|
They use numbers from previous elections (or whatever other source they want to use) to set their demographics, then make calls until they fill the pre-determined demographic.
Hypothetical example:
Shoofly Research looks at data from past six national elections (presidential and off-year) and determines that based on past numbers and perceived trends, the turn-out this year will be -- Dems - 36% Pukes - 34% Indies - 30% based on official voter registration.
They also determine that the percentages of registered voters who voted in at least 4 of the 6 elections have been Dems - 74% Pukes - 78% Indies - 84%
They determine that an accurate sampling can be taken from 2095 respondents.
They call 4290 randomly chosen phone numbers and reach 2095 respondents. Of those 2095 respondents, 714 identify themself as Dems, 824 identify themselves as Pukes, 557 as Indies. Since these percentages don't match their intended demographic, they make more calls until they reach 40 additional Dems and 71 additional Indies, and they discard at random 90 Pukes.
They fill the predetermined numbers, no matter how many phone calls they have to make. In a sense, they already know what the results will be, because they've "controlled" the sample to that extent.
In the course of these phone interviews, they ask each respondent "If the election were held today, who would you vote for?"
After they've received all the responses from the pre-determined set of subjects, they apply the "likely voter" percentage.
I won't do the math, because it's only a hypothetical. The point, however, is that the different polling organizations set up different definitions for their "likely" voters, some based on self-identification and some on past record and some on percentages and some on various combinations thereof. They use different demographics. Gallup is weighted heavily GOP -- but I'll bet they have some (albeit bizarre) justification for it.
And inherent in all this is a certain amount of guesswork. It's all a projection based on experience, but things change and the "trend" may not be 100% accurate. Over a ten year period, suppose Dem registration goes up 5% a year, Puke registration 4%, Indy registration 7%, based on a steady, predictable increase in population. But what happens when you have a sudden 24% over-all increase in registration in a single state or county? How can you project those "historical" percentages onto an aberrant increase? All you can do is guess.
This is what happens with the cell-phone-only group. The pollster determines, from past voter rolls, that 15% of the registered voters are in the 18-24 age group, and they traditionally vote at a 32% rate. They're going to call sufficient phone numbers until they reach the appropriate sample based on their total. It won't matter if they pre-determine that number of 18-24 year old voters to be 5, 50, or 578 -- they will call until they reach that number, and they will base their projections accordingly.
It's not because there are a bunch of 18-24 year olds who only have cell phones and can't be polled that is going to skew the numbers. It's the fact that the numbers can only be based on past data and whatever the pollster considers "reasonable" projections.
A good pollster can compensate for this, either by adjusting the sample or the percentge, or by including additional information. That information can come from just about anywhere. For example, if there are studies that show XX thousands of young people have cell phones and no land lines, that can be extrapolated into the sample. There are other ways of surveying besides phone polls: face-to-face polls, focus groups, even on-line polls.
The variances between the polls isn't indicative that one is "right" and another "wrong," but rather that they are defining their terms differently in an attempt to predict the future. They take a snapshot now and suggest that the way things are tonight, they will remain the same on Nov. 2. But we all know that some people change their minds, some haven't made up their minds, some who say they're going to vote won't and some who say they might or might not, actually will.
So the pollsters try, and they're a little better than Magic 8 Balls, but they still can't predict the future, because even the past isn't perfect, and they bring to it their own very human biases and errors and hopes and fears.
It may very well turn out that the cell-phone-only population will swing the election, or the fundies who are turned off by any positive talk about civil unions or the traditional pukes who can't stand deficits or the pro-life moderate <?> dems or whatever.
Each voter falls into lots and lots and lots of different categories, so trying to define each and every one isn't within the realm of a 1400-person poll sample. Just think about how many categories each of us falls into -- cell phone only, cell phone plus land line, land line only; internet dial-up, internet broadband, internet wireless, no internet; internet at home, internet at work, internet at school, internet at home and work, internet at home and school, internet at work and school, internet at work and home and school, etc., etc., etc. Gay, straight, bi, metro, celibate. Married, single, divorced, widowed, polygamous. Income level. Education level. Parents' income level. Parents' education level.
Each of us is, in fact, a demographic of one. And that's why each of us has one vote and --with any luck -- it will be counted. The pre-election polls can only guess, within their own pre-determined margin of error.
|