Thoughts on "early voting" poll results - significance of a sample of 90

[patcox2]

Someone posted, misleadingly, that "exit polling" showed a Bush lead among early voters. Well, for one thing it wasn't an exit poll, the results of which are never released before the polls close.

It also misled by suggesting that this was "early voting" result and therefore indicative of the results in those particular states where early voting is allowed. The fact is it also included absentee ballots, which are allowed everywhere.

What happened is that when NBC did its regular tracking poll, it found that 9% of respondents had already voted, either in "early voting," or by absentee ballot. (I have already voted by absentee ballot, by the way).

So, this calls for a note on statistical significance. I know many people find it hard to beleive that a sample of 1,000 can accurately predict the result in a total population of 250 million, but the accuracy of statistical sampling is an absolute fact with the only caveat being you have to have a truly random sample, and thats where Gallup, for example, goes wrong.

But back to these early voters. The numbers quoted in the article probably aren't even statistically significant, and if they are, they probably have a margin of error of 20% or so.

It works like this; the figure quoted was a cross-tabulation, it gave the horse race figure among those 9% or respondents who had already voted.

Well, when you do a cross-tab, you have to compute the confidence interval (+ or - 5% is an example of a confidence interval) and the statistical significance( whether there is any value at all) using the size of the cross-tab sample, not the size of the total sample. So, although a total sample of 1,000 will produce results which can be generalized to the population as a whole, a sample size of 90 (the number of people polled who had already voted) cannot. Its not a big enough sample. Even if it is statistically significant, the confidence interval is going to be huge.

Its journalistic malpractice to even mention such a figure, in light of their significance.

Given that, I still think its good news for Kerry. I would bet that absentee ballots are skewed for Bush, so him having only a 1% lead among absentees and early voters is just more evidence to me he will lose on election day.