About that American Crossroads poll…

ABOUT THAT AMERICAN CROSSROADS POLL…. One of the lead stories from Politico this morning was about a new poll from one of Karl Rove’s operations. It’s generated some additional attention, so I suppose it’s worth noting why it’s hard to take the results seriously.

The survey claims to offer proof that Republicans may be able to win 10 Senate seats this November, thus claiming a majority next year. There are a couple of problems here.

The first is that the poll, conducted for American Crossroads — a group launched in part by Rove and former RNC Chairman Ed Gillespie — combines results from different states to make inferences about specific contests. Dave Weigel noted how misguided this is.

Does anyone buy this? An “average” poll number across states with Senate races is not worthless, but it’s close. In Delaware, Arkansas, and Indiana, Republican candidates have better than double-digit leads over the Democrats. In Florida, Democratic candidates Kendrick Meek and Jeff Greene are basically also-rans, as Democrats gravitate to independent candidate Charlie Crist.

So you’ve got one 25-point lead (I’m guessing, based on other polls in Florida) warping the first five results, and three leads of 10 to 20 points warping the other eight results. If we had all the numbers we’d confirm what we already know — that Republicans can waltz into three (four, if you add North Dakota) open Democratic seats, but that the other 10 competitive Senate races are toss-ups. The poll averaging on messaging is more interesting, as all but three of these states voted for the Obama-Biden ticket. But the purpose of the poll is to take something that political junkies think — that sometimes, every close race breaks the challenger party’s way — and make it look like science.

Quite right. But there’s something else that bothered me about the credibility of the poll. This paragraph in the Politico piece seemed to make the rest of the article unimportant.

The Republican firm Public Opinion Strategies conducted the poll, testing 1,300 likely voters across the 13 states, for a small state-by-state sample of 100 respondents. That means that for the individual races, the survey’s margin of error is so wide as to render the results effectively meaningless. [emphasis added]

There’s simply no credible way to argue that a Republican pollster can survey 100 people in a state and offer worthwhile data about a competitive Senate race. That’s just silly.

This isn’t a poll so much as it’s a Karl Rove press stunt. I’m not dismissing the notion that significant Republican gains in the Senate are possible; I’m dismissing this poll as lacking all merit.