Monday, November 01, 2004

a quick disquisition on the polls-exclude-cell-phone-users-thing

Salon, whose electoral commentary tends to be so optimistically skewed that my rationalistic-cerebral-hygiene-practices usually dictates me to avoid it, is running some numbers on the omission of cel phones from political polls. Namely:
Will the cell phone voting bloc wind up becoming the November surprise? Zogby has just released a path-breaking presidential poll conducted exclusively on mobile phones. And the winner is John Kerry by a landslide margin of 55 to 40 percent... If the Zogby poll, which was conducted in partnership with Rock the Vote, is a reliable indicator, Tuesday night might not be such a drawn-out, nail-biter after all."
What annoys me about this discussion is that it is conducted with such little respect for the math involved, which is all the more annoying because it is very simple math. Three numbers are required:

a. The estimated % of persons-who-will-show-up-to-vote who use only cel phones and so cannot be reached by pollsters.
b. The estimated gap between Kerry and Bush among those who-will-show-up-to-vote and use only cel phones.
c. The estimates gap between Kerry and Bush among those who-will-show-up-to-vote and are not exclusive cel phone users.

The adjusted Kerry-Bush gap is just: (1-a)c+ab

The larger a and (b-c) are, the more of a difference cel phones will make, but the difference is determined by the product of the two and products of proportions get small quickly. I personally can't see the cel phone bias in polls giving Kerry even anything like a full percentage point in the polls. To which one can say, of course, that one percentage point is a big deal in an election this close. To which one can reply, of course, that the cel phone problem--in terms of the potential size of the bias--is almost certainly smaller than the general bias from refusals/other-nonresponders in telephone surveys; the only issue is that it is not so clear which direction that bias is. Bottom line is that it wouldn't surprise me at all if the final result is 2-3 percentage points (or--who knows?--maybe more) different from what the consensus-of-polls predicts on the day of election; the only problem is that it's anybody's guess what direction that discrepancy will be.

14 comments:

Anonymous said...

You make it seem like the authors of the surveys over at Zogby, or Gallup, or elsewhere, to say nothing of Salon.com and us mere mortals are all so mathematically/analytically challenged that they (we) would not even think of ever considering let alone understanding these factors. Wasn't one of the senior analysts at Gallup a former soc prof from Wisconsin (and a once-teacher of the graduate methods class)?
If your post generates 42 comments of the "right on" type from purveyors of "easy math," I'm going to have to suggest that an esteemed body of Psychologists consder tagging a new disorder -- the Analytical Superiority Syndrome, no acronym-generated insult intended.
Phew! And they say lawyers are magisterial in their manner! {from a not-so-anonymous-after-all JFW reader)

Anonymous said...

I for one am still waiting to read the analysis of that informal survey previously conducted in this blog. Most concur that the data shows athletes to be dumb, but personally, I have somes issues with the controls of this research - one respondent was not even a student and there was a janitor who cast some dispersions on our esteemed jocks. Prof. Freese should have jumped in and cautioned his readers not to let such bias influence their answer. The Probability 'hit' on this research can be blamed on only one person.

jeremy said...

I'm not sure what Commenter #1s problem is, but I always appreciate getting a new-candidate-for-the-DSM-V diagnosis first thing in the morning. Given various things that I have heard people say about the potential influence of cel phones on elections, including the excerpt from Talbot included in my post, it doesn't appear that it's all that well publicly understood. Indeed, I wrote the post because I was exasperated by reading yet another analysis/commentary making it sound like cel phones were the main source of potentially dramatic bias in polls, where I think that's obviously not the case. While I appreciate the commenter's faith in Zogby, Gallup, et al., the fact of the matter is that no pollster has any privileged information on how turnout in this-particular-election is going to differ from its predecessors, and so they all know there is a bit of murk in their crystal balls, but no one knows how much.

jeremy said...

More on Crazed Commenter #1. I mean, seriously, read the Talbot quote in my post. It certainly sounds like he's of the opinion, after seeing that the cel phone gap is only 15 points, that cel phones could take what looks like a really-close election and make it not that close after all. I argue that cel phones won't make that much of a difference, which is the opposite of Talbot's stance. And then I get criticized because I'm supposedly being condescending and saying something that everyone already knows and pretending like everyone doesn't. If everyone already knows this, then why do Talbot and I disagree?

nina said...

Sounds like a frustrated lawyer who actually IS mathematically challenged has been provoked into deep quandry by your weblog. That's good -- gotta get the legal profession thinking. I thought you made a good point, though I don't understand why in the last sentence you concluded that there is an absence of information on where we may be with the cell-phone users. Wasn't the initial poll at least providing a basis for a partial information on how those points might tilt?

goesh said...

I'm just curious why this polling mechanism hasn't been employed before. It would seem there are too many sharp Statisticians/pollsters around for this to have not been previously considered and/or used - 15 points is glaring, but then, what the hell do I know? I would be more interested in seeing what percentage of the respondents in this survey actually follow through and vote. If I were young and on the go in this target population, and was interested in voting only for the sake of removal for instance, I might be inclined not to vote, since the numbers clearly show the job will be done without my participation. On the other hand, there is no indication that the participants were interested in the results of the survey, or that the results were specifically and necessarily shared with them, other than being registerd in summary on the net. We may never know what % of this group actually voted - if Bush wins, who would want to throw mud on their own face? If Kerry wins, it becomes irrelevant and left behind as a dusty byproduct of polling innovation.

Tom Bozzo said...

Goesh, this is the first presidential election where the cord-cutters threaten to be significant. From what I've seen, the "a" percentage is somewhere in the mid single digits, so if "c" is close to zero, the cell phone effect ends up in the range of 2/3% to 1% -- maybe at the high end of Jeremy's expectations. In 2000, "a" would have likely been in the very low single digits.

Nina, I think what Jeremy is saying at the end amounts to an expectation that the sampling error effect (which is fundamentally unpredictable in direction), nonresponse effects (which is also unpredictable in the absence of information on response rates [thought to be very low] and characteristics of non-respondents) would tend to dominate sources of bias such as the cell phone effect that can be assigned a direction. Though if the LV models still have approximately the pro-Bush bias that they seemed to have had in 2000, the combination might allow more room for a pro-Kerry than a pro-Bush surprise. (Yes, from my comment to the electorate's voting devices.)

jeremy said...

Tom -- basically yes, except even though other nonresponse effects are "thought" to be small, I'm not so convinced the conventional wisdom will be correct here. It's kind of the point at which the science and alchemy of polling meet, or at least where extrapolation is as faith-based as some of the programs proposed by the current administration.

Anonymous said...

Does anyone here really, truly, honestly think that the "winner" won't be decided in a shenanigans match between lawyers?

jeremy said...

I don't. Perhaps I will be proven wrong.

Anonymous said...

Well now, it appears that Mr. Bozzo has done some ciphering in his day, yet nary' a peep about donut consumption and weightlifting. The latter survey projects every bit as robust a reliability coefficient as does interrupting the masturbation of young punks via cell phone and querrying them over political favoritism. Flutter that deck of punched cards through an IBM mainframe, buster.

Anonymous said...

i propose a content analysis of all JFW posts containing the word "buster." perhaps there's another DSM-worthy pattern in the deep structure of this posters' rants.

Anonymous said...

Just as I thought - not much sass gets posted when I hearken back to the old days when variables were clad in iron and Grads did it in long-hand during late night sessions and vied with each other to see who would fetch the first cup of coffee to the Master when he entered the classroom, after prostating to him of course - indeed, when control was wrought of steel and careers hinged on a single survey and lard didn't have to be rubbed on regressions for a bit of cooking. Toss the punks in this binary 'skit' of a survey a shot of the schematics of the computer program that runs the latest model of a chain gun along with a picture of Bush, ring their cell phones, and that 15 point spread gets cut in half on the spot. I'll wager a dozen glazed and a pound of wisconsin's finest smoked bacon on that.

Tom Bozzo said...

Last anonymous, you're on, but I get to choose the picture of Bush.