Even more on today’s Star

Even more on today’s Star Tribune Minnesota Poll: The Star Tribune’s Mondale/Coleman poll is obviously the story of the day and bears further examination. One of the mysteries of the the Star Tribune’s Minnesota Poll is the various formulas it employs to adjust raw survey results. I believe they are proprietary, like the formula for Coca-Cola. The formulas are in any event applied invisibly to the raw data to obtain the published poll results. But on occasion the published results themselves have visible problems, and today’s results provide a good example.
Today’s survey results are based on a survey sample that was 51 percent Democratic (16 points over the Republican share), as opposed to a survey sample that was 40 percent Democratic in the last published Wellstone/Coleman Minnesota poll.
Rob Daves is the director of the Minnesota Poll for the Star Tribune. Daves has received an irate e-mail message from the Coleman campaign noting that the Minnesota Poll survey sample for the poll reported in today’s Strib was 51 percent Democratic, versus 40 percent Democratic in the Strib’s last reported Wellstone/Coleman poll. The Coleman campaign has further noted that no Minnesota exit poll has shown Democrats to exceed Republicans by 16 points. We have contacted Daves regarding this issue and Daves responds as follows:
“The [Minnesota Poll survey] sample was 51 percent DFL and those who leaned to the DFL, 36 percent GOP and those who leaned to the GOP. Exit polls normally don’t break out leaners separately from independents, as we do…
“If you treat the leaners in our poll as independents, the way the VNS exit poll does, then our sample was 36 percent DFL, 26 percent GOP, and 38 percent Republican[sic–this should read “Independents”]. But remember that exit polls have a pretty big margin of sampling error, usually bigger than our poll, so their numbers are not exactly a gold standard. [Daves referred to and attached a table we are omitting.]
“As you’ll notice, the percentage of strong partisans is remarkably stable between the two polls, one done over a 5-night period with call backs, refusal conversions and all of the standard techniques we use to get a good sample…The differences come in the fluidity of the leaners, who typically bounce in and out of weak partisanship anyway. And our percentage of strong DFLers in the Oct. 28 poll is EXACTLY what the VNS percentage was in 2000 — hardly a Democratic bias.
“The 2000 VNS exit poll showed a 5-point advantage for DFLers over Republicans. Our Oct. 11-16 poll showed 5-point advantage for DFLers over Republicans, when comparable measures are used. Our Oct. 28 poll showed a 10-point advantage, but not because there were too many DFLers (that percentage was EXACTLY what the 2000 VNS percentage was).
“Moreover, a look at our unweighted demographics for the two polls shows a remarkable similarity in the proportions for education, ideology (liberal/moderate/conservative), and geography. The overnight sample slightly undersampled men and 35-44 year olds, compared with the Oct. 11-16 sample, but we weight on those variables to bring them into Census proportions anyway, just as I’m sure Glen does. If you just look at unweighted ideology, the proportions are virtually identical in the two polls.
“Sorry, but I don’t buy [the Coleman campaign’s] criticism of the poll’s sample.”


Books to read from Power Line