| Before anything else, the poll.
SurveyUSA for Firedoglake (1/17-19, likely voters):
Baron Hill (D): 41
Mike Sodrel (R): 49
Undecided: 10
(MoE: ±4.1%)
I'm not so sure if FDL made the right choice here. Sodrel is not the only option in the Republican primary -- attorney Todd Young has banked quite a bit money for the primary already, and teabaggers' choice Travis Hankins is also staking out a position as a real wingnut's wingnut. Sodrel has a lot of name recognition, sure, but you have to wonder if 9th CD Republicans are eyeing their options.
And now, for the other stuff: If you've been following SSP this week, you probably saw that we raised some questions about SurveyUSA's latest round of House race polls that they've conducted for Firedoglake. We were particularly concerned with SurveyUSA's sample composition in its poll of New York's 1st Congressional District. More to the point, we had problems with a poll that pegged the share of 18-34 year-olds among likely voters at just one percent of the electorate. SUSA founder Jay Leve has responded:
The sample is not "pretty weird." It is a sample of likely mid-term voters. There is no one "right" way to draw a sample in a congressional district 10 months from a midterm election, but this sample was drawn carefully and defensibly. To be included in the sample (SurveyUSA's criteria, not the client's): the voter had to be registered with the secretary of state; had to have a telephone; had to have voted in 2008 and had to have voted in either [2006 or 2002], and had to confirm that he/she resided in the district being surveyed. By design, this was not a survey of registered voters (which would have resulted in a younger sample).
Reasonable people can disagree about exactly what percentage of the electorate in 2010 will be age 18 to 34 , and I am not defending any specific turnout target. But most would agree that midterm voters are older. That's what these results show. When SurveyUSA re-weights the respondents in NY-01 to be younger, the survey results do not meaningfully change. This may seem to some counter-intuitive; it is not.
When, for internal analysis, SurveyUSA re-weights the respondents to be younger in AR-02, OH-01, and IN-09, the survey results do not change.
To respond to Leve's reply, I would first of all, with all due respect, point out that merely saying that a sample is "not pretty weird" does not actually make it so.
It is true that midterm voters are older. National exit polling for the last two presidential elections showed that voters between the ages of 18 and 29 made up 18% of the electorate in 2008 and 17% in 2004, but only made up 12% in 2006. Please note that these numbers are of 18 to 29 year-olds, and not the 18-to-34 bracket that SUSA uses in its demographic breakdowns, so the comparison is numerically kind here.
Next, SUSA attempts to address the concern by "re-weighting" their sample for the NY-01 poll, to bring up the 18-to-34 demographic from 1% to 3% of the electorate. The result actually benefits Republican Randy Altschuler, tightening his race against Democrat Tim Bishop to 47-46 from 47-45 in the original sample. While I question whether or not adjusting the sub-sample from 1% to 3% is a meaningful or satisfactory correction, this raises another issue: Just how exactly do you re-weight from a 1% sample? With such a small pool of sampled 18-to-34 year-old voters, aren't we dealing with an astronomical margin of error here?
People who read SSP with any degree of regularity know that we like and respect SurveyUSA, and we're not trying to suggest that anything untoward happened when these polls were drawn up. Leve seems to be confusing "drawing a sample" with "the sample you end up with." It's entirely possible to have a sound methodology that, for whatever reason, winds up with a sample that's not quite right. And that's all we're suggesting here. (Though I would also point out that SUSA's criteria that one has to have voted in 2006 means that there was no one younger than 21 or 22 captured in their polling.)
RaceTracker Wiki: IN-09 |
