ProPublica's project on presidential pardons benefited from a lawsuit filed in 2008, under the Freedom of Information Act, by George Lardner Jr., a former Washington Post reporter. Lardner's successful suit forced the Justice Department's Office of the Pardon Attorney to make public the names of pardon applicants who had been denied clemency.

The office does not reveal race or any other information that would identify an applicant, citing privacy grounds.

To do the first systematic analysis of presidential pardons, ProPublica reporter Jennifer LaFleur began with a list of all the people whose pardon applications were granted or denied during the George W. Bush administration - 1,918 names. From that list, a computerized random sample of nearly 500 was obtained for statistical analysis.

Working from the sample, reporter Dafna Linzer, LaFleur and researchers used public records and phone calls to gather data such as race, age, gender and marital status. Data were also collected on criminal records, bankruptcies, judgments and tax liens.

Statistical tests were used to determine the characteristics that made a person more likely to get a pardon. Variables that appeared to influence pardons included race, marital status, time passed since the crime, and whether the person received probation rather than prison time. Financial stability, as measured by the absence of a bankruptcy or lien, also increased odds of a pardon.

After statistically controlling for characteristics listed above, whites were nearly four times as likely as minorities overall to get a pardon. At ProPublica's request, the analysis was reviewed by several outside experts who endorsed the approach.

In raw percentages, the analysis found that 12 percent of white petitioners and 10 percent of Hispanics were pardoned. None of the 62 African Americans in the sample received a pardon. Based on their representation in the sample, as many as 4 percent of all black applicants could have been pardoned.

For more on our analysis, read our detailed methodology.