We don’t envy political pollsters in a race like the Pittsburgh Mayoral Primary. When an election is likely to be decided by the votes of fewer than 45,000 city residents, even getting 400 people to answer a phone interview is no minor (or inexpensive) feat. Consider, moreover, that other public polls identified as much as 16% of the electorate as “undecided” a week before the election. In a race like this, one campaign driving a couple extra bus-loads of voters to the polls could swing numbers wildly in one direction or another.  It reminds us that focusing our core business on consumer research is much better for our quality of life.
But, just for fun and with no commercial interest in the outcome whatsoever, we’re happy to give it a shot.
HOW WE GOT HERE
We started tracking the mayoral primary going all the way back to November 30th of 2011, when we first tested a list of candidates including Luke Ravenstahl, Bill Peduto, Mike Lamb, and Jack Wagner. That first wave of data showed Ravenstahl as a clear front-runner, with only a few points separating the other three, depending on the assumptions we made about turn-out (more on this later).
Over the next 16 months, we polled over 8,500 city Democrats to see the trend lines on a monthly basis. We also studied head-to-head scenarios between Ravenstahl and each of the challengers. Then, when Ravenstahl dropped out of the race, we even tested a battery of new or hypothetical names like Jim Ferlo, Darlene Harris, Chelsea Wagner, and David Caliguiri (the latter of which polled quite well, in fact) against the likes of Peduto, Wagner, and Lamb.
Now, after all of that, an essentially two-candidate race has emerged between Bill Peduto and Jack Wagner. With all due respect to Jake Wheatley and A.J. Richardson, their numbers have never gained enough momentum to put a dent in the support for the other two.
THE CURRENT FIELD
We’ve been tracking the current slate of candidates since April 1st, having polled a total of 7,493 city Democrats during that period. The movement in those numbers over the past six weeks has been astonishing. In mid-April, the race appeared to be a lock for Wagner, who held a lead as high as double-digits (again depending on the turn-out model). By early May, seeming to correlate with attack ads launched against Peduto by a group allegedly steered by Luke Ravenstahl the numbers began trending sharply in Peduto’s direction. And, in the past week, our numbers have shown a race much closer than reported by the other recent public polls.
And now we have the fun task of trying to predict how the race will end when the polls close at 8pm tonight. To formulate this prediction, we have to not only factor in the responses to our polls but also gauge far more muddled factors like the expected turn-out composition and the final decisions made by a large and lagging group of late Undecideds (who themselves may or may not vote).
TURN-OUT MODELS
The first challenge is to project the demographic distribution of the voters who will actually show up at the polls today. Given some clear differences between Peduto and Wagner’s support among voters of different age groups, in particular, this demographic distribution is a critical factor. One possible approach would be to use a model based on the known turn-out from the last Mayoral Primary in 2009. In that election, over 41% of voters were over the age of 65, a voting bloc that significantly favors Wagner.
We could also look at the 2011 election, when city voters went to the polls primarily focused on a referendum to fund the city library system. In 2011, voters over the age of 65 comprised 35% of participating voters. If this relative change from 2009 is a trend, rather than an anomaly, it would seem to bode well for Peduto, whose numbers improve using a younger turn-out model.
Using only our responses for the last week (746 likely voters) we see a slight difference based on the two turn-out models:
Under either model, our data suggests that the race between Peduto and Wagner is razor thin, well within the margin of error on a sample size of 746. For the purposes of our forecasting, we chose to use a hybrid model, taking the mean between the 2009 and 2011 turn-out figures. Here the numbers get even closer.
GAUGING THE UNDECIDEDS
The key factor, it seems, is the large portion of undecided respondents. Political experts would take this number and divide the Undecideds using a formula that says roughly 80% will vote for the challenger or “insurgent” on Election Day. One could certainly argue that Peduto is the “insurgent” in this race, given Wagner’s long political career in the region. But we would rather root our assumptions in data rather than political wisdom. You need only look to the 2012 Presidential Election, where the incumbent President picked up the majority of late undecided voters.
Our data from past elections (limited as it may be) has led us to a different model, one which may need a lot more data to fine-tune but that has been fairly reliable in our past forecasting efforts. In general, when we analyze the demographic and socio-economic profile of late undecided voters, we find that they tend to vote more frequently in line with other voters who look like they do. In the 2012 election, for example, the majority of late undecided voters were younger, more female, middle-income, and more moderate, looking much more like Barack Obama’s voters than Mitt Romney’s. In the end, these undecided voters broke for President Obama roughly 3 to 1.
So let’s look at the remaining undecided voters in the mayoral primary. According to our data, 59% of the Undecideds are women, who also represent 54.5% of the projected electorate. Women favor Wagner roughly 38% to 35%. Peduto wins among men 43% to 41%. To sum that up, if the Undecided voters vote along gender lines, it suggests a slight edge for Wagner.
Age is a bit trickier.  When we get down to this level of detail, the sample sizes are very small in a given cell. Our turn-out model assumes that 80% of the electorate will be over age 45. This same age group represents 62.4% of the late Undecideds. While Peduto wins among voters under age 44 and those between 55 and 64, Wagner wins among those over age 65 and between 45 and 54. If all Undecideds vote in line with their age cohort, they should likewise split slightly in Wagner’s direction.
Going deeper, we see that Undecideds are more likely than average to be grandparents, to have less than a graduate degree, and to be non-White. All of these characteristics are more common among Wagner supporters than Peduto’s.
AND FINALLY
Given all of this, we have no choice but to trust our data and our models, even though momentum, conventional wisdom around “insurgency,” and lots of other public polls say otherwise. Unless Bill Peduto can drive an unprecedented number of young voters to the polls (voters over age 65 need to represent less than 32% of the total turnout), we’re predicting a close victory by Jack Wagner.
Call it 2 points.