 Hi everyone, I'm Diyang and I'm actually from Penn State University. Many people mistaken it as a U-Pen, I wish it was U-Pen. But again, the presentation today is about social health insurance consolidation in China and its impact on urban rural inequality in medical utilization and financial risk protection in China. So first of all, let me give you some background information about the social health insurance programs in China. So there are three major social health insurance programs and Chinese government basically divided the population into three subgroups. So the NCMS, the new cooperative medical scheme is for rural residents and UEBMI, the urban employee basic medical insurance covers the urban employee population. And the last URBMI, the urban residents, the urban residents basic medical insurance covers the urban residents. The problem in this system is that it is creating inequality by policy design because these three programs, they have different levels of medical benefits. For example, someone caught a flu and go to see a doctor. For example, the medical cost is like 100 yuan. Then for the UEBMI, what's the laser? Which one is the laser? So for UEBMI, for the urban employee population, this is the one with best benefits. So for example, if the doctor visited costs 100 yuan, just give an example, the urban employee insurance will probably reimburse like 60 yuan. But for urban residents, probably only like 50 yuan and for rural residents, they're probably only like 40 yuan. So this system is creating inequality by policy, by design, and that is obviously not bad, obviously bad because China is a socialist country. So we're pursuing universal healthcare for the healthcare system to be a universal healthcare. Then everyone should have the same medical benefits no matter where you live, where you're in the rural area or you're in the urban area. But this system is creating inequality by design. So in 2016, there is a consolidation for the two programs. So for the urban resident insurance and for the rural resident insurance, it was consolidated into a new program in 2016. So the new program takes the better benefits for the two old programs. So nobody is worse off in this consolidation. As for the urban employees, their insurance program is not impacted. So they're not better off or worse off. They're just not relevant. And this is a pretty good situation for the study because I can use urban employees as the control group as I show you in the next few slides. So for the policy goal of the consolidation, it was gradually to reduce the urban rural inequality in healthcare, particularly among urban and rural residents. And obviously the natural research question is, did this consolidation really reduce the urban rural inequality, particularly urban rural gaps in total medical expenditure and reimbursement? Well, I choose total medical expenditure because it measures the medicalization and reimbursement measures the financial protection of the social health insurance program. And the data I use is China Family Panel Study. It's conducted every two years and it's pretty large, like 20,000 or 30,000 individuals sampled every time. And it's a panel study, so the same household is tracked every two years. And it asks a lot of questions across healthcare, demographics, your occupation, like they're asking everything. The question is like 50 pages, so super long. And we're using 2012 and 2014 and 2018 data sets. We're not using 2016 because the consolidation is in 2016. So we can't say it's pre-consolidation or post-consolidation. So we just decide not to use 2016 at all. So the method we use is a kind of a two difference in differences, just stack them together. It's an augmented method. So we're trying to get the difference into a different def estimator. So this is illustration of the method. So you see on the top green line, this is a control group. This is the urban employee population. So they're not impacted by the consolidation. So they're the control group. And there are two treatment groups here. These two groups are consolidated. So the first treatment group is the urban resident population. So this will be the first DID. So this will be delta 1, the first DID captured by beta 4 in the equation. If you look at the equation, it's pretty simple. It's just two DIDs put into the same equation. And the second DID is between the urban employees and rural residents. So this is the second DID captured by beta 5 here. So the gap, the changing gap between urban residents and the rural residents will be just delta 1 minus delta 2, which is beta 4 minus beta 5. And STATA actually has a very simple code that after you estimate this equation, whether it's OLS or other regression, you just do a post estimation test, just test beta 4 minus beta 5. And the STATA will tell you if it's significantly different from 0. And you will see if the gap is increased or not. And the result, spoiler alert, the result is not very good. The consolidation actually did not reduce urban rural gap in metro expenditure reimbursement or out-of-pocket, at least not in a statistically significant way. So I'll show you the table here. Well, we have three outcomes. So total metro expenditure, reimbursement, and out-of-pocket. So for the total metro expenditure, I use a two-part model because total metro expenditure, they have a cluster of zero because a lot of people are healthy and they don't really go to hospital at all. So they're not using any medical service. So their medical expenditure is zero in the previous year. So that's why we're using two-part model. And you can see its p-value is only 0.08, which is not significant. I think what's even more disturbing is that actually the magnitude is positive. It's 399, which is positive. That means the gap is increasing. But more importantly, it's not significant at all. The reimbursement also have some clustering as zero because, for example, if you utilize, if you use some medical service that is not covered by the insurance, then you're not receiving any reimbursement. So there are also some clustering at zero. So I also use two-part model. And you can also see it's not significant here. As for the out-of-pocket, so for out-of-pocket, there's no clustering at zero because there's no free healthcare. You have to pay more or less. So I just use OLS here. And you can also see it's not significant here. So none of this is significant. So the gap actually did not reduce in a statistically significant way. Because I use different diffs, so obviously I need to test the parallel trend. So basically we run the same regression using 2012 and 2014 data. And because of the data limitation, some outcome variables, namely reimbursement and out-of-pocket, they're not available in the 2012 data. So we're only able to test the parallel trend for the total medical expenditure, which is, again, a major limitation, which I will mention later. And before the total medical expenditure, the parallel trend is not violated. The parallel trend did not yield evidence that the parallel trend is violated. Well, I guess a very natural question is why? Because the policies seem to consolidate to different programs, so people would expect at least reduce more or less a little bit of inequality. So why the urban rule inequality did not reduce in a statistically significant way? So I have some hypotheses. The first hypothesis is the health characterization patterns change differentially for urban and rural residents as a result of the consolidation. For example, maybe urban residents, because of consolidation, they have a bigger increase in likelihood of doctor visits when they feel sick. Or because of consolidation, urban residents are much more likely to visit high-cost medical facilities like specialty medical facilities. This is particularly relevant in China because the expensive hospitals are always in the metropolitan area. And in urban area, the medical facilities are usually limited or underdeveloped, and they're obviously low-cost in rural area. So I tested these two possible explanations. So the dependent variable is a binary. So do you go to see a doctor when you feel discomfort? And I use the same specification of the regression equation, and you can see it's not significant. So, well, you can see... I think I can explain all of the coefficients here. Okay, so for the beta one, urban residents insurance... I'll just go here, start from here. So for beta three, it's significant and it's positive. So compared to before the consolidation, after consolidation, people are 77 percentage points more likely to visit a doctor when they feel discomfort. So the consolidation actually increased people's likelihood to visit doctor. So that's a good thing, but unfortunately the inequality doesn't really reduce. This is whether you usually visit a high-cost facility. So there's a question like what type of medical facility do you usually visit when you feel sick, and I classify them into whether it's high cost or not. And you can see the urban rural gap in that regard is also not reduced. But for this one, for beta one, this is an urban resident population. So the reference is the urban employees, right? So the gap between urban residents and urban employees actually reduced in the statistically significant way, a 13 percentage point, which is quite a big reduction. But for the gap between urban residents and rural residents, it's not statistically significant. So this hypothesis is not the case here. So I have the second hypothesis. It's the provincial heterogeneous effect. So it's possible that in some provinces the urban rural gap is reduced and in other provinces urban rural gap is widened. And on the national average, the gap probably doesn't change or doesn't reduce in a statistically significant way. So I group the provinces into four groups here by their physical spending on healthcare. So in the highest physical spending on health, for the highest healthcare physical spending provinces, you can see there the urban rural gap actually increased. So the rich provinces is making trouble here. So the second hypothesis is what we think is the root cause for not seeing a statistically significant reduction in urban rural gap. So the conclusion is that 2016 conservation did not reduce the urban rural gap in total medical expenditure and reimbursement, at least not in the short term. But we do not know if the gap will continue to get wider or it will reduce over time because we only have the data until 2018. The 2020 data is going to be released soon. So I guess I can revisit that. The policy recommendation here, I think it's more on the supply side because the urban, there are like many high cost and good medical facilities in urban area, but in rural areas really under invested. Also in urban area there should be more gatekeeping so that people can't just go to the high cost facilities no matter what kind of small units they have. For the flu, you really don't need to go to a specialty hospital at all. The limitation here, so first of all, it's a parallel train assumption. We can only test it on total medical expenditure but not for reimbursement and out-of-pocket. There could be other policies between 2014 and 2018 so that could be other potential policy changes that we were not able to tease out. But still we can say that between 2014 and 2018 the urban rural gap did not reduce. And also we can say that because we control for county fix effect, so we really just see the difference between the same county. And also we don't have reasons to believe these policies would impact urban and rural residents in a different way. So the last slide, I'm on the draw market. So if you know any potential job opportunities, please email me. I have my website, also my QR code. You can scan it to see my CV and my paper. And I don't really care about geographic or locations. So for example, if Oony on this want to recruit me, I'm very glad if that's an opportunity there. Thank you, everyone.