Bucky

• Bucky 28 Mar 2020 8:36 UTC

  2 points

  in reply to: Phil's comment on: COVID-19 growth rates vs interventions

  A cou­ple of ad­di­tional points to leggi:

  Eliz­a­beth calcu­lates roughly 25% of peo­ple are in es­sen­tial roles. Th­ese peo­ple are less able to re­duce num­bers of con­tacts.

  At least ini­tially many peo­ple don’t take so­cial dis­tanc­ing se­ri­ously so the effects are likely to ramp up over time.

  In that case it makes sense that ini­tially dou­bling times in­crease over 5 and over time they keep in­creas­ing.

  In China the dis­tance was en­forced and Kore­ans took it se­ri­ously right away so it didn’t take long for their dou­bling times to in­crease.

COVID-19 growth rates vs interventions

• Bucky 25 Mar 2020 20:53 UTC

  2 points

  in reply to: Jay Molstad's comment on: Break­ing quaran­tine is neg­li­gence. Why are democ­ra­cies act­ing like we can only ask nicely?

  You can’t suc­cess­fully sue Bob for giv­ing you COVID un­less you can prove it more likely than not that your COVID came from Bob. That’s ba­si­cally im­pos­si­ble.

  In South Korea where the con­tact trac­ing is work­ing this seems like it would be pos­si­ble. Pa­tient 31 has been more-or-less de­ter­mined to have lead to 5k+ peo­ple get­ting in­fected.

  If the num­bers come down in the US such that the au­thor­i­ties are able to con­tain via con­tact trac­ing then this would be­come rea­son­able there too.

• Bucky 25 Mar 2020 13:50 UTC

  25 points

  on: What will the eco­nomic effects of a 3 week quaran­tine be? 3 months?

  Here’s my rough sketch of what might hap­pen to large man­u­fac­tur­ing com­pa­nies due to shut­downs within the sup­ply chain.

  Costs of other peo­ple’s shutdowns

  Differ­ent coun­tries /​ com­pa­nies are likely to have differ­ent shut­down pe­ri­ods.

  Sup­ply chains are highly in­ter­na­tional so shut­downs in one coun­try have huge knock-on effects. Most com­pa­nies will aim to be dual sourced on crit­i­cal com­po­nents which re­quire long qual­ifi­ca­tion pe­ri­ods but this isn’t always sim­ple. Even if dual sourced, bring­ing up the sec­ond source to cover the loss from the other won’t hap­pen overnight, if at all.

  For in­stance, say you buy a crit­i­cal com­po­nent from Italy. It’s dual sourced but on an 80:20 ra­tio so if Italy shuts down then your the po­ten­tial out­put drops by 80%.

  In re­al­ity stock lev­els will of­ten sit at 30 days so for a shorter shut­down there is some slack. How­ever, 30 days is an av­er­age. With enough differ­ent com­po­nents you’re likely to be low-ish on stock for at least a cou­ple of com­po­nents and may be forced to drop out­put.

  The other side of this is cus­tomers. Some cus­tomers will shut their pro­duc­tion lines, oth­ers stay open. If your product in fairly ho­moge­nous this might be fine and match your in­com­ing com­po­nent lev­els. With a wider product range you may find that your sup­ply line fails in the place where you most need the com­po­nents.

  The above will ap­ply to a lesser ex­tent where com­pa­nies have to slow pro­duc­tion.

  Cost of your shutdown

  Whilst you’re shut down you aren’t pro­duc­ing any­thing. This ob­vi­ously has the di­rect im­pact of re­duc­ing turnover to 0 for how­ever long you’re closed. This is al­most cer­tainly the biggest im­pact. If the gov­ern­ment steps in to cover this (at least partly, e.g. the UK gov­ern­ment will pay 80% of wages) then large man­u­fac­tur­ers shouldn’t have an is­sue. Even if they have a tem­po­rary cash flow prob­lem, the banks are likely to step in to help out.

  Costs like rent­ing the space that you’re in might still need to be paid. How­ever if you’re strug­gling then who­ever you’re rent­ing off doesn’t re­ally have the op­tion to rent out to some­one else in the short term so there may be some rene­go­ti­at­ing be­ing done (I’m less con­fi­dent about this point—I’m not sure how the con­tracts would work out).

  When you shut­down you prob­a­bly already have lots of goods on their way to you by sea. My guess is that if pos­si­ble peo­ple will try to get these ac­cepted into the fac­tory al­though I know some de­liv­er­ies are be­ing turned away if they have come through a high risk coun­try.

  When you re­open, ev­ery­thing will be a mess and the first week will be chaos (al­though with peo­ple work­ing from home maybe this can be min­imised with good plan­ning). For a month or so things will be a bit mud­dled so effi­ciency won’t be op­ti­mal.

  You’ll have some cus­tomers chas­ing you to get product im­me­di­ately. I guess there’ll be a huge de­mand for airfreight—maybe this is how the air­lines can re­cu­per­ate some of their losses?

  There may need to be some work­ing with cus­tomers and sup­pli­ers on con­trac­tual terms of pay­ment etc. In nor­mal cir­cum­stances these are very tightly con­trol­led but I would an­ti­ci­pate that most com­pa­nies will be able to take the prac­ti­cal ap­proach and over­come the bu­reau­cracy which is in­her­ent in such ne­go­ti­a­tions, due to the ex­cep­tional cir­cum­stances. Com­pa­nies which are un­able/​un­will­ing to do this are likely to suffer ad­di­tional dam­age.

  Smaller companies

  The above mostly ap­plies to smaller man­u­fac­tur­ers but to a lesser de­gree.

  They are likely to be lower on the list of pri­ori­ties for banks to sort out emer­gency loans which could cause a num­ber to go out of busi­ness. This may be the tar­get of ad­di­tional gov­ern­ment in­ter­ven­tion.

  Sup­ply chains are likely less com­plex and so have fewer crit­i­cal point to go wrong. They are also prob­a­bly able to switch sup­pli­ers more eas­ily if re­quired.

  They will man­age to get things sorted out more eas­ily be­fore and af­ter­wards.

  Smaller com­pa­nies have less lev­er­age in ne­go­ti­at­ing new con­tracts. In pur­chas­ing this is offset by prob­a­bly be­ing able to be more flex­ible. In sales this is harder if they are sel­l­ing to larger com­pa­nies.

  Over­all economy

  So mul­ti­ply the above through­out the econ­omy and you get a large vari­a­tion across com­pa­nies de­pend­ing on how the in­di­vi­d­ual sup­ply chains which they are a part of are hit. Every­one will kind of mud­dle through as best they can but things will be far from effi­cient for as long as there are sig­nifi­cant parts of the world in lock­down, even for com­pa­nies which aren’t in lock­down.

  The ob­vi­ous cost of lock­down (lack of pro­duc­tivity) is likely to be the most im­por­tant and other con­sid­er­a­tions are likely to be large but con­sid­er­ably smaller.

  ***

  I wrote the above and then re­al­ised that this was based on the as­sump­tion that over­all de­mand for your product will be the same a year af­ter lock­down as it was a year be­fore. For many in­dus­tries this is prob­a­bly true but oth­ers (e.g. some lux­ury goods?) might not bounce back fully or might bounce back into a differ­ent shape than be­fore. This is a com­pletely differ­ent ques­tion that I’m not sure how to an­swer.

• Bucky 23 Mar 2020 9:45 UTC

  5 points

  on: Us­ing smart ther­mome­ter data to es­ti­mate the num­ber of coro­n­avirus cases

  I think it would be good prac­tice to add the re­sults of your sen­si­tivity anal­y­sis into your sum­mary here. If I’m un­der­stand­ing your sheet cor­rectly the range found from the sen­si­tivity to un­cer­tainty in baseline rate of fever gives COVID-19 num­bers of 0.9 − 7.0 mil­lion cases?

• Bucky 21 Mar 2020 16:47 UTC

  2 points

  in reply to: Ramiro P.'s comment on: Preprint says R0=~5 (!) /​ in­fec­tion fatal­ity ra­tio=~0.1%. Thoughts?

  If there is a 1:1 symp­tomatic:asymp­tomatic ra­tio and 2,000,000 odd in­fec­tions then there are 1,000,000 symp­tomatic peo­ple out there and only 40,000 iden­ti­fied. Of that 1,000,000 we ex­pect 200,000 to re­quire hos­pi­tal­i­sa­tion and 50,000 to re­quire ICU.

  If this was true I would ex­pect some­one to have no­ticed.

  There might be an­other ex­pla­na­tion for the figures that I’m miss­ing but, as I said, I think it’s up to them to ex­plain what they think is go­ing on.

• Bucky 20 Mar 2020 22:31 UTC

  15 points

  in reply to: Ramiro P.'s comment on: Preprint says R0=~5 (!) /​ in­fec­tion fatal­ity ra­tio=~0.1%. Thoughts?

  Di­a­mond princess is im­por­tant be­cause they did 100% test­ing so it gives us an idea of asymp­tomatic : symp­tomatic ra­tio. The re­sult was roughly 1:1, noth­ing like 50:1 or what­ever this pa­per sug­gests. The sci­ence study with 6:1 is at least plau­si­ble if you ac­count for symp­tomat­ics who weren’t iden­ti­fied.

  If South Korea hadn’t man­aged to test the ma­jor­ity of their cases then it is un­likely that they would have man­aged to re­duce their in­fec­tion rate so dra­mat­i­cally—their quaran­tine mea­sures aren’t mas­sively strict al­though I think the pop­u­la­tion are self-en­forc­ing good prac­tice pretty well. I doubt that Wuhan death rates could be be­low South Korean rates due to the ac­knowl­edged over­crowd­ing in Wuhan. Again, 0.6% is kind of plau­si­ble, the model here (0.1%) isn’t.

• Bucky 20 Mar 2020 20:33 UTC

  6 points

  on: Preprint says R0=~5 (!) /​ in­fec­tion fatal­ity ra­tio=~0.1%. Thoughts?

  Like oth­ers I doubt the in­fec­tion and fatailty rates be­cause of South Korea and Di­a­mond princess (if the au­thor knew about how much this re­sult con­flicts with those datasets then its up to them to ar­gue why the new pa­per is bet­ter).

  R0=5 isn’t com­pletely un­be­lie­ve­able. If the dou­bling time with­out con­tain­ment mea­sures is 2 days and the in­fec­tive pe­riod is 12 days (i.e. 5 days in­cu­ba­tion pe­riod and a week af­ter­wards) then R0=5. Un­for­tu­nately based on the rather un­be­liev­able in­fec­tion and fatal­ity rates I don’t think this pa­per re­ally adds any ev­i­dence for this—it sug­gests the model is fatally flawed.

• Bucky 20 Mar 2020 20:22 UTC

  6 points

  in reply to: jimrandomh's comment on: Preprint says R0=~5 (!) /​ in­fec­tion fatal­ity ra­tio=~0.1%. Thoughts?

  I was par­tic­u­larly be­mused by quot­ing cu­mu­la­tive in­fec­tions to 7 sig­nifi­cant figures where the 95% con­fi­dence in­ter­val spanned a fac­tor of 2. This did not fill me with con­fi­dence...

• Bucky 19 Mar 2020 15:44 UTC

  2 points

  in reply to: Harold Schranz's comment on: Growth rate of COVID-19 outbreaks

  Over the past week or so Aus­tralia has lost con­tain­ment and is run­ning at a dou­bling time of 3-3.5 days. I don’t know whether that cor­re­lates with higher con­cen­tra­tion of cases in big cities—my prior would be that most im­ported cases would ar­rive in the big cities in the first place but I haven’t checked this.

• Bucky 19 Mar 2020 15:14 UTC

  10 points

  in reply to: romeostevensit's comment on: Covid-19 Points of Lev­er­age, Travel Bans and Eradication

  It doesn’t mat­ter hugely whether hos­pi­tals are over­loaded by 3-5x for a long time or 20x for a rel­a­tively short time. In the former about 75% of peo­ple are un­able to get treat­ment, in the lat­ter 95%. 3-5x is bet­ter but isn’t as much bet­ter as it might seem.

  The UK gov­ern­ment has claimed that hy­giene/​so­cial dis­tanc­ing can flat­ten the curve by 20% but 20% doesn’t make much differ­ence un­less you hap­pen to be just over ca­pac­ity be­fore­hand.

• Bucky 19 Mar 2020 10:30 UTC

  4 points

  in reply to: jimrandomh's comment on: [UPDATED] COVID-19 cabin sec­ondary at­tack rates on Di­a­mond Princess

  If I as­sume the rate at which peo­ple in sin­gle-per­son cab­ins get in­fected (8%) is the rate of in­fec­tion out­side the cabin, and that the higher rate of in­fec­tion in two-per­son cab­ins is caused en­tirely by within-cabin sec­ondary trans­mis­sion, then it looks like each per­son would have to in­fect their part­ner an av­er­age of 1.5 times each. This also tells us that the trans­mis­sion rate be­tween el­derly cou­ples shar­ing a bed is likely to be ex­tremely high, and also that peo­ple in sin­gle-per­son cab­ins must be differ­ent in some way—per­haps they spent less time in the ship’s com­mon ar­eas.

  This was my origi­nal thought too. How­ever, as the 8% is based on only 6 pos­i­tive cases it isn’t a very pre­cise figure.

  As an ex­am­ple, the max­i­mum like­li­hood for any pair of vari­ables for my mod­els comes at back­ground in­fec­tion rate of 0.133, sec­ondary at­tack rate=0.55 with no ter­tiary at­tack (I didn’t men­tion this in the OP for fear of peo­ple tak­ing the 0.55 to be es­pe­cially rele­vant). In this case the prob­a­bil­ity of get­ting 6 or fewer in­fec­tions in 1-berth cab­ins would be 0.11 - un­likely but not mas­sively so.

  The cor­re­spond­ing prob­a­bil­ities for 2, 3 and 4-berth cab­ins are 0.68, 0.14 and 0.50. Those 4 num­bers seem fairly ran­dom, sug­gest­ing that there’s no need to stipu­late base rates which vary based on cabin size to ex­plain the data.

  In truth I sus­pect that there may be differ­ences in the base rate be­tween cabin sizes but wouldn’t have known in ad­vance which size would have had a higher base rate. With only 4 data points even us­ing 2 vari­ables in the model is push­ing it—if I used any­more I could have ex­plained al­most any­thing!

  ***

  Edit: Sec­tion be­low is no longer endorsed

  Re­gard­ing the effect of quaran­tine mea­sures, only 115 of the 536 pas­sen­ger in­fec­tions analysed had on­set af­ter the quaran­tine started. Figure 1 here sug­gests to me that al­most all of the in­fec­tions oc­curred be­fore quaran­tine and on­set was de­layed by in­cu­ba­tion pe­riod.

• Bucky 18 Mar 2020 23:34 UTC

  6 points

  in reply to: johnswentworth's comment on: [UPDATED] COVID-19 cabin sec­ondary at­tack rates on Di­a­mond Princess

  at­tack rate = 1 within a cabin would be ev­ery­one catches it at some point (but not nec­es­sar­ily im­me­di­ately) pro­vided that some­one brings it in in the first place—its a rate per sick per­son rather than per unit time. I don’t have data on whether this is the case al­though I doubt it.

  Tech­ni­cally I sup­pose hav­ing 18 cases in 4-berth cab­ins does rule that out. My model isn’t so­phis­ti­cated enough to catch some­thing like that—I look at av­er­age ill­ness rate as an in­put to the bino­mial dis­tri­bu­tion, I never check whether the to­tal num­ber is likely. Ad­ding that com­plex­ity might help nar­row down the true sec­ondary at­tack rate.

  I’ve added a log graph.

[UPDATED] COVID-19 cabin sec­ondary at­tack rates on Di­a­mond Princess

• Bucky 17 Mar 2020 23:43 UTC

  4 points

  in reply to: Donald Hobson's comment on: Bucky’s Shortform

  Hmm, I don’t think that’s what an up­vote gen­er­ally rep­re­sents. An up­vote is more of a gen­eral “I’d like to see more like this” rather than a spe­cific “I re­searched this point and found it to be cor­rect”.

• Bucky 17 Mar 2020 13:48 UTC

  10 points

  on: Good News: the Con­tain­ment Mea­sures are Working

  The most in­ter­est­ing case to me is South Korea who have man­aged to achieve con­tain­ment with­out mass lock­downs—in­stead they have an ag­gres­sive test­ing policy such that only 3% of tests are pos­i­tive.

• Bucky 17 Mar 2020 13:05 UTC

  4 points

  in reply to: avturchin's comment on: Rea­sons why coro­n­avirus mor­tal­ity of young adults may be un­der­es­ti­mated.

  Note that se­vere =/​= crit­i­cal (I think that post con­fuses se­vere with crit­i­cal in the con­clu­sions)

  In the three Chi­nese stud­ies which are quoted, se­vere in­cludes e.g. cases of short­ness of breath and high fever which, whilst pos­si­bly hop­si­tal­is­able un­der nor­mal cir­cum­stances, not ob­vi­ously fatal if hos­pi­tals are full. Se­vere doesn’t im­ply “re­quiring me­chan­i­cal ven­tila­tion or other in­ten­sive care.”

  A bet­ter es­ti­mate for ICU cases based on that ev­i­dence would be ~0.4% as (se­vere + crit­i­cal) /​ se­vere = 4 in the 44k per­son China data.

  Of course some se­vere but not crit­i­cal cases might be­come crit­i­cal if not treated, so death rate with­out any treat­ment would be be­tween the two.

• Bucky 16 Mar 2020 23:44 UTC

  4 points

  in reply to: Vaniver's comment on: Coron­avirus: Jus­tified Prac­ti­cal Ad­vice Thread

  Looks like v2 of the pa­per has cor­rected the er­ror.

• Bucky 16 Mar 2020 23:42 UTC

  7 points

  in reply to: gwillen's comment on: Coron­avirus Jus­tified Prac­ti­cal Ad­vice Summary

  V2 of that pa­per also cor­rects the half-life er­ror in v1 dis­cussed by Vaniver here.

  The OP has the in­cor­rect val­ues from the first pa­per. An­noy­ingly v2 doesn’t give ac­tual num­bers but it looks like ~2.5 − 5 hours to me.

  What links here?

  • Coron­avirus: Jus­tified Prac­ti­cal Ad­vice Thread by Ben Pace (28 Feb 2020 6:43 UTC; 231 points)

[Question] Why such low de­tected rates of COVID-19 in chil­dren?