This is an edited version of a blog entry I wrote on November 10, 2008 where I examined the accuracy of the predictions made by the various pollsters at that time. It turned out that "Kid Average" came in a respectable 3rd. This year I'm following Nate Silver's blog and reading his book BTW and as a result I am very confident of an Obama victory. We'll see.
November 10, 2008
I was recently watching a pre-election podcast of the CBS McLaughlin Group show in which John McLaughlin asked Pat Buchanan to predict the outcome of the presidential election.
He said: "The outlook is [a] probable Obama victory." I paused the podcast at that point and thought that what Pat said was not exactly a hard call to make given what the polls were saying. I wondered at that point what he would have said if he was also asked to predict the margin of victory in percentage points. (I found out the answer later in the show but I didn’t know that at this point.) Here’s a summary of the show's pollsters' predictions tha were in the video. The point spread ranges from a low of 3 to a high of 15. I've added an additional pollster "pundit" which I affectionately call "Kid" Average and include this numerical wonder as a candidate for the best estimate prize as well.
A question for you. If we could turn back the clock to before the election results were known, which polster would you have bet on to have made the best projection (i.e. came closest to the final outcome based on the polls provided by McLaughlin?) I would have bet my money on "Kid" Average who as I found out later did reasonably well. (See spreadsheet on left.) According to today's (11/10/08) CNN website Obama won by 6.6 percentage points. I used a simple average of the polls to come up with my prediction.
The pollsters not only use averages but also more sophisticated regression methods to make their projections. Here's what is written in the FAQ page at pollsters.com: "In most cases, the numbers are not an "average" but rather regression based trendlines.
The specific methodology depends on the number of polls available.
- If we have at least 8 public polls, we fit a trend line to the dots represented by each poll using a "Loess" iterative locally weighted least squares regression.
- If we have between 4 and 7 polls, we fit a linear regression trend line (a straight line) to best fit the points.
- If we have 3 polls or fewer, we calculate a simple average of the available surveys.
If you are one of those "I can't get enough of this" folks, the drama of the election is still playing out in some corners especially the race for the Senate in Minnesota. Currently Norm Coleman leads Al Franken by 206 votes! See http://www.fivethirtyeight.com/2008/11/more-minnesota-madness.html for the latest details.