As we head into the final stretch of the campaign, we're likely to be inundated by a spate of poll results from all kinds of different organizations, and if, well, everything is any indication, then different pollsters are likely to show very different things about the state of the race in key swing states. There may be no point in trying to make sense of mind-bogglingly contradictory survey results, but in case anyone is interested, I took a look at the polling done in the run-up to the 2010 midterms and constructed a little table on which pollsters came the closest to foretelling the election results in states that are now considered "battleground" territory and which missed the mark.
I measured polling accuracy by looking at the results of the last public poll of each race by each pollster and comparing the margin between the two highest-polling candidates (sometimes the only two candidates) to the eventual winner’s margin of victory over the second-place finisher. If there was more than one poll tied for most or least accurate by this metric, then I compared each candidate’s individual poll position in the survey results with their eventual share of the final vote. This is why PPP is listed as having the most accurate poll of the Florida governor’s race despite the fact that their final pre election poll had Alex Sink (D) leading Rick Scott (R), the eventual victor, 48% to 47%. This was rated "most accurate" notwithstanding Scott’s victory because it underestimated Scott’s eventual share of the vote (48.9%) by only 1.9 percentage points, while overestimating Sink’s vote share (47.7%) by only 0.3 points. The next-most accurate survey of this race was conducted by Rasmussen Reports, which had Scott leading by three points heading into Election Day.
Race
|
Most Accurate Poll
|
Least Accurate Poll
|
Ohio Gov.
| ||
Pennsylvania Gov.
| ||
Florida Gov.
| ||
Colorado Gov.
| ||
Colorado Sen.
| ||
Florida Sen.
| ||
Iowa Sen.
| ||
Nevada Sen.
| ||
New Hampshire Sen.
| ||
North Carolina Sen.
| ||
Ohio Sen.
| ||
Pennsylvania Sen.
| ||
Wisconsin Sen.
| ||
Wisconsin Gov.
| ||
Nevada Gov.
| ||
Iowa Gov.
| ||
New Mexico Gov.
| ||
Michigan Gov.
| ||
New Hampshire Gov.
|
Of course, not all these races were as close as the presidential race is this year. The Senate races in Florida, Iowa and New Hampshire, and the gubernatorial elections in Colorado and Michigan, were all blowouts. Take that however you like. Also notice that the above table does not include any data from Virginia, which had neither a U.S. Senate race nor a gubernatorial election in 2010. In the 2009 race for governor there, SurveyUSA was right on the money; its final poll before Election Day showed Bob McDonnell (R) leading his Democratic opponent, Creigh Deeds, by eighteen points 58% to 40%. McDonnell won that election with 59% of the vote to 41% for Deeds. That was three years ago, however, so it may not be worth much.
Finally, there are a handful of states that are assuredly not up for grabs in the presidential election but are experiencing heated U.S. Senate and/or gubernatorial races. For those who are interested, here are my findings on polls of 2010 races in those states.
Race
|
Most Accurate Poll
|
Least Accurate Poll
|
Connecticut Governor
| ||
U.S. Senate – Connecticut
| ||
Maine Governor
| ||
Massachusetts Governor
| ||
U.S. Senate – Washington
|
http://fivethirtyeight.blogs.nytimes.com/2010/11/04/rasmussen-polls-were-biased-and-inaccurate-quinnipiac-surveyusa-performed-strongly/
ReplyDeleteCheck that out, it is a more comprehensive way to look at it.