SSM under Nonignorable Nonresponse

ATE Coverage

The simulations are based on the the make_ssm_data-DGP with \(500\) observations. The simulation considers data with nonignorable nonresponse.

Metadata

DoubleML Version                                    0.9.0
Script                   ssm_nonignorable_ate_coverage.py
Date                                  2024-09-09 08:50:13
Total Runtime (seconds)                       5195.712764
Python Version                                     3.12.5

Table 1: Coverage for 95.0%-Confidence Interval over 1000 Repetitions

Learner g	Learner m	Learner pi	Bias	CI Length	Coverage
LGBM	LGBM	LGBM	3.667	15.499	0.966
LGBM	LGBM	Logistic	1.327	5.758	0.974
LGBM	Logistic	LGBM	1.232	5.472	0.959
LGBM	Logistic	Logistic	0.728	3.015	0.940
Lasso	LGBM	LGBM	2.892	12.159	0.975
Lasso	LGBM	Logistic	2.051	8.416	0.974
Lasso	Logistic	LGBM	1.187	4.970	0.950
Lasso	Logistic	Logistic	0.566	2.375	0.928

Table 2: Coverage for 90.0%-Confidence Interval over 1000 Repetitions

Learner g	Learner m	Learner pi	Bias	CI Length	Coverage
LGBM	LGBM	LGBM	3.667	13.007	0.919
LGBM	LGBM	Logistic	1.327	4.832	0.918
LGBM	Logistic	LGBM	1.232	4.593	0.897
LGBM	Logistic	Logistic	0.728	2.530	0.867
Lasso	LGBM	LGBM	2.892	10.204	0.931
Lasso	LGBM	Logistic	2.051	7.063	0.917
Lasso	Logistic	LGBM	1.187	4.171	0.894
Lasso	Logistic	Logistic	0.566	1.993	0.870