King develops solution to longstanding social science problem

By Debra Bradley Ruder

Gazette Staff

It was Nov. 16, 1994, and Professor Gary King was in an Ohio courthouse watching a federal voting rights trial unfold. The results would affect control of the state's legislature for years to come.

An expert witness was using the best available statistical method to gauge how many African-Americans had voted for the Democratic candidate in a recent election.

"His estimate of the percentage of blacks who voted Democratic was 111 percent -- which is ridiculous," recalled King, a professor of government in the Faculty of Arts and Sciences.

King played a minor role in the trial, but he came up with a major breakthrough in his field while waiting to testify.

He developed a solution to a problem that has vexed social science researchers for 75 years: how to learn about individual behavior when the only information available is on groups. It has been dubbed the "ecological inference" problem.

King's new algorithm for computer software is detailed in his new book, A Solution To The Ecological Inference Problem: Reconstructing Individual Behavior From Aggregate Data, published this month by Princeton University Press.

Earlier this month, the American Political Science Association honored King with its Gosnell Award "for the best methodological work in political science in 1995-96."

King's solution is expected to help researchers and policymakers who analyze data about a wide range of subjects, from dropout rates to consumer purchases.

An epidemiologist, for example, may have information on the degrees of radon exposure and the number of people with lung cancer in a particular county, but she wants to know the fraction of individuals with high radon exposure who are diagnosed with lung cancer.

"Ecological inference is required whenever surveys are unavailable, unreliable, or too expensive," according to King. "Surveys cannot address most historical questions unless they are conducted then and there. They are also unreliable for studying controversial issues, such as racial politics, since respondents do not always report their opinions and behaviors accurately."

A Long-Standing Problem

The ecological inference problem was discovered in 1919 by scholars trying to figure out how women were going to cast their ballots, just before they won the right to vote nationwide.

"Ecological inference was the first statistical problem raised in the new field of political science," said King. "Scholars soon recognized the same problem in other scientific disciplines. It's the problem that graduate students in political methodology cut their teeth on."

King's solution unifies two separate and hotly contested approaches -- the deterministic and the statistical approach -- in use since the 1950s.

Basically put, the deterministic method extends a simple insight. For example, suppose 40 percent of a district is African- American and 40 percent of the district votes for the Democratic candidate. A determinist would deduce that the share of African- Americans who vote for the Democrat must fall somewhere between 0 percent and 40 percent.

A researcher using the statistical approach, meanwhile, would examine the data across precincts. If the precincts voting most heavily for the Democratic candidate were those with the most African-Americans, he/she would infer that blacks are voting for the Democrat.

"One of the things my solution does is to combine these two approaches and to improve them both," said King, who has taught at Harvard since 1987. "Everybody in the field knows they don't work separately. What I have developed is a solution, not the solution. "

With funding from the National Science Foundation and the Guggenheim Foundation, he tested his method with data about known individual behavior, making more than 16,000 comparisons between his estimates and the known behaviors.

Word of King's solution spread through press releases from the National Science Foundation and a couple of other organizations, and King has received messages from a diverse set of scholars ranging from cardiologists to sociologists interested in learning more.

The first chapter of his book, and the software he developed, may be downloaded from King's homepage, at http://gking.harvard.edu. Browsers may also get reprints of his other scholarly work and information about the Harvard-M.I.T. Data Center, which King directs, or view a selection of book covers for Designing Social Inquiry and The Elusive Executive, which he co-wrote.

Homepage visitors may also check out photos of King's daughter Ella Michelle -- to whom he dedicated his latest book -- at four days, four months, and a year and four days, as well as a snapshot of the family dog, Willow.