The New York Times has a fascinating article on Steven Levitt, an economist with a very different take on the dismal science, The Probability That a Real-Estate Agent Is Cheating You (and Other Riddles of Modern Life). I remember reading about his famous abortion-and-crime paper a few years ago:
He is a prolific and diverse writer. But his paper linking a rise in abortion to a drop in crime has made more noise than the rest combined. Levitt and his co-author, John Donohue of Stanford Law School, argued that as much as 50 percent of the huge drop in crime since the early 1990′s can be traced to Roe v. Wade. Their thinking goes like this: the women most likely to seek an abortion — poor, single, black or teenage mothers — were the very women whose children, if born, have been shown most likely to become criminals. But since those children weren’t born, crime began to decrease during the years they would have entered their criminal prime. In conversation, Levitt reduces the theory to a tidy syllogism: ”Unwantedness leads to high crime; abortion leads to less unwantedness; abortion leads to less crime.”
How about this study, on drug dealers?
Venkatesh was Levitt’s co-author on ”An Economic Analysis of a Drug-Selling Gang’s Finances,” which found that the average street dealer lives with his mother because the take-home pay is, frankly, terrible. The paper analyzed one crack gang’s financial activities as if it were any corporation. (It was Venkatesh who procured the data, from a former gang member.) Such a thing had never been tried.
I love this bit:
He takes particular delight in catching wrongdoers. In one paper, he devised a set of algorithms that could identify teachers in the Chicago public-school system who were cheating. ”Cheating classrooms will systematically differ from other classrooms along a number of dimensions,” he and his co-author, Brian Jacob of the Kennedy School of Government, wrote in ”Catching Cheating Teachers.” ”For instance, students in cheating classrooms are likely to experience unusually large test-score gains in the year of the cheating, followed by unusually small gains or even declines in the following year when the boost attributable to cheating disappears.”
Levitt used test-score data from the Chicago schools that had long been available to other researchers. There were a number of ways, he realized, that a teacher could cheat. If she were particularly brazen (and stupid), she might give students the correct answers. Or, after the test, she might actually erase students’ wrong answers and fill in correct ones. A sophisticated cheater would be careful to avoid conspicuous blocks of identical answers. But Levitt was more sophisticated. ”The first step in analyzing suspicious strings is to estimate the probability each child would give a particular answer on each question,” he wrote. ”This estimation is done using a multinomial logit framework with past test scores, demographics and socioeconomic characteristics as explanatory variables.”
So by measuring any number of factors — the difficulty of a particular question, the frequency with which students got hard questions right and easy ones wrong, the degree to which certain answers were highly correlated in one classroom — Levitt identified which teachers he thought were cheating. (Perhaps just as valuable, he was also able to identify the good teachers.) The Chicago school system, rather than disputing Levitt’s findings, invited him into the schools for retesting. As a result, the cheaters were fired.
His bio amazes me:
He comes from a Minneapolis family of high, if unusual, achievers. His father, a medical researcher, is considered a leading authority on intestinal gas. (He bills himself as ”The Man Who Gave Status to Flatus and Class to Gas.”) One of Levitt’s great uncles, Robert May, wrote ”Rudolph the Red-Nosed Reindeer” — the book, that is; another great uncle, Johnny Marks, later wrote the song.
At Harvard, Levitt wrote his senior thesis on thoroughbred breeding and graduated summa cum laude. (He is still obsessed with horse racing. He says he believes it is corrupt and has designed a betting system — the details of which he will not share — to take advantage of the corruption.) He worked for two years as a management consultant before enrolling at M.I.T. for a doctorate in economics. The M.I.T. program was famous for its mathematical intensity. Levitt had taken exactly one math course as an undergraduate and had forgotten even that. During his first graduate class, he asked the student next to him about a formula on the board: Is there any difference between the derivative sign that’s straight up-and-down and the curly one? ”You are in so much trouble,” he was told.
”People wrote him off,” recalls Austan Goolsbee, the Chicago economist who was then a classmate. ”They’d say, ‘That guy has no future.”’
Levitt set his own course. Other grad students stayed up all night working on problem sets, trying to make good grades. He stayed up researching and writing. ”My view was that the way you succeed in this profession is you write great papers,” he says. ”So I just started.”
Another interesting paper:
Then he happened upon a political-science book whose authors claimed that money wins elections, period. ”They were trying to explain election outcomes as a function of campaign expenditures,” he recalls, ”completely ignoring the fact that contributors will only give money to challengers when they have a realistic chance of winning, and incumbents only spend a lot when they have a chance of losing. They convinced themselves this was the causal story even though it’s so obvious in retrospect that it’s a spurious effect.”
Obvious, at least, to Levitt. Within five minutes, he had a vision of the paper he would write. ”It came to me,” he says, ”in full bloom.”
The problem was that his data couldn’t tell him who was a good candidate and who wasn’t. It was therefore impossible to tease out the effect of the money. As with the police/crime rate puzzle, he had to trick the data.
Because he himself had typed in the data, he had noticed something: often, the same two candidates faced each other multiple times. By analyzing the data from only those elections, Levitt was able to find a true result. His conclusion: campaign money has about one-tenth the impact as was commonly accepted.
An unknown graduate student, he sent his paper to The Journal of Political Economy — one professor told him he was crazy for even trying — where it was published.
Read the whole article.