Monday, September 29, 2014

The Real Battle in #MAgov: WBUR vs. the Boston Globe

Look at a graph of Massachusetts governor polls this year and you'd think it's been an extremely volatile race: the last five polls have gone Baker +2, Coakley +10, a tie, Coakley +3, and Coakley +9. Problem is, you'd be wrong. The horse race has actually been remarkably stable; it's the campaign's two premier pollers that have diverged.

Almost all of the polls that have shown a statistically tied race have come from SocialSphere and been sponsored by the Boston Globe. SocialSphere conducts a weekly tracking poll of the race, and its last three results have been Baker +2, Coakley +3, and Baker +1. On the other hand, the MassINC Polling Group runs a weekly tracking poll of its own, sponsored by WBUR, Boston's public-radio station. MassINC's results have consistently given Democratic Attorney General Martha Coakley a solid lead: Coakley +10, Coakley +9, Coakley +5, and Coakley +9 in the same time span as the Globe polls.

So even though the polling average suggests a "Leans Democratic" race, this contest is probably either "Likely Democratic" or an outright "Tossup." But which pollster is right? Let's poke around into each poll to see where their discrepancies might originate. The most recent Globe/SocialSphere poll is here with crosstabs here. The latest WBUR/MassINC polling memo is here, and crosstabs are here.

The differences aren't immediately obvious to find. Both polls use registration-based sampling and called both landlines and cell phones. Both firms conduct live telephone interviews, and the sample sizes were not significantly different (502 likely voters for MassINC, 400 likely voters for SocialSphere). The similar methodologies produced similar samples, too. The MassINC poll breaks down as 36% Democrats, 12% Republicans, and 52% unenrolled. The SocialSphere poll's partisanship is 35% Democratic, 13% Republican, and 52% unenrolled. MassINC's sample was 84% white, and SocialSphere's was 85%.

There were some slight differences. The SocialSphere sample was 55% female, while MassINC's was 52%. And although the age categories did not line up perfectly between the two polls, it looks like SocialSphere surveyed a much older electorate. For MassINC, 38% of respondents were age 44 or younger vs. 29% for SocialSphere. Half of SocialSphere's sample was 55+, while just 29% of MassINC's was 60+. But the biggest difference of all was in level of education. The electorate projected by SocialSphere was 30% postgraduate degrees, while only 16% had some college, but no degree. For MassINC, only 21% had an advanced degree, and a full 29% said the highest education they had reached was some college but no degree.

However, none of these explains the 12-point discrepancy. SocialSphere's bigger pool of female voters should have helped Coakley do better in the Globe poll than the WBUR poll, not worse. (More to blame is the fact that Coakley led women by a whopping 25 points according to MassINC and "just" 15 points according to SocialSphere.) And MassINC found that the candidates were more or less tied among "some college, no degree" voters but that Coakley was way ahead with holders of postgraduate degrees (the SocialSphere poll was too small to provide crosstabs for each specific education category), which, again, should have underestimated her support. Finally, although both polls agreed that Martha Coakley did best with young voters (52% to 28% among 18- to 29-year-olds per MassINC; 57% to 19% among 18- to 34-year-olds per SocialSphere), Coakley won every age group in the MassINC poll, so her lead wasn't just due to millennials. It was more that the polls wildly disagreed on how older voters felt. SocialSphere gave Baker a whopping 51%-to-29% lead among voters ages 55–64.

As you can probably tell by now, the real disagreement between the polls has to do not with their sampling, but with how each candidate is performing with different segments of the electorate—the most confounding and unresolvable way to disagree. Is Baker winning independents 45% to 23%, as SocialSphere says? Or is it basically a tie game, 43% to 38%, as MassINC says? How do women voters genuinely feel—massively pro-Coakley, or does Baker still have a fighting chance with them? And are voters making $150,000 or more really spurning the millionaire Baker to vote for Coakley, 46%–38% (MassINC), or is Baker the one up by 47%–38% among those making $100,000 or more a year (SocialSphere)? We won't truly know until Election Day.

Normally I would defer to the better pollster, but neither firm in this case has really given us a reason to doubt it. MassINC is an extremely reputable firm up in the Bay State, with a long record of excellence and a solid B rating from FiveThirtyEight's newly released Pollster Ratings. SocialSphere, meanwhile, is a newish firm, so it doesn't have a pre-2014 track record to base any judgments off (indeed, FiveThirtyEight doesn't even give it a rating). And if you're looking to outside pollsters to break the tie, you're out of luck; since the beginning of September, two national firms polled the state, and one agreed with MassINC and one agreed with SocialSphere. When it comes to how Massachusetts's key constituencies are feeling in this surprisingly close race, the best any of us non-professionals can do right now is guess.

UPDATE: Maybe we don't have to wait until Election Day after all. A spate of polls Monday all agreed with the Globe results, and then WBUR's tracking poll released on October 1 also moved in line with the SocialSphere numbers.



No comments:

Post a Comment