Something About Exit Polls Over The Years That I Do Not Understand

[ThomWV]

Exit polls get it wrong in this country. For some reason they work elsewhere, but not here. I don't know why that is but that's not what is bothering me right now.

The Polls seem to have been getting it wrong for about 8 years now. Before 2000 they worked pretty well, now they don't. Lots of reasons are given; cell phone use tops the list. Pollsters are smart people and statistics is an exacting branch of mathematics; probabilities are determined with a rosy level of certainty - this shit does not happen by accident.

So here is what I don't understand, why is it that if the polls are wrong and getting wronger that the "margin of error" has not been adjusted to match the increased uncertainty in poll results? Why aren't these cell-phone-users, and Bradley-effect projections factored in? If a poll in the year 2000 got 300 positive answers to a question asked of 400 people we might say that the likelihood of the same answer coming up elsewhere is 75%, with a margin of error of 4% or something like that; so today why isn't the answer 75% likely with a margin of error of 20% because we can't reach people like we used to be able to?

What gives? Is the arithmetic of polling so sacrosanct that it can't be adjusted to more accurately model society?

[DS1]

1. In 2000 Exit Polls were suddenly deemed 'useless' by big media

They knew what was up, and started talking them down well before the election.

It's all part of the plan.

[amdezurik]

2. they were "wrong" in that they reflected

the real vote vs. the one we were allowed to have...cheating always distorts the results.

[frogcycle]

3. You seem to be confusing exit polls and pre-election polls

the latter require all sorts of complex algorithms in effort to "get it right" and they vary all over the place.

Exit polls are where you ask people how they voted. The only way they are wrong is if the people lie. Or if the pollsters lie (or can't tell race/gender/etc :) ) Edit: or if the votes aren't counted - but then the exit poll is right and the count is wrong.

[Bucky]

4. This is why I only ever rely on pre-election exit polls. It's always 0% to 0% in those.

I used to work as a data analyst for a market research company--I ran the polling operation and the number crunching--and there's damn little algorithming going on, complex or otherwise, in election forecasting. We watch for demographic variables and sometimes set quotas for regional, racial, or other demographic factors. But mostly it's just random calling after households and hoping for the best. It's part of the nature of random sampling selection that the numbers just sorta come close to the reality within the margin of error. It's stunningly uncomplicated.

By the way, don't disrespect random sampling--it's the same thing that protects the food you buy and the toys your kids chew on.

[frogcycle]

5. the trick is to make it actually random

Call lists are developed from somewhere. And when they try to id "likely" voters, they are introducing filters that may not be valid.

[applegrove]

6. I wonder if the GOP have not trained followers to not stop to talk to pollsters or something.

Wrong exit polls have apparently followed Bush from Texas. Early wrong results favouring the democrats would result in a bounce to the GOP. I wonder.