The third straight September poll shows Minnesota’s anti-gay-marriage amendment leading — but for the second straight time, with a constitutionally insufficient plurality.

This time, the Star Tribune’s Minnesota Poll shows 49 percent yes, 47 no, and 4 undecided. The 2-point gap mirrors Public Policy Polling’s 1-point margin a week earlier, leaving SurveyUSA’s 7-point Sept. 6-9 margin the outlier for now.

Conservatives regularly flay the Strib poll’s alleged pro-Democrat bias, and the partisan breakdown here — +13D compared to SUSA’s +7 and PPP’s +5 — won’t mollify the right.

One complication in the conspiracy theory: the paper dumped its 2008-11 pollster, Princeton Survey Research Associates, in favor of Mason-Dixon, which formerly polled for the Pioneer Press and MPR. (The PiPress later dropped polling entirely; the latter saddled up, questionably, with Larry Jacobs.)

Strib deputy digital editor Dennis McGrath says cost, not bias charges, dictated the switch. Still, back in the day, GOP partisans touted Mason-Dixon’s numbers as a counterpoint to the Strib’s; M-D managing director Brad Coker says his poll had a slight Republican “house effect” in the 2008 presidential  year.

Mason-Dixon’s Minnesota poll split — 41 percent Democrats, 28 percent Republican, 31 percent independent — should give Vote No partisans pause. That includes those fighting the lesser-noticed Photo ID amendment, who received the shocking news that they’re losing only 52-44 when SUSA & PPP had them 2-to-1 behind.

Why? The Strib’s survey show anti-gay-marriage leads 55-38 and pro-Photo-ID 61-35 among independents. If the Strib and Mason-Dixon have too many Dems, Vote No may get crushed twice.

Coker, who founded Mason-Dixon 29 years ago, is actually pretty refreshing on polling’s uncertainty.

Mason-Dixon uses all live operators, making it easier to poll cell-phone-only voters. Still, Coker (whose survey was 80/20 landline/cell), cheerfully admits, “Any pollster who tells you they know how many cell phones to have in their poll, they’re B.S-ing you.”

He’s not worried about the Minnesota Poll’s partisan split. “The number fluctuates,” he says. “I’ve polled Minnesota a long time, and over the years, it ends up 40-30-30 (D-R-I). If I’m in the ballpark, I’m not too worried. If I had 50 percent DFL, I might go back and check what I’m doing.”

OK, but it’s worth noting a Mason-Dixon poll just six months ago — on the Vikings stadium* — found a mere 33 percent Democrats, 26 percent Republicans and 41 percent independents. The +6D gap is more in line with the other recent polls.

However, Coker contends that closer to Election Day, voters tend to sort themselves into a party. Also, the March poll measured registered Minnesotans; Sunday’s only tallied “likely voters” who say they are definitely or probably voting in November, the two voting-ist responses on a 6-point scale.

“We’ll use a tighter screen the closer we get to Election Day,” Coker adds. “When early voting starts, we’ll ask people if they’ve already voted, and when we get to the last week, we’ll ask people, ‘Do you intend to show up on Election Day and same-day register?’”

Right now, that last group is not represented in the newest Minnesota Poll: Coker says his operators are working off third-party data-mining lists that link previous and newly registered voters to phone numbers.

Therefore, if you haven’t voted before, or haven’t yet registered, Mason-Dixon and the Strib don’t yet know you exist. Given that these voters tend to be young, and tend to be liberal, one could argue the poll is missing an anti-amendment group.

Unlike some pollsters, Mason-Dixon doesn’t weight results. Some pollsters will actually count a response as more or less than “1,” to achieve, say, Census balance for age or income. Instead, Mason-Dixon operators stop accepting responses if either gender goes outside a narrow 50-50 range, and age, based on loose, historic voting norms.

Coker adds that Mason-Dixon doesn’t offer a “not voting” response to the amendment, even though voters who pass on the amendment are counted as no’s. His 49-47 Yes result also had 4 percent undecided; based on the many other polls he’s done on gay marriage, he predicts most will break for the amendment.

So there are your methodological notes. What can we conclude now that we have three marriage polls?

First, to belabor the obvious, Vote Yes is ahead. Not by much in two of three cases, not enough by the Constitution’s “50 percent of all voters voting in the election” standard (non-votes on the amendment = no), but still.

Second, Republicans really love the marriage amendment. Each pollster has GOP support around 80 percent and opposition at 20.

Third, every pollster has voters ages 35-65 split roughly 50-50.

Fourth, voters over 65 support the amendment by roughly 15 percentage points, according to all three polls.

Where’s there’s disagreement:

The youngest voters. The most pro-amendment poll, SUSA’s, has a “huh, what?” as big as Mason-Dixon’s partisan split. SUSA has 18-34-year-olds supporting the amendment 50-39. The Strib has the same group 57-37 against. PPP has a slightly younger 18-29 cohort 44-50 against.

Independents. It’s basically a continuum — SUSA shows a near-even +1 Yes, PPP has it +9 and the Strib +17.

Gender gap. The three polls show men supporting the amendment more than women, but that’s where it ends. SUSA has the gap +3, PPP +25 and the Strib +34.

Caveats:

Margin of sampling error, usually 3-4 percentage points plus or minus for the overall result — is higher for subgroups like age, partisanship and gender. Therefore, minor disagreements shouldn’t be overemphasized. What’s minor? That’s where human judgment comes in.

* — This story originally said Mason-Dixon’s Vikings poll was for KSTP; it was done for pro-stadium group Home Field Advantage.