BT et C

Monday, July 23, 2007

On Scientific Polling

I'm a little out of practice, but as a once-trained philosopher I thought I'd hold forth on the topic of "scientific" polling that is intended to predict the outcome of an election or primary. Let's begin with a hypothetical situation.

Suppose you would like, for whatever reason, to know which restaurant people are going to go to for dinner. Here are some relevant parameters

-There are six restaurants, and all require reservations. Making a reservation is not very difficult, and is not *more difficult for one restaurant than for any other.
-Three of the six restaurants -- McRomnald's, Julie's, and the Thompson Steak House -- are discussed frequently on the Food Network. The others are rarely mentioned.

You call a few hundred people and ask "At which of the following restaurants would you like to eat?" And you list the six restaurants. The results come in

30% say "Julie's"
30% say "McRomney's"
22% say "Thompson Steak House"
The other 3 each get 6%

You type up your results and call it a day. Congratulations!

Now... I hate to rain on your parade but there's a glaring problem with your method.

A lot of the people you just called are not going to *any restaurant tonight. They're going to eat at home. They might want, or even expect, to go to one of those places ... but if something comes up they won't bother.

Time to blow my cover: I'm talking about presidential primaries. As a matter of fact, a *very small number of respondents are going to vote in the primary they're being asked about. The epistemological problem is that one action (picking up the phone and answering a question) doesn't "map" directly enough onto the second one (making the reservation, getting in the car and going to the restaurant/polling booth).

All is not lost, though. You can try to make your list of phone numbers better by restricting it to to "likely diners". Perhaps you only call people who own a car. Or you might get a list of people who ate out last year. This is the usual method -- but I submit that there's a *much better one.

As it turns out, you have a great source of data that can help you, but before we get to it I have to throw another wrinkle in your survey.

Now you're not calling about tonight; you're calling about "Super Friday" -- a Friday that's occurring next February. So there are more parameters:

-The people you call have to look up, then remember, the date of Super Friday, and go to the restaurant on that day.
-They obviously can change their minds between now and then.

It's starting to sound impossible. Suppose you had a bunch of money to invest in a restaurant, and you got paid back depending on how well they did on Super Friday. You might be inclined to call it a crapshoot, or you might just spread your investment among the top 3 restaurants and consider your bets hedged.

But don't forget I told you about a great source of data that can help you:

You can get the reservations lists from all these restaurants. You can even go tonight and see which one(s) are full and which ones are empty. If you do this every night from now until Super Friday, you discover something astonishing:

People are just packing it in at the Ronpaul cafeteria. The reservation list is as thick as a phone book. There's a line out the door and down the street every night. Clubs are booking their private functions there. Some of the people going there are nuts, but some are good amateur cooks, and a few are professional chefs.

McRomney's and the other restaurants also have some people in them, most of the time.

For whatever reason, the Food Network isn't talking much about the Ronpaul cafeteria. When they do mention it, they postulate that there's a glitch in some people's restaurant behavior, because after all why would they keep going to this place if it's never really mentioned on the Food Network?

I am 100% open to discussion of flaws in my metaphor, but it seems to me that the following activities "map" more closely than the ones mentioned earlier:

1. Going out and talking to people, hanging up signs, etc.
2. Going out and voting in the primary.

I think 90% of the people currently doing #1 are doing it for Ron Paul, even if only 2% of the people answering their phones are in his camp.

As a result, I am one of the (apparently few) who will not actually be that surprised if Ron Paul wins the nomination, which is to say if the current 'shape' of political action continues to go as it is currently going


  • I completely agree and I believe your analogy is appropriate. I won't be surprised either and I'm continually amazed at all the people who say he has no chance. People seem to have a problem looking at evidence empirically. They see a picture of a tube on TV and someone telling them its a WMD and they believe it. Or they see candidates getting all the MSM attention and they assume those are the only candidates worth paying attention to. I think alot of people are gonna be surprised when his fundraising totals come in next quarter... and I think even more will be surprised at the NH primaries.

    By Blogger Tommyhawk, at 10:49 AM  

  • This is an excellent piece. It is very true that Ron Paul supporters are the most enthusiastic of any presidential candidate out there. We are dedicated, and you can bet that the vast majority of us will show up for the primaries. This is one of several reasons that I believe Ron Paul will win the nomination, and spark a revival of conservative republicanism.

    By Anonymous AMack, at 7:00 AM  

  • Exactly!

    These straw polls are proving that the telephone polls can only predict other telephone polls.

    Even ignoring Ron Paul's encouraging results, where were the pollsters regarding McCain vs Huckabee in the Iowa Straw poll?

    During the Aug 5th debate, George made great snark out of putting McCain and Huckabee dead even at 8% among likely straw poll voters but where did they really end up?

    Hukcabee with 18% and McCain with 1 or 2 or something, I can't remember (not sure if that was votes, or percentage either...)

    Landline telephone polls have a near-perfect record of inaccurate results. In cases where they do match, it's either dumb luck or cherry-picking from the known results.

    But no one ever mentions this the day after the election.


    By Anonymous Anonymous, at 1:01 PM  

Post a Comment

<< Home