References & Citations
Economics > Econometrics
Title: Proxy Controls and Panel Data
(Submitted on 30 Sep 2018 (v1), revised 15 Mar 2020 (this version, v3), latest version 21 Nov 2023 (v8))
Abstract: We present a flexible approach to the identification and estimation of causal objects in nonparametric, non-separable models with confounding. Key to our analysis is the use of `proxy controls': covariates that do not satisfy a standard `unconfoundedness' assumption but which are informative proxies for variables that do. Our methods are presented in sufficient generality that they apply to both cross-sectional and panel models. Our identification results motivate an easy-to-implement nonparametric estimation method. Under our identifying assumptions the estimation problem is `well-posed'. We describe the estimator and derive convergence rates. When applied in panel settings our methods provide a novel approach to identification with non-separable general heterogeneity and a fixed time dimension. In the panel case, observations from different periods serve as proxies for unobserved individual heterogeneity and our key identifying assumptions follow from restrictions on the serial dependence structure of the data and latent variables. We apply our methodology to two separate empirical settings that respectively showcase the cross-sectional and the panel implementations of our approach. We estimate causal effects of grade retention on performance in a range of cognitive tests using cross-sectional variation in the data and we estimate a structural Engel curve for food using panel data.
Submission history
From: Benjamin Deaner [view email][v1] Sun, 30 Sep 2018 00:38:11 GMT (57kb,D)
[v2] Tue, 31 Dec 2019 01:40:48 GMT (17kb)
[v3] Sun, 15 Mar 2020 23:03:15 GMT (69kb,D)
[v4] Mon, 27 Jul 2020 16:38:49 GMT (93kb,D)
[v5] Fri, 9 Oct 2020 01:47:27 GMT (103kb,D)
[v6] Mon, 19 Oct 2020 21:56:28 GMT (99kb,D)
[v7] Tue, 5 Jan 2021 18:14:49 GMT (104kb,D)
[v8] Tue, 21 Nov 2023 17:23:48 GMT (159kb,D)
Link back to: arXiv, form interface, contact.