What Have I Learned from COVID-19 Data So Far...

Everyone knows that there is an argument about the data of the number of confirmed cases. Many papers were written about understating the number of infections. The typical story is that, with the absence of reliable and timely tests, we may have understated the infections by a factor of 10. Because the US government pays hospitals for treating infected patients, and also pay them more if the patients need to be on ventilators, some hospitals were overstating the number of COVID-19 deaths related cases. Minnesota was one state circulating in the news. So, if the numbers of confirmed cases, which we are using from Oxford University, the Johns Hopkins, or else in our research, are understated, researchers must deal with these measurement errors. Ordinary Least Squared regressions estimates are biased and inconsistent. IV estimators should be used. Testing for COVID 19 matters for reducing the number of deaths even though millions remained untested. See my paper

The second thing I learned from COVID-19 data is that the modelling of the infection using the Gompertz (1825) function overestimates the peak infection. Usually, we try to model the data as they arrive. The data have a steep upward slope. The Gompertz function is a very suitable model for this kind of events. However, it is a statistical function, which has a couple of fixed parameters. It does not account for policy. So if we have data from time t to t+k and we fit the function up to time t+k+1 without having accounted for policy, we will overestimate the peak infection. Policy (stringent) reduces the number of infections, but the Gompertz function does not take this into account.

Figure (1) plots my estimates of the New Zealand curve, see my paper ...The data that I used in this paper were from Feb 28 to Mar 27. Figure (2) use the same graph but add the actual data up to Apr 23. As you can see in figure (1), I predicted the peak infection to be 2630 cases on April 3. Then we learned when the actual data arrived that the number of infections on April 3 was 772, see figure (2). The peak, probably did not occur until April 22...and much lower than my estimate.

Figure (1)

Figure (2)

I also learned that policy responds to the number of infections positively, and the latter responds to the former negatively. Policy response, however, is endogenous and country-specific. New Zealand and Australia responded quite differently to the infection, but the outcomes of the two countries are pretty much similar. I also learned that if country A adopts country B policy response, country A cannot achieve the same outcomes of country B. I tested whether, or not, the New Zealand policy response, which achieved zero infections, could reduce infection to zero if it is adopted by Denmark, Sweden, and the USA.I found that it is effective in reducing the infections significantly, but not to zero as it did in New Zealand. There are omitted factors that need to be taken into account in such analysis. Culture might be an important missing variable. Although the Swedes and the Danes are seemingly Scandinavians, they followed different polices and the people have been reacting differently. The outcomes are very different. See my paper.     

I am sure that we will learn more from doing more research.