UNIVAC and the 1952 Presidential Election
This article was initially published in Today's Engineer on November 2012
The story has been told and retold for decades: how CBS Television News used a UNIVAC computer to predict the 1952 U.S. Presidential election returns and — when the computer accurately predicted the Eisenhower landslide at around 8:30 in the election night broadcast — the prediction was doubted, and only hours later did CBS reveal that the prediction had been accurate. It has become a classic cautionary tale of the dangers of allowing human preconception to interfere with logic and evaluation of facts.
There is more to the story. The exact timeline of when UNIVAC's initial prediction was made is not certain, but what is important is that UNIVAC's correct prediction of a landslide victory was ostensibly ignored until much later in the broadcast because of journalistic prudence and lack of confidence in the accuracy of the results.
Dr. Ira Chinoy, whose doctoral thesis examines the use of computers in broadcast journalism, estimates that the celebrated initial prediction of the Eisenhower landslide was made closer to 9:15. At 8:30, only slightly more than one million votes had been tallied; it took until at least 9:15 pm for three million votes to be transmitted from CBS to the Remington Rand factory in Philadelphia. CBS was receiving vote tallies from the wire services and teletyping them to Remington Rand’s factory in Philadelphia. Additional time was needed to input the data and to run the programs.
The 8:30 CBS segment merely gave the television audience a visual tour and introduction to UNIVAC; the second UNIVAC segment of the evening at 9:30 asked for a prediction, but the machine was not yet ready. By that point in the television coverage, the human commentators were already commenting on the surprising Eisenhower strength in the early returns. On the basis of pre-election polls, the race between Eisenhower and Stevenson had seemed to be close (Eisenhower held a slight edge), so the use of a state-of-the-art computer to predict what was expected to be a very close election had generated a lot of popular interest.
At some point relatively early in the evening, UNIVAC predicted an Eisenhower landslide victory. However, the UNIVAC programmers decided that the prediction was too risky to release because it contradicted what the pollsters had been saying previous to the election about a tight race.
At 10:30, which was the third on-air UNIVAC segment, the computer predicted twenty-eight states for Eisenhower and twenty for Stevenson. This was a softer prediction, and was in line with what the CBS commentators had already been telling their television audience. It was the initial correct prediction of an overwhelming Eisenhower win that the UNIVAC programmers decided not to release because it contradicted the poll numbers.
The 11:30 UNIVAC on-air prediction caused more drama. It reversed its earlier prediction, calling 24 states each for Eisenhower and Stevenson, and a slim 270 to 261 Electoral College vote margin for Eisenhower. But by 11:45, the prediction had been corrected and UNIVAC predicted 100 to 1 odds of an Eisenhower victory.
UNIVAC made its predictions based on the difference between vote tallies and the expected vote in cities and counties, based on a statistical model extrapolated from past elections. By applying this deviation in places that had already voted to those which had not yet voted, an estimate of the present election could be obtained based on past tallies in those places. One of the ironies of the election of 1952 was that the returns from Massachusetts, one of the crucial early-reporting states, were incorrectly reported to UNIVAC. That UNIVAC was nonetheless able to make accurate predictions.
The UNIVAC used by CBS was the fifth UNIVAC machine made. In the autumn of 1952, UNIVAC-5 was still in the Philadelphia factory of Remington Rand waiting for its future installation at the Lawrence Livermore Laboratories. Because UNIVAC itself was too large to be moved conveniently, a dummy control console was set up in the CBS studio in Grand Central Terminal, New York City for visual effect, its lights blinking evocatively thanks to delay switches ordinarily used for making Christmas tree lights flash on and off.
There was some irony that a machine which debuted in the public spotlight of national TV would go on to do classified weapons work. UNIVAC contained mercury delay lines, which allowed it to store 1,000 words (45 bits each) as electric pulses in tubes of mercury. Up to one million characters could be stored and accessed on magnetic tape. It was these tapes, replacing punched cards, which made the UNIVAC revolutionary, and which gave it a tremendous speed advantage because it could access its own data instead of needing to wait for cards to be loaded. It could perform four hundred and sixty-five multiplications per second and had a clock speed of 2.25MHz.
A brief video of the CBS prediction can be seen here.