Sound Politics: Accuracy of Presidential Polls

November 06, 2008

Accuracy of Presidential Polls

Conservatives will not be surprised to see Zogby, CBS/New York Times, and Newsweek bringing up the rear in accuracy. Indeed, polls sponsored by media organizations fared worse than the independent pollsters.

Likewise, conservatives will not be surprised to see that Rasmussen tied for first - along with Pew, in a bit of a surprise.

Other notables: Gallup under-performed its regular reputation of being close to the money in their final surveys. The bi-partisan GWU/Battleground poll gets points for consistency (they were also among the leaders for final accuracy in 2004).

Posted by Eric Earling at November 06, 2008 06:31 PM | Email This

Comments

1. I think the real winner of this cycle, state and national, is easily Scott Rasmussen. I haven't seen the final numbers, but by and large, he was on or near the money in every race I remember.

I was disappointed by my 2004 favorite, Mason-Dixon. Not only did they cut down their poll count, but they did rather poorly. Gold standard to bottom half of the pack is a long way to fall. Things change, I suppose.

Posted by: Benjamin Johnstone-Anderson on November 6, 2008 07:54 PM

2. I think these are not surprising facts. The customers of professional polsters have a very different set of expectations than the customers of the media.

Posted by: mykela on November 6, 2008 08:29 PM

3. Individual polls aren't where the action is anymore. The new game in town is aggregators -- 538.com predicted 52.3%/46.2% and it turned out 52.5%/46.2%, for instance.

Whatever you think of Silver's politics, the takeaway is pretty clear: don't bet against the biggest baseball nerds in the world when it comes to statistical analysis.

Posted by: brent on November 6, 2008 09:48 PM

4. Dr. Larry Sabato is probably in a class of his own when it comes to predicting the outcome. His prediction for Electoral votes the weekend before the election was very close. He actually called the race back on July 24th, but keep quiet about it until the weekend before.

For strategy, Dick Morris is probably the best. He could have possibly made McCain/Palin a narrow winner, but then again - in the words of Dr. Sabato, no Republican could have won this year against a decent Democrat, whether it was Hillary Clinton or BHO.

Posted by: KS on November 6, 2008 10:16 PM

5. After all the hype from 'experts' there was no 'Bradley effect.' I think Obama did slightly better, not worse, than most of the polling.

To me the take of James Tarranto of the WSJ on the Bradley effect made most sense. He said that it wasn't from people lying to pollsters for fear of being thought racist(?!). Instead, many voters want to vote for a minority candidate, but as the election draws closer and they learn more about the candidate's policies, they might decide not to vote for him/her. Thus minority candidates can seem to underperform on election day.

The fact that so many pundits bought into the validity of the 'Bradley effect' says something.

Posted by: russell garrard on November 7, 2008 03:53 AM

6. Actually, there was indeed a Bradley Effect in this election and Sen Obama was the driver! :)

Posted by: Duffman on November 7, 2008 06:53 AM

7. The winner is Five-Thirty-Eight. It had the final popular vote percentage almost exact, and before the final decision in NC, it had the Electoral College exact. Pretty good for a first time simulation model. Its only fault was the noise factor in the too-close to call states (MO, IN, NC), otherwise it called all the rest, including factoring in the convergence of polls the last day or two (adjustment of poll sampling models??).

It also came very close, depending on the final outcome of the remaining Senate races, in predicting the Senate outcome.

Posted by: tc on November 7, 2008 07:24 AM