We might be in for a surprise in South Carolina
Why I think a Biden win is more likely than (some of?) the pundits think
The South Carolina Democratic presidential primary on Saturday will be a crucial test for Joe Biden. It will either be his final disappointing contest—if he doesn’t win there, his campaign is probably over—or the first of a series of victories he’ll need in order to emerge as 2020’s comeback kid. From a non-horse-race perspective, South Carolina is also important because it’s the first state where African American Democrats—the second largest of the party’s various demographic constituencies, and roughly the majority in SC—will have a big say over who they want to carry the party’s banner into the November general election.
It is because the results in South Carolina are so important that the history of handicapping the contest (it ain’t great!) is so concerning. Let’s take the state’s 2016 Democratic primary as our prime example.
Then, polls missed the outcome of the contest. According to the record-keeping at RealClearPolitics.com, eight of the nine polls conducted in the final 10 days before the contest pegged Hillary Clinton’s lead over Bernie Sanders at under 30 points. In reality, she won by 48 percentage points. Their polling average underestimated her actual vote margin by 20 percentage points (yikes).

The only poll to come close was a survey from Clemson University (yes, that Clemson), which found Clinton ahead by 50 points. Here, we might have a chance at an edge in handicapping for this year’s primary.
Clemson has a new poll out in advance of the primary that shows Joe Biden leading his competitors by nearly 20 percentage points. According to the university’s “Palmetto Poll,” Biden has earned the support of 35% of Democratic likely voters in the state. Tom Steyer is on 17%, Bernie Sanders on 13%, and everyone else below 10%.
The results from Clemson’s poll are important for two reasons. First, they are different than both the polling averages and the narrative being presented by political pundits. The average of polls has Biden up by just about 10 percentage points at the time I’m writing this (half his lead in the Clemson poll). The media has been characterizing the race as even closer than that. Take this article from CNN’s Chris Cilizza, for example. In it, he reports almost entirely on a poll from NBC News and Marist College that showed Biden up by just 4 points, and it only mentions other data in passing in the penultimate paragraph.
To be fair, we shouldn’t lean into one poll too much either, but it is worth noting that the Clemson poll has performed better than others (or maybe it just got lucky!) in the past and that it shows a very different race than some pollsters are finding (and that many analysts aren’t acknowledging).
The horse-race findings from the Clemson poll are also important because they suggest a pretty high likelihood that only Joe Biden and Tom Steyer emerge from the South Carolina Primary with the lion’s share of congressional district (CD)-level convention delegates. That would happen if no other candidate reached the 15% threshold in any of South Carolina’s seven congressional districts. To be clear, there’s a lot of CD-level heterogeneity in demographics (which map clearly onto each candidate’s vote share) so most candidates should be viable in at least one district—but only one (or maybe two) will be viable in most, if Clemson’s poll is correct.
Here’s another thing. I suspect that most of the polls of the 2016 South Carolina Democratic primary were wrong because they got the turnout mix in the state incorrect. But if we don’t want to, we don’t actually have to make guesses about likely voters here. Instead, we can model the result in South Carolina by training a statistical model to predict results using precinct-level demographic and political data in states that already voted. We could use that model to create predictions for precincts in South Carolina (given we have demographic data there). Nate Cohn over at the New York Times has done just this. The results, which he tweeted, are pretty interesting:

Well, turns out that the estimates Cohn is presenting here are pretty different from Clemson’s. This could be because Clemson is picking up on some state-level differences in how demographic groups behave, or it could be because their poll is bad. Cohn’s model comes very close to what the average of Palmetto-state polls say today (surprise, surprise!).
So maybe all my groveling about polls and demographics and turnout models will be for naught. Maybe the results in the state will actually be pretty foreseeable after all (IE: by the average of polls and Cohn’s model). But it would be good for pundits to take all this information—rather than just one or two eye-catching polls—into account.
Editor’s note:
Thanks for reading my thoughts on this subject. And thanks for subscribing! Your membership adds up and makes all this newslettering possible (reminder: I do all this work independently). Please consider sharing online or with a friend; the more readers, the merrier!
As always, send me your tips about what you’d like to read about next, or your feedback otherwise. You can reach me via email at elliott@thecrosstab.com or @gelliottmorris on Twitter.
—Elliott