Advances in Causal Inference With Continuous Exposures

University Of Massachusetts Amherst, Amherst MA

Investigators

Abstract

Determining cause and effect is one of the fundamental goals of scientific inquiry. For instance, does a vaccine reduce risk of disease? Is a chemical such as lead or ammonia in drinking water harmful to human health? Causal inference is the area of statistical research concerned with developing methods for using data to answer such questions. The majority of causal inference research has focused on binary exposures; that is, exposures that can only take two values, such as treatment and control. However, many exposures of interest can take a large or even infinite number of values, such as the dose of a drug or vaccine received, or the concentration of a substance in drinking water. These are called "continuous exposures", and are commonplace in many disciplines, including biomedicine, epidemiology, public health, and economics. In this project, the PI will develop flexible statistical methods for assessing the causal effects of continuous exposures. The PI will develop three methodological innovations for causal inference with continuous exposures. In the first two aims, the PI will focus on the causal dose-response curve, which describes how the causal effect changes as a function of the exposure level. In order to make valid statistical inference regarding the shape of this curve, the PI will develop a uniform confidence band for the causal dose-response curve and develop tools for assessing the fit of parametric models for the dose-response curve. These methods will permit researchers to understand the qualitative effect of a continuous exposure while making minimal assumptions. In the third aim, the PI will address nonparametric inference on the effect of incremental shift interventions, which provide useful one-number summaries of the causal effect of continuous exposures under weaker assumptions than those necessary for the dose-response curve. User-friendly software implementing the methods developed in each of these aims will be made freely available. 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.