Thursday, July 14, 2005
Yesterday, I finished reading Freakonomics. Before I comment on the book, I should admit that I found myself continually telling Brian about different stories and data from the book. As promised in the introduction, the book does provide a good deal of cocktail party fodder. Beyond that, it provides some dumbed-down summaries of economic research done by Levitt and other economists. It's a light and very short read; it's interesting and entertaining. Rather than write a full book review, I will offer some general comments and observations.
I'm sure it's a marketing ploy by the publisher, but I question labeling Levitt a "rogue" economist. Other economists, even those cited in the book, research off-beat topics. See also.
It's not like Levitt is challenging or questioning economics fundamentals or work by esteemed economists. [In this regard, I'd say Krugman is more of a rogue; rogues make enemies.] I'd bet almost any economist could outsmart the conventional wisdom.
The clever part of Levitt's approach is bucking the academic system to turn economics on everyday or non-economic questions. The fact that he has good economics establishment credentials gives him the freedom to ask and answer those questions. Levitt is clearly very much part of the economics establishment. He went from undergrad to full professor in an elite department in less than a decade. That's certainly impressive, but not roguish. (I suspect if an untenured econ prof at a mid-range state school tried to study similar questions it would be considered a waste of time rather than clever.)
Levitt and Dubner really do try to explain economic principles and research in a way that non-economists and non-social scientists can understand. I think they succeed at doing that. I'm thinking the book would make a good gift for my dad, and I plan to recommend the 3-4 page description of regression analysis to my graduate (master's) methods class in the fall.
They also try to cover a variety of topics that many non-academics would find interesting: cheating, real estate, the KKK, gangs, crime, child car seats, guns, swimming pools, and baby names. Each topic is presented in a similar fashion: a couple of anecdotes followed by summaries of economic research by Levitt or others.
Sometimes the anecdotes are interesting. Sometimes not. Sometimes they help demonstrate the point. Sometimes not. For instance, the authors want to demonstrate that one of the principle goals of the Klan as an organization was to generate income through dues and robe sales. They present surprising data that lynchings declined in the U.S. at the same time that Klan membership began to rise. This is used to support the claim that the KKK served other purposes.
Then, the authors digress into an interesting, but not relevant, discussion of how Stetson Kennedy infiltrated the Klan and provided the producers of the Superman radio program with the code words and secret handshakes of the Klan in an effort to discredit the organization. Interesting, but not directly related to their argument. [Listen to the interview on This American Life about this part of the book and Stetson Kennedy.] Stetson Kennedy is certainly important because he provided information on the dues of the Klan and hurt the organization's ability to recruit, but I'm not sure the digression really supports their argument.
Returning to the data, the authors present annual lynching data (page 61) and indicate that there is no correlation between growing Klan membership and declining lynching rates. Initially, I was surprised by the lynching numbers. My general impression was that lynching was still a problem into the 1950s, and I found it problematic that the definition of lynching for their data was not clearly defined. Lynchings drastically declined dramatically after the 1940s. Research on lynchings suggests that the decline in the 1940s was in part a result of the national debates and federal anti-lynching laws proposed in 1938/1939 in the national Congress. An interesting NPR segment interviews an Af-Am scholar on this topic. (As an aside, the data mentioned by the interviewed scholar differs from that presented in Freakonomics.) Freakonomics does not discuss or mention the research explaining the decline in lynchings. Nor do the authors acknowledge that perhaps the Klan (in addition to dues-generating) was more about intimidation and fear, which was quite possible without lynching. By using the drastic decline in lynching numbers to support their claim that the Klan was about making money, they underestimate the extent to which the Klan engaged in other reprehensible activities. In some ways, it seems that using the lynching data to make their argument about the other organizational goals of the KKK is misleading. (Dr. History does a better job explaining this lynching data problem.)
In other parts of the book, the presentation of the data is ambiguous. For authors who are trying to explain or discount the conventional wisdom using hard data, it seems unlikely that such ambiguities are accidental. Take, for instance, the following section where the authors discuss crack and crime rates:
The violence associated with crack began to ebb in about 1991. This had led many people to think that crack itself went away. It didn't. Smoking crack remains much more popular today than most people realize. Nearly 5 percent of all arrests in the United States are still related to cocaine (as against 6 percent at crack's peak); nor have emergency room visits for crack users diminished all that much. (134)Now, the authors have just spent several pages explaining how crack is different from powder cocaine. Crack is a rock created from baking powder cocaine with other ingredients. Crack is cheaper than powder cocaine. Crack had a different clientele than powder cocaine. Then, why do the authors use data regarding overall cocaine arrests to suggest that crack cocaine is still popular? What they need to present is data on crack arrests. It could be that overall cocaine-related arrests have not declined, but the composition of those arrests (i.e., powder vs. crack) could have changed in the 1990s. How we know that crack use didn't decline while powder use (and arrests) among dot-com millionaires or others enjoying rising incomes in the 1990s increased? A plausible hypothesis could be that economic growth and rising incomes in the 1990s led to an increase in powder cocaine use and crack decline, while the overall cocaine arrest rate changed very little. In any event, the data do not necessarily support the argument. Such statements are why we try to teach our students not to lie with statistics and how to spot such lies in the press.
Here's another example of data presented to support their argument that left me asking many questions about the data:
In a given year, there is one drowning of a child for every 11,000 residential pools in the United States. (In a country with 6 million pools, this means that roughly 550 children under the age of ten drown each year.) Meanwhile, there is 1 child killed by a gun for every million-plus guns. (In a country with an estimated 200 million guns, this means that roughly 175 children under ten die each year from guns.) The likelihood of death by pool (1 in 11,000) versus death by gun (1 in 1 million-plus) isn't even close.(149-150)In a country with 6 million residential pools? Who knows? One child killed for every million-plus guns? Are those only privately-owned guns? Does it include legal or illegal? And the 200 million gun estimate? Are those legal or illegal? Are those privately-owned? I'm not sure these data can be used to support this conclusion. You need to compare the likelihood of accidental drowning to accidental shooting in the home. To do that, you need to have data on those deaths and residential pools. You need to have data on accidental shooting deaths (excluding those of intentional homicide) and guns in the home. Maybe that's what these data are, but it's certainly not clear from the presentation.
These are just two examples from the book that left the scratching my head. They seem symptomatic of the way journalists (mis)use data to justify an argument. It wouldn't have hurt readability of the book that much to be a little more careful with data presentation, especially since the point of the book is to use data to question conventional wisdom.
These criticisms aside, I liked the book. It was interesting. I found myself telling Brian about most of the main points in the book. I'm sure Levitt is really smart. I'm sure Dubner is a good journalist. It had a nice beat that I could dance to. I'd give it a 8 out of 10. Krugman, however, is still my favorite economist to read for pleasure.
[I'd like to know what Bitch, Ph.D. thinks of the abortion-crime rate argument. A search for Freakonomics and Levitt didn't turn up a hit on her site.]
See also this interesting discussion on Crooked Timber. Should have read it before I wrote my post, but "oh well."
posted by Michelle @ 1:35 PM,
- At 7/15/2005 5:12 PM, Schroeder said...
Michelle. Nice review. I read something about the book a few weeks back that paralleled your interpretations.
I don't know if this is on your radar or if it's of any interest to you, but I saw this today:
- At 7/16/2005 7:34 AM, said...
I'm not sure how you came up with that critique of the lynching analysis. The authors say that the lynching data is provided by the Tuskeegee Institute. They also put forth the hypothesis that the KKK (as a special case of white racism) was all about intimidation and fear --
"The most compelling explanation is that all those early lynchings worked. White rascists -- whether or not they belonged to the Ku Klux Klan -- had through their actions and their rhetoric developed a strong incentive scheme that was terribly clear and terribly frightening..."
The lynching frequency declines monotonically while Klan membership oscillates wildly over the 80 years of Tuskeegee data, which is why they claim no correlation necessarily exists. As to the digression, it was what finally pushed me over the line to buy the book tonight (at a local HP VI event).
Looking forward to the crack-cocaine analysis, to see if we agree about the strong and weak points or not. And I'm enjoying the blog alot after stumbling onto it tonight.
- At 7/18/2005 2:50 PM, Ken Houghton said...
The crack/cocaine problem is not Leavitt/Dubner's directly. Despite different sentencing guidelines for the two chemically-identical substances, they are still grouped in Federal reporting as "cocaine." (Just as being a "parent" doesn't mean you specifically have one, two, or ten children.)
So they used the most accurate information available--emergency room visits. (One might suggest that the ERs have a vested interest in categorizing "crack" separately. Noting the clientele difference, it is more likely that a crack abuser will be uninsured than a Cokehead. [Who is more likely to be insured, the Chris Rock character in New Jack City or Larry Kudlow?])
Having lived in Washington Heights during the time of the "epidemic," I'm more inclined to argue that crack went through the traditional "Business Cycle" for new products, reaching "maturity" ca. 1991-1993. (For all the talk of an epidemic, the damage was largely limited to neighborhoods and large cities, which is not to minimize the effect--it caused the only time Senator Pothole ever visited my 'hood, and another neighborhood that was significantly altered for the worst is the one right above where Dubner lives--but to note that it compared poorly to the current meth problem, which appears to be more EPIdemic.)
- At 7/18/2005 2:52 PM, Ken Houghton said...
Btw, completely agree with you about Leavitt's not being a "rogue" economist. If nothing else, he's following directly in the footsteps of Gary Becker.
- At 7/19/2005 1:20 PM, Ken Houghton said...
Yikes. Having now checked Crooked Timber, I see that Henry Farrell already made the Becker analogy, with details.
- At 7/21/2005 10:26 PM, Michelle said...
Sorry for the delay in responding; always bad form to make a long post and then disappear for a week.
Anonymous, I guess my original point was not clear. I just meant that the authors seemed to set up a straw man in the way they constructed their argument. The logic was something like: "look, there's no correlation between fluctuations in KKK memberships and the lynching data....therefore, KKK membership must serve some other purpose." If an undergrad turned in a paper with similar logic, it would be resoundly rejected. I would ask the student to think about what question they want to answer and then construct their argument around that question.
What is the authors' question about the KKK? Well, they aren't interested in explaining lynching rates. They are interested in Klan membership and what drives the organization to expand membership (and here I'm working backward from their answer to the question).
So, if the question is "What drives Klan membership?," it's not clear that using the lynching numbers is an effective way to get at that question. Other data, such as unemployment for poor whites or Southern white unemployment, might have been a better foil for setting up their argument that Klan membership served other revenue-generating purposes.
I'm sure a rhetoric teacher would have a more eloquent phrase to describe the problem with the argument's construction, but something is troubling about the way they used the lynching data as an alternative hypothesis or explanation of KKK membership.
And Ken, I see your point(s), too. I'm sure crack went through a business-cycle and I understand that you can't always get the data you need to support your argument/test your theory. But, in the book, the authors talk about arrests for cocaine (in general) and then make an argument about crack. (I haven't read the academic research upon which the book is based.)
In this case, my problem was with the presentation. It was more about presenting data as if it justifies or supports your point, when really the data aren't necessarily specific enough. This is really common in journalism, and usually readers don't pick up on it. But it's the type of presentation that gives statistics a bad name (lies, damn lies...and all that).