Saturday, April 5, 2008

A suspect poll

The Charlotte Observer, News&Observer, and two North Carolina tv stations teamed up for a detailed poll on the primary in North Carolina. The poll was conducted by Braun Research, which Justin Guillory profiled when he was still at PPP, and apparently does little election or campaign polling.

In one sense the poll is a good thing because they were able to delve a little more deeply into voters' attitudes toward the race and the candidates than most public polling on the race has so far.

At the same their horse race results make no sense, and unfortunately that casts a bit of a pall over the worthwhileness of the entire poll.

In the Presidential race it shows a 35-26 lead for Barack Obama among 'likely' voters and that 39% undecided rate really makes me wonder how seriously any of the poll can be taken.

Five different companies that specialize in elections- PPP, Survey USA, Rasmussen, ARG, and Insider Advantage- have released polls on the North Carolina race in the last month and the undecideds in those polls ranged from 4% to 17%.

It has been months, maybe over a year, since I saw any poll in any state that used any sort of likely primary voter model that showed this level of undecideds.

In the race for Governor they show Richard Moore leading Bev Perdue 25-19. I'm not necessarily surprised that Moore is in the lead considering the recent direction of the race, but the poll showing over 50% undecideds is a major red flag. That just is not correct.

In the US Senate race they have Kay Hagan at 10%, Jim Neal at 3%, Marcus Williams at 2%, and Duskin Lassiter and Howard Staley at 1%. Again, 60% undecided in that race? Sure. 80% undecided in that race? No.

I think the fatal flaw with this poll is the 'likely voter' model or relative lack thereof. It looks like they called everyone in the phone book irregardless of voting history and simply asked them if they're going to vote.

The problem with that is that virtually everyone is going to tell an interviewer they intend to vote, even if they don't really.

Most good election pollsters either call folks based on voting history to identify likely primary voters, or ask a series of questions to really nail down whether someone intends to vote instead of asking a simple yes or no.

My problem with this poll is the same as my problem with when the Elon Poll releases horse race numbers. If you're going to do the horse race and pretend like your numbers matter, you need to try to get the poll to really reflect what you expect to happen at the polls on election day. Either stick to the issues and skip the horse race, or employ a legitimate likely voter model so your horse race numbers make some sense.

Here are the poll questions and poll results.

I certainly hope the McClatchy papers will not start reporting these numbers like they're the gospel truth. Because the simple reality is that they are wrong.

Other issues with the poll:

-Unless I missed it they did not release the topline demographics. I would be interested to know if their racial, gender, and age distributions are anywhere in the ballpark of where they should be.
-Where are the full crosstabs? Media polls should be a beacon of transparency but I do not see these on the website of any of the four outlets that commissioned this.

No comments:

Web Statistics