On October 7th, a peer of mine collected 20 surveys from self-selected (volunteered) members of the political science department (from first year graduate students to professors) on an electoral challenge. The survey presented seventeen different questions in regards to electoral outcomes and those who were able to predict the most accurately 14 of the 17 questions (with 3 of the questions serving as tie breakers) win an undisclosed prize for the top performer. There is no prize awarded for second place.
This is neither a random sample nor a sample of electoral experts, but instead represents a combination of strategic and sincere beliefs of the upcoming election (grounded in the week of 10/7/08) by individuals who are quantitatively and politically savvy. Given this caveat and framework, there are a few potentially interesting results that might intrigue readers since are one week away from election day.
The national election results are not overly exciting to put
in graph form. 18 of the 20 (90%)
entries suggested that Obama would win.
One of the tie-breakers asked for a prediction of the spread. The question asked for a spread in percent,
but an entry contained 3,000,000 as an answer; I gather that this prediction is the number of votes cast for the particular candidate.
I converted this to 2% given the turnout of 2004 having
roughly 125 million voters and we expect a higher turnout this year. Given McCain
spreads a negative value (to move away from the absolute value that is
currently coded), the spread of the predictions can be seen in the below
table. The mean prediction is 4.87
(positively in Obama’s favor) with a standard deviation of 2.66.
The Battleground States
The next table shows the results for the six battleground states tested. Generally, it appears that there is favoritism towards Obama in four of the states. North Carolina appears to be clear McCain ground and Nevada has a bit of uncertainty around the prediction. Technically, this is the same variance that North Carolina has and should be regarded as the same probability of either, but I believe, given the other results, that Obama might be over-favored in this sample.
Senate Races
Four races were questioned and the most contentious race was Begich-Stevens. If this question was asked today, I believe the distribution would be more skewed, leaving the Minnesota race as the most likely for an upset. The histogram below shows the predicted Democrat gains in the Senate. The data provide a mean of 4.75 seats witha standard deviation of 1.8.
House Races
The four races here are a bit tighter. This may occur from actual close races for the house or increased general ignorance about the state of the races. As such, New York’s 26th and Connecticut’s 4th are not deterministic. The predicted Democrat gains in the House has much more variance, as would be obvious, than the Senate. However, 25% of the respondents belived 15 to be likely. The mean was just under 13 seats for the Democrats.
The summary statistics are relatively simplistic and do not leave room for causal interpretation. However, given that it is an aggregate of predictions, it is an interesting poll of sorts. You can view the data that I have here if you are so inclined: Download election_prediction.csv
On Wed. you should compare the performance of the IR and Comparative-American folks. I predict IR faculty and grad students will prevail.
Posted by: Matt Digiusepe | November 02, 2008 at 05:53 PM