[4O3-J-7] Agents: learning agents

[4O3-J-7-03] Causal Inference between LGBT Measures and Profit using Propensity Score and Causal Tree

〇Mayuko Tahara¹, Yasuhiro Hasegawa¹, Ryoka Furusyo¹, Chie Murai¹, Takato Mori¹, Takahiro Hoshino¹ (1. Keio University)

Keywords:causal inference, propensity score, causal tree, semi-parametric analysis, LGBT

This paper aims to examine the effect of LGBT measures on profit. Previous researches are limited to the effects of general diversity and company-internal LGBT measures, and no mention have been made toward the effect of general LGBT services provided by companies. Therefore, this paper carries out an empirical analysis under the hypothesis that LGBT measures do have effect in raising firm revenue.

Rental property data of real estate information site has been used for demonstration. Given LGBT-friendly tags to some of its rental properties, the paper compares the effect on the number of browsing and inquiries between the dwellings with tags and those without. The paper analyzes the effect in consideration of endogeneity, using IPW estimator and doubly robust estimator derived from propensity score. Specifically, by consolidating multiple covariates into one variable known as the propensity score, the paper successfully estimates the pure effect of the tag. In addition, a machine learning technique called Causal Tree was used to infer the heterogeneity. Our hypothesis is supported and revealed that LGBT-friendly tag gives a positive effect on both the number of views and number of inquiries.