Sunday, August 22, 2004
Letter to a pollster
Earlier today I gave the Rasmussen Electoral College vote projections and noted that by their poll, New Mexico had slipped into the "toss-up" category for Kerry. I had hardly hit the "Publish post" button when I came upon another poll by American Research Group showing Kerry significantly ahead in New Mexico.
I was fascinated with your description of the latest New Mexico poll results. In a three-way race with Nader1 your poll found Kerry leading Bush 49% to 42%, and in a two-way race Kerry leads Bush 52% to 42%. The margin of error is 4%.
Even though in the three-way race Kerry leads Bush by almost twice the margin of error, and in the two-way race by more than twice the margin of error, you describe Kerry's lead as "slight."
Then, in the breakdown by party, you hasten to add:While Kerry performs worse among Democrats than Bush among Republicans, registered Democrats account for nearly half of all likely voters, thus giving the overall edge in the ballot to Kerry.
Kerry polled at 80% among Democrats and Bush at 83% among Republicans. If the margin of error for the overall sample is 4%, the margin of error for the subsample is certainly no less. Hence, there is no basis for asserting that Kerry performs "worse" among Democrats than Bush among Republicans. A more accurate statement would be "Within the margin of error, Kerry and Bush are tied for support from their respective parties."2
On the other hand, you show a favorable-unfavorable rating for Bush of 43%-50%, as compared with 45%-48% on July 19—a move from a net 3% unfavorable to a net 7% unfavorable. While I can't tell from the data as shown, I presume this rating is based on the overall sample. I would have been interested to know if this move in Bush's favorability rating was significant.
If I didn't know better, I might get the impression that you were spinning the poll results.
2 I'm glossing over the fact that the question of whether Kerry leads Bush, based on a sample of 600 likely voters, is a different statistical problem from the question: "Does the support for Kerry among Democrats (226/282) in the sample represent a significant difference from the support for Bush among Republicans (169/204) in the sample?" But I'll stick by the assertion that the margin of error is greater, and leave it at that. The point of all this is spin, not statistics. [back]