Development of statistical methods for analyzing causal effects with population-based register data

Project title

Development of statistical methods for analyzing causal effects with population-based register data

Project summary

Linked administrative databases are powerful resources that provide data on the whole population for investigation of important policy and public health concerns. Register studies commonly involve drawing causal conclusions about investigated relationships. The purpose of the research program described in this application is to develop new statistical methods for the analysis of causal effects with population-based register data and to promote advances in statistical methods in the register research community.

The research program has three distinct projects:

1. We will develop new methods for causal inference with register-based data. Targeting a causal parameter in a register-study is related to selecting a study population from a longitudinal population-based register. In practice, a study population is formed by making and applying a series of inclusion/exclusion criteria. Here, questions concerning selection bias, conditional and marginal parameters and design-weighting will be addressed.

2. Estimators of average causal effects include working models summarizing the information necessary to control for confounding, e.g., propensity score models. Since the main objective is the average causal effect, it is important that model-building strategies favor the performance of the causal effect estimator rather than inference for parameters in the nuisance models. Here, we study sensitivity to model misspecification and propose methods for model-building that are robust for estimation of the average causal effect.

3. Controlling for confounding in a register-based study means comparing individuals with similar background characteristics. The propensity score is commonly used to control for confounding, but other scores have also been studied and proposed in the literature. Here we aim at describing summarizing scores and covariate scores, as well as their properties when used in an estimator of an average causal effect, e.g., a matching or stratification estimator

Project duration

2017-2022

Main applicant

Ingeborg Waernbaum

Co-applicant

David Seim

Funder

Swedish Research Council

Amount received

SEK 12,000,000

Last modified: 2021-03-09