AAAS Seminar Explores How Statistics Can Be Used to Support Human Rights

Millions of pages of national police records from the long civil war in Guatemala are bundled haphazardly in four huge warehouses in the heart of Guatemala City. The records, discovered in 2005 and now available to researchers, are expected to yield evidence of police abuses and murders during the nation's bloody civil conflict.

Human rights researchers have been sifting through the estimated 80 million pages of documents with the help of Benetech, a data analysis company based in Palo Alto, Calif. But Benetech, faced with a volume of records too large to fully assess, called on members of the American Statistical Association for help.

"We gave general guidelines on how to sample" the records, said Gary Shapiro, a statistician with Westat, a research corporation based in Washington, D.C. said recently at a AAAS seminar. In doing a sampling of the storage space, using probability methods, he said, specialists could get a better handle on the character of the archive and the differences among the records. Such a structured search, he said, can give a better picture than a haphazard retrieval of material from the shelves.

The sampling project is still underway, and it is too soon to draw conclusions about patterns of police behavior, according to Shapiro and his colleagues.

But the there is little doubt about the value of statistical techniques in the investigation of complex human rights cases or the impact of civil conflicts and natural disasters around the world, according to speakers at the 17 May seminar organized by the AAAS Science and Human rights Program and the Washington Statistical Society's Human Rights Section.

In addition to the Guatemala case, the seminar at AAAS headquarters dealt also with recent work on estimating the number of deaths in the war-torn Darfur region of Sudan and lessons learned in efforts to survey residents of New Orleans about their experiences in the wake of Hurricane Katrina.

Paul Zador, a statistician who also is with Westat, noted media estimates of from 200,000 to 450,000 deaths in the Darfur region, where conflict among African rebel groups, the Arab-dominated government and militia who back them has led to a situation the United States government has called genocide. Zador said the accuracy and credibility of such estimates can be enhanced by good statistical methods.

Late last year, the U.S. Government Accountability Office (GAO) reviewed Darfur death estimates in six studies, ranging from 35,000 to 370,000 deaths for periods from 11 to 31 months long. The report concluded that the most credible estimate came from the Center for Research on the Epidemiology of Disasters, a nonprofit research institution based in Brussels, Belgium. The center estimated about 142,000 deaths during a 22-month period.

All such estimates are difficult to derive and often are subject to revision, Zador said. In 1980, the U.S. Central Intelligence Agency estimated that there had been 900,000 excess deaths in Cambodia from 1975 through 1979 during the time of "the killing fields." A more recent study, in 2005, now estimates there were 2.2 million excess deaths for the same period.

Estimating the number of deaths in conflict areas such as Darfur is "conceptually challenging and operationally daunting," according to Zador. It is difficult to reach dangerous or restricted areas. There can be multiple reporting of the same deaths by different members of a family in different refugee camps. Fear of retaliation may lead to low response rates in surveys. Survey teams may need to decide how to classify starvation deaths when a conflict has worsened conditions that were already dire due to a drought.

Teams also must decide what to estimate. They can estimate total deaths, such as the number who died during a specified period in a defined area. But Zador said the most relevant number is excess deaths—those who would not have died had there been no conflict.

In an effort to better understand the situation in the Darfur refugee camps, volunteers from Westat and elsewhere began crafting a survey plan for the camps that included probability sampling methods. But because of renewed violence in the Darfur region and other reasons, no effort has been made to undertake the survey. Zador urged establishment of a Blue Ribbon Committee to "better estimate the number of excess deaths in Darfur: past deaths, current deaths and, I am afraid, future deaths."

David Banks, a professor of statistics at Duke University, said that survey and statistical methods can be valuable during both the rescue and recovery phases of a disaster such as Hurricane Katrina. But officials should lay some legal groundwork before a disaster occurs, he said. For example, Banks said that the Census Bureau, with proper authorization from the Secretary of Commerce, could waive confidentiality provisions and give hurricane rescue crews the home address of every person over age 70 living alone. Such information could help them target those most likely in need of quick attention and could have helped reduce the death toll, Banks said.

Statisticians also could develop a registration system to help process and manage the displaced in a developed region. Scientists have the capability, he said, "to construct a very effective data base" on where people are coming from and where they are going, Banks said. No such system was in place in New Orleans, and the result was chaos, with people being loaded onto planes without knowing their final destination and some families finding it difficult to reunite after being separated during the storm.

Banks, along with colleagues from the University of North Carolina, Charlotte and Tulane University in New Orleans, conducted a post-Katrina survey of residents. It was fraught with difficulty and potential biases due to the inability to reach many residents who had left the city and never returned. Banks said that those who were contacted -- in person, by telephone or by cell phone -- were surprisingly cooperative. The response rate for in-person interviews was 77 percent, he said. During the survey, teams were directed to random locations in the city, identified by green dots on a map, and told to interview those in the closest home if it was occupied.

The survey found "too many rich people, too many homeowners, too many women and too many older Black women" compared to the pre-storm population of New Orleans, Banks said. Despite the uneven sampling, the survey did provide insight on decisions to evacuate, with 28 percent of respondents who left the city saying they did so because of concern about family members and 13 percent because of fear about the hurricane's size, strength and direction. Whites, with more ready access to transportation, were 2.4 times more likely to evacuate than non-whites.

But Katrina was "a remarkably egalitarian disaster," Banks said. "Certainly it helped to be rich, certainly it helped to be white, certainly it helped to be evacuated before" the storm hit, he said. "But no matter how you slice it, there were still significant numbers of people who had really rough post-Katrina experiences."

Those with the most diverse social networks tended to fare the best, he said. "On the other hand, if everybody you know is like you -- doesn't have a job, doesn't have financial assets, doesn't have transportation -- then, basically, you're in trouble," Banks said.

Erik Voeten, an assistant professor of political science and international affairs at George Washington University who served as a discussant during the seminar, said statisticians can provide a valuable sense of credibility in human rights cases. "There is still this perception out there in the world that human rights advocacy groups have incentives to exaggerate human rights abuses," Voeten said. With outside groups doing solid statistical research, he said, "you sort of remove that suspicion."

Among the roles statisticians can play, he said, is helping to point out the underlying patterns of violence in civil conflicts and the groups that may be most vulnerable. During the Bosnia conflict, advocacy groups emphasized the need to evacuate women and children first. The statistics were to show that young men and women were the most likely to be killed, Voeten said.

Accurately estimating deaths during conflicts is extremely important, he said, but he also urged a focus "not just on death but on disease." He said there are long-term health consequences of wars and civil conflicts, including higher infant mortality in displaced populations. He cited a study that estimated there were almost as many residual deaths in 1999 from previous civil conflicts during the 1990s as there were from active wars underway in 1999.

Voeten also noted that "sometimes the effects of war on health and death rates take place in places where we wouldn't immediately look." He mentioned a study last year in the American Economic Review that estimated the years of potential life lost due to smoking by those who served in the U.S. military during World War II and the Korean War. Cigarettes were made freely available to the troops abroad and smoking was widespread. The study found that life-years lost due to smoking-related deaths among the veterans has been "roughly approximate, " Voeten said, to the number of life-years lost by U.S. troops due to battle deaths in World War II and Korea.

Earl Lane

30 May 2007