Whilst We Wait

I crunched the data earlier this morning for this week's Horserace Update, but I haven't had the time to perform the rather meticulous acts of assemblage needed to blog it. I have received an email from a reader linking me to this article which is interesting and related to the Friday Feature.

The gist of the article is that surveys may be registering increasing levels of discrepancy because of changes in the ways Americans use their telephones. I broadly agree with the article's thrust - polls do not represent the totality of their purported object (the electorate), but rather a very broad subsection of that totality.

But conceding the point, I still think it's too soon to throw the polls out with the bathwater. First of all, they continue to measure a non-negligible slice of the American electorate. And secondly, watching the same poll week after week will still likely reflect actual movements within the thinking of that subset. It's possible that a telephone-based survey will reflect a broad movement among the phone-answering public which is not mirrored by those who block calls... but there's certainly no reason to suppose that those who are blocking calls are likely to move in a different direction than those who are not.

The article mentions disparities in answering skew demographics older, whiter, and maler than the population at large. But a reputable pollster should still be able to adjust his sample to compensate for that disparity. The fact of a skewed population is only a problem if like-to-like comparisons between demographics which are identical in all aspects save phone-usage patterns would yield large differences. For example, if call-screening black females were twice as likely to vote for George W. Bush as call-answering black females, then your poll would really start to become fatally compromised.

I think polls should always be taken with a grain of salt, especially as projections of ultimate totals. At root, they are little more than snapshots of an electorate in flux. Many do misread the polls as signs that one candidate does in fact have a one point margin over another, and I've seen many who find the whole concept of polls fraudulent because they often miss the winner of a race. The one is a fallacy of over-confidence in the poll's accuracy, and the other a fallacy of presuming that polls describe a future reality rather than the present one of the time the poll was conducted.

So, with regards to the Friday Horse Race, I'd like to point out that it should always and everywhere be taken with a grain of salt. But the wideness of the polled lead in a state, and the movement trends from repeat samplings by the same polling outfit DO yield information that is of value.

Anyhow, a good article, and well worth the read.