Ayanna Pressley's insurgent victory over 10-term incumbent Rep. Michael Capuano in the 2018 Democratic primary is an example of the big shifts in voter preference and turnout that can lead polls to miss the mark. (Photo by Michael Jonas)

IT’S PRIMARY SEASON across the country, bringing a close to months of intraparty wrangling – and uncorking the predictable seasonal scolding directed at pollsters who released surveys in the run-up to the elections that widely missed the mark.

In the last two weeks, Democratic primaries in Michigan and Wisconsin featured outcomes that were very different from the margins shown in late primary polling. In the primary for US Senate in Michigan, progressive Abdul El-Sayed eked out a 1-point win over the more moderate Rep. Haley Stevens after the final two polls in the race showed him with a double-digit lead. Meanwhile, on Tuesday, in the Democratic primary for governor in Wisconsin, Francesca Hong, a democratic socialist member of the state legislature, was narrowly defeated by the Milwaukee County executive, David Crowley. While there was little polling in the final stretch, what there was seemed to show a comfortable victory for Hong.

Here in Massachusetts, so far there is relatively little in the way of public polling or even leaked internal poll results in the run-up to the September 1 primary, Just one public poll has been released in the Democratic Senate primary between incumbent Ed Markey and Rep. Seth Moulton over the last three months, same as the GOP gubernatorial contest. There were a few earlier in the cycle but nothing recent, even as primary day rolls closer. Polling in the US House districts with competitive primaries is even more sparse. But as we enter the final weeks, we may see a bit more, which makes it worth pausing to reflect on what they actually tell us.

First and foremost, primary polls are very imprecise, missing the final margin by an average of 13 points according to a recent analysis by Strength in Numbers. Primary polling challenges stem from a fairly durable set of reasons that have little to do with the newer methodological challenges of election polling.

There are no partisan or ideology cues for voters (or pollsters). In general elections, an enormous amount of the potential variation in outcomes is accounted for simply by the fact that some people prefer voting for Democrats and some prefer voting for Republicans. They will vote that way in the vast majority of cases. These preferences are fairly easy to address in poll results, through the process of weighting the final poll results by party. In primaries where all or most voters are members of one party and clustered in one part of the political spectrum, sorting voters into the correct ideological distribution is much less precise.

In primaries, there are more candidates in voters’ preferred ideological space. With two or more candidates in a given party primary, many voters will be comfortable with any of them. This can be mysterious to close political observers who see stark differences between the candidates and can’t imagine drifting from one to another. But the fact is that voters can and do move from one candidate to another in very large numbers, since they may see several as potentially appealing. Think about any recent presidential primary and you’ll see examples of this, with candidates growing and fading in succession as the primary continues.

There is more likely to be last-minute upheaval, which can also have bigger impacts. In general elections, there are relatively few huge shifts in the field at the last possible minute. Even replacing Joe Biden as the Democratic presidential nominee in 2024 happened months before Election Day. Primaries, however, are notoriously volatile. In the Wisconsin primary, former lieutenant governor Mandela Barnes, who was polling in second place, dropped out of the race with less than two weeks to go. Meanwhile, David Crowley’s victory was remarkable not only because of the huge polling lead Francesca Hong enjoyed just weeks earlier, but because Crowley had actually dropped out of the race and only reentered it on July 18. When it comes to primary upheaval, perhaps the biggest national example in recent years is Joe Biden’s primary consolidation in 2020 after his New Hampshire debacle in 2020, when support shifted to him at lightning speed.

Voters tune in late and change their minds late. Even as political junkies are riveted by primaries for months ahead of time, primary polls often show huge swaths of voters undecided even days before voting. In both of the Wisconsin polls conducted after Barnes dropped out, Crowley showed a huge jump in support, which was still growing even in the final days. In one of the late polls, a full 34 percent of likely primary voters were still undecided or did not select a major candidate. That was an internal poll from the Crowley campaign, likely distributed to emphasize Crowley’s huge jump in support and potential path to victory if the undecided vote broke his way.

Overall turnout varies enormously in primaries. Even in a stable environment without last-minute surprises, we often see huge swings in turnout in party primaries. A good example of turnout affecting margins is the 2014 Massachusetts gubernatorial primary. Nearly every poll, including ours, vastly overstated Martha Coakley’s margin of victory. In our case, it was due to greatly overestimating turnout. High-engagement voters who reliably cast ballots no matter what were more likely to support her main opponent, Steve Grossman, leading to his stronger-than-predicted performance in what proved to be a low turnout affair. Because Coakley nevertheless won, the pillorying we suffered was less severe than what takes place when polls peg the wrong winner, though the polling error was just as large.

Estimating turnout demographics in an atypical primary is very imprecise. In the 2018 Democratic primary in Massachusetts for the US House between 10-term incumbent Michael Capuano and Ayanna Pressley, Pressley drew support from a flood of younger, atypical primary voters in her insurgent victory. These voters only became likely voters over the summer, in the weeks before the primary, hearing about an exciting new candidate and deciding to cast a ballot. The one public poll done in Wisconsin after Barnes dropped out suffered from the opposite problem: too many young people and would-be Hong supporters ended up in the final sample compared with the demographic profile of the Democratic electorate that actually voted last week. If vote choice is closely linked to one of the demographics the poll gets wrong, it can have a very significant impact on the poll’s accuracy.

Here in Massachusetts, the Democratic primary for Senate, the Republican matchup for governor, and a handful of US House primaries are the major races coming in just over two weeks. With little or no recent polling and a very unsettled electorate, there are few tea leaves to even try to read. The best thing to do is just go vote and then watch election returns rather than look to prediction markets, since nobody actually knows anything.

As races from other places and times have shown us, when it comes to primaries, anything can happen.

Steve Koczela is president of the MassINC Polling Group.

Steve Koczela is the President of The MassINC Polling Group, where he has grown the organization from its infancy to a nationally known and respected polling provider. During the 2014 election cycle, MPG conducted...