Research Fellow at Open Philanthropy.
Previously at the Forethought Foundation for Global Priorities Research.
isabel
Karma: 215
Quick question about reputation scores: “Every time a question resolves, the reputation is updated depending on how many Metaculus points a user got relative to other users (with a mean of zero and a standard deviation of 10)”—does this mean that predicting on questions late in the life of a question is harmful for one’s reputation? Because predicting late means that you’ll typically get fewer points than an early predictor.
re: the edit at the top of the article, were the unresolved questions resolved? I’d be interested in seeing an updated version of this post! (either with just those questions or also including more questions from the past year and half)
Great analysis, I really enjoyed reading this. I’m also excited to see how Metaculus and Manifold compare on the 2023 ACX predictions! I think that’ll be a great set of identical questions on both platforms that will avoid any selection effect issue of which platform the questions started on.
this might be due to a change on EA forum since you initially posted this post, but the left and right columns of the table of studies are quite unreadable for me on desktop, on both Chrome and Edge. see screenshot for what it looks like from my end. is there any other format I can read this post in?
I was commenting because I’ve been curious about it and it seemed like info that would often be present alongside whether or not there was a prison sentence at all, so it seems like it wouldn’t have been much marginal work to collect it on your first pass (though obviously much more work now, unless you were going back through the list for some other reason).
There do exist questions about how long people will be sentenced to prison around this (like this Metaculus one), but it also wasn’t obvious to me that you were going for exclusively decision relevant info—how is jurisdiction of crimes decision relevant?
Though maybe I should have just said interesting rather than useful.
length of prison term (sentenced? served?) seems like a useful column in the spreadsheet.
was the Cuban Missile crisis higher risk than actual nukes going off? actual nukes seem to me to be more salient.
while I’d agree with the advice of “don’t go if they don’t provide any funding”, not providing quite enough funding to actually cover expenses seems more ambiguous?
Metaculus performed decently well anticipating the Russia-Ukraine conflict, so I do think they can provide information about conflicts
if you’re able to apply directly into PhD program, that might be worthwhile as they usually provide funding
This (https://discord.com/invite/nFsNfaH) is an invite link to EA Corner, which I think is the current main EA discord server?
this time, the WPP does show a population decline before the end of the century, though they still have a later and higher peak than Vollset et al.
prior to the update, the UN projections were clearly worse than the Vollset ones, now that their projections are closer together, I’m less confident which one is likely closer to the truth. but lean towards Vollset still being better and the UN not having revised down enough yet.
also, fun observation: two of the eight countries contributing most to growth in the next three decades already have shrinking birth cohort sizes, because even though cohorts are getting smaller than previous years, they’re still much larger than the elderly population which has the highest mortality rates. (India and the Philippines, though note that there is a really wide discrepancy between Philippines estimates of births and WPP estimates of births, which is wider than what the birth registration gap is purported to be)
I’m pretty sure Kelsey didn’t drop out, though she did post about having a very hard time with finishing.
drafting a post which uses this feature, and I’m a bit disappointed in terms of how much smaller the map is compared to using the downloaded figure from the OWiD. I think it would be better if the map were took up more of the available width.
thank you for writing this! your personal experiences were extremely relatable. I laughed at “Remind yourself that it’s someone else’s job to figure out whether you’re meeting their standards as an employee, so you don’t need to constantly feel angst about whether you should quit to save them the trouble of firing you.” (I will not type up the counter-argument my brain is attempting to come up with)
one of the ways in which ~imposter syndrome has made me worse at accomplishing things is that it has made me worse at asking questions when I was confused, because I was afraid that the questions I would ask would give me away as an imposter. this isn’t all downsides: it probably resulted in me being better at working through confusions that I have, but also cost me a lot of time. I personally found reading https://danluu.com/look-stupid/ helpful for that specific facet.
you’re right that St. Mary’s has an exceptionally high birth rates, but they are both very small (pop ~2600) and as of recently not growing (2015: 2664 → 2020: 2652). (Kansas Department of Health) which makes me suspect that they’re experiencing some retention issues.
more broadly, this article suggests that while there’s a fairly large number of attendees (~150,000) of Latin Mass in the United States, I expect that the Pope’s recent restrictions on Latin Masses discussed in that article will hurt their popularity + growth.
Love this feature! Very excited to use this in my next post.
Oh, good idea. Reverse image searching resulted in me finding the same version claiming that it’s 2007 data. So that’s maybe partially reflecting differences in how people responded to the financial crisis in particular?
Decided to find some TFR data from Eurostat and recreate this map for some more recent years. The France-Germany gap has been decreasing in visual saliency: 2014 is still pretty visible but but 2019 is less so (though there is still some aggregate TFR difference between France and Germany). Data doesn’t go far enough back for me to be able to check the original map but it doesn’t seem particularly implausible.
2014 TFR—Eurostat 2019 TFR—Eurostat
I think your comment does a really good job of illustrating the difficulty in determining which groups and circumstances are selecting on what traits, as the two examples of unusually strong selection on fertility that you bring up are the Amish and the French, which have been on opposite ends of fertility behavior. It’s not impossible that both of these groups are selecting more strongly on fertility than everyone else, but it is somewhat counterintuitive.
I agree that the Amish are selecting on something, but that something isn’t necessarily a preference to have more children. The paper you linked also lists “affinity for work, perseverance, low status competition, respect for authority, conscientiousness, and community orientation” as other characteristics that may be being selected for among the Amish. If the Amish are being selected for ~conformity and community orientation rather than desire to have more kids irrespective of circumstances, then if the circumstances change at the community level (for example, if it becomes more difficult to purchase farmland, as is already happening to the community in Lancaster County, or if the Amish stop being exempt from the requirement that children need to stay in school until they are 16, which some people are pushing for) then the Amish fertility rate could decline further than it already has.
The French case seems somewhat more compelling: because of contraception and norms around family sizes, the people who had larger families in France would be people who intrinsically valued larger families, and so selection would in the direction of higher fertility preferences more directly, rather than high fertility being a result of conforming to local norms. That being said, two centuries of selection in the direction of people who want kids more than average hasn’t been enough to bring French fertility above replacement, merely to above average for Europe.Do you know which year the map from Breeder’s Revenge that you linked is for? I don’t see a year on it, but it shows more départements at above replacement than the paper that I’ve found on Recent Demographic Developments in France, which has a map for 2015. Note that the country-wide TFR of 1.96 shown below has declined a bit since then, reaching 1.86 by 2019, and 1.83 in 2020 and that non-immigrant TFR is even further below replacement. While I think it’s plausible that TFR in France will start going up again in the near future, it’s also quite plausible that it won’t.
TFR by Departement, 2015
It’s also possible that something has changed recently that makes this a better practice now than it used to be — to use your example, maybe water access is better now than it was historically. Maybe beans weren’t prevalent in this region or a strain of beans that were worse for soaking were previously more common.