Tuesday, January 10, 2012

Volatility and Free Parking

Nate Silver and Micah Cohen had an interesting piece last night about the extraordinary volatility in polling during the invisible primary this year; turns out that the three most volatile candidates ever (since 1984 at least) were Newt Gingrich, Herman Cain, and Rick Perry this year, and even Michele Bachmann finished an impressive seventh, while at the same time Mitt Romney and Ron Paul both wound up on the all-time top ten most stable list. Interesting!

Silver attributes it to campaign organization, or rather the lack thereof for the unstable group. Without strong campaign organizations, they're more dependent on media attention, which comes and goes.

Plausible! But then again, two of the candidates on the most-volatile list, Howard Dean 2004 and Bill Clinton 1992, had pretty strong organizations. Not saying he's wrong, but a couple of other things occurred to me as I looked at his list. Also, it's not quite clear from the article whether which polls we're looking at, but if it's national polling then it's hard to see how organization would have much effect on the bulk of voters, since there's so little early organization in large, late-voting states.

My guess is that volatility is an effect of the lack of free parking. Let me explain. Most voters aren't focused on the nomination race at all during the year before the election. Therefore, they'll park their vote -- in the unlikely event that they need to, which is one of the effects of a pollster asking the question -- with the first candidate they can think of. When there's a heavyweight candidate, they park there and stay there; in fact, they'll likely stay there right through their primary, whenever it is. Otherwise, it's going to be whoever has been in the news latest, which moves around a lot. So I'd expect a lot more volatility when there's no obvious place for inattentive voters to park their votes.

If heavyweight candidates are people who were recently (last couple of cycles) on a national ticket, then we are talking about on the Democrats side 1984 (Mondale), 2000 (Gore), 2004 (Lieberman) and 2008 (Edwards, plus a First Lady ad hoc inclusion for Hillary Clinton). On the Republican side, 1988 (Bush)...and that's it, although subjectively I'd certainly include 1996, with Dole a three-time candidate, the previous runner-up, and a former VP nominee.

Going to the chart, eight of the most-volatile candidates are from years with no heavyweight. The stable candidates include the three big names from the 2008 Democratic contest, and one other, so four for ten.

But of course the problem is how to account for Ron Paul and Mitt Romney this year. The clue? Perhaps it's 2008 stable candidate John McCain (although -- really? that's not my memory of it...might be an artifact of the particular months they're using). The answer might be that repeat candidates are less likely to exhibit wild swings. Accounts for Pat Buchanan 1988, too, if you don't like my rigging that year by including Bob Dole as a heavyweight. That makes sense: people who haven't yet focused on the current cycle will bring their leftover feelings from the last cycle, therefore either parking their support with a candidate they remember liking or being more resistant to minor fluctuations in publicity for that candidate if they don't like him or her. Note that Edwards 2008 is on the stable list, which makes four of ten stables as repeat candidates, compared with zero for ten on the volatile list.

If I played with it a bit more...it doesn't affect the results here that much, but 2004 for the Democrats certainly turned out to behave more like a year without a heavyweight. It's possible that the quicky formula I'm using here could be tweaked to exclude Lieberman...although of course you need to be very careful about that sort of thing. I would exclude Dan Quayle's 2000 campaign, since he withdrew during the period that Silver and Cohen are using.

So that's my guess. Volatile/stable in the year before voting is a function of two things: whether there's a heavyweight in the field, and whether the candidate has run before. I strongly suspect that if Silver and Cohen run that on their database, those are going to be the big drivers, even if they had a good measure of campaign organization and tossed that in.


  1. What if it's even simpler?

    Name recognition.

    If people knew who Herman Cain was, they never would have parked their vote there.

    Romney, Paul, McCain, Dole...these were known names. Forget "they ran before" and focus on "people know who the hell they are"

  2. Maybe. Holy Joe had high name recognition in 2004, and Newt is #1 on the volatile list with presumably decent name recognition...I'm not sure that Obama (#1 stable) had higher name rec in 2007 than Newt in 2011.

  3. How about even simpler than that?

    Polls are not elections. Political professionals TREAT them as interchangeable, but they are not. At least many people tend to learn something and pay some attention before casting a vote, so their expressed preference reflects an actual intellectual involvement with the question being asked.

    Whereas a poll result does not.

    Polls are a completely manufactured reality. They measure something (preference among presidential candidates well before an election) that often doesn't actually exist until the poll creates it. And politicians, their consultants, the media, and political scientists NEED it to exist, so they create it.

    But given that what polls are measuring is a completely manufactured reality, nobody should expect them to be stable.

    More importantly, nobody should pay attention to them.

  4. Thanks, Dilan, however I'd put it in a more nuanced version: polls seem to be a tangible reality, however they are manufactured out of a much more nebulous and hard-to-get-facts-on reality (the actual states of mind of millions of individual voters).

    I'll put this link up again, as I do believe that our host Jonathan B. may have misunderstood my point when I put it out months ago. Polls are not as scientific as claimed, because of the impossibilities of getting a truly random sample in an American telephone poll. It's not just cell phones, pollsters are taking measures like using cell phone lists and calling random numbers, it's more the not-at-homes, won't-answer, foreign-language speakers, busy signals and everything else that forces the pollsters to call 4 or 5 thousand numbers to get 1000 completed interviews. It's just not random enough in practice, and the net effect is that true margins of error in telephone polling should be two to three times higher than the stated MOE. This is supported by the evidence from multiple polls asking very similar questions, which show variations running at two to three times the claimed margin of error of any one poll. Full discussion at www.philsophical-ron.com/opinion-polls-arent-scientific/ .

  5. Misspelling my own name even after proofing, again, that's www.philosophical-ron.com/opinion-polls-arent-scientific

  6. Ron:

    You are right about margins of error. They are also compounded by the sheer number of polls taken-- nothing is sillier than the 1/3 of an already too-small sample "tracking" poll, which tracks nothing but statistical noise.

    The bottom line is that when there is a market for information and there is no good information to be had, people are going to make money selling bad information, which will then be analyzed as if it is actually good information. That's the polling-media-consultant complex in a nutshell.


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

Who links to my website?