Princeton Molecular Biology Professor Tracks Polling With Algorithm

This story was written by Ryan Martin, The Daily Princetonian
(UWIRE) --

Princeton molecularbiology professor Sam Wang believes he will know the winner of the 2008presidential election by 8:30 p.m. Eastern Standard Time on Nov. 4.Should early returns from New Hampshire and Virginia show the samecandidate ahead, that man will be inaugurated come January, Wang said.

Wang,who runs the Princeton Election Consortium, a poll aggregator andpolling analysis website, uses polls from pollster.com to calculate theprobability of a candidate to win any given states electoral votes bylooking at the median and standard deviation of the state polls.

Polls are, on average, very good predictors of election outcomes, Wang said.

Additionally, Wang projects various electoral map scenarios by throwingswing states to either of the two presidential candidates. Theseprojections, which ultimately yield a tally for each candidate betweenzero and 538, involve almost 2.3 quadrillion individual calculations.

He added, however, that looking at individual polls is moderately reliable ... The simplest thing to do is to average polls.

Wangssite currently indicates that, were the election held today, it wouldbe a victory for Sen. Barack Obama, D-Ill., over Sen. John McCain, R-Ariz., by 320 electoral votes to 171 electoral votes, with Ohio andFlorida being too close to call.

Wang, unlike some pollsters,said he does not believe that current polling is influenced by what iscommonly called the Bradley effect. This stipulates that voters willtell pollsters theyll support a black candidate to avoid beingperceived as racist but in fact have no intention of voting for thatcandidate.

Statistically, in races [since 1996] where there is data available, the Bradley effect appears to be zero, he noted.

Wanglaunched PEC in 2004 to provide a clear daily update of the close racebetween incumbent President George Bush and Sen. John Kerry, D-Mass. The PECwebsite received more than one million visits during the 2004 campaign.

Onthe eve of the election, Wang predicted that Bush would winre-election, with 286 electoral votes to Kerrys 252. He turned out tobe correct.

The correct prediction shows the algorithm is good,he said, though he added that at the time I thought that the undecidedvoters would split in favor of the challenger, but in 2004 that was notthe case.

Since Wang resumed updating the site in July for the2008 campaign, PEC has received more than 771,000 total site views,with about 20,000 visitors each day, he said.

PEC tries tominimize the effect of outliers on poll results, he said, adding thatoutliers may accurately represent voters intentions, or they mayresult from voter screenings that pollsters use to determine whetherthe person who is contacted will actually vote.

Inability tocontact a representative sample of voters can skew data becausepollsters may choose to count one poll response as if it had come frommultiple respondents, Wang explained.

The shift of some households to only cell phones has made these users less accessible to some pollsters, he noted.

Andrew Ferguson 08, who assists with the coding of the website,explained in an e-mail that the goal [of PEC] is to understand thestate of the election todayclearly, the polling data is out there,it just needs to be organized.

A really important part of thesite is that we keep our calculations as simple and straight-forward aspossible ... because we believe that polling is influenced by outsideevents which cant be predicted, he explained.

PEC is not theonly website providing statistical analysis of polls Other popularsites this year include fivethirtyeight.com, realclearpolitics.com andelectoral-vote.com.

Its turned into this funny cottageindustry, Wang said. We live in an age when its possible to becomepopular by being a data geek.

As for what Wang will be doing on Nov. 4, he said, I think the thing to watch on election night will be Senate races.
  • CBSNews

Comments