Thursday, October 30, 2008

Karl Rove is Wrong

Today, in the Wall Street Journal, Karl Rove counsels voters not to, "let the polls to affect your vote. They were wrong in 2000 and 2004." There are several things wrong with this, and the most glaring is actually that in the column, Rove writes only about exit polls. His argument is that exit polls can lead to erroneous assumptions about who the winner will be be (that's certainly true, and is especially true of "early exits" of the type that were leaked in 2004).

But the implication of this column is that we can't really trust public opinion polls, so we don't know if Obama is really ahead. This insinuation is popping up all over the place. Take a look at the bottom story on today's msnbc website:


Now, you can probably guess where I'm going with this.

I downloaded electoral-vote.com's entire cache of 2004 polls and then I took a simple average of each state's polling from the last five days of the election. I used a simple average because, unfortunately, the data did not include sample sizes. Here are the results (note that in 16 states there was no public polling in the final five days) :

State Kerry Avg Bush Avg
Alabama 42.0 53.0
Arizona 41.0 56.0
Arkansas 44.5 51.0
California 53.5 43.0
Colorado 46.3 48.9
Connecticut 52.0 42.0
Florida 47.3 47.9
Georgia 41.3 54.0
Illinois 53.0 42.0
Indiana 38.0 57.5
Iowa 47.0 46.7
Kentucky 38.0 59.0
Maine 52.0 44.0
Maryland 53.0 43.5
Michigan 48.4 46.1
Minnesota 48.4 45.9
Missouri 45.3 50.3
Nevada 45.4 50.4
New Hampshire 48.3 47.0
New Jersey 48.2 41.5
New Mexico 45.6 49.0
New York 54.3 37.7
North Carolina 44.0 53.0
Ohio 46.7 48.3
Oklahoma 34.0 64.0
Oregon 51.5 45.0
Pennsylvania 48.4 46.4
South Carolina 42.0 54.5
Tennessee 40.0 58.0
Texas 34.5 58.5
Utah 24.0 69.0
Virginia 46.0 50.5
Washington 51.7 45.3
West Virginia 43.0 51.0
Wisconsin 48.2 46.2

Of these 35 states, the polling correctly "predicted" the winner in 34 cases. The only exception is Iowa, where the final average was closest of all the states. Moreover, the average difference between the final margin in the polls and the actual final margin was less than 2.4 points. Indeed, in 21 of these states, the polls "predicted" the margin of victory to within 3 points.

The story is even better when looking only at "swing states." Among the fifteen "showdown states" identified by CNN, the average difference between the final polling margin and the actual margin was less than 1.9 points. In Pennsylvania, for example, the average of the final polls gave Kerry a lead of 2.1 points. The final margin was 2.2 points.

Can polls be wrong? Yes, of course, there is always a chance that individual polls are off base. But when taken together, polls paint an accurate picture. Even a simple averaging of polls yields very reliable results (though I prefer to weight my averages when I have the data to do so). 2004, despite Karl Rove's ominous column, is great evidence of that.

No comments: