In one sentence: A Chicago economist and a journalist point serious measurement at questions economists were not supposed to care about, and produce the book that created a genre, whose best work is rigorous and whose most famous claim, that legalized abortion drove the 1990s crime drop, was seriously challenged afterward.
At a Glance
Author: Steven D. Levitt and Stephen J. Dubner
First published: 2005, in this Revised and Expanded edition 2006 (William Morrow)
Category: Economics / Society
Length: about 315 pages, about 90,000 words (William Morrow Revised and Expanded paperback)
ISBN-13: 978-0-06-123400-2 (William Morrow Revised and Expanded paperback)
Summary reading time: about 12 minutes
Book reading time: about 6 hours
Notable adaptations: a 2010 documentary film of the same title, plus the Freakonomics Radio podcast and a sequel, SuperFreakonomics
Content note: the book’s best-known claim, that legalized abortion was a major cause of the 1990s crime drop, was challenged after publication on a coding error in the underlying paper and on how the model was specified. This summary reports it as the authors’ claim and reports the dispute rather than presenting it as settled. The crack chapter describes gang violence and quotes language some readers will find offensive. One passage sets out a deliberate thought experiment weighing fetuses against homicides. The Revised and Expanded edition bolts a magazine profile, seven newspaper columns and fifteen blog entries onto the original six chapters.
Steven D. Levitt is a University of Chicago economist who had just won the John Bates Clark Medal when Stephen J. Dubner, a journalist, profiled him for the New York Times Magazine in 2003. That profile became this book. The research is Levitt’s, much of it from his own papers, and the prose is Dubner’s. Written in a joint “we,” it is six case studies, each opening with an odd question, held together by a method rather than a subject.
Read it if you want to see what careful measurement can do with questions that look unanswerable. Skip it if you need every conclusion to carry only the weight its evidence can bear.
The Big Idea
Levitt and Dubner set out five ideas. Incentives are the cornerstone of modern life, and understanding them is the key to almost any riddle. The conventional wisdom is often wrong. Dramatic effects often have distant, even subtle, causes. Experts use an informational advantage to serve their own agenda, and can be beaten at their own game. And knowing what to measure, and how to measure it, makes a complicated world much less so.
Underneath sits a claim about what economics is: above all a science of measurement, a set of tools rather than a subject, so no question is beyond its reach. Morality, they write, represents how people would like the world to work, and economics represents how it actually does.
Key Ideas
1. Incentives come in three kinds, and a small one can backfire
Levitt and Dubner sort incentives into economic, social and moral, and their opening case shows one crowding out another. Two economists studied ten day-care centers in Haifa over twenty weeks. A fine of three dollars per child for being more than ten minutes late, on a monthly bill of about 380 dollars, roughly doubled late pickups, and removing the fine did not bring them back down. The authors’ reading is that the fine was too small, so an economic incentive displaced a moral one and parents could buy off their guilt cheaply. They pair it with the 1987 IRS rule requiring Social Security numbers for dependents, after which seven million “children,” about one in ten, vanished from tax returns.
2. Cheating leaves fingerprints in the data
After Chicago tied promotion and school probation to test scores in 1996, Levitt built an algorithm and ran it across every student in grades three to seven from 1993 to 2000: more than 700,000 answer sheets, nearly 100 million individual answers. It looked for blocks of identical answers on the hard questions at the end of a test, and for classrooms whose scores spiked far above prediction and fell back the next year. The algorithm flagged more than 200 classrooms a year, about five percent. Then the prediction was tested. The school system let them pick 120 classrooms for a retest under its supervision, more than half suspected cheaters and the rest controls. The controls held or improved. The suspected classrooms fell by more than a full grade level on average. About a dozen teachers were fired, and cheating fell more than thirty percent the next year.
3. Sumo wrestlers throw matches when the rankings say they should
Levitt and Dubner examined nearly every top-division bout from January 1989 to January 2000, about 32,000 matches among 281 wrestlers. The eighth win in a fifteen-bout tournament is worth roughly four times a typical victory, so a wrestler on seven wins and seven losses on the final day has everything at stake against an opponent with nothing. Those wrestlers won almost eight times out of ten, when their past record against the same opponent predicted just under half. At the pair’s next meeting, with neither on the bubble, the previous winner wins only about forty percent of the time, which looks like a favor repaid. And when the Japanese press ran allegations of rigging, the final-day figure fell from about eighty percent to fifty.
4. Information is an asset, and giving it away destroys power
The Ku Klux Klan chapter and the real-estate chapter make the same argument twice. On the authors’ telling, the Klan’s power lay largely in its hoarded secrets, and Stetson Kennedy broke it by handing the passwords and rituals to a radio show. Real-estate agents, on their data, keep their own houses on the market longer and sell them for more than their clients’ houses, because the agent’s share of an extra few thousand dollars is too small to be worth the work.
5. A crack gang is a franchise, and its foot soldiers earn less than minimum wage
The chapter rests on Sudhir Venkatesh’s six years inside a Chicago gang and four years of its financial notebooks. The structure was a franchise: about a hundred branches under a board of roughly twenty men, with the local leader paying nearly twenty percent of revenue upward. The leader earned about 100,000 dollars a year tax-free. His officers took roughly seven dollars an hour. His foot soldiers earned about 3.30 dollars an hour, and over four years a member faced an average of 5.9 arrests, 2.4 nonfatal injuries and a one-in-four chance of being killed. The authors’ point is that this is a tournament, a few enormous prizes and a crowd competing for them.
6. The crime drop had three ordinary causes and, on their argument, one extraordinary one
Violent crime had risen eighty percent in the fifteen years to about 1990, and experts predicted worse. Instead it fell everywhere. Levitt and Dubner keep three explanations: more imprisonment, roughly a third of the drop, more police, roughly ten percent, and the collapse of the crack market, roughly fifteen percent. They dismiss the rest with arithmetic. Then comes the argument everyone remembers. Their claim is that legalized abortion led to less unwantedness, that unwantedness leads to high crime, and that legalized abortion therefore led to less crime. The first post-Roe cohort reached its late teens in the early 1990s, exactly when crime turned. Their evidence: five states legalized before Roe and crime fell earlier there, and higher 1970s abortion rates went with larger 1990s crime drops across states even controlling for imprisonment, police and the economy. They concede the relationship might be correlation rather than cause and spend a page arguing otherwise. They close with a thought exercise, offered with no right answer, weighing fetuses against homicides, and find that on any plausible exchange rate the trade is terribly inefficient.
7. Parents matter for who they are, not what they do
Their regression on the Early Childhood Longitudinal Study finds eight factors that correlate with early test scores and eight that do not. What correlates is largely what parents are: highly educated, of higher socioeconomic status, older at the first birth, with many books in the house. What does not correlate is largely what parents do: an intact family, a move to a better neighborhood, a stay-at-home mother, Head Start, museum visits, spanking, television, being read to daily. Their line is that technique looks to be highly overrated, and that a book in the house is less a cause of intelligence than an indicator of it. On names they argue, drawing on Roland Fryer’s work, that a distinctively Black name is an indicator, not a cause, of outcomes.
Context and Analysis
A lot of this book holds up, and the teacher-cheating work is its best piece of method. Levitt made a prediction from the data, the district tested it with a supervised retest, and the prediction held. The sumo result has a pre-specified threshold, a large sample and two independent confirmations. The typology of economic, social and moral incentives is a durable tool. The insistence on asking what the data can actually identify, rather than what the story suggests, is the right instinct. Its real legacy is institutional: it moved applied microeconomics into general reading and started a genre that is still producing.
The criticisms matter as much, and the abortion-crime argument leads them. It is a correlational design, state-and-year abortion rates against state-and-year crime, and “unwantedness,” the variable the whole mechanism turns on, is never measured directly. After publication, and this is outside the book, the underlying paper was challenged on two separate grounds: a coding error in the regressions, which the authors acknowledged, and whether the specification could separate the effect from other changes happening in the same states at the same time. The blog entries reprinted in this edition concede that a published table lacked controls the paper had described, then rework it with new data. The slippage runs wider. The authors write that a regression can show correlation but not cause, then treat their own regressions as proof, and they sum the crime-drop percentages into one ledger as though each were independent and measured. Single datasets carry very large conclusions: one bagel man’s customers supply an eighty-seven percent honesty figure, one gang’s notebooks supply the wage structure, and several headline studies are Levitt’s own, reported as if by a neutral narrator. Throughout, the charm does work the evidence does not.
On this site it pairs naturally with Thinking in Bets, Annie Duke’s book on separating the quality of a decision from the quality of its outcome and resisting the story that fits, which is the discipline Freakonomics preaches and sometimes breaks. It also reads well against The Tipping Point, the other book that made social science a mass-market genre in the same decade, and a useful contrast in how much weight each puts on a single striking study.
What It Teaches
Five lessons that outlast the disputes.
1. Before asking why people do something, ask what they are paid, praised or shamed for doing. 2. A price can crowd out a conscience. Once the Haifa parents could pay for lateness, they stopped feeling bad about it. 3. Make a prediction, then test it. The teacher retest persuades because it could have failed. 4. Ask what the data can identify. The crime chapter’s own warning that correlation is not cause applies to its most famous claim. 5. Be suspicious of a story that fits too well, including your own.
Memorable Lines
“The conventional wisdom is often wrong.” (Levitt and Dubner)
“Incentives are the cornerstone of modern life.” (Levitt and Dubner)
“An incentive is a bullet, a lever, a key: an often tiny object with astonishing power to change a situation.” (Levitt and Dubner)
“Cheating is a primordial economic act: getting more for less.” (Levitt and Dubner)
“Dramatic effects often have distant, even subtle, causes.” (Levitt and Dubner)
“A regression analysis can demonstrate correlation, but it doesn’t prove cause.” (Levitt and Dubner)
Should You Read the Full Book?
Verdict: Recommended
This summary carries the findings, the figures and the shape of the dispute. The full book adds the storytelling, which is the point: the answer sheet laid out line by line, Venkatesh in the gang’s stairwell, and the bonus columns in which the authors argue with their critics. Read it for the method and the pleasure, knowing that the chapter everyone remembers is the one most seriously challenged afterward, and that the prose is good enough to make every chapter feel proved whether or not it is.
The Freakonomics book page has the full details and where to get a copy.