Friday, June 13, 2008

Freakonomics

I was a bit disappointed by this book, but that may not be its fault.

Freakonomics: A Rogue Economist Investigates the Hidden Side of Everything makes the following proposition:
Economics presents a very powerful set of tools, particularly the study of incentives and statistical analysis. (Though other disciplines might also like to lay a claim on many of these tools.) The trouble is that they've been put to use on very dull issues.

Steven Levitt (economist) and Stephen Dubner (journalist) try to rectify this by turning these tools on to questions such as why drug dealers live with their mothers and what difference good parenting makes. The book is a collection of studies of assorted subjects, rather than an attempt to formally argue the initial proposition or anything else.

Now part of the reason I didn't like the book all that much is probably because I'm already familiar with this kind of approach, mostly from reading the Freakonomics blog. So it hardly has the shock and perspective changing value it might otherwise have.

That aside, the book is rather hit and miss. It's written in a journalistic style, which is fine really, but grated with me a bit. The chapters are very hit and miss. In particular the last chapter on names is horribly boring. The most interesting is almost certainly the chapter on drugs gangs. Coincidentally, this is also a chapter that relies heavily on the research of someone else... It's based on the work of Sudhir Venkatesh, a sociologist who basically spent a long time hanging out with and kind of writing the biography of a drugs gang. That produced some fascinating results, but it's a method quite different to Dubner & Levitt's.

The economics data they take from Venkatesh comes from the gang's carefully recorded account books. What these reveal is that in the four tier structure that the gang was part of those at the top earned a lot - $500,000 a year for the top tier, $100,000 for the gang leaders, both tax free (naturally) - but those at the bottom earned only $7 an hour - less than minimum wage. The top level did have a 1 in 4 chance of being in prison at any particular point, but the lowest level people had a 7% chance of being killed each year. Levitt & Dubner conclude the gang members are living the same dream as millions of americans hoping for a career in Hollywood or the NFL - working a crappy job in the hopes of striking it big. But they fail to address why they have chosen this particular career (waitressing while hoping for acting work isn't great, but it has a much lower mortality rate) or even whether these odds make any sense in the first place. Without more sociological data the conclusion is seriously lacking.

Not that Levitt & Dubner's approach isn't a decent one - there are good examples of what statistical analysis can discover, unintended consequences (particularly teachers getting the incentive to cheat on behalf of their students). If there are two lessons you can take from the book, they are probably: 1) incentives are very powerful motivators; 2) always consider how people will 'cheat' to get them. You can see those issues very clearly in criticisms of government policy here - how they've crudely tried to incentivise public sector workers, but not how creatively they'll 'cheat' to get the rewards. There are some lessons about devising incentive schemes tucked away in Freakonomics, I'm sure, but they're not spelled out much.
Another lesson is: 3) learn how to do (multivariate) regression analysis.

I've got the revised and expanded edition of the book, which includes Stephen Dubner's original piece on Levitt for the New York Times (which I didn't read), and a selection of posts from their blog. I don't really know about their value. They bulk out the book (which is under 200 pages, quite large spaced) without adding much, and nothing that's not freely available on the web.

My advice is: don't bother with the book, but read the blog. You'll soon pick up the key points, and you can happily skip over all the bits that don't interest you or aren't quite so good.