential to substantially bias results.
This Work Package aims to contribute to the quality of research using biomarkers in longitudinal studies by tackling a number of methodological issues such as:
- What are the current practices when dealing with different mechanisms of missing data in this context? What are the assumptions these practices make and how plausible are they? What could researchers do to make the assumptions more plausible/realistic?
- What are the effects of nurses and interviewers on the probability of participating in the survey, the probability of giving blood and on data quality? How can we model such effects?
- Which are the best statistical methods (e.g., multiple imputation, propensity weighting, pattern mixture models, etc.) of dealing with informative missing data in the context of biomarkers in longitudinal surveys? How can we jointly model the different stages of missing data (non-response, attrition, consent refusal) in this complex setting?
- Administrative data and other data sources that can be linked to surveys have the potential to compensate for missing data, but may require consent from the respondents and may be prone to errors. What statistical methods can be developed to compensate for linkage errors?