If you’re an electoral politics junkie, you may have already read something about how Iowa could actually produce three different winners (!) on Monday night. What the heck does that actually mean? Let me explain — while also giving you some insight into how our forecast model handles these different versions of the vote count, and which candidates might benefit in one count versus another.
The Iowa caucuses work by voters entering a caucus site such as a high school gymnasium at 7 p.m. and physically forming preference groups — all the Elizabeth Warren supporters stand in one corner of the room, and all the Joe Biden supporters in another, for instance. Voters may also join an uncommitted group.1
This is called the first alignment of voter preferences — and for the first time this year, Iowa Democrats will count and record how voters are aligned at this stage of the process. Thus, the first alignment is the first of the three ways that the caucus vote will be reported. It’s the most straightforward metric of candidate support, and it’s the closest that the Iowa caucuses come to the voting as it will take place in most other states, where voters get just one choice and there are none of the complicated rules that I’ll describe in a moment.2
Iowa’s process doesn’t stop there, however. Voters in candidate groups that fail to meet a viability threshold in the first alignment — typically 15 percent of the vote, but it can be higher in smaller precincts — have one of three choices:
- They can join a viable group.
- They can try to combine with voters in other nonviable groups to achieve viability. For example, if the Warren group initially has 14 percent of the vote, it’s not necessarily dead in the water. Rather, it could recruit a few Tom Steyer voters, or a Michael Bennet supporter, etc., to get to 15 percent. It can’t recruit voters from groups that already achieved viability, however. Once a group is viable, it can continue to add new voters and grow larger, but it can’t lose them.
- Or they could just go home. Nothing forces them to participate in further rounds of caucusing.
The vote count after this stage goes by several names, but I’ll call it the final alignment. It’s the second of the three ways that Iowa will report its vote. And it has a lot of things going for it. Since only candidates who get at least 15 percent of the vote are eligible for delegates to the Democratic National Convention, it makes sense to give voters who initially choose a candidate who can’t get 15 percent a chance to express a preference for a candidate who does. And the realignment process is a highly traditional and distinctive part of the caucuses that the candidates plan around.
‘Don’t reduce (Iowa caucuses) to a 2-way race yet’: Nate Silver
Incidentally, the final alignment is the version of the vote count that our model simulates. When we say that Bernie Sanders’s projected share of the vote runs between 13 and 43 percent, that’s a projection of his final alignment vote.
But the model does make a few simplifications. For instance, because the process of simulating realignment is somewhat computationally intensive, we calculate it for 80 hypothetical caucus sites of various sizes — 20 in each congressional district — rather than the full contingent of roughly 1,700 precincts that will caucus on Monday. And to estimate how widely initial preferences might vary from precinct to precinct, we used the precinct-by-precinct vote from the Republican caucuses in 2016 as our guide.3 Incidentally, you should expect there to be quite a bit of variation in the vote from precinct to precinct, especially given that most precincts are fairly small. If Warren is at 15 percent of the vote districtwide, she’ll have some precincts where she’s at 25 percent of the vote or more and others where she has very few votes. There is not necessarily a bright line at 15 percent exactly.
In simulating the process, we assume that voters make decisions in order of the smallest nonviable group to the largest. (Essentially, this is the equivalent of tabulating votes under ranked choice voting.) If Bennet has 1 percent of the vote for example and is the smallest group in a particular precinct, his voters will be assigned to other candidates first. The process continues until all groups either become viable or have their voters reassigned.4
I’ve skipped over a rather important detail, however. How does the model decide which voters realign to which groups? This is based on a combination of three factors:
- Larger groups are assumed to be more capable of attracting new supporters, other things being equal.
- Whether the group was initially viable matters. If a group was not viable at first, it may be at risk of losing members to other groups at the same time it is trying to recruit its own new members. So the model hedges its bets. In some simulations, supporters of nonviable candidates may be willing to join with other nonviable candidates in an effort to achieve viability. Other times, they’ll resist this.
- Finally, we use a proximity rating, which estimates how far apart the candidates are from one another along ideological and other dimensions. For instance, Sanders supporters are assumed to be somewhat more likely to go to Warren than the other major candidates and vice versa. And Biden voters are assumed to be fairly close to Amy Klobuchar voters and to be more likely to prefer one another as a second choice.
None of this is ideal — there’s a lot of guesswork involved. Second choices for unconventional candidates such as Andrew Yang aren’t easy to capture under this method, for instance. (All of the data that Iowa publishes this year should hopefully make things easier in 2024.) Nonetheless, our method does a pretty good job of replicating polls that show voters’ second choices. Note that our assumptions aren’t terribly aggressive, either. For instance, if a group of Warren voters is trying to decide between Sanders and Biden, the model would have them going about 3:2 to Sanders, but not in a more lopsided ratio than that.
As you can probably infer from the above, this yields some fairly complicated patterns as far as second choices go:
- Sanders isn’t a natural second choice for any of the moderates (Biden, Klobuchar or Pete Buttigieg). But he is a good second choice for Warren voters, and Warren may be under 15 percent in many precincts.
- Biden’s case is somewhat parallel to Sanders’s, except with Klobuchar serving in the role of Warren. If Klobuchar stays under 15 percent in most precincts, Biden is an obvious second choice for her voters. But a Klobuchar surge could hurt him.
- Buttigieg tends to pick up a lot of second-choice support both in polls and in our model, and to draw it from a fairly wide-ranging group of candidates. This is because he’s sort of in the middle of the spectrum: more progressive than Biden and arguably Klobuchar, but less so than Warren and Sanders.
- Warren often polls well as a second choice. However, much of her second-choice support comes from Sanders supporters, and Sanders figures to be above 15 percent in most precincts. So while there’s an upside case for Warren if Sanders underperforms his polls, his recent surge has hurt her. Still, she is also a somewhat logical landing spot for Klobuchar and Buttigieg supporters as all three candidates overperform among college-educated voters.
- Klobuchar, like Buttigieg, could theoretically be a logical second choice for supporters of several other candidates. However, she’ll need to beat her polls to do so, as her current position in our Iowa polling average (10 percent) will have her under the viability threshold in many precincts.
Acknowledging that our model has to make a lot of educated guesses about voter behavior, let’s look at some of its predictions. First, here’s a comparison between the first alignment and the final alignment for each candidate, averaged across 10,000 simulations from a model run late Sunday night:
|Candidate||First Alignment||Final Alignment||Change|
As you can see, the topline effects of realignment are fairly simple. Candidates comfortably above 15 percent tend to gain votes and see their leads grow — for instance, Sanders’s share of the vote grows from 23.6 percent (first alignment) to 28.1 percent (final alignment) in our projections. Candidates near 15 percent tend to tread water. And candidates well under 15 percent can lose the large majority of their vote, perhaps almost all of it, to other candidates.
With that said, these topline effects conceal a range of possible outcomes. For a taste of this, here is a chart showing how Buttigieg’s final alignment is projected to turn out given various levels of first alignment vote:
If Buttgiieg’s first alignment vote is in the single digits or the low teens, then realignment will hurt him. But if he gets around 20 percent of the vote in the first alignment, he’s projected to wind up with an average of somewhere around 24 or 25 percent of the vote in the final alignment. That’s actually a fairly big gain. It’s slightly better than Sanders, for instance, who would be projected to get closer to 23 percent of the final alignment vote with 20 percent of the first alignment vote:
Granted, these are not exactly huge differences — 23 percent as compared with 24 or 25 percent. But an extra percentage point or two added or lost as the result of the realignment process could make the difference in a race where the top four candidates remain fairly closely bunched together. Moreover, our model’s assumptions about who might be helped or hurt by realignment may be too conservative. Having well-trained precinct captains can help a candidate in the realignment process a lot, and there’s also the possibility of strategic alliances between the candidates. Klobuchar and Buttigieg, for example, could agree to try to send their caucusgoers toward one another in precincts where they hadn’t achieved viability. The one thing you probably don’t want to do is enter into an alliance with one of the front-runners.
This isn’t the final stage, though. Instead, the final alignment is translated into something called state delegate equivalents, which is the third way that Iowa counts its vote. Until this year, in fact, this was the only way that Iowa counted its vote. It comes from a now archaic process where the caucuses technically served to elect delegates to county conventions, which in turn elected delegates to district conventions and the state convention, which in turn elected delegates to the Democratic National Convention.
This process is no longer used — delegates to the DNC will be elected based on caucus night results. However, state delegate equivalents are still used to determine essentially how much weight each precinct gets in determining DNC delegates.
The catch is that these weights are based on the number of votes that Hillary Clinton got in the general election in 2016 and that Democratic gubernatorial candidate Fred Hubbell got in the 2018 general election. And this can create some distortions. General election turnout is much higher than caucus turnout; 630,000 Iowans voted for Hubbell in 2018 and 650,000 did for Clinton in the general election in 2016. By comparison, 172,000 Democrats participated in the 2016 caucuses. Additionally, a higher share of caucus turnout tends to come from more liberal, upscale, urban and suburban counties and/or counties with colleges and universities, and proportionately less of it from rural counties.
Essentially, then, rural votes are likely to give candidates more bang for the buck in terms of state delegate equivalents. Candidates whose voters are mostly concentrated in liberal, highly populous counties are liable to underperform, conversely. There are various estimates out there as to who this is liable to help or hurt. But my read on the evidence is that Warren and to a lesser extent Sanders are likely to be hurt by it. Warren’s vote is mostly concentrated in upscale suburbs, and Sanders is hoping for big turnout surges in counties with a lot of young voters. Conversely, Klobuchar and Buttigieg have put a lot of emphasis on covering Iowa’s entire map and could benefit from this process. So could Biden, whose voters should be older, more working-class and more rural. To repeat, none of this is reflected in our model, which stops at the final alignment stage.5
But despite being somewhat convoluted — and do Democrats really want to give more weight to rural areas given the party’s complaints about the non-representativeness of the U.S. Senate and the Electoral College? — state delegate equivalents are likely to get a fair amount of attention on election night. It’s the metric that the various networks and news agencies are used to using to declare winners. And it is technically the measure that is used to determine delegates to the Democratic National Convention, although the importance of Iowa comes from the momentum it produces rather than the number of delegates it has.
As you can probably infer, we personally don’t buy that state delegate equivalents is necessarily a superior measure to the other two. So we intend to refer to all three metrics fairly often in our coverage of the caucuses on Monday night.
We can’t control what the rest of the media does, though, and the media will probably still lean into state delegate equivalents fairly heavily. Knowing how the media covers the race is important to being able to project the bounce that Iowa may be expected to produce for candidates who perform well there.
So in predicting the bounce that each candidate will receive, our model will essentially make a compromise. Bounces are projected based both on the share of the vote each candidate gets, and a binary variable indicating whether he or she won the state. For the vote share part of the calculation in Iowa, the model will use state delegate equivalents. But the winner bonus will be split, if necessary, between the three different measures, reflecting the fact that different candidates may have reasonable competing claims to victory.
In any event, a disputed or ambiguous outcome in Iowa is likely to dampen the bounces that candidates get out of Iowa, especially given that the media’s attention span may be limited by the State of the Union on Tuesday and the conclusion to President Trump’s impeachment trial later this week. So every candidate’s goal in Iowa should be not just to win, but to win clearly enough that they sweep all three metrics.