Rethinking causal inference for recurring exposures: The incremental propensity score approach with lavaan

TL;DR

This paper introduces a new method for understanding how recurring events like family violence affect outcomes like depression.

Contribution

The paper introduces the incremental propensity score intervention (IPSI) for causal inference with recurring exposures.

Findings

IPSI allows for more realistic causal questions about recurring exposures.

IPSI requires fewer assumptions than existing methods.

IPSI was applied to a real-world dataset on family violence and adolescent depression.

Abstract

Scholars are often interested in evaluating the causal effects of a recurring exposure (e.g., family violence) on behavioral and psychological outcomes. However, causal inference of recurring exposures is challenging. Conventional analytic approaches target causal quantities lacking practical relevance, such as mandating everyone to uniformly always be exposed or unexposed to family violence. Estimation further relies on everyone having a non-zero probability of being either exposed or unexposed at each occurrence, which is frequently unrealistic when past exposures perfectly predict future exposures. In this paper, we introduce a novel approach from the causal inference literature for drawing causal conclusions about recurring exposures: the incremental propensity score intervention (IPSI). IPSI frames causal questions more realistically by assessing how changing the propensity of a recurring exposure may influence an outcome. To facilitate the adoption of IPSI for recurring exposures, we develop an estimation procedure using lavaan, a widely used structural equation modeling software in R. We demonstrate the application of IPSI with a real-world dataset investigating the impact of recurring family violence on adolescent depression. IPSI requires fewer assumptions than existing approaches while offering more meaningful insights into the causal effects of recurring exposures.

Figures