Our model takes into account the state-specific delegate selection rules and the state-specific poll averages (if they’re available). In roughly half of the plausible scenarios we run, Donald Trump comes up around 100 delegates short of the 1,237 he needs to win on the first ballot. In real life, this probably requires more strategy than #NeverTrump actors have heretofore realized, but makes Cook Political/FiveThirtyEight’s “two-front war” look like what the electorate is already planning.
The chart above is a reasonably likely outcome: Trump fails to exceed his poll averages and John Kasich wins congressional districts in which the House member won by over 10 percentage points in 2014.
Above is the math behind that projection. A couple of notes: Colorado is just a guess for delegate outcomes, the state will be a black box. The model has Trump winning 100% of the Congressional districts, in New York’s district winner-take-all contests based on the strength of his state-wide polling (he’s a good 35 points ahead of Kasich), but it’s hard to imagine Ted Cruz and Kasich failing to pick up. That’s probably balanced by Katich’s stronger showing in California, where the model shows him picking up 3 delegates in 30 of the 53 House districts.
Closer examination shows a big gap in Pennsylvania’s delegates. That’s because voters select unbound delegates on the district basis. West Virginia follows a somewhat similar model, where delegates may or may not declare their intention on the ballot. There are already 75 unbound delegates heading to Cleveland, and another 174 bound to candidates who have “suspended” their campaigns. That’s a big wildcard: enough to push Trump over the edge if a majority break for him.
This outcome actually requires Kasich to stick around and remain viable through June. Kasich can’t win the 1,237 needed on the first ballot; there just aren’t enough points left on the board, so a vote for Kasich is just a “none of the above” vote. Sure, he beat Trump in Utah, but he lost to Rubio – who had dropped out a week earlier – in Arizona due to the prevalence of early voting. It isn’t too big of a stretch to see the “mathematically unviable” argument swaying enough voters in Pennsylvania, New York, and California to make his spoiler status not that big of a deal.
It’s also not a sure thing that this is the direction late-deciders break. The models above show Trump picking up 0% of undecided voters. If Cruz picks up 60% and Trump picks up just 20%, the race looks a lot different:
In that projection, Trump falls just 3 delegates short, and it’s easy to imagine him picking those up among the unbound delegates.
If, based on that slightly weaker projection for Kasich, he loses more of the Democrat-heavy congressional districts in California, with Trump and Cruz picking up a proportional share, Trump wins outright:
None of this takes into account the political realities that govern the later elections and a brokered convention itself. As Trump has picked up some endorsements (and Cruz only some half-hearted ones), GOP leaders could fear the “riots in the streets” outcome, or fear that by suppressing Trump the party could permanently lose the voters he’s won. It also assumes that almost no delegates jump on the Trump bandwagon in subsequent ballots. Nevertheless, it shows there’s still a reasonable set of outcomes in which Trump fails to secure 1,237 delegates before Cleveland, if the GOP plays its cards right.