A brief history of election forecast computers
By John Lister
1952: CBS uses UNIVAC (Universal Automatic Computer), from a range originally designed to process data from the national census, to forecast the outcome based on early results. It predicts Dwight Eisenhower will defeat Adlai Stevenson in the Electoral College by a 438-93 margin. With opinion polls pointing to a clear Stevenson win, CBS officials refuse to announce the prediction on air. When the final tally comes in with Eisenhower a 442-89 winner, the network apologizes on air for not airing the forecast.
1960: With ABC and NBC having followed CBS’s lead in 1956 and used computers, NBC unveils RCA-501 (pictured) which can do the work of “60,000 clerks”. (In a sign of how times change, viewers learn the computer takes into account 200,000 factors including the voting “of Northern and Southern negroes”.) Once 10 percent of the results are in, the computer predicts John F Kennedy will beat Richard Nixon by 51.1 percent to 48.9 percent in the popular vote, a forecast it sticks to throughout the night.
The network proudly proclaims that RAC-501 is the only computer which at no point predicts a Nixon victory; says one NBC anchor, “If they ever teach this machine to talk, we’re out of business.” (The actual vote results are slightly different because RCA-501 doesn’t include third-party candidates; of votes cast for the two leading candidates, Kennedy prevails by 50.1-49.9 percent)
1980: Driven by the race to be the first to call a result, the networks begin using exit polls as a source for computer predictions, allowing them to make forecasts much earlier (in some cases, before polls have even closed). Early accuracy problems lead to the networks establishing Voter News Service, a standard system for conducting shared exit polls in 1990; during the following decade the poll only forecasts one race incorrectly from a total of around 5,000.
2000: In one of the most embarrassing computer-based predictions, the networks all project Al Gore the winner in Florida based on exit poll data with some projections coming more than an hour before polling closes in some parts of the state. George W Bush actually wins the Florida (and the Presidency) though the result is so close that it leads to a prolonged recount process. While many Democrats dispute the outcome, some Republicans argue that the early and erroneous network call may have dissuaded potential Florida voters from bothering to vote.
2002: Voter News Service suffers a disaster in Congressional elections when its computer system, redesigned after the 2000 problems, suffers technical issues which leave networks unable to rely on exit poll data. CBS and NBC fall back on the Associated Press’ reporting of actual announced results.
2004: The latest cyber-trend is websites predicting the final outcome based on pre-voting polls. ElectionProjection.com wins the battle, correctly forecasting George W Bush beating John Kerry 286-252. There’s embarrassment for the University of Virginia’s Center for Politics, which predicts a 269-269 tie, and Electoral-Vote.com (run by Andrew Tanenbaum, who developed a pre-cursor to Linux) which goes for a Kerry victory by 306-218.
Related:





Stumble It!
