14Healthcare and Insurance Among the Elderly in China: Evidence from the CHARLS Pilot¹

Insurance Coverage

In the five years before 2008, a major change began in terms of the availability of health insurance to the Chinese population. We can see this in Table 14-1, which shows the fraction of the men and women separately who claim they have some health insurance (public or private). Overall, 91% of our sample over 45 has some insurance. By age, the fraction with insurance stays fairly constant, with a slight drop among those over 75 years. There are no major differences between men and women. Among those who do not have insurance, a small percentage, 15% of men and 13% of women, lost their insurance recently,⁵ but most never had any after the collapse of the old system. Five years prior to 2008, very few people would have had insurance coverage, especially in rural areas.

We examine the different types of publicly provided insurance separately in Table 14-2, for men and women respectively. Three new public insurance programs had come into existence in the five years prior to 2008.⁶ The NCMS is designed for rural areas and is the most prevalent of the three. Two insurance schemes dominate urban areas: the Urban Employee Medical Insurance and the Urban Resident Basic Medical Insurance, with the former being more prevalent in our sample. The employee insurance is given mainly through employers, while the resident insurance is a public program provided through the community. As seen in Table 14-2, the rural insurance scheme is targeted to people with a rural hukou,⁷ not necessarily those living only in rural areas.⁸ While in rural areas there is little spillover from the rural to the urban insurance schemes, in urban areas many residents have the rural insurance because they are farmers and still have rural hukou. For example, Table 14-2 Panel B shows that 48.7% of men in urban areas have NCMS insurance, while only 7.4% have it for those with an urban hukou. Almost no difference exists for men in having some insurance by rural or urban area, or by rural or urban hukou. For women, Table 14-2 Panel C shows that having a rural hukou makes it a little more likely that they will have some coverage. Women with an urban hukou are much more likely to have the Urban Resident Basic Medical Insurance, and men almost only have the Urban Employee Medical Insurance.

If we define a migrant as a person whose hukou is in a different county than where he or she currently resides, migrants are severely disadvantaged. One-third of migrants, both male and female, have no insurance whatever, up from the 9% in the general population.⁹ Clearly, migrants are a vulnerable population in this regard. If a migrant has insurance, it is most likely the rural NCMS insurance. This makes sense because their hukou are in most cases rural places, even though they may live in an urban area.¹⁰

It is of some interest to examine whether the insured are in better or worse health than the non-insured. No causality can be attached to these correlations in this chapter. We have many available measures of health to use; here, we use a self-reported measure of general health on a scale of excellent, very good, good, fair, or poor. While self-reported general health measures have biases (see Strauss and Thomas, 1998, for example), they also have signal, since they predict future mortality well (Banks et al., 2009). As shown in Table 14-3, there are no significant differences in insurance coverage among those in poor health versus those not in poor health. This is of interest for several reasons, among which it appears that adverse selection is not a major problem in our sample, at least not for all programs taken together. Lin et al. (2009), however, do find adverse selection in their analysis of the Urban Resident Basic Medical Insurance, but they do not focus exclusively on the older population.

Table 14-4 provides regression results for having any insurance for men and women, respectively. We start in column 1 with a model that includes major socioeconomic status (SES) variables: a set of dummy variables for age group (under 55 years the left-out group), for education levels completed (no schooling the left out group), and a linear spline in log of per capita expenditure (pce).¹¹ Pce is preferred to income because income is measured with much more error than pce (see Lee, 2009, for instance) and because pce is a better measure of long-run resources because it is smoothed in the face of annual income shocks. The knot point for the spline is at the median of log pce. We also include dummies for province interacted with rural area of residence. In column 2, we add a set of variables that are arguably endogenous: a dummy for migrant status (not a migrant being left out), and dummies for being widowed and being divorced or never married (the two are both very small in number and cannot be statistically distinguished in our data). Column 3 replaces province-rural dummies with community fixed effects.¹² The idea here is that each community has factors that will affect insurance and healthcare utilization that are not captured by the provincial dummies interacted with rural or urban. These factors will include healthcare prices, inherent healthiness of the area, public health infrastructure, and other factors. F-tests for all combinations of dummy variables are reported as well.

Regressions are all ordinary least squares (OLS). For binary dependent variables, this represents linear probability (LP) models. LP models consistently estimate average treatment effects, which is what we are interested in. Furthermore, all of our standard errors are calculated for robustness to heteroskedasticity and allow clustering at the community level. The robustness makes our standard error estimates consistent when we are estimating LP models (Wooldridge, 2002).

For men, age dummies are weakly significant (at the 10% level), except when we add community fixed effects, and show that men aged 55–64 are about 3% more likely to have insurance than younger men. For women, the age dummies are highly significant (at the .001 level), and older women over 75 years are 10–14% less likely to be insured, a large impact.

For men, education does not seem to have significant effects on possessing any insurance, but for women, it raises the likelihood of having insurance. On the other hand, for men, higher pce is associated with a higher probability of having some insurance, but not for women.

Reporting poor general health is not correlated with having insurance for men. It is weakly (at the 10% level) positively correlated for women, with some evidence of adverse selection for women.

Being a migrant male is associated with a 22% decline in the likelihood of being insured, a very large difference; the difference for women is similar in magnitude and significance. The NCMS generally does not reimburse migrants for their medical expenses incurred in urban areas where the migrants are working, so the incentive for them to participate in the NCMS is low. The two major urban insurance schemes also do not cover migrants. In May 2006, a separate medical insurance pooling fund was set up to cover expenses for migrants, but the effectiveness of this program has not been evaluated to date.

Being widowed also is associated with a large (10%) decline in the odds of being insured for men, though a much less large difference for women and not significant in most cases.

Men and women who live in rural Gansu have a somewhat higher chance of having health insurance, once individual and household SES factors are controlled. This is interesting, and one might have expected the reverse. However, even the unconditional probabilities of having insurance are a bit higher in rural areas of Zhejiang and Gansu. When we add the community dummies in column 3, they are jointly significant at the .001 level for both men and women.¹³ Now the migrant dummy loses magnitude and significance for women and its magnitude drops by 30% for men. Apparently, important, unmeasured factors exist at the community level that affect access to insurance for those who do not get it.

In results not shown, we repeat the regressions for respondents who have a rural hukou and are thus eligible in principle to join the NCMS.¹⁴ Most of the results for the rural insurance are similar to the main results: Older women, migrants, and male widowers are substantially less likely to be insured with this scheme. However, the associations of education and pce with belonging to the NCMS are generally not significant, except for women for being able to read and write, but not for those with more schooling.

Healthcare Utilization

About 16.5% of men and 21% of women aged 45 and older said they used outpatient services at least once over the past month (see Table 14-5).¹⁵ Most of those who went used a hospital or a village or private clinic. Township hospitals and healthcare posts were also important destinations, especially in rural areas. Table 14-5 shows that inpatient use over the past year was less frequent, 6.7% for men and 6% for women. Virtually all of the inpatient use was at hospitals, general hospitals being the most important destination by far, followed equally by specialized hospitals, Chinese medicine hospitals, and township hospitals.

Regressions for outpatient utilization (see Table 14-6) show that for men, being older and more educated have positive associations with use, while for women, having higher pce is positively related to outpatient service use, as is living in rural Gansu, controlling for SES. The community dummies are highly significant for women, though not for men. Again, unmeasured factors at the community level are important for women seeking outpatient care.

For inpatient care (see Table 14-7), income (pce) is highly significant and positively related to care for both men and women. Other factors being equal, being in Gansu, rural or urban, is associated with more inpatient care use for women and living in urban Gansu for men. It is, of course, of prime interest to examine how having insurance is associated with outpatient and inpatient care. The problem is that having insurance is endogenous, and there is no good way with these data to identify a structural relationship. Still, looking at the correlations is of interest. Insurance is positively and significantly correlated with inpatient use by men, though it is not significant for women and it is not significantly related to outpatient use. A man with insurance is 4.5% more likely to seek inpatient care than a man without any insurance. These results are consistent with the results of Gao, Raven, and Tang (2007) and Wagstaff and Lindelow (2008), who also found a positive correlation of insurance with inpatient use.

Health Insurance Parameters and Reimbursement

Having health insurance does not tell anything about the nature of that insurance. The CHARLS pilot collected data about some of the characteristics of the insurance, in particular about the premiums paid, as well as, for those who had insurance and who went for inpatient or outpatient services, what fraction of their total costs they anticipated would be reimbursed. For the Urban Resident Basic Medical Insurance, not enough respondents reported their premiums so our sample is too small to report. For the Urban Employee Medical Insurance and the NCMS insurance, we report mean premiums in Table 14-8.

The mean reported premium for the NCMS is about 20 RMB per year. The mean premium is lower in rural Gansu, only 13 RMB, and higher in rural Zhejiang, about 28 RMB. These are remarkably low premiums even compared to low rural incomes in Gansu of 5,000 RMB pce. Even in Zhejiang, both rural and urban, about 20% say they did not pay any premiums for their NCMS insurance. For the Urban Employee Medical Insurance scheme, premiums are much higher, around 300 RMB per year in urban Gansu, about the same for women in urban Zhejiang, but much higher, 600 RMB per year, for men in urban Zhejiang. Even these levels, however, are not large relative to annual pce of between 7,500 RMB and 10,000 RMB per year in Gansu and Zhejiang. We have to be a bit careful because our cell sizes are not large, but it is quite interesting that for this insurance plan, unlike the rural plan, premiums are lower for women, particularly in urban Zhejiang. We do not know why as yet. The other point to note from Table 14-8 is that in urban Zhejiang, a little more than 10% get their Urban Employee Medical Insurance without any premium.

Another important parameter of the new insurance schemes is the reimbursement rate. Reimbursements depend on the plan and on the parameters, which are set at the county or district level.¹⁶ They vary greatly across county-units and depend on many factors, such as total medical expenditures, which type of facility the respondent went to and whether that type of facility is covered by the particular insurance policy, and the specific treatment received. The NCMS has four models to reimburse patients for in- and outpatient services (Lei and Lin, 2009). The most frequent model, used in two-thirds of the rural counties, uses a medical savings account. Each household has its own medical savings account, with household members depositing their contributions into this account and then spending money from it. Only household members are entitled to the funds in the account, which is used mainly for outpatient services. There is a deductible and a reimbursement cap for using a medical savings account.

Traditionally, the reimbursements for the NCMS have put more emphasis on inpatient than outpatient services. However, a more recent, general trend includes outpatient services and physical check-ups in patient reimbursements (Du and Zhang, 2007). In addition, some counties provide insurance coverage for an annual physical examination. Overall, there has been an increasing trend in the amount of coverage per capita and an increase in the range of services offered (Du and Zhang, 2007).

We calculated individual reimbursement rates by first asking the respondent if they went for outpatient service in the past month or inpatient service in the past year. If the answer was yes, then we got details about the last visit, including the total costs of the service, including medicines. We then asked how much he or she expected to pay, not just what he or she had paid to date.¹⁷ One could reasonably worry how well respondents can answer these cost questions, but they were thoroughly pretested, and it is our judgment that these answers are reliable. For the cost repayment estimates, we aggregated men and women to maximize our cell sizes, which are small. We also used only those respondents who reported that they have insurance, as otherwise this question does not make sense.¹⁸

The reimbursement rates for inpatient service are 40% and 30% in urban and rural areas, respectively (not, however, significantly different at standard levels), similar between Gansu and Zhejiang (see Table 14-9). By hukou status, the reimbursement rates are 58% for urban hukou holders and 28% for rural hukou holders (these are significantly different at the .01 level). By hukou, there were some important differences across provinces; the rates for rural hukou holders were higher in Gansu (35% versus 23%), while the rates among urban hukou holders were much higher in Zhejiang (64% compared to 46%). Furthermore, the standard error of the reimbursement rate was much higher for urban hukou in Zhejiang than for Gansu (12.4 versus 7.3).¹⁹ This is important because the county-level units are allowed to set the parameters for the health insurance, and apparently there is a wider variance of experience related to reimbursement rates in urban Zhejiang than in urban Gansu. Also since Zhejiang is far richer than Gansu, there is more scope for higher reimbursement rates should the county-unit choose.

We checked the variation in reimbursement rates between and within counties, using an analysis of variance (ANOVA), and find that counties account for 24% of the variation for inpatient rates in Gansu and 56% for Zhejiang. Our finding that more variation exists between counties in Zhejiang than in Gansu makes sense given that there seems to be more experimentation in insurance policy parameters among counties in Zhejiang than in Gansu, in part perhaps because of its greater wealth.²⁰ When we disaggregate plans and run an ANOVA for inpatient respondents with the NCMS, we find 41% of the variation in Zhejiang is between counties versus 30% in Gansu. We do find important variation between counties, which we expect since counties have control over setting parameters of the programs, but even more variation within counties. Reasons for this variation include not only that we are aggregating over different insurance schemes (because of sample size consideration), but also because even for a given program, reimbursement depends on factors such as the particular reason (illness) for going to a health facility, the type of facility visited, and the total cost. These factors vary by individual and are apparently important.²¹

Yi et al. (2009) report that in the five provinces in which they did their study (which did not include either Gansu or Zhejiang), as total costs of inpatient treatment increased, reimbursement rates fell for the NCMS, often to quite low levels, around 10%. If total costs are taken as a proxy of severity of the health problem, then this strongly suggests that the major new rural health insurance scheme is not covering catastrophic illnesses well and is certainly not adequately covering catastrophic health expenditures.

The CHARLS pilot data show a very interesting, and somewhat different, story. We start in Figure 14-1 by nonparametrically plotting reimbursement rates for inpatient services against total inpatient costs, using data on respondents with any insurance who reported inpatient utilization over the past year.²² It is clear that reimbursement rates rise, not fall, as total costs rise. Reimbursement rates rise from 31% to a plateau of 40% at a total cost of 10,000 Yuan, which is right at the 75th percentile for the total cost distribution. Mean pce for our households is approximately 8,000 Yuan, and the median is 5,800, so an expense of 10,000 is quite large compared to mean pce. For costs above 10,000 Yuan, though, reimbursement rates stay flat at 40%. If the system were effective in covering catastrophic health expenditures, then the reimbursement rates should continue to rise; they do not.

So the glass seems to be half full. There is nontrivial coverage for inpatient expenses that does rise some as total costs rise, at least up to a point. Furthermore, when this situation is compared to the period before 2003, this represents a large improvement. Finally, we note that further reform and experimentation is ongoing, so this is not the final story.

For outpatient reimbursement, the story is quite different. The new health insurance schemes, particularly the rural scheme, were initially designed to help with inpatient service, so many county-units are apparently choosing not to reimburse for outpatient service (see Table 14-10 Panel A). The mean rural reimbursement rate is only 8.5%, being slightly higher in Zhejiang. The urban rates are considerably higher, a mean of 30%, but more than double in Zhejiang (36%) than in Gansu (14%). The urban reimbursement rates are significantly higher than the rural ones in Zhejiang and aggregating both provinces. Again, the standard error is much higher in urban Zhejiang than in urban Gansu, suggesting more experimentation in urban Zhejiang.²³

If we split the sample by rural and urban hukou, the differences in reimbursement rates become starker (see Table 14-10 Panel B). Now, urban hukou holders face 63% reimbursement in Zhejiang and 17% in Gansu, which is still double the reimbursement rates for rural hukou holders in Gansu. Urban-rural hukou differences are significantly different at .01.

If we look only at reimbursement rates for people holding rural insurance through the NCMS (results not shown), we see a similar pattern to the rural population. Rates of reimbursement are about 10% for urban residents and 8% for rural, and the provincial differences are quite small. Clearly, this insurance scheme was not designed to cover outpatient service, just inpatient. However, it remains to be seen in the future whether these same patterns hold. It may be that over time, reimbursement rates may rise for outpatient service, as they have for the urban plans in Zhejiang.

Table 14-11 reports regressions on the cost share paid, separately by men and women, for outpatient service.²⁴ The sample is those persons who used outpatient services in the last one month and who have insurance that could be used to get reimbursement. The SES variables, education, and pce are insignificant in all cases. Age is important for men, but not women. Older men have lower out-of-pocket costs as a percentage of the total. The regional dummies are significant, with lower repayment rates in urban Zhejiang, as we saw in the previous tables. The community dummies are also significant.²⁵ This is interesting because, as discussed in Brown and Theoharides (2009), most of the choice regarding insurance parameters is set at the county-level. Thus, there is great scope for experimentation, which is evidently ongoing.