arXiv subjects

Myoung-jae Lee

Publications and source records attributed to Myoung-jae Lee.

4 recordsLinked to original sources

arXiv2025

Easy-to-Implement Two-Way Effect Decomposition for Any Outcome Variable with Endogenous Mediator

Given a binary treatment D and a binary mediator M, mediation analysis decomposes the total effect of D on an outcome Y into the direct and indirect effects. Typically, both D and M are assumed to be exogenous, but this paper allows M to be endogenous while maintaining the exogeneity of D, which holds certainly if D is randomized. The endogeneity problem of M is then overcome using a binary instrumental variable Z. We derive a nonparametric "causal reduced form (CRF)" for Y with either (D,Z,DZ) or (D,M,DZ) as the regressors. The CRF enables estimating the direct and indirect effects easily with ordinary least squares or instrumental variable estimator, instead of matching or inverse probability weighting that have difficulties in finding the asymptotic distribution or in dealing with near-zero denominators. Not just this ease in implementation, our approach is applicable to any Y (binary, count, continuous, etc.). Simulation and empirical studies illustrate our approach.

stat.ME↗

arXiv2025

Finding network effect of randomized treatment under weak assumptions for any outcome and any effect heterogeneity

In estimating the effects of a treatment/policy with a network, an unit is subject to two types of treatment: one is the direct treatment on the unit itself, and the other is the indirect treatment (i.e., network/spillover influence) through the treated units among the friends/neighbors of the unit. In the literature, linear models are widely used where either the number of the treated neighbors or the proportion of them among the neighbors represents the intensity of the indirect treatment. In this paper, we obtain a nonparametric network-based "causal reduced form (CRF)" that allows any outcome variable (binary, count, continuous, ...) and any effect heterogeneity. Then we assess those popular linear models through the lens of the CRF. This reveals what kind of restrictive assumptions are embedded in those models, and how the restrictions can result in biases. With the CRF, we conduct almost model-free estimation and inference for network effects.

stat.ME↗

arXiv2023

Difference in Differences and Ratio in Ratios for Limited Dependent Variables

Difference in differences (DD) is widely used to find policy/treatment effects with observational data, but applying DD to limited dependent variables (LDV's) Y has been problematic. This paper addresses how to apply DD and related approaches (such as "ratio in ratios" or "ratio in odds ratios") to binary, count, fractional, multinomial or zero-censored Y under the unifying framework of `generalized linear models with link functions'. We evaluate DD and the related approaches with simulation and empirical studies, and recommend 'Poisson Quasi-MLE' for non-negative (such as count or zero-censored) Y and (multinomial) logit MLE for binary, fractional or multinomial Y.

econ.EM↗

arXiv2021

Path-Free Decomposition for Direct, Indirect and Interaction Effects in Mediation Analysis

Given a binary treatment and a binary mediator, mediation analysis decomposes the total effect of the treatment on an outcome variable into direct and indirect effects. However, the existing decompositions are "path-dependent", and consequently, there appeared different versions of direct and indirect effects. Differently from these, this paper proposes a "path-free" decomposition of the total effect into three sub-effects: direct, indirect, and treatment-mediator interaction effects. Whereas the interaction effect has been part of the indirect effect in the existing two-effect decompositions, it is separately identified in our three-effect decomposition. All effects are found using conditional means, but not conditional densities, and are estimated with ordinary least squares estimators. Simulation and empirical analyses are provided as well.

stat.ME↗