Modelling hospital outcome: problems with endogeneity

Table 1 Model performance estimates for the base models

Model title	Logit1	Logit2	Probit1	Probit2	Logit3	Probit3	LPM1	LPM2	LPM0
Regression method	Logisitc	Logit1 + site FE	Probit	Probit1 + site FE	Logit 2 + RE	Probit2+RE	LPM_ldm	LPM_ldm+siteFE	LPM[0,1]
Indices
Number of patients	92,693	92,693	92,693	92,693	92,693	92,693	92,693	92,716	68,264
Number of parameters	110	234	110	234	107	107	110	234	110
ROC AUC	0.915	0.917	0.915	0.917	0.917	0.917	0.912	0.915	0.884
H-L statistic; P-value	0.173	0.044	0.000	0.000	0.073	0.000	0.173	0.103	0.000
Out-of-sample shrinkage %	0.940	1.600	0.940	0.580			0.390	0.360
In-sample-shrinkage %	0.380	0.360	0.380	−0.510			0.000	0.360
Overfitting %	0.560	1.250	0.560	1.090			0.390	0.000
Calibration belt: P-value	0.850	0.733	0.000	0.000	0.593	0.000	0.850	0.987	0.000
AIC	33,867.39	33,712.31	33,897.66	33,756.53	33,758.82	33,799.85	33,863.39	33,712.31
BIC	34,792.25	35,665.78	34,803.62	35,710	34,674.21	34,715.24	34,769.35	35,665.78
Development set
CITL					−0.002	0.000
C-slope					1.005	1.009
AUC					0.917	0.915
E:O ratio					1.001	1.000
Validation set
CITL					0.002	0.021
C-slope					1.005	1.006
AUC					0.916	0.916
E:O ratio					0.989	0.987
ICC: unconditional					0.201	0.154
ICC: conditional					0.018	0.016
ICC: unconditional					0.201	0.154
ICC: conditional					0.018	0.016

ISSN: 1471-2288