Empirical and Causal Models for Heterogeneous Data Fusion

University Of Colorado At Denver, Aurora CO

Investigators

Abstract

This research project will advance the use of causal inference methods in situations where individual-level data are not available due to practical, ethical, or legal constraints. There has been a lot of work on the development of innovative methods to evaluate policy effects in observational databases, the area being termed causal inference. However, many of these methods require individual-level data. For a variety of reasons, it might not be possible to obtain individual-level data due to reasons such as maintaining patient privacy or other logistical issues. This project will extend statistical methodologies to accommodate practical real-world scenarios in a wide variety of disciplines, including medicine, the social sciences, and public health. There are a variety of important problems the new methods could be applied to, such as evaluating the effects of climate change on COVID19 incidence and deaths. Graduate students will be trained, and software and curricula in causal inference will be developed. This research project will develop new methods for combining heterogenous databases. Such data have become commonplace with the vast expansion of databases in various types of scientific and epidemiological applications. First, the project will develop new approaches to estimate empirical associations for heterogenous data fusion problems. The investigator will leverage model misspecification theory in conjunction with resampling/perturbation-based methodology. Second, the project will develop new causal inference approaches for heterogeneous data fusion problems, primarily focusing on constrained estimation, simulation-based approaches, and sensitivity analysis techniques. The results of this research should lead to new theoretical underpinnings in various areas of the mathematical sciences, including statistical theory and causal inference. Primary subfields of statistics that will be addressed in this research include likelihood theory and inference, estimating equations, model misspecification, and causal inference. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.