Playing With Numbers, Health Care Edition.
In the United States, a shocking number of people have no health insurance and poor access to health care. Many go bankrupt because of medical bills. We desperately need reform. What we don’t need, however, is bad charts.
There are some major problems with this chart.
It’s got two different units superimposed. On the left hand side, there is the number of “Uninsured Americans” and on the right, “Unemployment Rate.” This problem is exacerbated because the scales are chosen so that the two lines track with the same slope. If you notice, the range of values on the right is from 44 to 50 million. That’s a difference of 12%. The range on the right is 4% to 8%, or a difference of 100%. Additionally, the lines are smoothed (this is monthly data, but the lines smoothly move between each sample, implying data that doesn’t exist). If instead of charting these absolute numbers with smoothing, you charted how they changed over time, you’d get something like this:
Chart provided by Visible Certainty
That’s a lot less dramatic. This leads us to next problem: the data is a little fishy.
Notice, there are three sources, the Urban Institute, the Bureau of Labor Statistics (BLS), and the Census Bureau. The BLS is the source of the unemployment rate. There are well known flaws in BLS methodology (e.g., Wikipedia’s explanation), but that’s not my main beef. My main problem is that the uninsured figures are guesses. In fact, no one collects monthly figures for the uninsured (that’s a problem in it’s own right!). According to the report, they rely on an earlier study that found for every 0.1% increase in unemployment, 110,000 people lost health insurance. That may be true, but it’s only part of the picture.
Their figures assume a 1:1 correlation - for every 0.1% the unemployment rate changes, the number of insured changed by 110,000. The flaw with this is that there are other factors that determine insurance levels. A simple thought experiment: if unemployment went to 0%, that would mean, based on March 2007 figures, that 4.8 million people would gain insurance (44 * 110,000). But that would still leave 40 million people uninsured. Clearly something more complicated is going on here than just employment rates. After all, many employers don’t offer insurance. In fact, I suspect uninsured rates are even higher with the economic crisis because many employers are probably eliminating health benefits.
Now, clearly the massive increase in unemployment, which shows no sign of abating, is going to make the situation worse. But there’s a lot more to the story. For example, what if we looked at the change in aggregate national output (GDP) and the change in the uninsured? I could only get figures from the Census for 1999-2007. The results are a little shocking:
Chart provided by Visible Certainty
Now that is definitely an interesting story. The larger point is: make your point honestly and accurately. Don’t fudge the charts (with dual axes) or the numbers. And when it comes to the scandalous state of health care in this country, there’s no need to tell anything but the verifiable truth.
— Jason Y., Verifiable.com
