Metadata record for The Future Strikes Back. Using Future Treatments to Detect and Reduce Hidden Bias
104060
Inter-university Consortium for Political and Social Research
ICPSR metadata records are licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.
V1
The Future Strikes Back. Using Future Treatments to Detect and Reduce Hidden Bias
104060
http://doi.org/10.3886/E104060V1
Felix Elwert
Fabian Pfeffer
Please see full citation.
This work is licensed under an Other license created by the PSID, namely that you agree not to redistribute data or other materials without the written agreement of ICPSR, unless: 1. You serve as the OFFICIAL or DESIGNATED REPRESENTATIVE at an ICPSR MEMBER INSTITUTION and are assisting AUTHORIZED USERS with obtaining data, or 2. You are collaborating with other AUTHORIZED USERS to analyze the data for research or instructional purposes.
Ann Arbor, MI: Inter-university Consortium for Political and Social Research
Elwert, Felix, and Pfeffer, Fabian. The Future Strikes Back. Using Future Treatments to Detect and Reduce Hidden Bias. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], 2018-07-25. https://doi.org/10.3886/E104060V1
PSID
Conventional advice discourages controlling for post-outcome variables in regression analysis. Here, we show that controlling for commonly available post-outcome (i.e. future) values of the treatment variable can help detect, reduce, and even remove omitted variable bias (unobserved confounding). The premise is that the same unobserved confounders that affect treatment also affect future values of the treatment. Future treatments thus proxy for the unmeasured confounder, and researchers can exploit these proxy measures productively. We establish several new results: Regarding a commonly assumed data-generating process involving future treatments, we (1) introduce a simple new approach to reduce bias and show that it strictly reduces bias; (2) elaborate on existing approaches and show that they can increase bias; (3) assess the relative merits of approaches; (4) analyze true state dependence and selection as key challenges; and (5) demonstrate that future treatments can test for hidden bias, even when they fail to reduce bias. We illustrate these results empirically with an analysis of the effect of parental income on children’s educational attainment.
United States