On May 21, 1996, a press release came across the PR Newswire from an outfit called Taponline, an internet site then run by the MarketSource Corporation. It announced that in their online straw poll of college students, Bob Dole led Bill Clinton by two points in the race for the U.S. presidency. These results beggared belief, as Clinton dramatically led Dole in every other poll at the time, so I tracked down Taponline's online poll and cast a vote for Alan Keyes. I am not, for the record, a supporter of Alan Keyes, but I find him entertaining and felt that he needed all the help he could get.
At this point, the rankings were roughly as follows:
Dole: 43%
Clinton: 41%
Keyes: 4%
For kicks, and to explore the possible origin of the taponline poll results, I attempted to cast a second vote. Their server was smart enough to catch this and disallowed the second vote. However, its error message made clear that it tracked voters by IP address, rather than account. Since my primary account lived on dozens of machines, I proceeded to vote many, many times from many, many IP addresses. After an hour, I had adjusted their results to the following:
Dole: 29.1%
Clinton: 29.1%
Keyes: 29.7%
I then quit from the field, printing out the poll results to memorialize them, as I figured they would soon revert to something like the original distribution.
Here is where the story gets interesting. On the 28th I checked back on the taponline site to discover that things had not developed at all as I had expected. No stream of Dole or Clinton votes had tilted the balance back to the mainstream candidates. To the contrary, the totals looked like this:
Dole: 28.9%
Clinton: 29.1%
Keyes: 30.5%
And the Keyes lead grew substantially until I stopped checking.
As I see it, there are two possibilities: I voted as a joke, other people took it as a joke, and joined in as joke. Alternately, perhaps I voted as a joke, other people took it seriously, and voted with the perceived groundswell of Keyes support.
It is no surprise to me that poll results are, in some sense, autoregressive. People prefer to back a winner, so previously public poll results must play some role in the outcome of current polls. It does, however, bother me that the results would be so overwhelmingly dominated by this effect. It might make sense if votes for Keyes had come in at a rate higher than the 4% "natural" level of his support among taponline customers. However, for his lead in the poll to increase past my farcical 29.7% suggests that people’s votes were completely controlled by the standings when they arrived at the gate.