Thursday, August 5, 2010

Breaking it Down: The Polling Mess

Yesterday, the Dispatch wrote this about the new Rasmussen poll showing Rob Portman up by 4 on Lee Fisher:

Rasmussen's explanation of the party ID crosstabs hardly seems to lead to a 4-point Portman lead, unless its sample is overwhelmingly Democratic:

"Portman, a former congressman and George W. Bush administration official, has support from 85% of Republicans, while Fisher is backed by 69% of Democrats. Portman leads by 20 points among voters not affiliated with either major political party."

Well, with Rasmussen's numbers out today showing Kasich up only by three on Strickland, do these crosstabs look any better from a Party ID weighting perspective?

Not so much.

Kasich wins 83% of the GOP vote.
Strickland wins 72% of the Dem vote.
Kasich wins Independents by 19%.

And we're supposed to believe Kasich is up by only 3%? Once again, that's only possible if the sample is overweighted with Democratic votes.

So let's try a little experiment of our own.

If we use the highest profile Exit Poll from the 2008 election, as provided by CNN, that broke down the number of GOPers, Dems, and Independents that voted in the Presidential, and applied those ratios to this result, what would happen?

In 2008, per CNN, 5,608,789 Ohioans voted. 39% were Democrats, 31% were Republicans, and 30% were Independent.

As you can see, if we apply Rasmussen's gubernatorial choices by Party to the Party ID model provided by CNN in 2008, we get a vastly different result. And that doesn't even take into consideration the all too likely probability that the 2010 breakdown by Party will likely be more favorable to Republicans than the Year of Obama.

Now, is this formula perfect? Of course not. We're using two different polls, and we're unable to determine whom the undecideds in Rasmussen would vote for. But this does give us a more realistic perspective of what to expect if Rasmussen's breakdown by Party ID is correct.

The one pollster so far that has most closely reflected Party ID as it broke down in 2008 has been the Democratic polling firm PPP. Unfortunately, we haven't heard from them since late June.

This all goes to show you, gauging a race by a single poll, while very interesting, isn't likely to give you a real reflection of the race. Instead, we need to gauge polls as an aggregate. The bad news there? Four of the past six polls have been Rasmussen.

PPP, Quinnipiac, Ohio Poll, and Survey USA need to step it up.


  1. Yes, if there's anything Rasmussen is known for... it's oversampling Democrats....


    And what about the last poll that showed Strickand gain three points two weeks ago?

    Face it, the RGA and Kasich ads "didn't get the jobs done."

  2. What a totally and completely unresponsive comment.

  3. Boy, you're cranky today... That faceplant on Coleman really irrated you.

    The idea that Rasmussen oversamples Democrats is just nuts.

    You didn't seem to have any problems with them two weeks ago when it showed Strickland gaining three points. Now, it shows him gaining another five.

    The only conclusion is that Rasmussen shows Kasich and the RGA's ads not only didn't work, but Kasich's angry tone turned voters off.

    But if you want to cite CNN's infallible exit polls, then Al Gore must have been President after 2000.

  4. Ha!

    Still unresponsive. And I called out Rasmussen months ago for these same problems.


No profanity, keep it clean.

Note: Only a member of this blog may post a comment.