Wednesday, November 09, 2011

Lies, Damn Lies and Pre-Election Polling

First off, the post title was not meant to be as pejorative as it comes across.  The reference and apologies are to a great recent paper by Walsh, Dolfin and DiNardo(2009).

Remember it was a very short time ago that polls were saying Raja was making this a race with 'polls' saying it was barely a 10 point gap between the two candidates (41% to 31% with 28% undecided). If you looked at it from an angle it might have been a 9 point gap. Single digits! A race!

Just 5 days ago the same said polls had at least said it had widened to 18 points

The final results.. 62-38 or a 24 point gap.

I have a simple observation. There were not even enough people paying enough attention to this race to explain any shift in the polls over the last few days, or certainly over the last few weeks. People voted today as they would have if they had been asked to vote last month. It is not that there really were 28% undecided last month.  Just consider if that were true. The core votes for Raja certainly remained core Raja voters and that early poll matched his results at just about 31% exactly. It means that virtually 100% of the 'undecided' voters all decided en masse for Fitz in just the last couple weeks. No, of course they didn't.  Or if that is really true most need to rethink the interpretation of what 'undecided' means.  

So as the next mayoral race officially kicks off today, maybe the mayor does not have as much to worry about as one might have inferred.

There were just a shade under 230K votes cast for Allegheny County Executive.  229,419 to be precise in the preliminary count.  I predicted 240K votes for that race which isn't bad given other numbers that were being floated. Turnout overall was just barely under 27%.  I am a bit unclear why the county really predicted between 30-35% which would have meant between 10% and  30% more people at the polls than actually showed up. You can't even say weather kept people away since it was pretty amazing weather for a fall election.  Imagine if there has been some more typical weather what that turnout would have been.


Yeah, yeah.  Maps.  There will be maps.  But I think you know what they look like. 

5 Comments:

Anonymous Anonymous said...

Wolosik(sp.) said he based his turnout estimate on previous ACE election turnouts.

Wednesday, November 09, 2011 9:01:00 AM  
Anonymous MH said...

I think 'undecided' means 'will vote for Irish last name' in at least 50% of the cases.

Wednesday, November 09, 2011 9:33:00 AM  
Blogger Bram Reichbaum said...

Would love to see a map on the Library question.

Wednesday, November 09, 2011 10:44:00 AM  
Anonymous John Dick said...

Chris, thanks for the offer to post a comment to your blog. Naturally, we get a bit defensive when people critique our methods and capabilities without fully understanding them. Our doors are always open to you and your many readers to learn about our admittedly unique approach.

As for your story above, what seems to have happened here is a "game of telephone," where you reflected on Jon Delano's interpretation of our findings instead of the source of those findings directly (our blog at http://blog.civicscience.com). You will see that we never characterized our findings as predictions but, rather, simply as snapshots of local sentiment during given points in time.

Many voters, consciously or unconsciously, identify themselves as "undecided" for myriad reasons. The number of those voters naturally dwindles as campaign strategies unfold, advertising efforts are waged, and the conveniences of procrastination wane in the hours approaching election day.

As for our findings, we did identify a trend of undecided voters moving into the Fitzgerald camp throughout October. Our final analysis (both manual and algorithmic), suggested that the remaining 22% of undecided voters would break approximately 2:1 for Fitzgerald and Raja, respectively, on election day. Roughly, this would have indicated a final tally of 62.75 (Fitzgerald) to 37.25 for Raja. Where we erred was in overlooking the 1+% of write-in votes that were tallied. This was an embarrassing mistake on our part and not one that we will make again. Nonetheless, our final forecasts were fairly precise.

We do not believe it is appropriate for pollsters to publish their predictions within a certain proximity to an election, as those results themselves may have unintended influence on people's voting behavior. In this case, we shared our predictions privately with select people in the media and political sector on Tuesday afternoon. You will have to take our word for that (or not).

We appreciate your inherent skepticism of public opinion research but please do not associate us with conventional methods. We are different. And, we believe, better.

Thursday, November 10, 2011 11:06:00 AM  
Anonymous Anonymous said...

John, The blog you referenced states that "Our main goal now is to use the final wave of data to measure our predictive capabilities and hone our models to be even more effective in 2012." And yet, you are trying to say here that your findings were never characterized as predictive? And then also say that "We do not believe it is appropriate for pollsters to publish their predictions..." So which is it? Are they predictions or not?

Friday, November 11, 2011 11:08:00 AM  

Post a Comment

Links to this post:

Create a Link

<< Home