Draft

CATEs

CATE Coverage

The simulations are based on the the make_heterogeneous_data-DGP with \(2000\) observations. The groups are defined based on the first covariate, analogously to the CATE IRM Example, but rely on LightGBM to estimate nuisance elements (due to time constraints).

The non-uniform results (coverage, ci length and bias) refer to averaged values over all groups (point-wise confidende intervals).

DoubleML Version                        0.9.0
Script                   irm_cate_coverage.py
Date                      2024-09-09 10:07:30
Total Runtime (seconds)           9833.500947
Python Version                         3.12.5
Table 1: Coverage for 95.0%-Confidence Interval over 1000 Repetitions
Learner g Learner m Bias CI Length Coverage Uniform CI Length Uniform Coverage
LGBM LGBM 0.148 0.795 0.972 1.690 1.000
LGBM Logistic Regression 0.059 0.281 0.942 0.599 0.996
Lasso LGBM 0.158 0.766 0.949 1.634 1.000
Lasso Logistic Regression 0.062 0.295 0.941 0.629 0.997
Table 2: Coverage for 90.0%-Confidence Interval over 1000 Repetitions
Learner g Learner m Bias CI Length Coverage Uniform CI Length Uniform Coverage
LGBM LGBM 0.148 0.667 0.936 1.689 1.000
LGBM Logistic Regression 0.059 0.235 0.889 0.596 0.996
Lasso LGBM 0.158 0.643 0.896 1.632 1.000
Lasso Logistic Regression 0.062 0.247 0.889 0.630 0.998