Explaining provincial government health expenditures in China: evidence from panel data 2007–2013

China Finance and Economic Review

Table 7 General method of moments (GMM) models

	Model 5	Model 6
Instrumented variable	Real per capita budgetary deficits	Economy
Instruments	L(1/2). real per capita budgetary deficits	L(1/2).economy
Endogeneity test	p value = 0.6055	p value = 0.0161
Real per capita budgetary deficits	0.037*** (0.014)	0.034*** (0.008)
Economy	167.859*** (54.612)	248.886*** (44.731)
Openness	−12.046 (9.079)	−9.525 (9.356)
Industrial Structure	−26.148 (17.685)	−32.773** (14.302)
Unemployment rate	−1844.576 (1324.935)	−2493.487** (1064.039)
Dependency rate of the aged	804.165*** (198.447)	1018.315*** (191.3554)
Province dummies	Included (29, Qinghai and Tibet excluded from data)	Included (29, Qinghai and Tibet excluded from data)
Time dummies	Included	Included
Under identification	p value = 0.0094	p value = 0.0045
Weak identification (Cragg-Donald Wald F statistic)	53.349	71.133
Over identification (Hansen J statistic)	p value = 0.4606	p value = 0.1140
N	145	145
R-squared	0.9412	0.9374

Notes: Standard errors are given in parentheses after the coefficients. The individual coefficient is statistically significant at the *10%, **5%, or ***1% significance level. The GMM estimates reported are all two-step results. The statistics are robust to heteroscedasticity and autocorrelation. The null hypothesis for the endogeneity test is that the specified endogenous regressor can be treated as exogenous. A significantly low p value suggests endogeneity. The null hypotheses for the under-identification test are that the model is under identified. For the weak identification test, the null hypothesis is that the model is weakly identified. Stock-Yogo weak ID test critical values are 19.93 for 10% maximal IV size, 11.59 for 15% maximal IV size; and 8.75 for 20% maximal IV size. The Cragg-Donald Wald F statistics for both model 5 and model 6 are significantly higher than the critical values, therefore rejecting the null hypothesis. The null hypothesis for the over-identification test is that the model is identified