Two notes about the final round of surveys now available for Michigan.
First, on the Democratic side, please note that we decided to stop updating our chart, since the last round of polls reports a very different trial-heat question than Michigan pollsters had been using previously. The earlier polls asked questions that included all of the Democratic candidates. However, the names of Barack Obama and John Edwards will not appear on the ballot that Democrats will use in Michigan on Tuesday, though voters will have the option to support an "uncommitted" slate of candidates. As such, our chart page for the Democratic contest in Michigan reports the most recent polls in a table at the top of the page, but does not include the new results in the chart.
Second, on the Republican side, we have something of a puzzle. Most have the Romney-McCain margin well within the margin of sampling error, although one (MClatchy/MSNBC) gives Romney a statistically significant eight-point lead while another (ARG) shows McCain with a significant seven-point lead. Needless to say, both cannot be right. Given the conflicting results, and the 39% that say they "could still change their minds" on the McClatchy/MSNBC poll, the outcome is still clearly in doubt.
I agree with many of our commenters that differences in the likely voter "models" -- and particularly the degree to which they include independent leaning voters -- probably explains much of the variation. MIchigan's primary is open. Eight years ago, when there was no contest on the Democratic side and John McCain defeated George W. Bush among an electorate swelled with independents and Democrats. The network exit poll showed 48% identifying as Republicans, 35% as independents and 17% as Democrats (Michigan has no party registration).
Here is what we know: According to the Zogby release, "Republicans made up about 51% of the sample, Democrats 22% and independents 27%." However, Republicans made up 70% of the respondents on the McClatchy/MSNBC/Mason-Dixon sample (and Democrats only 5%), 73% of respondents on the ARG survey, and 75% of the respondents on the automated Mitchell Interactive survey. The Rasmussen story does not report the party results, but the Republican percentage appears to be approximately 70%, given their crosstabulations showing Romney winning twice the support among Republicans (31%) as he gets from independents (16%). The Free Press story on the Free Press/Local4/Selzer survey says only that "relatively few Democratic voters plan to vote in the Republican primary," although it confirms the pattern present in the other polls:
Independent voters in the Michigan Poll favored McCain over Romney by 5 percentage points. But Romney holds an 11-point edge among self-identified Republicans.
Age may be another factor. Both the Free Press and Rasmussen surveys show Romney doing far better among older voters. The Free Press shows Romney winning the support of 33% of Republicans over 55 years of age but only 9% of those under 35. On the other hand, the McClatchy/MSNBC/Mason-Dixon survey -- the only one to report age composition data (45% of their respondents were 18-50) -- shows only a slight difference in Romney's support between those over (32%) and under (28%) 50 years of age.
When the results are in, we will learn yet another lesson on the perils of primary polling.