{I} t’s common wisdom that Donald Trump will beat Nikki Haley in the South Carolina primary by a hefty margin. There’s no reason yet to doubt that assessment. Public polling nonetheless likely will have a systematic pro-Trump bias that observers should account for.

This bias comes from the inherent difficulty of polling in a party primary that, in the case of South Carolina’s and others’, is not limited to voters registered in that party. Pollsters’ work is only as good as the accuracy of their samples. If the sample to interview with is not representative of the actual electorate, their results will be off.

Pollsters correct for this somewhat in general-election polls by a process called weighting. Simply put, in the modern age people do not respond to pollsters’ queries randomly. This means that any poll sample is likely to have more of one voter type and less of others than the real electorate will have. Pollsters handle this by adjusting the number of responses they include in their estimates by weighting them according to known statistical benchmarks, such as age, education, and gender.

That produces decent results in a general election where these voter shares are largely known. Primaries, however, can be difficult to weight for because there is less reliable data on who votes in party primaries. That’s one reason why one can often find polls off significantly even in high-profile races.

This difficulty is compounded in contests like South Carolina’s because anyone can vote in the Republican race so long as they did not vote in last Saturday’s Democratic race. There’s a chance that people who normally vote in Democratic races or voters who usually don’t cast primary ballots might vote in this year’s GOP slugfest.

This difficulty in determining weights and a sample’s composition could be one reason why the New Hampshire GOP primary polling was significantly off. The final RealClearPolitics average showed Trump ahead by over 19 points, and the polls taken after Ron DeSantis dropped out showed Trump with leads over 20 points. Trump won by only a bit more than eleven. Underestimating the share of the vote cast by people who rarely if ever vote in a GOP primary is likely a reason for this error, given how well Haley did among this cohort.

New Hampshire, however, was somewhat easy to poll compared with South Carolina. The Granite State registers voters in parties, and registered Democrats were barred from voting in the Republican race. This allowed pollsters to exclude any registered Democrats from their samples. South Carolina does not have any partisan registration, which means theoretically no registered voter should be excluded from the sample.

Pollsters likely won’t do that, however, because it dramatically increases the poll’s cost. They would have to contact many voters who will simply reply they are not likely to vote, forcing many more expensive voter contacts to get usable replies. Poll sponsors, often cash-strapped media outlets, won’t pony up the cash for that.

That means pollsters have to make educated guesses up front as to who is likely to vote. Those guesses will almost certainly start with the premise that the likeliest voter is someone who has voted before in a GOP primary. That’s a valid decision, but it inherently leaves out the very voters Haley is trying to entice to cross party lines.

One post–New Hampshire poll of South Carolina clearly makes this assumption. The Monmouth/Washington Post poll explicitly acknowledges the difficulty of assessing the electorate’s composition. It states that its sample includes any South Carolina voter who has voted in any GOP primary since 2016, plus any person who has registered in the state since 2020 and who has not voted in any party’s primary. Based on that sample, they show Trump walloping Haley by 26 points.

There’s no reason to doubt the poll’s findings as far as they go. Monmouth is a respected pollster, and the poll’s cross tabs show the exact breaks that one has found in exit polls and polls in other states. Trump destroys Haley with very conservative voters, beats her soundly with somewhat conservatives, and loses among moderates. Demographic differences are one reason that Trump’s lead is so much greater than his New Hampshire margin: The southern state simply has fewer Haley-friendly college-educated voters and moderates than New Hampshire in the Northeast.

But the sample design itself automatically excludes two groups of voters who will, if they show up, be overwhelmingly for Haley. Traditionally Democratic voters who didn’t vote last week can participate, and there are potentially a lot of them. Over 539,000 people voted in the 2020 Democratic race, but only 131,000 voted this year. That’s a lot of potential GOP voters, many of whom likely were never contacted for the poll.

The other group is even larger: voters who have been enrolled prior to 2016 who have not voted in any primary. It may strike readers as odd, but most voters never participate in presidential primaries. In 2016, for example, South Carolina’s primaries were heavily contested by both parties, drawing a combined 1.11 million votes. That is only roughly half of the 2.1 million votes cast in November. This group would likely tilt toward Haley because it contains fewer intense Republican partisans, a demographic that Trump clearly dominates.

There’s no reason to think these facts mean that Haley is actually ahead, or even that she’s all that close. Even her campaign isn’t predicting victory, merely saying that it wants to do better than the eleven-point New Hampshire defeat. There’s every reason, however, to think that the margin won’t be as large as 26 points if her efforts to get Democrats, independents, and nontraditional primary voters to polls succeeds at all.

Early voting will provide a clue as to whether that is happening. That process begins on Monday, February 12. If early-voting turnout seems to be running higher than previously in areas known to be less Republican, then we can infer that the crossover-voting effort is bearing fruit.

Pollsters generally do a fine job at a difficult task, and we should praise Monmouth for openly acknowledging the sampling challenge it faced. Savvy observers should look at the polls that will be released soon to see how other pollsters address this substantial hurdle. The more polling firms limit their sample to traditional GOP primary voters, the likelier they are to overstate Trump’s already substantial edge.