Background on the study

• A consortium of the following media organizations commissioned the Florida ballot study: The Associated Press, The New York Times, The Wall Street Journal, CNN, the St. Petersburg Times, Cox Newspapers, The Washington Post and the Tribune Co., which includes the Los Angeles Times, the Chicago Tribune, the Orlando Sentinel and the Baltimore Sun as well as other papers.

• The National Opinion Research Center (NORC) at the University of Chicago conducted the study.

• NORC researchers examined 175,010 ballots in a survey of all 67 Florida counties. NORC used experienced researchers on their permanent staff to recruit, train and supervise a larger pool of people called coders who examined the ballots. Although these coders were not permanent NORC staffers, all went through rigorous training, and many of them moved from county to county for the study's duration.

• For undervotes, teams of three coders examined each ballot under the direct supervision of an NORC supervisor. For overvotes (except in three counties), one coder only was employed. This decision was taken after testing the results from the three-coder system in three counties to assess inter-coder reliability. The result of this study was a recommendation from NORC that coders agreed with each other at such a high level that using three coders instead of one did not produce a significant gain in precision.

• Coders were instructed to record what they saw for each chad (in punch-card counties) or each ballot position (in paper ballot counties) in the presidential and Senate race. They used codes developed by NORC to describe the chads, marks, etc., that they saw. Coders also recorded additional information such as what -- if anything -- was in the write-in field, whether there were similar marks elsewhere on the ballot, ink color, etc. All coders used standardized forms that NORC developed and supplied. NORC entered all data to create the final database.

• Coders made no attempt to determine whether a ballot would have counted as a valid vote for any candidate. They were instructed to record what they saw, not to make any judgments about the ballot itself. Coders were not allowed to handle the ballots; instead, the ballots were held up for the coders' review by county employees.

• There are limitations to this study:

1. Due to clerical inefficiency on the part of county election officials, some counties were unable to produce the same number of overvotes and undervotes that they certified to the Florida secretary of state's office last year. In other words, NORC did not examine every ballot that was considered spoiled during the official vote count, and there is a high likelihood that some ballots that were counted for George W. Bush or Al Gore last year were presented to NORC as undervotes or overvotes. This was not due to any mistakes made by NORC or the consortium.

2. Using three coders on undervotes increased the reliability of the study but led to some statistical uncertainties because coders frequently disagreed.

3. The human factor: The study relied on human beings -- not computers -- to evaluate each ballot, leaving the study's findings vulnerable to both human judgment and error. Just as important, any hypothetical recount assumes that human beings would have examined the ballots and made their own judgments. There is no guarantee that the judgment calls made by county election officials would have matched the judgment calls made by NORC coders.

• Bottom line: Human judgment and error, combined with some counties' inability to produce the same undervotes and overvotes that they saw last year, creates a "margin of error" that is greater than most of the margins of victory produced by the various scenarios. That's why the scenarios are really "too close to call."