At the start of GJ Open I was of a mind to look at the most likely outcome and vote 0% or 100% on each question, assuming that I was looking forward, needed to make a decision now, and needed to make a definite decision. This is in line with the idea that forecasting should have forward utility and not function only as a more effective summary of today’s or yesterday’s news.
I wasn’t seeing that there was planning utility in having more shades of grey. I.e. situations where the most effective choice of action now depends on the range of possibilities and their weights, rather than the possibility you see after turning the Contrast button way up.
For example, I own a Chrysler Pacifica 2007. It has a little problem with random power steering shutdown. It happens rarely, but when it does happen, it is scary. So say it shuts down 10% of the time? That means 90% of the time it doesn’t, should I maximize to 100% and not get the car fixed? Obviously not. What if it is 1 out of 100 drives, should I get it looked at? Most likely.
So what I was missing in my initial view was the payoff of the event times the probability. A 1% chance of losing steering in heavy traffic at high speed is obviously highly unacceptable, because the potential loss to me would be infinite. (Or less according to some analyses, but very high for me.)
So I’m going back to estimating granular probabilities, for questions of the “slightly pregnant” variety, such as Korean nuke test, where the payoff value is highly negative. For other questions such as “Will Hillary win NH”, I am more satisfied to make an either/or forecast, where the payoff value is immaterial for me (I support Bernie but the 538 endorsement poll is telling me that doesn’t matter, Hillary is an overwhelming favorite to win the nomination).
UPDATE: It rained today, and coincidentally the dealership was finally able to duplicate the problem to their own satisfaction. Coincidentally, other Pacifica owners noted this problem primarily in wet conditions.
PS: This thing from Stratfor, just FYI, kind of relevant, at a certain level.