Discussion of ‘Causal inference with misspecified exposure mappings: separating definitions and assumptions’

原文摘要

We thank Sävje (2024) for a thought-provoking article and appreciate the opportunity to share our perspective as social scientists. In his article, Sävje recommended misspecified exposure effects as a way to avoid strong assumptions about interference when analysing the results of an experiment. In this discussion, we highlight a key limitation of Sävje’s recommendation. Exposure effects are not generally useful for evaluating social policies without the strong assumptions that Sävje seeks to avoid. Exposure effects are commonly used as an organizational tool. The idea is to decompose a complicated interference pattern into individual units that are easier to manage separately. For instance, consider an experimenter who wants to learn whether assigning a treatment to every unit is expected to increase outcomes on average. The effect of this policy may be difficult to estimate directly when there is interference. However, it can often be written as a combination of exposure effects, each of which may be straightforward to estimate using standard techniques. See, for instance, Manski (2013) and Aronow & Samii (2017).

摘要来源：OpenAlex 索引摘要（稿件版本未注明）