@elspeth Not even just selection bias. He predicted Gore would win in 2000 and Trump would win 2016. Either he's predicting electoral college or popular vote, and either way he missed one of those two.

I'd be interested in the related question "how many states does 538 miss?". As I recall, in 2016, their model output included a prediction of number of missed states -- i'd be interested to see if the community thinks that the model's predictions of its own mistakes were approximately correct.

Recall campaign claims to have gathered 1,825,000 signatures as of this morning.

https://recallgavin2020.com/recallgavin2020-r…

Based on an 82.1% validity rate, that would be sufficient for the recall to happen.

— edited by cd

I think >60% is overconfident.

Yang has a lead in head-to-head polls. But this is a ranked choice election. A 25-15 (what it seems to be?) lead in polls is just not that hard to overcome with peoples second and third choice picks.

I don't think it would be shocking if Yang won. I just don't think it would be shocking if he lost either. I'm sceptical of people putting down 70+%. I doubt the polls are really high-quality enough to get us there in this setting, and I don't see anyone citing other reasons.

@eibbett I think you're right.

But it does make it hard to argue that Yang is out of touch if he had the 'most correct' answer. As this was one (large) piece of opposition case against Yang, it may be useful for him to advertise.

@cd Further update. OWID puts the first date of 100mil fully vaccinated (across all vaccines) as March 20th.

This... represents a lower bound. If all the vaccines they are counting qualify, then March 20th (or April 3rd with a uniform 2-week lag) would be the earliest date this could resolve positive.

I predict a 40% chance. My forecast is based on the (updated) numbers by @AABoyles. But I believe in a substantial correlation. #Senate probability As indicated in that discussion, the upper bound on this percentage, is the lowest probability of any of the events. In this case, Dems winning the senate is the lowest probability event, with a probability of ~45%. Other sources indicate possibly higher probabilities (predictit implies 49/102 = 48% at the moment). #Model beyond P[Senate] The question then, is how often do I think the dems win the Senate, ...

Perhaps a silly question, but when we consider the amortization, how do we think about the timeline? What I mean is the following. In your example, building a plant and removing one ton cost $1billion, so the cost per ton was $1billion. Now (for simplicity) suppose the marginal cost is 0. Once the plant has removed 20 tons, the cost of removing a ton is $1billion/20 = $50million -- for every ton _including the first one_. Once the plant removes 20 million tons, the cost was $50/ton _for every ton including the first one_. So now my question is -- did th...

# Model As with the dem trifecta, I'm going to start with the least likely event, and think about how likely I think it is that it happens, without the other two events happening in order to make my forecast. This is the easiest way to comprehend the correlation between the events that I've seen. # House Odds By far the least likely of the three events is that the republicans win the house. [PredictIt puts this at a 16% chance](https://www.predictit.org/markets/detail/4365/Which-party-will-control-the-House-after-2020-election), [JHK puts this at 9%](h...

## Trend Uncertainty Agree with @nagolinc, it seems likely that the [global trend](https://ourworldindata.org/extreme-poverty) will continue. I think reason 1 -- that companies seek out cheap labor -- is important. In the link to global trend above, we can see this pattern play out -- first decreasing rates in East Asia, then South Asia. Consistent with that, I think it is likely that Sub-Saharan Africa will start to see improvements in the next ten years. I'd say there is a ~1/3 chance I'm wrong about this, and structural regional factors prevent the t...

@(JavierSouto) There have been a few people saying you misunderstood the 538 model. I'm going to elaborate for them. No polls are suggesting that Biden has a 73-27 % lead in the popular vote. Most polls are suggesting a popular vote lead between 5-15% for Biden. 538 takes that poll information, and uses information about likely polling error sizes, electoral college divergence from popular vote, etc. With all that information, 538 concludes that Biden is about a 7:3 favorite to win the election. You are of course right. If the polls were even marginal...

Wow, I've never had so much trouble coming up with a reasonable point estimate.

PredictIt puts GOP senate chances at 40%.

I refuse to bet on this and related scenarios.

Personally, I feel that in-depth gaming out of these situations normalizes them and brings them way too far into the overton window for my preferences. Self-fulfilling prophecies are real, and as it is said, we must snip "the threads of destiny one by one".

There is of course value in the forecasts and discussion -- and I certainly don't intend to disparage or discourage others. I merely intend that forecasters know of another filter (of many) in who participates.

Nate Silver says 10%. https://twitter.com/NateSilver538/status/1308…

updating upwards. ongoing silence from nyt staff is not a good indicator.

@Pablo Personally I thought that for questions without much participation, (1) was an active choice to incentivize participation.

@(Pablo) I think you've summarized this all very aptly. Just adding my 2 cents. I also don't think the 'incentive' issue here is very large. I don't have a great understanding of the actual design of the score bump for young questions, but in general, economists tend to say that lump sum payments don't change incentives. So if the setup is "the 3rd predictor gets an extra 10 points", that seems totally incentive-compatible to me. Not sure if Zvi has insider info that the design is actually bad or a proof that lump sum payments are not incentive compatib...

@kievalet I think you're fairly spot on.

except that the CDC numbers have a 5 day lag before they stop being updated.

So that by the time the CDC number reads "82.5 million", the actual date of completion will have been several days earlier.

ETA: the posted Question is very specific about resolution at 82.5 million, not 85 million (the number you discuss)

Edit 2: Current number on CDC site is 82.47 million.

— edited by cd

@(krtnu) I believe the 10% only applies to profits, not to revenues. So you get $7 of revenue on $6.92 of costs, for profits of $0.08. Then you get taxed (by PI) 10% of that $0.08, or about $0.01, for net profits of $0.07, or almost exactly 1%, risk free, over a 6 month period. Then of course you hit two more barriers. I believe PI also has some kind of 'redemption fee/minimum', so you need your overall profits across bets to make that worthwhile. Then you hit federal/state income (or capital gains?) taxes. Almost all of this is designed to hit profits...