Prognostication and Prophecy for Mini Tuesday

A number of people have written to ask me about my predictions for today. I was going to sit this one out (for very good reasons), but I will cave to peer pressure and at least tell everyone why my best guesses are little more than casting runes.

The polls are so wrong, so often, and getting worse. Plus, no one is bothering to poll the nation, so I have no 'norm' signal to work from. Add in republican heads exploding the number of open primaries, well...

Here are the five approaches I have been looking at, including how well they have done this year. (Note: when multiple models explode on the launch pad, you have a phenomenon that is technically referred to as @%$#!)

Approach 1: Model the polls as noisy observations, each with their own bias and variance. 

Accuracy to date: Piss poor.

Computer output for Mar 15: You're lying to me. There's no way that all of those polls are coming from those pollsters, or from those races.

Approach 2: Model the whole nation using national and regional polls to treat each state as a local deviation from a norm. 

Accuracy to date: Roughly accurate, but highly uncertain.

Computer output for Mar 15: [An error 0xc0000100F has occurred transferring execution. If the problem persists, please contact your election administrator.]

Approach 3: Pick the largest outlier polls, and use momentum to decide whether you use the largest or smallest spread for the apparent leader.

Accuracy to date: Fairly decent. Which is kind of scary.

Computer output for Mar 15:

   Florida: Trump >25 (or Trump <6)

   Illinois: Trump <4 (or Trump >13)

   Missouri: No data available

   North Carolina: Trump <6 (or Trump >20)

   Ohio: Kasich >6 (or Trump >6)

Approach 4: Determine historical correlations between state results, and use exit polls and actual results from states that have already voted to predict.

Accuracy to date: Fairly good for Super Tuesday. Not a lot of data for previous states.

Computer output for Mar 15:

   Florida: No one votes like Florida. Not even a little bit.

   Illinois: None of the good correlating states have voted yet

   Missouri: Trump or Cruz by <10 (Based on states like MN, OK, TN, KS)

   North Carolina: Trump +8-14 (Difficult, since it normally votes so late, but it reflects nearby southern states)

   Ohio: They're weird. Anyone they kind of resemble hasn't voted yet. Maybe Kasich +2 (based on a bunch of weak correlations on the county level)

Approach 5: Use a Tarot deck

Accuracy to date: No worse than celebrity "statisticians" and pundits

Computer output for Mar 15:
   Florida: Justice [Justice in your future indicates that you will examine your past, and how the actions of your past have brought you to your present. In doing this, you will realize how what you do right now will influence who you become in the future.]   Illinois: Seven of Cups [Flights of fancy and daydreaming are indicated by the Seven of Cups. In the picture we see the many visions of happiness that the character sees in the cups, but they are all unrealistic goals at this time.]   Missouri: The Page of Swords [Symbolizing changing circumstances, unpredictable behaviors, and wavering opinions at the root of the situation. Flexibility and adaptability are key.]

   North Carolina: The Fool [The Fool indicates that the root of your issue lies in a time when you needed to look at your situation with optimism. You had to make a leap of faith, taking only the baggage that benefited you, with the ultimate trust that you would make it alright. You went beyond your boundaries, and worked past your fears. From this experience, you learned to believe that anything is possible.]

   Ohio: The Star [The Star in represents your hopes and dreams. This may indicate the moment that you realized what you wanted to become, or it may remind you of dreams you used to believe could be your future someday.]

Super Tuesday Results

No Surprises on Super Tuesday
The last of the Super Tuesday results are trickling in now, and the net results are about what we expected. Many pundits and campaigns are generating much sound and fury over who "won" the different states. However, I am a pragmatist and the only winner to me is someone who gets closer to winning their party's nomination. With that in mind, it doesn't matter if Trump got 32.7% to Kasich's 30.4% of the vote in Vermont when they both get 6 delegates. With our eyes on the prize, let's see how the candidates on each side did:

While we could quibble over the particulars, this is basically what we were expecting. Although there are several delegates still not decided until all of the results come in, Clinton took approximately 60% of the available delegates, which is exactly what we predicted. The Republicans split the delegates three ways with Trump at 42%, Cruz at 38%, and Rubio at 16%.  Our predictions were Trump at 43%, Cruz 29%, Rubio 25%. While Cruz stole a lot of the delegates we expected to go for Rubio, it was almost completely due to Rubio narrowly missing the threshold in Texas, and being denied all of the 46 delegates (8% of the overall total) that we were projecting there. We talked about that risk, and honestly it was a coin toss whether he was going to make it or not. From the perspective of a data geek and number cruncher, yesterday's vote merely fulfilled expectations.

From here, we can expect Clinton to maintain course, and head into the convention with about 2/3 of the Democratic delegates. Trump is likely to capture 45%-50% of the delegates, depending on how things go in some of the bigger winner-take-all states, like Ohio and Florida. Exactly where he falls along that continuum will decide whether there is a contested convention. If that happens, we are out of the world of statistics and into the realm of psychology/sociology, which makes my ability to predict the outcome somewhat akin to a magic eight ball.

The Polls Still Suck
The total results for Tuesday may have hit the mark, but the individual polls continue to be pretty poor.

• In Texas, twelve polls were taken in the past two weeks, predicting a Cruz win by anywhere from 0-15%, with an average prediction of 8%. He ended up winning by 17%.
• Three polls in Alabama had Trump ahead by 13, 17, or 23%. He ended up winning by 23%.
• In Virginia, the polls claimed Trump leading by 13, 14, or 23%. His final result was just +3%
• Polling the Democrats in Texas looks better at an average of +26% for Clinton versus an actual result of 33%, until you realize the polls were fairly uniformly distributed over 10-42%. An estimate at +/- 16% is not a prediction, it's a guess.
• Georgia was predicted at 28-39% for Clinton, who ended up getting +43%
• Alabama was polled twice in the last two weeks, at +28% and +48% for Clinton. Final result: +60%

Across the board, the polls were awful at doing their job, with individual errors well beyond double the margin of error. In fact, in most cases your best bet is to take one the most extreme outlier poll you can find, and go with that result. 

So how can we still make any sense of this data at all, much less make meaningful predictions like we saw yesterday? Solid modeling from the bottom up. We learn the correlations, we account for evidence, and we keep track of the aggregate uncertainty. The data quality is nuts, and the net result is a much softer number than it has any right to be, but eventually a signal can be extracted from the noise. Which leads us to our two lessons for the day. First, don't trust anyone who is quoting a single poll or telling you how an individual area is going to turn out. Second, fire the pollsters. 

Chris Christie is Taken Hostage Behind Enemy Lines
In the midst of Super Tuesday results rolling in, Trump called a press conference. It started about 45 minutes late, and was basically just a chance for Drumpf to gloat. However, the one very odd component of the whole thing was Chris Christie's introduction. That was the most unenthusiastic introduction I have ever seen, with Christie looking beaten and depressed, mouthing his lines with the same intonation and body language I have seen from hostages and POWs trotted in front of the cameras by their captors. A televised confession from a foreigner in North Korea has more sincerity. It makes you wonder what is the gun that the Trump campaign is holding to his head.