As of Friday, you’ll see a shiny new face in our Democratic primary forecast: former New York Mayor Michael Bloomberg. We’re now featuring Bloomberg more prominently in our forecast interactive and in our polling averages; he joins four other candidates (Sen. Bernie Sanders; former Vice President Joe Biden; former South Bend, Indiana, Mayor Pete Buttigieg; and Sen. Elizabeth Warren) who get their own color in the forecast (in Bloomberg’s case, gold). Bloomberg has always been in the underlying calculations and the detailed output behind the model,1 but he was lumped in with “all others” on many of the charts, making him hard to find.
Bloomberg is a tricky candidate to forecast, given that his strategy of essentially skipping the first four states but then spending enormous amounts of money on the race is fairly unprecedented. Although Bloomberg is at only 11 percent in national polls right now — below the 15 percent threshold required to pick up delegates in states and congressional districts — he’s getting close enough to the threshold that the model actually has him picking up a decent number of delegates in its average simulation.
On the other hand, the model thinks it’s quite unlikely that Bloomberg can get a majority of delegates because he’s getting off to a late start. It’s not that skipping out on Iowa, New Hampshire, Nevada and South Carolina is itself all that costly; those states have relatively few delegates. Rather, it’s that Bloomberg is unlikely to have a huge surge before Super Tuesday.
Here’s why that matters. Bloomberg could certainly do reasonably well on Super Tuesday and get a surge in later states. But at that point, 38 percent of delegates will already have been chosen. Say Bloomberg wins 30 percent of the delegates on Super Tuesday; that would certainly get him some attention, probably make him a real contender, and perhaps knock other moderate candidates out of the race. Bloomberg, however, would need to get 64 percent of the delegates in all the states beyond Super Tuesday to earn a majority of pledged delegates, which is an awfully high bar to clear.
Bloomberg getting a plurality of pledged delegates, on the other hand, is more likely. (There’s a 1 in 40 chance of that, or about 3 percent, according to our model — as compared to a roughly 1 in 100 chance he gets a majority.) More likely still is that Bloomberg appears to be the strongest candidate at the end of the process, even though he doesn’t necessarily have a plurality. There’s a 5 percent chance that Bloomberg will be leading in national polls at the end of the race, our model estimates.2 Being able to point to indicators like that could be helpful to Bloomberg in the not-at-all-unlikely event of a contested convention.
These distinctions matter because – it seems like I can’t emphasize this enough, as I see people misquoting our forecast all the time — we are not actually forecasting the identity of the nominee. Rather, we are forecasting the chance of each candidate getting a majority of pledged delegates (or a plurality) after the Virgin Islands casts the final votes of the primary season on June 6. Bloomberg could easily become the nominee at a contested convention — and a contested convention is a reasonably likely possibility — but our model does not try to predict how a contested convention would turn out.
We’ve also made two subtle changes that should slightly help the model’s handling of Bloomberg, although they make little difference to the top line forecast.3 First, in the state-by-state regression analysis that we conduct to help forecast states with little polling, we are no longer using Bloomberg’s Iowa results as an input for him. In the regressions, the model doesn’t use a candidate’s performance from states in which they weren’t on the ballot. Iowa technically didn’t have a ballot, however. (Indeed, Bloomberg won a very small number of votes there.) But since he never set foot in the state after launching his campaign (his last visit there was in December 2018) nor made any other effort to compete there, the regression will ignore his performance in Iowa.
The other small change is in how we calculate what we call the “fundamentals,” which are a combination of indicators based on a candidate’s fundraising, endorsements and level of experience in elected office. (See Step 3 in our methodology guide for more about this.) We find that candidates who are strong in these areas tend to see their polling improve on average, and candidates who are weak tend to see their polling get worse. However, their effects are quite subtle and plenty of candidates defy the trends (see also: President Trump). The model now randomizes how much weight it puts on these categories in each simulation, instead of always treating them as equally important. Likewise, it randomizes the amount of weight it puts on the three categories of fundraising we track (small-donor contributions, all individual contributions, and all contributions from any source including self-funding). Thus, in some simulations, the model treats Bloomberg’s enormous spending as a relatively important factor in the race, and in other simulations, it gives it very little weight. This reflects the fact that the evidence is quite mixed on how self-financed candidates do as compared with candidates who raise money from individual donors.
There’s certainly a lot to think about here, so I may do some longer stories about Bloomberg after New Hampshire. But we’ll leave it there for now. Suffice it to say that I think there’s sometimes a lack of rigor when analyzing Bloomberg’s chances. The more you try to run through specific, realistic scenarios for exactly how Bloomberg wins the race — which is what our model is doing in trying to simulate all the possibilities — the harder you’ll find it is for him to get a majority of pledged delegates. It isn’t necessarily so hard to conceive of him accumulating a lot of delegates and winning the race via a delegate plurality or at a contested convention, however, and the model is more agnostic about those possibilities.