Friday, November 9, 2012

Guam General Election Results & Exit Polls

Guam's 2010 General Election is now, presumably, over (this is written before the recount). The Guam Election Commission intends to do recounts on a few precincts, so that may affect the election on the margin, but I want to, instead, give a little insight into the differences between the exit poll and the unofficial results. To start with, look at the relationship between the exit poll result for a candidate (represented on the x-axis) and the unofficial tally (as reported by KUAM and represented on the y-axis). What we should see, if the levels in the poll were perfectly repesentative on the unofficial results would be a series of 30 points aligned along a perfect 45 degree line (slope of 1). That would mean that the only difference in the results would be on the scale. Instead, we see this:
It is actually pretty close to a slope of 1, although there are pretty big outliers (Judi Won Pat, Judi Guthertz and Ben Toves). In fact, if you compare the poll results to the election results, you find that all of the candidates who were predicted by the poll to rank at least 3 places higher than the unofficial ranking were Democrats; and all of those who were predicted to rank at least 3 places lower than the unofficial result were Republicans. To me, this strongly suggests that there was a non-random bias in the poll. I am not saying it is intentional. In fact, let me make up an arbitrary statistic: I feel 99% certain that the exit poll was not intentionally skewed. First, there'd be no point in it because the results would not affect the outcome. Second, I believe that Dr. McNinch wants his student-conducted polls to be unbiased, as much a possible. Third, Dr. McNinch would want the polls to reflect the actual outcome of the election because it would be embarrassing if they did not.

What kind of things could affect a poll that would bias its results? Any sort of demographic over/underrepresentation could add bias, as could any selection bias (such as survey workers approaching certain types of people by behavior or dress). It could also be affected by where or even when individuals are surveyed, since that could also determine who are most likely to be asked. It would be hard or maybe even impossible to eliminate all bias. Also, people have to be willing to participate in the survey.  Thus, even with a random selection method, willingness to participate may not be random. And consider this: any of the characteristics I alluded to earlier could affect willingness to participate, too.  Okay, enough talk about bias.

What seems most strange, to me, are the bigger outliers, like the Judis and Ben Toves. While it is not unusual to have an occasional outlier, it seems odd that there would be 3 outliers, all in the same direction.

I have one slight caveat to add. Logically, another reason that the poll might be off from the unofficial results is that the unofficial results may be affected by the mistakes of voters. You can void a race on your ballot by overvoting, but you cannot void your selections for a survey because, generally, the survey worker won't let you cancel out all your votes by trying to vote for 16+. They'll tell you and make your votes add up to no more than 15. That can introduce a slight bias, too. I'm, personally, in favor of letting people vote for as many candidates as they like. If you vote for 30, it doesn't really count, if you vote for 29, then you're voting against one candidate. My guess is that most people would still vote for 15 or fewer, but at least those who vote for more won't have their votes nullified. I'm not sure if this idea is organic/constitutional or not, nor do I expect anyone to adopt it.



Sort-of related posts:

Guam general election 2012

Memo about polling and Guam's election
A few thoughts on elections from the demand-side
Guam Legislature from a labor perspective and a rallying call
Old hands, new hands and flashbulbs
Comments on the Calvo administration's 'spending cuts' and the debt ceiling
Possible response from McNinch (different topic)
Lee Webber admits why he wants a part-time legislature
Functions of the Guam Legislature
A view of Guam's Primary election

2 comments:

  1. FANTASTIC ANALYSIS!!! This is why McNinch has zero credibility - this is real work!

    ReplyDelete
  2. I think that Dr. McNinch's work is valuable and he has real insight. That being said, I do not always agree with his analysis. When I do, there's not much to write about.

    ReplyDelete