Monday, November 03, 2008

The Real Margin Of Error

I was over visiting The Volokh Conspiracy's discussion of the polls. And I Said something like:

America is not a predominately socialist/communist country.

As to why the polls are wrong: the pollsters are cooking the books. Operation Chaos screwed voter ID numbers. Republicans don't like being polled. PUMAs are being seriously under counted.
Commenter LM came back at me and said that saying that America is not a socialist/communist country was a veiled smear. Well that set me off for another round. And this is what I said (revised and extended):


Interesting - saying America is not a "share the wealth" country is a smear. OKy doake.


How about all the predictions of Obama wins in the primaries that saw 10, 20, 30, and in some cases 40 point swings (polls vs reality) in Clinton vs Obama.

Big Obama Win Predicted.

Of course the pollsters have gotten all that fixed and now their predictions of a big Obama win are right on the mark. Why take Pennsylvania for instance. Despite a big Obama loss in the primaries, Obama has converted all those Clinton voters and Republicans and is now poised for a big win there.

Sure it is possible. What are the odds?

The polls this year are GARBAGE. Let me spell it out for you: G. A. R. B. A. G. E.

Their connection to reality is tenuous at best. They called a bunch of people. They got answers to questions. They then adjusted the numbers to match their proposed model of reality. Which is all good if their model is correct.

But suppose most of the D surge in registrations was Operation Chaos people? Those should be shifted from D to R. Or perhaps their view of defecting Democrats is about 20% of Hillary voters and it is actually 40%. Obviously if you get "too many" of them in a sample you have to scale it back 60% or 70% of that demographic.

And how about the 80% that won't answer the pollsters. Does their demographics match the 20% that do answer?

With those kind of response rates you have a self selecting sample. A no no in statistics. Not random. All opinions don't have an equal chance of being sampled. So what is the weighting for that? Well you sample the people that don't want to be sampled and adjust your samples to match the total pool. But how do you randomly sample people that don't want to be sampled to find out what the bias is?

But it is worse. All these numbers are shifting all the time and from place to place. And sample demographics vary from day to day. How many church goers are you going to sample from 9AM to 1 PM on Sunday? How many employed people will you get from 8 AM to 5 PM on workdays? If you sample in the evenings how many 2nd shift workers do you get? How many long haul truckers do you get on any day?

In Hillary vs Obama the polls at least about 1/2 the time weren't even close. No where near the "margin of error". All we know is the statistical margin of error given the sample size. We know nothing about the real margin of error.

Cross Posted at Classical Values

No comments: