How to Lie with Statistics

What would you conclude if I showed you the following graph?

If you know how to read a scatterplot, you’d probably say that changes in the tax rate have no effect on the number of hours worked. If you follow economic policy debates, you might be surprised by this graph because it contradicts a very common belief: Europe is less productive than the U.S. because their taxes are higher.

But I didn’t show you the whole graph. Here’s how it really looks:  

Do you see the difference? Look closely. The second graph has one dot that the first doesn’t. See it on the upper left?

That’s it. One dot, and suddenly the conclusion is, “a clear negative relationship” where “countries with larger increases in taxes also have larger decreases in hours worked.”

The author of those quotes and the second graph is Arizona State University economist Richard Rogerson. Apparently this work forms the foundation of his new book on how taxes affect the labor supply. Suffice it to say I won’t be buying that book.

Statistics is a tricky business. Regressions rarely tell a simple, clean story. So many factors affect our economic decisions that it’s damn near impossible to pinpoint one cause of any effect.

Statisticians always expect a few datapoints to deviate from the herd, so the rule is that you shouldn’t trust any regression whose conclusions disappear if you delete one or two points.

Rogerson has a PhD. He knows this stuff. But he’s hoping you don’t. Because that’s the only way to convince you that raising taxes will turn us into a bunch of lazy old Europeans.