Reply #3: NO NO and NO (or how polls really work) [View All]

[Tansy_Gold]

I'll come back in a few and edit this, but allow me to first say that the shift to cell phones DOES NOT IN AND OF ITSELF skew the polls.

Okay, on edit:



When they get ready to do a poll, the pollseters set up a demographic profile BEFORE they poll. Hypothetically, then, they would look at the electorate they want to poll -- State of Ohio, for instance -- and determine that in the last election, 40% of the voters were R, 42% were D, and 18% were I or other. Or they might take an average of the past two or three elections. Or whatever. They determine the breakdown BEFORE they do the polling.


They then determine what a valid sampling would be -- and I have no idea how they do it -- and say it's 1500 opinions. It might be 500; it might be 5000. Whatever.

So they decide to use that as their model and they start calling. They have a list of demographic questions -- Are you registered? Did you vote in the last election? What's your registered party affiliation? etc. - and then they start in with the actual polling questions.

When they fill their "quota" of 40% R's, they stop polling any more of them. Once they reach 600 R's, when they get another person who says "I'm a registered idiot," they either ignore that person's responses or they just say, "Okay, thank you for your time." When they get 630 D's, it's the same, and so on.

Obviously, this is all very simplistic. But the argument about cell-phone-only users isn't a valid one, not even close. If the sampling matches the determined demographic, it doesn't matter if they have a cell phone, a landline, or whatever.

The cell-phone-only IS a specific demographic that may be significant, and it may alter the actual demographic, but that would normally be taken into consideration only IF and WHEN it was determined AFTER THE FACT to be significant to the outcome of election vs. poll correlation.

What Chuck Todd has been saying on MSNBC about under-sampling of registered Dems is a valid criticism. If there has been a HUGE uptick in the number of people registering as Dems in a given area, then a poll taken only based on PAST voting patterns will be inaccurate. But that has nothing to do with cell phone usage. It has to do with registration patterns. And since there's no valid historical data on how people who have never registered before will vote, it's difficult to design a survey model to predict it.

All of this is why the arguments about "Only pukes sit at home and answer the phone" is equally false. Pollsters don't take the responses of the first 1500 or 5000 or whatever number of people they call. They have a pre-defined distribution BEFORE they ever make the first call.

IF the demographic changes significantly and IF the voting pattern alters from the past, pre-election polling CANNOT be accurate. It is not in and of itself the election. If it's used as a tool for targeting campaign ads and efforts, then it's doing its job. But it's not necessarily a picture of how things WILL turn out.

I am not a statistician, but I checked this with someone who is. You can believe it or continue to believe the totally erroneous notion that polls only reach old white pukes on their land lines. I assure you, the pollsters aren't THAT stupid.


And neither is



Tansy Gold