Control Variables

After compiling the panel datasets, the only thing left to do is updating the control variables. Controls are things like average personal income, number of elderly persons, number of persons of a given race or gender in a county and the number of beds of a hospital. Control variables are extremely important! Remember, we want to find the effect of unemployment on the number of Medicare patient procedures. So, let’s say we find a positive relationship between them (when we’re not using controls). It could be because counties with more minorities have higher unemployment and minorities are more likely to have Medicare and go to the hospital than whites. If this is the case, unemployment actually has no effect on procedures and is just picking up the effect of race. So last week I searched through the US census and Florida websites in search of the needed control variables and now we have as many control variables for as many years as we can. All that’s left is for the 2009/2010 data to come and be cleaned, and then we’re ready for business. I can’t wait!!!


Last week we made one awesome breakthrough with the Florida inpatient data. Unfortunately, the data collectors in Florida changed how they recorded doctor ID’s in 2006, which means that we can’t can’t follow a doctor’s behavior throughout the whole time period even though they have worked for the whole period. We broke the Florida data collector’s code, and I’ve been busy this past week running checks to make sure it actually works. I ran a test on the coded vs original doctor ID to see if they are working in the same hospital and preforming the same procedure. So far it looks good!

Health Economics

I am working for the economics professors in the Health Center. They are trying to determine 1) if people with insurance go to the hospital less during recessions and 2) if as a result, doctors substitute for this loss of income by either preforming more procedures on their medicare patients. The data set we’re working with tracks all patients in Florida hospitals from 1997 to 2008 (about 14M patients). My job is to “clean” the data, which basically involves getting it into a format that the Professors can easily run regressions on.