I've significantly improved the Candidate Scoring System which looks at the 2020 US presidential election.
There is now a permanent link to reduce the confusion of different versions. When making updates, I will save over the same PDF. And within this document, the links to supplementary CSS files (the Excel model, the draft for shared editing and updates, and supplemental collections of evidence/arguments) will remain permanent as well.
https://1drv.ms/b/s!At2KcPiXB5rkyABaEsATaMrRDxwj?e=VvVnl2
The report itself has been improved in a variety of ways - more comprehensive, more accurate, etc. The conclusions have changed but not to such an extent that our previous recommendations could be considered harmful. Some major updates are: (1) Very helpful analysis from FiveThirtyEight which suggests that we should focus much more on major candidates and much less on minor candidates, as the former seem much more tractable. Thus Buttigieg becomes the top recommendation now. (2) Lower opinion of Andrew Yang, due to placing less weight on domestic tech policy and more weight on international relations for the purpose of mitigating x-risk. This was not an arbitrary change of mind, but an implication of a very sensible revision of how the issues were divided. If someone wants to increase the weight of "Emerging Tech" then we can have that debate.
I am still not confident that it is robustly argued and defended well enough to be worth broadly sharing as outreach to mass audiences (e.g. "vote for so-and-so, here's the EA proof") lest they find problems with it and then condemn EA to the detriment of other causes like global poverty, etc. However, for EAs and any EA-adjacent people, it should be useful and at least somewhat persuasive - not just for this election but also as a broader summary of ideas on political policy. It's also probably good to share with people who will happen to agree with some of its conclusions.
As always, I will take a careful look at all input and am open to collaboration.
What do prediction markets add that polls don’t? Is there any history of them being more accurate than polling-based models such as 538’s? To the extent they are relevant my intuition is that it would be as to the lesser-known candidates, where the type of person participating in prediction markets has priced in info the average voter doesn’t have. But where average voters do have more complete information, I would rely exclusively or at least more heavily on polls.
For one thing, it's not clear how to translate polls into probabilities. Let's assume for sake of argument that when Jack is at 5% in the polls and Jill is at 4%, then Jack is necessarily more likely to win the primaries than Jill. But we still don't know: how much more likely? What are their probabilities of victory? Translating the polling into expectations (which is necessary for calculating expected value) is difficult. Better to let the prediction markets do that efficiently than to rely on my own guess.
Second, prediction markets take i... (read more)