Monday, October 27, 2008

Poll crosstabs can show bogus results

If you want to know if a poll can be relied upon, you can't just use your "gut" instinct about how you feel the race is going, or assumed all along. There needs to be a factual basis for judging the quality of the results:

Take the IBD/TIPP poll which purports to have been the most accurate in 2004. I can guarantee you this poll is not the most accurate survey of 2008, and here's how I know:

They have Obama leading women by 9: 50 to 41

They have McCain leading men by 5: 48 to 43

If men and women voted in exactly equaled numbers the results should be Obama by 4.

They have Obama by 2.8 %! Obama 47 McCain 44.2.

That would mean that more men than women will vote.

No one projects or pedicts that.

The question is only how many more women than men will vote.

If you take the generic data, women have "outvoted" men for the past 28 years.

So these results make no sense.

A good assumption would be women to be 53% and men 47%.

Using that metric, in this poll Obama's lead would be almost double what is reported.

I'll be interested to see if this issue gets "fixed" in upcoming polls.

btw. today is the first time that this poll has had Obama leading among young people 18 - 24.
which is an unbelievable result. Literally unbelievable. They keep placing an asterick to say that there weren't enough in that category to make reliable conclusions, but that in and of itself should be worrisome to the the poll's conclusion.

No comments: