Wendy Davis Gets Good News in SD 10

We've been tracking the campaign of Wendy Davis for state senator in District 10 pretty closely here at the Blue. Some recent polling data — some reliable, some not so much, but all of it good news — has come out that should have incumbent Senator Kim Brimer sweating over his seat.

We'll start with the polling numbers released by the Davis campaign just yesterday. They ran a poll through the national firm BPN that had Brimer at 39% to Davis' 35% with a 4.9% margin of error. The Brimer campaign can't be happy that an incumbent Senator is already polling at a statistical tie with his opponent nearly half a year prior to the election — particularly when his numbers have nowhere to go but down the more that word gets out about his "rent-to-own" scheme (PDF). Further, 59% of Brimer's voting constituents hadn't heard of him, and only 25% thought he deserved to be reelected.

The release numbers echo the polling done by the Lone Star Project late last year that had his re-elect numbers at 27.4% and his name ID at 50.3% — and that was after being identified in that poll as "state Senator Kim Brimer."

The other set of good news unfortunately takes a little more digging. That's because it's from a poll by Texas Poll Watch (subscription required), a very new Texas-based polling outfit that's been taking some licks for below-par polling. The poll they released yesterday was, sadly, no different.

The poll had a sample size of 466, for a margin of error of 4.54%, and showed Kim Brimer at 44.27% and Wendy Davis at 35.35% — barely within the margin of error. Unfortunately, the crosstabs show some serious sampling error in the raw data. In a district that is nearly one quarter Hispanic, the calling pool was just over 9% Hispanic, and the sample itself was barely over 5% Hispanic — that's 24 Hispanics in a sample of 466 respondents. Female respondents also outnumbered male respondents by a factor of nearly 2 to 1.

Texas Poll Watch tried to compensate for this by weighing the responses back so they'd be more representative of their original, more balanced proportions in their calling pool (52.8% female to 47.2% male, 9.15% Hispanic). And that's where they hit their statistical snag.

Because male Hispanics were the least represented, they got by far the highest scaling factor at nearly 300% — each of their responses effectively counted for three. And 50% more Hispanic males surveyed voted for Brimer than for Davis — because Brimer got three votes, and Davis two. The total size of the male Hispanic sample was 7. You can imagine the sky-high margin of error on a 7-person sample. And the huge scaling factor multiplies that error dramatically going into their final tally. Non-Hispanic males, which also broke in favor of Brimer, got the second biggest bump at about 30%. Hispanic females (sample size of 17 here) got a 27% bump, and non-Hispanic females got docked about 25%. As you can imagine, the latter favored Davis.

A much more accurate result would have come from combining the Hispanic and non-Hispanic polling numbers and scaling only based on that criterion, cutting the margin of error we got with the male Hispanics by over half while making the scaling factor smaller. Unsurprisingly, when that is done the numbers change dramatically — and interestingly, they mirror the campaign's numbers much more closely, with Brimer at 42.7% and Davis at 37.7%.* This is also nearly identical to the result Texas Poll Watch would've gotten if they didn't scale the numbers at all — the poll without "massaged" numbers resulted in Brimer getting 43.1% to Davis' 37.1%. Both sets of numbers, of course, are well within the margin of error and point to a very competitive race in the district.

So it takes a little digging to get an accurate read from Texas Poll Watch's numbers — but when you do, it is clear from both their polling and the campaign's that Kim Brimer is an incumbent basically unknown by his constituents and unpopular with the ones that do know him, polling at a statistical dead heat with his challenger. Add to that a scandal that may well chase him all the way down to the general election, and you have a recipe for a Democratic flip brewing in SD 10.

* It should be noted that while combining the Hispanic polling numbers, as I did above, does drop the margin of error dramatically — but at 24 people within a sample size of 466, it remains quite high. That may be the "best attempt" at correcting for the huge sampling error, but there is only so much that can be done with so small a sampling of Hispanics. There is no substitute for an appropriately large sample set, and it is legitimate to argue that the corrected numbers may well be less accurate than the raw polling. As they say, garbage in, garbage out.