How did news networks get every presidential race projection correct in the 1990s?

[kcguy]

I remember hearing some stories about bad data on election night in Florida in 2000.

A Google search turned up this report from 2001.  https://www.cbsnews.com/htdocs/c2k/pdf/REPFINAL.pdf

Here are some highlights.

• The exit polls of 45 survey precincts suggested Gore was winning by 6.6 points.
• In the first 6 survey precincts to report, Gore did 1.7 points better than the exit poll.  When all 45 eventually came in, including Miami and Tampa, Gore actually did 2.8 points worse.
• The absentee vote later turned out to be historically high, at 12% of the total vote, rather than the estimated 7%. Bush carried the absentee vote statewide by 24 points.
• In order to figure out statewide patterns, CBS chose the 1998 gubernatorial race (Jeb Bush's) as its baseline, which turned out to be a bad predictor.  In retrospect either the 1996 presidential or the 1998 senatorial results would have produced a better model.
• After the call was made, a keypunch error for Duval County gave Gore an extra 40,000 votes, making the networks even more confident of their call.  It wasn't for another hour that the mistake was noticed.

• By 2:00am EST, Bush was up statewide by 29,000 votes with 96% of precincts reporting.
• Shortly afterwards, the numbers for Volusia and Brevard Counties were added, giving Bush a 51,000-vote lead.  However, both results included computer errors, and Bush's lead should actually have been closer to 27,000.
• The error was quickly noticed and corrected by the Associated Press, but none of the networks were getting their numbers from AP.
• By now, the number-crunchers were estimating that there were only about 180,000 votes left uncounted, but the votes were coming from the most Democratic parts of the state.  In actuality, the number of outstanding votes was nearly double, partly due to Palm Beach County counting its absentee ballots and its largest precincts last.
• Half an hour after Florida was called for Bush, the networks were reporting that the official results had Bush with a 55,000-vote lead.  At the same time, the AP had his lead at 14,000.
• Eventually, the networks discovered their error.  By the time the networks retracted the Bush call, his lead was less than 2,000 votes.