Skip to main content
. 2021 Feb 27;22(3):473–483. doi: 10.1007/s10198-021-01268-2

Table 3.

Robustness checks on the impact of the NRCMS on health care utilisation and medical expenditure

Formal care Preventive care Folk doctor use Inpatient care Village clinics Township health centres County hospitals City hospitals Pr(OOP > 0) Log of OOP if positive
(1) IV of county NRCMS status and number of years of the NRCMS coverage
  NRCMS treatment effect − 0.000 (0.022) 0.005 (0.014) − 0.037* (0.022) 0.024 (0.034) 0.116* (0.068) − 0.029 (0.085) 0.016 (0.054) − 0.077* (0.046) 0.023 (0.082) − 0.298 (0.279)
  F statistics for weak identification test 242.467 248.267 262.863 166.822 190.580 190.580 190.580 190.580 251.227 183.438
  Sargan’s over-identification test (prob) 0.753 0.127 0.718 0.208 0.303 0.113 0.288 0.594 0.796 0.597
  N 17,397 17,467 14,361 1754 1744 1744 1744 1744 2688 2021
(2) Non-linear IV models
  NRCMS treatment effect − 0.024 (0.118) 0.188 (0.243) − 0.443** (0.219) 0.218 (0.321) 0.418** (0.206) − 0.185 (0.301) 0.019 (0.233) − 0.603 (0.372) 0.073 (0.249) − 0.320 (0.277)
  N 20,324 20,431 17,322 2048 2042 2042 2012 1932 3136 2389

Notes: The first panel shows results from IV estimations in linear probability models (LPM) using 2 IVs: the introduction of the NRCMS at county level and the number of years covered by the insurance, with F statistics for weak identification test and probability of Sargan’s over-identification test. Sargan’s over-identification test is calculated as N*R-squared from a regression of the IV residuals on both instruments. The joint null hypothesis is that the instruments are valid instruments; a rejection casts doubt on the validity of the instruments. The second panel shows the IV probit estimation results for health care utilisation. Two different models are applied to the two parts of the OOP payments: an IV probit model for the probability that an individual made any OOP payment and OLS, applied only to sub-sample with non-zero OOP payments, for the log of OOP payments. Robust standard errors clustered at county level in brackets.

Other independent variables include age, gender, household size, marital status, ethnicity, eastern region, central region, household income, asset index, education level, occupation, number of major diseases, severity of illness in the last month, health risk variables and urbanicity index at community level.

*Indicates statistical significant at the 10% level; **indicates statistical significant at the 5% level.