The recount after the recount

After the election, eight news organizations commissioned NORC at the University of Chicago to examine Florida’s rejected presidential ballots. NORC’s own report says its purpose was research into ballot reliability, rather than naming a president. That distinction matters: “NORC declared the winner” misidentifies what the organization did. [1]

The final examined total was 175,010 ballots: 61,190 undervotes and 113,820 overvotes, according to legal scholarship citing the project’s contemporary documentation. An undervote registered no presidential selection; an overvote registered too many. About 65 percent of this reviewed set were overvotes. These were not 175,010 additional lawful votes waiting to be assigned to somebody. [2]

County workers displayed ballots while three observers independently described visible features: detached corners, indentations, holes and marks. Their job was to record observations without negotiating a shared interpretation. The data let analysts ask how different definitions of a countable mark would affect totals. A research coding team was not a legally empowered county canvassing board. [3]

The surviving archive includes raw and aligned data, repeated codings, comments and separate Orange County material. It is possible to inspect the research design rather than relying on a partisan summary. This edition links the archive but has not performed a fresh ballot-level reanalysis. Historical specificity is more useful here than claiming a new definitive recount. [4]

The first question was whether Gore would gain enough from the limited manual recounts he requested in Broward, Palm Beach, Miami-Dade and Volusia counties. The consortium’s analyses indicated that Bush would remain ahead. A second question modeled the statewide undervote recount ordered on December 8. Those modeled limited-count scenarios also favored Bush. These are meaningful findings against the simple claim that finishing either described count necessarily elected Gore. [1]

The third question was broader: what if investigators reviewed rejected ballots statewide, including overvotes that could contain a single identifiable choice? CNN reported a scenario using the standards county officials said they would apply in which Gore finished 171 votes ahead. A voter who filled Gore’s oval and also wrote “Gore” on the write-in line could trigger machine rejection without expressing support for two competing candidates. [2]

FactCheck’s retrospective reported a Gore advantage of 42 to 171 in broad-review scenarios and emphasized uncertainty. Other familiar summaries give 60 to 171; numbers should be attached to their specific modeled rules, rather than combined into an imaginary official result. [3]

The reasonable conclusion is conditional. Some broader counts supported a Gore win; the limited recounts modeled by the consortium supported Bush. Selecting only one result misleads. Nor should “Gore never sought it” be stretched into “no broader interpretation was legally imaginable”: contemporary legal scholarship discusses ambiguity in the December 8 order and the significance of including overvotes. [1]

An independent review was far better evidence than guessing from television pictures of chads. It was still a reconstruction. NORC’s annual report explicitly explained that repeated counting can produce variation; independent observers reduced that problem without making human judgment disappear. The sponsoring news organizations then made analytical choices about which observations should become votes. [1]

The archived project description gives a concrete custody problem. Orange County initially could not separate all ballots reported rejected on election night from ballots that had been accepted. More than 400 originally rejected ballots could not initially be distinguished, and the archive includes separate supplementary files. That is a documented limitation, not a license to assume those ballots all belonged to either candidate. [2]

CNN also reported discrepancies between expected and produced rejected-ballot totals across counties. FactCheck explained why missing material and small differences between observers prevent exact certainty about the result an official recount would have produced. The issue is especially consequential when a modeled lead is much smaller than the number of records requiring qualification. [3][4]

This does not make all the published results worthless or equally speculative. It means a modeled Gore lead is evidence for a possible different outcome, while a modeled Bush lead is evidence about that particular recount’s scope. Neither automatically demonstrates fraud, nor reconstructs the private intention of every confused voter. The correct response is to state the scenario, show its assumptions, and preserve its uncertainty.