Note: This blog's templates will be updated this afternoon to a responsive design bringing it in line with MiamiHerald.com.

At that time, we will also change to the Facebook commenting system. You will need to log in with a Facebook account in order to comment.

« House stadium funding package advances with Cuban baseball player provision | Main | Anti-soccer port groups begins an ad blitz »

Unskewing, averaging, rejiggering the FL governor’s race polls, Pt. 1

@MarcACaputo

Yeah, it’s April, but it sure doesn’t feel like spring-training season in the race for Florida governor.

Gov. Rick Scott is on pace to spend $6.5 million from mid-March to mid-May on TV ads. He and Charlie Crist are pulling in major-league money. And the fans have their box scores: myriad polls, many of which show Crist has a decent lead. But that’s either not really the case now or it likely won’t be so by Election Day.

Crist leads Scott by about 2.4 percentage points, based on averaging six polls released since Scott began advertising in March.*

One of the most-surprising recent polls came yesterday when conservative-leaning Rasmussen Reports showed Crist leading Scott 45-39 percent, or 6 points (the exact result of a St. Leo University Poll last month).

Like ref-working fans, Republicans complained the poll undersampled/under-represented Republicans. It probably did. We heard the same in 2012. Then, the refrain gave rise to conservatives “unskewing” polls to adjust the surveys’ demographics. The Unskewers were dealt a blow, however, when President Obama won. Many of those polls showing a Democratic win were right, after all.

Unskewing, though, has merit. It gives us a chance to look at poll data in different scenarios. But it's important to note that it should not be taken as being better than the survey/s conducted by professional pollsters. This is merely back-of-the-napkin reverse engineering using Excel. A major reason many pollsters don't manipulate the turnout is that a poll is supposed to be randomized -- once you get your target (say 800 respondents randomly called) then that's what you have. If it happens to lean one way or the other, then so be it. That's why they're called "random samples."

Rasmussen’s poll, generally speaking, had Democrats beating Republicans by 4 percentage points in turnout; 30 percent of the poll was independent. Generally, there should be more Republicans and fewer independents in a Florida mid-term. The complexities of polling and partisan breakdowns aside (background), here’s what the poll would have looked like if it had been scaled to reflect a more-realistic mid-term election model:

Crist 44.4 percent; Scott 41.5 percent (a 2.9 point margin).

The above is based on a turnout model of Democrats and Republicans at 41 percent each; independents at 18 percent. That, however, might still short Republicans because, in averaging the 2006 and 2010 mid-term elections, GOP voters comprised more than 43.7 percent of the electorate and Democrats less than 41 percent. By using the mid-term turnout averages, here’s what Rasmussen’s poll could look like:

Crist 43.6 percent; Scott 42.7 percent (a 0.9 margin).

But this is one poll.

And all polls (especially the cheaper, public ones here) have margins of error, unintentional biases and glitches. One way to strip out some of that noise and counteract the partisan sampling issue: take an average of the polls' partisan crosstabs (i.e., the percentage each candidate draws in support from Republicans, Democrats and independents/other-party voters).

Again, this is relatively simple math, another way to view data in an effort to provide a more-complete picture. If polls are truly bad and skewed, some of that skewy-badness will bleed into these numbers. Pollsters also use different sampling techniques and technology and therefore report different results. Some polls are better than others, but they’re treated here as if they’re all the same.

Based on cross-tab averaging of the six recent polls, Crist leads Scott 44.8 to 42.4 percent, a 2.4 point edge.

But that’s based on R=41% D=41% and I=18%. If we take the 06-10 averages (R=43.8%, D=40.6%, I=15.7 percent), here’s what happens:

Crist 43.9 percent, Scott 43.5 percent, a 0.4 point lead for Crist.

Incidentally, under a 2006 turnout model, Crist does slightly better. Under a 2010 model – a high watermark for Republicans – Scott wins by 0.6.

Numbers like these are why we keep believing the race will be close -- assuming the margin is not razor-thin already.

While we can’t be sure what the results are, it’s certain the fans will fight over this post as well. Libertarians, for instance, will note that candidate Adrian Wyllie is rarely if ever polled. 

Oh well.

Get some popcorn, folks. It’s a long way to November. More box scores, and more fighting over them, are to come. There's a lot of ball to play.

*Polls used: Rasmussen Reports, Mason-Dixon, SurveyUSA, Voter Survey Service, Public Policy Polling, Saint Leo Polling Institute. The only recent poll omitted: University of North Florida, which doesn't provide crosstabs online and couldn't on deadline.

Comments