Correcting for Inaccuracy in Polls May be Inaccurate
|
|
| Also listed in: Cruel and Illogical Twits: |
As has been pointed out many times throught this presidential campaign season, polls are inaccurate. We get that. The reason that many people have blogged about going back to before the Iowa Caucus is that pollsters such as Rasmutten only dial land lines. So at best, the polls ONLY represent the opinions of randomly selected Americans who have land line telephones with listed numbers.
It's all well and good to point out that fact as a potential source of poll inaccuracy, but some go further to provide the demographics of people who are NOT being reached because of the land line issue, and to suggest that it is valid to correct the poll in a certain direction and by a certain amount - based on information NOT gathered.
I'm willing, for the sake of argument, because I have nothing better to do for the next 10 minutes, to assume that you can legitimately correct gathered information based on NON gathered information... I would STILL argue against a valid correction based on simply the land line error and nothing else. I'd say it's REALLY difficult to accurately poll people in Florida and Ohio who will become confused at the voting booth and vote for a can of beans for President, later claiming they couldn't understand the ballot. That's gotta be like, close to a 5% correction. I'd say it's pretty darned difficult to accurately correct for the the voting dead - Democrats who will vote in November despite being incorporeal. I expect they're planning to come out in numbers this year. I'd give that a 3% correction. And here's perhaps my favorite, I doubt if everyone answering pollsters are being 100% honest. Look, it's the age of telemarketers and other unsolicited phonecallers troubling us at the LEAST convenient times with things that "will only take a minute." There are people who will participate JUST for the opportunity of screwing with poll results as payback for the intrusion. I'd call that at least a 2% correction. Then there's a random, let's say... 5% correction needed for all the other possible errors that I didn't think of.
Let's work the math. I've heard people say 8% for the landlines, plus another 15% for the other things I mentioned.... multiply by 2 just cause it sounds good, and take the square root because I know how to do that too, and you come up with a number THAT MEANS NOTHING!!!
You can't adjust gathered information based on non gathered information.
It's all well and good to point out that fact as a potential source of poll inaccuracy, but some go further to provide the demographics of people who are NOT being reached because of the land line issue, and to suggest that it is valid to correct the poll in a certain direction and by a certain amount - based on information NOT gathered.
I'm willing, for the sake of argument, because I have nothing better to do for the next 10 minutes, to assume that you can legitimately correct gathered information based on NON gathered information... I would STILL argue against a valid correction based on simply the land line error and nothing else. I'd say it's REALLY difficult to accurately poll people in Florida and Ohio who will become confused at the voting booth and vote for a can of beans for President, later claiming they couldn't understand the ballot. That's gotta be like, close to a 5% correction. I'd say it's pretty darned difficult to accurately correct for the the voting dead - Democrats who will vote in November despite being incorporeal. I expect they're planning to come out in numbers this year. I'd give that a 3% correction. And here's perhaps my favorite, I doubt if everyone answering pollsters are being 100% honest. Look, it's the age of telemarketers and other unsolicited phonecallers troubling us at the LEAST convenient times with things that "will only take a minute." There are people who will participate JUST for the opportunity of screwing with poll results as payback for the intrusion. I'd call that at least a 2% correction. Then there's a random, let's say... 5% correction needed for all the other possible errors that I didn't think of.
Let's work the math. I've heard people say 8% for the landlines, plus another 15% for the other things I mentioned.... multiply by 2 just cause it sounds good, and take the square root because I know how to do that too, and you come up with a number THAT MEANS NOTHING!!!
You can't adjust gathered information based on non gathered information.
There are a lot of us who still expect to get to our voting booth in November, look at the ballot, and feel a feeling of warmth, light, and peace wash over us as we hear the calming refrains of the heavenly host singing... for when we look at our ballot, we will see the name Al Gore.
You can't poll us, because you don't KNOW what we'll do.... WE don't know what we'll do. And our name is Legion, for we are many.
My entire and main point is: THESE POLL ARE ABOUT AS ACCURATE AND RELIABLE IN THEIR PREDICTIONS AS ARE HOROSCOPES.
My 8% guesstimate is no more or less accurate than their polls. Both are just silly. But the unfortunate reality is that many people understand these poll results to be the true and accurate.
I think it is good to continue to point out from time to time how and why they are not accurate.
Once is not enough. This is the first time in over a month that I posted this reminder, but you can count on me to do it again in a few weeks.
I did use your 8%, but seriously, I've seen other people give demographics, and really try to put a numerical value on it.
That the polls are not totally accurate, and that there are many reasons why, I completely agree with.
I got lost reading and following along with you. Just kidding. lolol. You are right. We cant know how many. just that we and others we know have not been polled. Because we cant be reached by the old fashion ways of communication techniques. So oh well.
I will know for sure soon enough. I'm just waiting for time.
love
For example, Scott Rasmussen who owns Rasmussen polls is a conservative evangelical who belongs to a religious group who have been called on the carpet more than once for discriminating against gays and blacks.
Gallup is owned by NBC which in turn is owned by GE, the largest war contractor in the world.
Scientific polling can be extraordinarily accurate, but still not predictive. All it tells you is how people feel NOW, not how they will feel next week or next month. If a poll showed Obama down 10 points the night before the election, that would be alarming and I wouldn't be laying bets on Obama winning. However, if the polls in, say, August 2007 showed Obama down 30 points (just to pick a random date and margin), I wouldn't bet he would lose the primaries.
nlycbyg
This is actually true and a very important tool. (Quick, simple example: you're playing hide-and-seek, and you've searched the whole house except one room with two closets. You search one, and the person is not there. Based on all the information you have gathered, you know the person is in the other closet, even though you've not gathered that information yet.)
Based on your explanation, the answer here is that we don't know enough information to adjust for the non-gathered information. The uncollected information might make things even worse (or better, depending on your point of view)how an un-polled segment of the population feels.
For example: let's say that 10% of African Americans use cell-phones and not land-lines. It's possible that it so happens that if you're Black and Republican, you are almost certainly going to get rid of your land-line. So assuming you have 80% of that last 10% might well be false.
In order to be able to adjust for un-gathered data, one has to have sufficient gathered data to reduce the unknown unknowns into a manageable size.