Table 5.

Overall balance test and matching results from three kinds of matching method.

Sample	Overall Balance					Matching Results			Bootstrap Results
	Pseudo R2	LR chi2	P > chi2	Mean Bias	Median Bias	ATT	S.E.	T-stat	S.E.	z-Value	p-Value	95%CI (lower, upper)
AA
Raw sample before matching	0.275	1321.76	0.000	32.2	25.7	−0.01	0.02	−0.68
kernel matching	0.003	16.50	0.927	2.8	2.0	−0.03	0.03	−2.05	0.03	−2.42	0.042	−0.08 to 0.02
k-nearest neighbor matching (k = 1)	0.004	22.07	0.896	3.7	2.8	−0.04	0.03	−2.18	0.04	−2.62	0.038	−0.11 to 0.03
local linear regression matching	0.006	19.16	0.919	3.3	2.6	−0.02	0.03	−1.59	0.04	−1.77	0.073	−0.07 to 0.02
AD
Raw sample before matching	0.233	879.57	0.000	30.2	21.3	0.22	0.01	16.57
kernel matching	0.003	3.42	0.943	2.2	1.4	0.19	0.03	13.87	0.02	14.28	0.000	0.17 to 0.25
k-nearest neighbor matching (k = 1)	0.012	4.06	0.933	2.9	2.1	0.17	0.05	11.15	0.04	11.56	0.000	0.15 to 0.19
local linear regression matching	0.019	12.79	0.712	5.4	5.6	0.16	0.02	11.00	0.02	11.48	0.000	0.14 to 0.20

AA, the appropriateness of admission; AD, the appropriateness of disease; S.E., standard error; CI, confidence interval. All results are computed using the Stata module of psmatch2. The S.E. reported by matching result did not consider the fact that the propensity score was estimated (that is, assuming the propensity score is the true value, and then deducing the S.E.), so we considered to get a more accurate S.E. and 95%CI using the bootstrap method.