Polling is all the rage, and a real problem with polls is getting the information you need to determine how realiable it is. DFLers are hooting it up over the new Minnesota Poll (via the StarTribune) which has Patty Wetterling up 48-40 over Michele Bachmann in the Sixth Congressional District Race. Jeff Kouba places the poll in perspective and argues the number for Wetterling is too high. Larry Schumacher notes:

Reporter Eric Black mentions in his story that the respondents to the poll broke down pretty evenly between Republicans, Democrats and Independents. Now, that's not the 6th Congressional District that I know, at least not exactly. Historically the district has broken down about 40 percent Republican, 30 percent Democrat and 30 percent Independent this decade.

Well, through some investigation I have found the tabs that Larry and Jeff could use.

Republican 30%
Democrat 34%
Independent 23%
Other/None/DK/Refused 13%

When they pushed the Independent and other voters to identify leanings, you get 9% more Republican and 16% more Democrat. Now, as my friends at the SCSU Survey will remind me, party ID is meaningless, just a point-in-time identification. But in a district that voted 57-42 for George Bush in 2004, how likely is it that the district in two years moves that much? Ergo, the 33-33-33 split Jeff thought when he studied the race is actually 30-34-36. And when Larry wonders "are Republicans being undersampled or is there a shift going on that is chipping away at the Republican advantage in this district," I think you have to say at least some of it is undersampling. But read on, because I'm saving the best for last.

Some other information from the data I have:

• Bachmann's favorable unfavorable ratings are 44/36, with 9% not knowing the name; Wetterling is at 56/35, with 1% not knowing her. Bachmann's advertising therefore has room to make ground, and Wetterling may be capped on her support.
• Along the same lines, 56% of survey respondents said they had a "great deal" of interest in the race as opposed to 35% who had spent a "fair amount". 8% said they had only a little interest, but somehow are still considered likely voters.
• Comparing demographics of the poll (see the StarTribune's pretty pictures) with the 2000 Census would indicate either many more college graduates vote or respond to polls, or that the sample has too many relative to the district. The poll reported 39% of respondents being either college graduates or having post-graduate education. The Census in 2000 reports less than a quarter of area residents over 25 years old have earned a bachelor's degree or more.
• And lastly, 58% of respondents were female. In all previous polls that Jeff and Larry mention, there has been a female advantage in the poll for Wetterling, so the more females in your sample the more likely you'll find a Wetterling advantage. The Sept. SurveyUSA poll was 49% female, the October SurveyUSA poll was 50% female, and the Majority Watch robopoll was 53% female. Now why hasn't anyone noted this? The only way this can be representative is if you thought that if the election were held today, 58% of voters would be female. Why would you believe that?
  

The Minnesota poll is an outlier. Whether or not they did this intentionally is not the issue -- a poll that was so overweighted with female voters in a race with a notable gender gap should have raised a flag. They fished a pond that had too many of one kind of fish.