Read on Twitter
Oh man, so many things:
On hospitalizations
1. I thought US hospitals would be more overloaded.
2. I thought hospital overload would be a bigger factor on death rate
3. I thought ventilator access would be a bigger issue. https://twitter.com/mckaycoppins/status/1261379754214649856
On hospitalizations
1. I thought US hospitals would be more overloaded.
2. I thought hospital overload would be a bigger factor on death rate
3. I thought ventilator access would be a bigger issue. https://twitter.com/mckaycoppins/status/1261379754214649856
ventilators:
4. I thought invasive ventilation would have a high death rate but still on the better side of 50% (like base rate from ICU wards in general)
5. I thought non-invasive ventilation would save more lives.
6. I encouraged my friends to work on ventilators (sorry!)
4. I thought invasive ventilation would have a high death rate but still on the better side of 50% (like base rate from ICU wards in general)
5. I thought non-invasive ventilation would save more lives.
6. I encouraged my friends to work on ventilators (sorry!)
Disease prognosis:
Up until ~ early April:
6. I thought asthma will be a large comorbidity.
7. I thought smoking would be a large comorbidity. (I'm still skeptical about nicotine)
8. In general, I overemphasized the respiratory aspects despite insufficient evidence to believe it.
Up until ~ early April:
6. I thought asthma will be a large comorbidity.
7. I thought smoking would be a large comorbidity. (I'm still skeptical about nicotine)
8. In general, I overemphasized the respiratory aspects despite insufficient evidence to believe it.
Personal strategy:
9. I "saw the smoke" relatively early compared to most non-professionals but didn't do much other than take personal precautions and a few small public warnings, until late Feb/early March
10. I vacillated between working on/not working on covid in Mar/Apr...
9. I "saw the smoke" relatively early compared to most non-professionals but didn't do much other than take personal precautions and a few small public warnings, until late Feb/early March
10. I vacillated between working on/not working on covid in Mar/Apr...
...which probably wasted critical time. I should've just "shut up and committed"
11. I emphasized respectability a bit too much over getting true beliefs, especially early on. (See attached screenshot). I think there was enough evidence in late Jan that this had >5%
11. I emphasized respectability a bit too much over getting true beliefs, especially early on. (See attached screenshot). I think there was enough evidence in late Jan that this had >5%
chance of effecting my life personally in significant ways, but I think I was implicitly too worried about fearmongering to publicly commit to this.
12. I focused too much on my local area/developed countries, instead of thinking hard about which places would be most affected.
12. I focused too much on my local area/developed countries, instead of thinking hard about which places would be most affected.
Meta-forecasting.
13. I trusted @metaculus medians too much early on (Jan, Feb), even though it should've been clear that their predictions didn't look anything like a martingale.
14. I thought the stock market was much correctly calibrated (Weak EMH).
13. I trusted @metaculus medians too much early on (Jan, Feb), even though it should've been clear that their predictions didn't look anything like a martingale.
14. I thought the stock market was much correctly calibrated (Weak EMH).
15. I was excited when the IHME models first came out. I got skeptical of them fairly early on, but didn't commit to writing them off.
16. I treated "domain experts" as a class too homogeneously; I didn't try hard enough to identify and follow all the *right* experts.
16. I treated "domain experts" as a class too homogeneously; I didn't try hard enough to identify and follow all the *right* experts.
Lockdowns:
17. I thought something like lockdowns will happen, but underestimated the speed at which they'd happen
18. I especially did not think that my local area (SF Bay) would be the first in America to do large-scale SAH.
19. I didn't think lockdowns would be lifted quickly.
17. I thought something like lockdowns will happen, but underestimated the speed at which they'd happen
18. I especially did not think that my local area (SF Bay) would be the first in America to do large-scale SAH.
19. I didn't think lockdowns would be lifted quickly.
Essentially, I over-emphasized inertia bias and thought that it'd be hard to get lockdowns to happen but after they happened, they'd stay for long enough.
20. I thought gov'ts (esp in US) would try harder to get Rt <<1, rather than have them hover at just slightly below 1.
20. I thought gov'ts (esp in US) would try harder to get Rt <<1, rather than have them hover at just slightly below 1.
Misc:
21. I over-emphasized unilateralist's curse, "do no harm" etc principles:
21a. I joined EpidemicForecasting later than I should have (because I was worried I'd do harm with bad forecasts)
21b. I'm too reluctant to engage with experts (to avoid wasting their time)
21. I over-emphasized unilateralist's curse, "do no harm" etc principles:
21a. I joined EpidemicForecasting later than I should have (because I was worried I'd do harm with bad forecasts)
21b. I'm too reluctant to engage with experts (to avoid wasting their time)
22. I had a belief that us nerds are the ones who cared explicit knowledge and statistics, and people "on the ground" will know "what's actually going on" and rely less on confirmed case statistics and more on seeing the real world and noticing whether their friends/family...
are getting sick or dying. I now think this is mostly wrong, if anything most laymen care more about case statistics and explicit information in the news than forecasters; I've grown pessimistic that if there's bad news+testing etc, ppl will even notice an epidemic happening now
Case forecasts:
23. I thought Japan would be better at containing it early on, and then I thought Japan would have a harder time containing it after some hospitals were overloaded.
24. I thought China would update statistics eventually but assigned very low % it'd happen in April
23. I thought Japan would be better at containing it early on, and then I thought Japan would have a harder time containing it after some hospitals were overloaded.
24. I thought China would update statistics eventually but assigned very low % it'd happen in April
25. I was very bad at predicting things in the Bay Area (essentially I thought there was a lot more cases here than actually were)
26. I was systematically underconfident in my predictions for US overall.
26. I was systematically underconfident in my predictions for US overall.
27. I predicted 10% on this question in mid-April (by which time the S'Pore app Trace Together already had 1.1 million users), and never updated my prediction: https://pandemic.metaculus.com/questions/3915/by-may-1-will-there-be-an-ios-or-android-app-that-shares-an-individuals-covid-19-infection-status-with-more-than-1m-other-users/
28. I thought there'd be more recorded deaths in India today than was actually the case.
29. After my terrible China prediction (see #24), I overcorrected and overestimated prob. of data updates happening in other places.
29. After my terrible China prediction (see #24), I overcorrected and overestimated prob. of data updates happening in other places.
30. I assigned only an 8% chance of the final number being >5k here: https://pandemic.metaculus.com/questions/3924/what-will-the-seattle-police-department-report-as-the-total-number-of-criminal-offenses-in-march-2020/
Overall, I think I did okay. Excited to keep making mistakes and (hopefully) correcting them more quickly so I can be less wrong in the weeks and months ahead.
