There are plenty of sites out there that do poll averaging, but this blog has the added bonus of JULYSSA forecasts. For those of you who have not read the about page, JULYSSA is the election forecasting AI I developed with Alice Everett at UC Bakersfield. JULYSSA has won several prizes for statistical forecasting, including MIT's Wilson-Williams Prize last month.
JULYSSA's model is very complex, and forecasts from the primaries through the general election. In the Democratic primaries:
(For more information on the candidates, see the new "Meet the Candidates" feature on the sidebar.)
The conventional wisdom is that Abdullah is well on his way towards winning the nomination, and JULYSSA agrees: her basic, "Polls-Based" forecast gives him a nearly 3-in-5 chance of victory.
This basic model has only two components: horse-race polls, and name recognition figures. We have shown that, all things being equal, a candidate with lower name recognition actually has a better chance of winning a nomination than a more-recognized candidate with the same polling numbers. This can be seen, more or less, with Stephen Dahl and DeShawn Andrews: their polling figures are nearly the same, but Dahl, who is much more famous, has a much lower chance of winning the nomination.
In short, a less well-known candidate who is very well liked by those who know them is more likely to win than a well-known candidate who isn't all that liked.
Now, this basic model is more of a snapshot of the race at present. But the convention is more than a year away, and a lot can change at that time. JULYSSA uses Monte Carlo event simulations to develop a forecast of the entire primary process, accounting for how candidates' polling figures are likely to develop over the rest of the primary--and who is likely to drop out.
(You can ignore ARMSE; it's a technical, statistics thing.)
Again, Abdullah comes out on top; JULYSSA's median case has him getting a large plurality, and the vast majority of the delegates. In the median case, Dahl, Hogan, and Kennedy drop out before the primaries begin, while Ridge, Suarez, Zhang, Wilcox, and Wong drop out very early, receiving few votes.
The most unusual simulation result, in my opinion, is Hancock's. Look at both tables: he goes from 3.5% support now to 6.2% at the end, outpacing Andrews, Fairchild, and Hillman. Indeed, in nearly 8% of the simulations, he comes in as high as third (!) place or higher.
The reason for this can be found in one of the components of the events-based forecast: Polling USA's American Values Survey. JULYSSA processed the survey for likely primary voters, and found that Hancock was actually closest to the values of the activist wing of the party by the Euclidian Distance Criterion. If other candidates favored by that wing drop out early, he consolidates his support by that wing and makes substantial gains. His odds of winning the nomination are still quite low, however.
I haven't done any forecasts for the Republicans yet; as I mentioned in my last post, no polls have included Harper Henry to date. As such, JULYSSA doesn't include her in simulations of the general election, which is why I'm not posting them yet. Hopefully, I can include preliminary results by this time next week.
EDIT: An earlier version of this post just read "Indeed, in nearly 8% of the simulations, he comes in as high as third (!) place."
No comments:
Post a Comment