The Impact of the Urban Resident Basic Medical Insurance on Health Services Utilisation in China

Five (seven) out of nine sample cities of the URBMI Survey reimburse regular (critical) outpatient services. In addition, three out of nine sample cities reimburse outpatient services due to accidents for young people [4]. In sum, the URBMI scheme covers at least part of outpatient health services in all nine sample cities reviewed.

The simultaneous equation model is more efficient than the two-stage least squares (2SLS) estimator since the potential errors correlation is not taken into account in the 2SLS [18].

As noted at Dardanoni and Li Donni [23], owing to the presence of multidimensional private information, this analysing framework could estimate the incentive effect of medical insurance accurately; however, it may be difficult to clearly understand the selection effect.

We thank an anonymous referee who pointed to this explanation.

It would be ideal to let u _i freely correlated with independent variables; however, for the non-linear models, the strategies that are widely adopted in the linear regression model to eliminate individual heterogeneity could not be directly applied when an independent variable is a binary response. The conditional fixed-effects logit estimator using conditional maximum likelihood estimator is one solution to eliminate the u _i ; however, in that case, only the observations that change the dependent variable status would be used [24]. As a robust analysis, we have also adopted the fixed-effects logit estimator to estimate the utilisation equations and compared results with random-effects probit estimates.

Nine cities which are included in the URBMI Survey are located across Western China (Baotou City, Chengdu City, Xining City, Urumqi City), Central China (Changde City, Jilin City, Shaoxing City), and Eastern China (Xiamen City, Zibo City).

Seven indicators were adopted in the cluster analysis, including GDP per capita, population size, population density, average number of hospital beds, average number of physicians, average financing cost of URBMI, and average financing cost of UEBMI social pooling account in 2006 [4].

One potential worry of using self-reported income is that respondents are likely to under-report their true income. An alternative strategy is to use household expenditure to proxy the wealth. To avoid the endogenous issue, we calculated the net household expenditure in the past year, which is defined as total household expenditure minus health expenditure [31]. We tested for both of the above two variables and it transpires that the results are comparable, with only slight differences in the magnitudes. Considering that more missing values exist in expenditure variables, the results reported here only come from the self-reported income information.

It would be ideal to further include supply-side factors into the independent variables; we did not include those variables because within each city, the city-level office defines the single plan (eligibility criterion, financing and benefit package) for the URBMI candidate throughout the city. So each eligible consumer can only choose whether or not to purchase the insurance but not how much to pay. The supply-side effects are captured by the city dummies.

There could be some worry about multicollinearity between household head’s drinking behaviour and household members’ drinking behaviour. We calculated the variance inflation factor (VIF) after the probit estimate on the URBMI demand equation using the pooled four-wave data. The mean VIF was 2.24 and the highest VIF value (7.9) was found for the household size variable. In one case [Table 4, Column (8)], the household size was excluded from the regression to help ensure concavity of the log-likelihood.