Nonparametric Estimation of ATE and QTE: An Application of Fractile Graphical Analysis

Table 1

ATE Monte Carlo simulations.

	MSE					MAE
Estimator	100	200	500	1000	2000	100	200	500	1000	2000
FGA ATE^(a)	0.221	0.157	0.079	0.057	0.036	0.372	0.308	0.221	0.187	0.147
FGA ATE (𝑅×2)^(a)	0.190	0.109	0.053	0.037	0.023	0.346	0.261	0.185	0.155	0.121
FGA ATE^(b)	0.686	0.657	0.630	0.607	1.699	0.464	0.456	0.369	0.343	0.336
FGA ATE (𝑅×2)^(b)	0.646	0.615	0.612	0.597	1.683	0.462	0.450	0.368	0.341	0.335
Hirano et al. [12]	0.732	0.679	0.644	0.618	1.710	0.472	0.460	0.371	0.344	0.337
PS matching
Nearest neighbor (1)	0.467	0.331	0.202	0.145	0.099	0.540	0.443	0.358	0.298	0.251
Nearest neighbor (2)	0.290	0.192	0.126	0.091	0.063	0.429	0.347	0.285	0.238	0.198
Nearest neighbor (4)	0.170	0.124	0.074	0.057	0.039	0.329	0.281	0.217	0.188	0.157
Kernel	0.285	0.226	0.127	0.074	0.033	0.418	0.369	0.283	0.217	0.146
Spline	0.233	0.139	0.058	0.034	0.021	0.384	0.297	0.192	0.149	0.117

^(a)𝛿̂_𝜏, ^(b)𝛿̃_𝜏. MSE: mean squared error. MAE: mean absolute error. Monte Carlo simulations based on 1000 replications of the baseline model.