Friday, November 13, 2009

Freakonomics

I recently read "Freakonomics" the 2005 bestseller by economist Steven Levitt and Stephen Dubner . Dubner wrote a very favorable profile of Levitt for the New York Times magazine. This book is a greatly expanded version of the profile. Recently the authors have written a sequel "SuperFreakonomics" which I have not read.

Freakonomics was easy and entertaining to read. It is not surprising that it was a success. However I have doubts about how reliable it is.

The most controversial claim in the book is the assertion that the drop in crime in the 1990s was in large part a delayed result of the legalization on abortion in the 70s. The idea being that legal abortion reduces the number of unwanted children who are more likely to become criminals. The claim comes from this academic paper by Donohue and Levitt. This claim has not been generally accepted. I have serious doubts just based on this homicide data . Looking at the graph for black males homicide rates for the 14-17 age groups and the 18-24 age groups rise together, peak in the early 90s and then fall. If legal abortion were causing the fall it should have showed up first in 14-17 group. The data seems much more consistent with an epidemic of violence (plausibly caused by the invention of crack cocaine) which affected both age groups similarly and then burned out. Foote and Goetz found errors in the Donohue and Levitt paper which appear to invalidate the results. Donohue and Levitt admit errors but claim a somewhat different analysis produces similar results to their original claim. Without getting too deep in the weeds here, I consider this pattern of coming up with a new argument that produces the same results as an earlier refuted argument to be a bit of a red flag as it suggests the method of analysis may have been chosen to produce the desired results. It is more objective of course to choose the method of analysis first and then accept whatever results it produces.

The same pattern was seen with the claims in the book about more police reducing crime. They are based on this 1997 Levitt paper. Justin McCrary found Levitt had made a serious programming error which when corrected invalidated his results. Levitt admits error but claims the results were correct anyway.

Levitt admits in the NYT profile that "... I'm not good at math, ...". This is a problem when you are doing complicated statistical analysis where it is easy to go wrong. Perhaps Levitt needs a coauthor who is good at math to keep him from error. Incidentally McCrary acknowledged that "Steven Levitt provided both data and computer code.". This is greatly to Levitt's credit and unfortunately far from universal among academics.

There are other claims in the book which have prompted criticism. One is the claim that a swimming pool in your home is more dangerous to a small child than a gun. However this seems correct to me although saying more that swimming pools are dangerous than that guns are safe. Another is an unsourced throwaway assertion that the average prostitute earns more than the average architect which seems dubious.

In general Levitt's work is interesting although somewhat peripheral to the main concerns of economics. This doesn't mean the issues are unimportant, the causes of crime are certainly important. But you won't learn much standard economics from Freakonomics. And as Daniel Davies says in this review Levitt has a bad habit of presenting his way of viewing complicated issues as only way (although unlike Davies I didn't find the pool gun comparison particularly objectionable).

So in summary if you want an entertaining look at some interesting and provocative topics Freakonomics is a good choice. But don't take it as the final word.

1 comment:

  1. Nice review.
    Crime could also increase if girls become addicted, then become pregnant, and produce crack babies. These children grow up with little self control or judgement. This may imply an earlier time for the beginning of the souce of the crime pattern, since the children would need to be about 14-17 years old.

    ReplyDelete