Investigations
Of sumo wrestlers and
other mysteries
Steven Levitt uses economics to solve conundrums
of human behavior.
Economist Steven Levitt delights in proving
what others suspect but don’t believe can be confirmed.
For years the Japanese tabloids accused sumo-wrestling matches
of being rigged, but Levitt was the first to demonstrate
it. Most studies are murky on whether more police actually
reduce crime in U.S. cities, but it was Levitt who showed
that the nation’s finest do, in fact, help protect
its streets. And though some teachers have been suspected
of helping students cheat on required standardized tests,
no one knew the prevalence of such activity until Levitt
quantified it.
Photo by Dan
Dry |
The Alvin Baum professor
of economics, Steven Levitt, on sabbatical at Stanford
this year, hopes his work can “answer relevant
questions.” |
Levitt, at Chicago since 1997, studies
problems that differ from the typical economist’s
purview because they don’t fit the standard formula.
“Usually in economics you have two variables,”
he says. For example, “when wages go up, people work
more hours. You have the data for the thing you want to
study.” Levitt, however, is drawn to “problems
that ostensibly seem difficult, but with a trick one can
find a simple solution.”
Those tricks vary from project to project.
In the sumo-wrestling study, published in the December 2002
American Economic Review, Levitt and assistant
professor of economics Mark Duggan “tricked the data
into telling us the answers” by studying the distribution
of wins and losses. In a tournament each wrestler competes
in 15 matches. If a wrestler wins eight matches he moves
up in the official rankings—and no longer must perform
menial tasks like cleaning out the gym or cooking and carrying
bags for the other wrestlers in his beya, or stable.
“So the eighth win,” Levitt
says, “is very important.” He and Duggan noticed
that in a given tournament many wrestlers won six matches,
“shockingly few” won seven, and a “huge
number” won eight. That statistically unlikely distribution
suggested to Levitt that wrestlers with seven wins facing
opponents with either eight wins or six or fewer wins had
paid a bribe to ensure a victory.
But before alleging corruption he weeded
out other scenarios. Perhaps, for instance, the win gap
reflected increased effort. After all, the wrestlers really
wanted that eighth win and simply might have tried harder
during that match. But when Levitt looked at subsequent
matches between the same pair, the winner of the crucial
eighth match usually lost quite badly the next time around.
“I have a hard time reconciling that with any other
model,” Levitt says, “except that they have
worked out some sort of explicit or implicit deal.”
To further corroborate Levitt’s claim, he compared
an inside whistleblower’s list of alleged cheaters
with his analysis. They matched.
In the police study Levitt’s trick
was to find a new way to examine the issue. Studies showed
that rises in crime correlated with increased police forces,
but whether one caused the other was arguable. So Levitt
looked for another variable that correlated with police-force
growth. Inspired by the 1993 New York mayor’s race—in
which incumbent David Dinkins expanded the city’s
police force to appear as tough on crime as his opponent,
U.S. Attorney Rudy Giuliani—Levitt used the timing
of mayoral elections. Often when big-city leaders are up
for reelection, he found, they hire a disproportionately
large number of police. His research, published in the June
1997 American Economic Review, showed that a year
later, when the new recruits had some time to get to work,
crime dropped.
To study teachers cheating on their students’
standardized tests, Levitt and Brian Jacob of Harvard’s
Kennedy School of Government looked at elementary-student
test scores in Chicago Public Schools. Because there is
no standard measure for cheating, the trick in this case,
Levitt says, was to “construct a variable”—that
is, make assumptions about what cheating classrooms look
like. So the researchers assumed that the more student test
scores fluctuated from year to year, or the more suspicious
answer patterns were found within a classroom, the more
likely it was that teachers were cheating. They may have
changed students’ answers, filled in blank sections,
provided students with correct answers, or obtained copies
of the test beforehand and taught students using that knowledge.
Creating an algorithm to check test-score
patterns in 10,000 third- through seventh-grade classrooms
between 1993 and 2000, Levitt and Jacob found evidence of
cheating in 4 to 5 percent of classrooms per year—a
figure they believe is low because the formula catches only
the most blatant cases. Their results led the city to fire
a few school employees and to question several more.
To Levitt, the sumo-wrestling, police,
and cheating studies offer models for solving mysteries.
They’ve shown him—and others—that “things
that seem impossible may be possible. The answers are there,”
he says, “when one thinks about questions the right
way.”
That means thinking like an economist
and understanding incentives. Sumo wrestlers want to move
up in the hierarchy to get out of doing grunt work. Mayors
hire more police in election years to convince the public
they’re doing a good job. Many states reward schools—including
merit pay for teachers—for improved test performances,
and sanction those that perform poorly.
Yet some of Levitt’s studies have
no tricks. On sabbatical this year at Stanford’s Center
for Advanced Study in the Behavioral Sciences, he is researching
“why black parents and white parents give their children
different names.” Citing Chicago economist Marianne
Bertrand’s recent study showing that job applicants
with white-sounding names get more callbacks than those
with black-sounding names, Levitt says, “It seems
like having a black name is very costly.”
So he and Chicago postdoctoral fellow
Roland Fryer are studying possible benefits of black-sounding
names. Rather than simply a matter of culture or taste,
Levitt says, “A parent might give a kid a distinctively
black name because she expects the child to grow up primarily
interacting with blacks.” In that case a child might
be better off with a name like Jermaine rather than Neil.
Influenced by Chicago economist and sociologist
Gary S. Becker, AM’53, PhD’55, Levitt’s
interests transcend number crunching and reach into human
behavior. Eventually he hopes to use his creative methods
to tackle larger problems, such as money laundering, tax
evasion, disability fraud, and antiterrorism. Although he
focuses more on empirical data than on Chicago School theories,
he says, his colleagues have taught him to think much harder
about those economic models—and the tricks they suggest.
—A.B.