After projecting one incorrect result on Super Tuesday, in the state of Minnesota, I was able to refine my forecasting models further. There does seem to be some variability in these outcomes that I am currently unable to account for, e.g. if the models predict a win in Minnesota, a win in Iowa also should have happened. Iowa could without a doubt be a special case for our purposes, as it was indeed the first state to hold a Democratic Caucus, both candidates had been campaigning there relentlessly for months, and so on. Therefore, it stands to reason that, looking backwards, perhaps it was Iowa that was the anomaly, and not Minnesota. Sanders outperformed my estimates in every state except Texas, Alabama, and Arkansas; with these states of course having in common the characteristic of being southern and having a larger minority population.

Fortunately, these next four states seem to be firmly in one or the other candidate’s favor. Here are the new projections:

Hillary Clinton will win Louisiana by a significant margin, but the subtle and interesting characteristic of this estimate (as my colleague Matt pointed out to me) is that the estimated margin of victory seems to be smaller than other similar states. For instance, Georgia, Alabama, and South Carolina are almost identical in demographic makeup as Louisiana, yet Clinton won all those states with greater than 70% of the vote. This could signal that Bernie Sanders is becoming increasingly more popular with the minority community.

Bernie Sanders is projected to win the other three states, Kansas, Nebraska, and Maine. Though these states have relatively few delegates up for grabs, this will still be a victory for his campaign insofar that it should create some positive momentum for his campaign after he lost the majority of the Super Tuesday states. Honestly, I expected the estimates for Kansas and Nebraska to signal a more hotly contested race, but the data from the past three days shows that the residents of these states are certainly feeling the Bern.

Special thanks to Andrew, Phil, and Matt for their collaboration and thoughts.

-Tyler