Covid-19 4/​30: Stuck in Limbo

Link post

Epistemic Sta­tus: As with all my Covid-19 posts I’m not any kind of ex­pert and am doubtless mak­ing a lot of mis­takes. Yet it seems worth­while to per­sist.

In pre­vi­ous posts I’ve been fo­cused my di­rect mod­el­ing on New York. That was for var­i­ous rea­sons. It’s the epi­cen­ter, it’s where I live, it is the place where herd im­mu­nity po­ten­tially mat­ters already, and data is bet­ter than try­ing to rec­on­cile 50 differ­ent states.

Now it’s time to start fully ex­pand­ing that to the coun­try. Wikipe­dia’s data is slightly differ­ent than my main source, and has the more gen­er­ally fatal weak­ness that it doesn’t provide the nega­tive test counts, but it pro­vides it in eas­ier-to-ac­cess form for deaths and in­fec­tions, so I’m go­ing to use it for the fol­low­ing charts I cre­ated (see calcu­la­tions in my spread­sheet here).

Deaths by Week in the Big 5 Re­gions:

WEST MIDWEST SOUTH NORTHEAST ex-NY NY
Mar 19 – Mar 25 164 450 182 143 364
Mar 26 – Apr 1 424 1894 667 856 1988
Apr 2-8 764 3873 1331 2248 4694
Apr 9-15 890 4853 1596 3605 5318
Apr 16-22 1033 5189 1727 5165 3716
Apr 23-29 1128 5234 1694 4742 2666

Pos­i­tive tests by Week in the Big 5 Re­gions:

WEST MIDWEST SOUTH NORTHEAST ex-NY NY
Mar 19-Mar 25 7176 7812 9927 11923 33106
Mar 26 – Apr 1 16665 22121 28412 37339 55123
Apr 2-8 19925 31460 38989 60433 67556
Apr 9-15 16291 29267 35570 62293 64463
Apr 16-22 20065 34130 33932 65080 43437
Apr 23-29 21873 42343 34683 62355 42475

Tests each week:

USA Pos% USA ex-NY Pos % NY Pos % USA Tests USA ex-NY Tests NY Tests
Mar 19-25 16.2% 10.7% 32.0% 347577 258695 88882
Mar 26-Apr 1 20.2% 15.4% 45.1% 728474 611073 117401
Apr 2-8 19.8% 15.8% 45.5% 1067220 922947 144273
Apr 9-15 20.1% 16.5% 40.1% 1039790 878931 160859
Apr 16-22 15.7% 13.8% 30.2% 1253535 1109565 143970
Apr 23-29 13.7% 12.6% 21.0% 1480101 1277602 202499

For com­plete­ness, the to­tal pos­i­tive tests:

USA Pos­i­tives USA Ex-NY Pos­i­tives
Mar 26-Apr 1 56198 27769
Mar 19-Mar 25 146888 93987
Apr 2-8 211095 145491
Apr 9-15 209067 144604
Apr 16-22 197111 153674
Apr 23-29 202934 160459

I de­cided to go by week be­cause re­port­ing on differ­ent days of the week is differ­ent. Tues­days and Wed­nes­days have a big­ger share, while week­ends have lower shares. There’s also a lot of day-to-day noise in gen­eral. Num­ber of tests fluc­tu­ates wildly. By com­bin­ing weeks, we get a much eas­ier to un­der­stand pic­ture, and there’s now enough data that we can do this.

New York: We Can Make It There

The pic­ture for New York is clear and ex­cel­lent. The new R0 calcu­la­tion comes in as R0 ~ 0.78, if we as­sume a five day se­rial in­ter­val. Us­ing deaths and us­ing in­fec­tions each give us the same an­swer, so it seems very solid.

It does raise one im­por­tant note about timing.

New York’s rate of in­fec­tions de­tected peaked in the first week of April as did its pos­i­tive test per­centage. They then de­cline a lit­tle in the sec­ond week of April and fall dra­mat­i­cally in the third week.

Deaths peaked the sec­ond week of April then fall off dra­mat­i­cally. This dates the in­fec­tion peak to about 21 days ear­lier, which would be on or right af­ter the lock down date. That makes sense.

In­fec­tions peak­ing only a few days be­fore deaths im­plies a very de­layed test­ing sched­ule. Per­haps more so than is even plau­si­ble, but cer­tainly more than I was pre­vi­ously tak­ing into ac­count. I think we have to pre­sume that if deaths lag by 21 days, a pos­i­tive di­ag­no­sis has to be at least 10 days de­layed, un­der the con­di­tions at the time. It should be some­what less now.

That’s ex­cel­lent news. What about other places?

Other Re­gions: Can We Make It Any­where (Else)?

We see the pos­i­tive test per­centage fal­ling around the coun­try. That’s very good news. It does re­flect in­creased test­ing, and may re­flect in­creased fo­cus of test­ing on places that need it the least. That wor­ries me.

If look at the West re­gion, we see a slow but steady in­crease in deaths. Cal­ifor­nia is steady from this week to last week, but in­creas­ing from be­fore that. It has a very low baseline, which is great, but things are not im­prov­ing.

We also see a similar pat­tern of things slowly get­ting worse in the Mid­west.

The non-NY North­east re­gion looks like it is fi­nally get­ting some­what bet­ter. But it’s hap­pen­ing later and slower than New York. The timing is weird, and there are some funky ar­ti­facts in the data con­tribut­ing to that. Penn­syl­va­nia’s data is a real mess and New Jersey’s isn’t much bet­ter, and since we’re ex­clud­ing New York that’s a huge per­centage of the whole re­gion.

The South was also start­ing to show slow im­prove­ment, but now it’s start­ing to re­open. As I’ve noted be­fore, the ‘re­open’ crite­ria be­ing used makes no sense. So it seems un­likely that the im­prove­ments will be sus­tained. There is ex­treme differ­ence be­tween states here. Virginia’s cases are sud­denly go­ing way up, whereas there seem to be clear pre­vi­ous peaks for some other states such as Louisi­ana. Louisi­ana, and New Or­leans in par­tic­u­lar, seems im­por­tant to un­der­stand­ing the gen­eral pic­ture.

The over­all pos­i­tive test per­centage does con­tinue to slowly de­cline. That is good news, but the de­cline is small, comes along with mod­estly in­creased test­ing, and likely it partly re­flects changes in the dis­tri­bu­tion of who is do­ing the test­ing. I’ll learn more about that if/​when I get to dive fur­ther into the states and cities one at a time and see if that’s true. It would be good to try nor­mal­iz­ing for test lo­ca­tions, but I haven’t seen any­one do that.

Get­ting Worse Be­fore It Gets Bet­ter: Louisiana

What’s go­ing on with Louisi­ana?

Louisi­ana con­firmed pos­i­tive tests and deaths by week:

Pos­i­tive Tests Deaths
Mar 26-Apr 1 1913 73
Mar 19-Mar 25 6845 227
Apr 2-8 9133 392
Apr 9-15 4921 451
Apr 16-22 3307 370
Apr 23-29 2402 329

Louisi­ana’s data is a royal mess. I’m not list­ing pos­i­tive test per­centages, be­cause they’re non­sense. From April 18 to April 25, the num­ber of cu­mu­la­tive re­ported nega­tive tests went down. There’s no later day where it looks like their re­port­ing caught up. We re­al­is­ti­cally have no idea how much test­ing is tak­ing place. De­spite this they claim an over­all pos­i­tive rate of only 21.4%, and be­low 20% each day since April 11 (not count­ing days when they re­ported zero or less than zero nega­tives).

The drop-off in deaths we’ve seen so far seems nowhere near dra­matic enough to be com­pat­i­ble with the drop in pos­i­tive test re­sults, un­less test­ing col­lapsed, which the nega­tive test counts claim didn’t hap­pen.

Still, this seems like a clear ex­plo­sion and fast peak, fol­lowed by a clear nega­tive trend.

We can do a similar calcu­la­tion to the one I pre­vi­ously did for New York. We take the deaths, back­date them three weeks, as­sume IFR of 1%, a se­rial in­ter­val of 5 and an R0 that matches the data, and see how many in­fec­tions we get. In this case, we can use the de­cline in deaths on a weekly ba­sis to es­ti­mate the R0. We get about 0.8. If we check that against the pos­i­tive test counts from April 9 on­ward, we 0.775. Close enough. Let’s split the differ­ence and say 0.79. Cor­rect­ing for 50% un­der-count of deaths we get 328,344 in­fected out of a pop­u­la­tion of 4.65 mil­lion. That’s only 7.1% of the state in­fected.

A big differ­ence is that only 391,000, or less than 10% of that amount, live in New Or­leans, ver­sus al­most half of New York state be­ing in New York City.

Re­do­ing the calcu­la­tion for only Or­leans Per­ish, which has 27.4% of the deaths (found by Google giv­ing me stats di­rectly, and me di­vid­ing Or­leans’ count by the to­tal from all per­ishes) and only 8.4% of the state’s pop­u­la­tion, I get 22.5% in­fected in New Or­leans. That’s lower than my es­ti­mate for NYC, but it’s more than halfway there.

If we re­fer back to the log of On R0, even the 7.1% in­fec­tion rate is a big game. The differ­ence be­tween 7.1% for Louisi­ana, and much smaller num­bers for most of the United States, ex­plains how Cal­ifor­nia can be in limbo around R0=1 while Louisi­ana has turned the cor­ner.

Is there an­other ex­pla­na­tion for why the epi­cen­ters seem to be im­prov­ing, while places that are do­ing ‘ev­ery­thing right’ but were not epi­cen­ters are not see­ing that? I don’t see a case available that Louisi­ana is tak­ing things es­pe­cially se­ri­ously, or im­ple­ment­ing bet­ter or stric­ter poli­cies than the West.

Limbo

The ex­pla­na­tion that makes sense to me is that lock downs in the Amer­i­can style, by de­fault, cre­ate R0 very close to 1.

We see im­prove­ments in places with some herd im­mu­nity, or which are es­pe­cially well locked down or oth­er­wise situ­ated. While oth­ers, where we’re fal­ling short, get worse.

Which is all, in some sense, the worst pos­si­ble situ­a­tion.

You can’t squash. You’re forced into a hold­ing pat­tern that dev­as­tates your econ­omy with­out build­ing the herd im­mu­nity that would let you come out of it. Even­tu­ally, you’re forced out, and you end up with the same wave you would have had ear­lier, ex­cept now you’ve used up your abil­ity to lock down your pop­u­la­tion. You don’t have the ca­pac­ity to squash.

How­ever, once you have sub­stan­tial amounts of in­fec­tion, that pushes things over the top. The peo­ple in­fected are not ran­dom, they re­duce risk more than pro­por­tion­ally, and thus are able to drag places with ear­lier higher in­fec­tion rates into a sta­ble place.

Thus, we need to do some com­bi­na­tion of im­prov­ing the effec­tive­ness of our lock downs, and al­low­ing peo­ple to be­come in­fected fast enough to mat­ter.

Un­less we think we can af­ford to sus­tain things un­til a vac­cine or very effec­tive treat­ment comes along. That doesn’t seem that likely at this point.

If we can’t come up with some­thing bet­ter and can’t ex­per­i­ment, the log­i­cal re­sponse to this situ­a­tion is var­i­o­la­tion. We could in­fect the young and healthy on pur­pose, with low viral loads, to cre­ate enough im­mu­nity to turn the cor­ner. Alas, we are highly un­likely to do this.

Thus, sec­ond best solu­tions prove nec­es­sary.

Reopening

Which means that a par­tial re­open­ing soon, as crazy as it seems and as mis­guided as the rea­son­ing be­ing used might be, starts to make some sense.

As does open­ing tat­too par­lors and gyms in the first wave, as in Ge­or­gia. At first I thought, what the hell, that doesn’t make any sense. Lob­by­ists at their worst. That per­haps makes sense for gyms, but I doubt any­one is es­pe­cially in the pocket of Big Tat­too.

Open­ing tat­too par­lors, where so­cial dis­tanc­ing seems ob­vi­ously im­pos­si­ble, and which are ob­vi­ously com­pletely inessen­tial, is a way of se­lect­ing for peo­ple most likely to go around get­ting them­selves in­fected, and also most likely to be rel­a­tively young and healthy, and gives them some­where to ex­pose each other. Gyms can serve a similar func­tion.

There’s also the fact that given what we know, a lot of the things be­ing closed make things worse rather than bet­ter. Or at least, a full clos­ing of them is worse than a par­tial clos­ing.

If you open restau­rants at 25% ca­pac­ity, it’s not ob­vi­ous at all that this in­creases risk. I think there’s a strong case that it de­creases it, in­stead. Gro­cery stores are over­loaded. By go­ing to the restau­rant in­stead, you re­duce den­sity at the gro­cery store, re­duc­ing wait times for oth­ers and the num­ber of in­ter­ac­tions while gath­er­ing items. You re­duce strain on the sup­ply chain, which re­duces how of­ten peo­ple need to make trips to get what they need. You also en­able places to get enough busi­ness to stay open that would oth­er­wise have to close, al­low­ing them to offer take­out and de­liv­ery.

You might already be win­ning at that point. If you could also con­vince ev­ery­one to eat in silence, and or­der off of apps on their phones? I’m guess­ing you’re win­ning. If you can do out­side seat­ing, you’re definitely win­ning.

It re­ally could go ei­ther way. It po­ten­tially helps a lot in in­creas­ing our abil­ity to sus­tain other mea­sures for a longer pe­riod.

Next Steps

Due to the speed pre­mium, I’m get­ting what I have out there now rather than wait­ing for next steps. It’s pos­si­ble all of this is a lit­tle se­lec­tive, yes­ter­day (4/​29) was an aber­ra­tion of some kind, and things are bet­ter than they look.

The next step would be to look at the data state by state, area by area, and try to make bet­ter sense of it. I worry I’m mostly talk­ing into the wind with these analy­ses. I hope not. I do know I’ve made enough of a differ­ence, in a few small places, to jus­tify con­tin­u­ing to try.

If I know you, or I should know you, and you’d be in­ter­ested in chat­ting about what’s go­ing on and helping figure things out, and per­haps even helping get bet­ter de­ci­sions made, or just want to let me know I’m not shout­ing into the wind, give me a buzz how­ever you typ­i­cally reach me. I’m at the point where it starts to make sense for me to start writ­ing code. If you know me at all, me writ­ing code is a sign that di­vi­sion of la­bor has prob­a­bly gone hor­ribly wrong.