Tuesday, May 04, 2004

A commenter who goes by the name "GDubya" mentioned in passing in this comment thread on the Command Post that the CBS/NYC poll from late last month had not been norm-balanced for party membership.

The two major national parties are roughly at parity these days, after several decades of decay in Democratic membership and increase in Republican membership numbers. The nation is roughly split, one third Democratic, one third Republican, and one third Independent. (I went from liberal Republican to sort-of-conservative Democrat last year, for those of you following from home, so I, personally, am swimming idiotically against the tide. Don't ask.)

In order to represent a proper, statistically balanced picture of the polity, a poll should approach the census numbers in general outline. In a *political* poll, they should be especially careful to make sure that their sample is not structurally biased on particularly political lines. Political party affiliation, for instance.

The CBS/NYT poll shows Republican/Democratic/Independent/NA numbers of 29/35/27/9. It *should* be closer to 36/36/22/6. (GDubya gives directions on how to find the relevant table - which is buried on page 31 of a 33 page pdf on the NYT link given above.) A perusal of the rest of the party-affiliation numbers, for polls stretching back into the mid-Nineties, shows that this polling group has always had a Democratic-Party bias in their polling. Whatever method they're using to select their survey subjects, it doesn't seem to have adjusted itself as the population has gotten less Democratic and more Republican. I would guess that they settled on some artificial line of procedure back in the days when there were more Democrats, and haven't checked their assumptions since then.

Bottom line conclusion? Don't trust polls from CBS/NYT. Intentionally or not, they've got some sort of systemic bias in their polls. Their results will be consistently somewhat more favorable to Democratic policies than Republican.

No comments: