Coronavirus Work from Home Model | Hacker News Reader

Coronavirus Work from Home Model

docs.google.com

83 pointsSQL221943 comments

I’d genuinely like to hear about it if I’m off base, but I took one look at the 2nd tab named something like “When to Close The Office” and it appears to contain dozens and dozens of assumptions. I can’t imagine that this calculation is meaningful in any way. The errors don’t average out here; the curve we’re modeling is exponential, tiny errors in assumptions can lead to roughly 100% error in the output. Isn’t there a word for when a lot of math and complexity and specific numbers with names imbue an article with some extra sense of being true or having authority?

The virus is blooming right now. If a business intends to close or start a work-from-home phase, then it should be happening right now, there’s no calculation to make and no risk model that’s going to tell you to wait another week or two.

There are legitimate reasons some businesses shouldn’t close, but those reasons don’t depend on how many people live nearby, how many are currently infected, etc..

>Isn’t there a word for when a lot of math and complexity and specific numbers with names imbue an article with some extra sense of being true or having authority?

That's Data Science isn't it?

Also forecasting adoption of new technologies when there's a lot of uncertainty about basic assumptions. Organizations that do this sort of thing regularly have these sophisticated methodologies and models. But so much comes down to qualitative assessments.

>Isn’t there a word for when a lot of math and complexity and specific numbers with names imbue an article with some extra sense of being true or having authority?

Quantitative fallacy?

That’s not bad, and I think it does apply here. https://en.wikipedia.org/wiki/McNamara_fallacy

> I can’t imagine that this calculation is meaningful in any way. The errors don’t average out here; the curve we’re modeling is exponential, tiny errors in assumptions can lead to roughly 100% error in the output. Isn’t there a word for when a lot of math and complexity and specific numbers with names imbue an article with some extra sense of being true or having authority?

The official term for the issue is False Precision [1], though I usually just say “false sense of precision”.

If someone does actually carry through and calculate margins of error due to their varied assumptions, it lets the reader decide. Unfortunately, I’ve rarely seen folks treat errors as anything other than independent, unless they’re actually building a model that they evaluate with a Monte Carlo method (e.g., system temperature and cpu busy-ness are correlated when machine failure rates).

[1] https://en.m.wikipedia.org/wiki/False_precision

That’s pretty good, and does apply here I think. But FWIW I was thinking about something slightly closer to Occam’s Razor and/or the Law of Parsimony, but specific to how adding more data or assumptions or ‘evidence’ can sometimes make something seem more true, even though it’s the opposite.

Yeah, that’s how you ended, but I think the root cause in the piece you linked isn’t “deception” but false precision. The false precision is what gives the sense that it seems true.

Perhaps this is a variant of curse of dimensionality for errors? That is, if each piece of evidence is only 80% “likely”, 80% to the Nth drops like a stone.

Perhaps Dave’s Law? :).

Ooh, I like the sound of Dave’s Law. :P

More seriously, I love the idea of the curse of dimensionality for errors. Random tangent, but phrasing it that way totally reminds me of the phrase ‘flaw of averages’ and the story about US jet fighters.

I wasn’t thinking the author was being intentionally deceptive. It can feel like accounting for more things makes something more accurate. Sometimes it’s actually true, sometimes failing to account for all the relevant inputs is a real problem. It’s legitimately difficult to know when you’ve accounted for too little or too much, right? I guess this is just one of those times when each piece being accounted for is making the result less accurate rather than more accurate, because the data truly does have a false precision, like you said.

Didn’t Google go on work-from-home a week ago? I did mid-week last week and I’m feeling stir-crazy already, but starting to wonder if the most responsible thing to do wouldn’t be to catch the virus right now and quarantine myself; guarantee I don’t spread it around.

Before the pivot to Covid content, the previous articles from the same creator included ...

"How to Deliver Your Funny Speech" "What The Rise of Skywalker Can Teach about Storytelling" "8 Reasons Why Your Ad Attribution Approach Is Wrong" "What I Learned Building a Horoscope That Blew Up on Facebook"

Not for me, Clive.

FWIW, there's no sophisticated science in that Medium article. All of the things in there come from basic understanding of how exponents work, plus thinking for 10 seconds about what the graphs from Wuhan from February showed. Author's biggest contribution is quality explanation of these points that's comprehensible by a wide audience. That doesn't require skills in epidemiology, but precisely the ones the author has.

Nobody seems to be taking real estimated number of infections, instead they take number of confirmed cases. Confirmed cases don't include - infected people in incubation phase, infected people with non-severe symptoms that were not scanned (age weighted on probability and mapped back on age weighted population) - then you calculate dubling on this estimate, you may have surprise when geometric growth shoots above hospital capacity.

All of above can be calculated with one exception - kids will have to be excluded because we don't know if they carry/infect and don't get symptoms or they don't have symptoms because they are immune.

But this calculation is easy enough to give rule of thumb estimate formula to apply to official confimed cases to get back real estimated number of infections.

I also read:

Infected people measured also includes dead people who would have died anyway and just happen to be infected before their death.

And more than 50% of people without symptoms were tested false positive.

It's pretty hard to get a false positive from the Coronavirus PCR test we're using. False negatives happen, but false positives are extremely rare.

Good to know, thanks.

No and no.

1) Epidemiological papers usually estimate the true number of cases, based on an estimate of the percentage of cases getting tested. The definitely include this in their outlook, and have correctly predicted the outcome of the epidemic in China. Saying "nobody" is doing that is totally incorrect. There's literally people's jobs who it is to model epidemics, who have PhDs and devoted their entire careers to it. The amount of ignorance about the quantity and quality of scientific research out there just floors me.

2) We do know now that kids are just as susceptible to be infected as adults [1]. We don't know exactly what their role in transmission is, but you could conservatively assume that they probably spread it just as much as everyone else (speculation).

Some clinical characterization of cases in children are available in refs [2] and [3]

[1] https://www.medrxiv.org/content/10.1101/2020.03.03.20028423v...

[2] https://www.nature.com/articles/s41591-020-0817-4

[3] https://www.nejm.org/doi/full/10.1056/NEJMc2003717

I find "When to Close The Office" with "risk you want to take" set at 1% profoundly disturbing.

Sure, from a company perspective it makes sense: 2 weeks of medical leave for everyone now + a few new hires later. If you multiply it by the risk factor, it may make good business sense.

Yet, if there is an infection, it will spread to their families, and to other people, accelerating the infection spread (and potentially killing orders of magnitude more people than within office). It is irresponsible, to put it mildly.

Author found here:

https://medium.com/@tomaspueyo/coronavirus-act-today-or-peop...

That's quite a lot of sensationalism for one article. While it's true that the situation needs to be actively dealt with, phrases like the below help no one:

> Now it’s a pandemic that nobody can stop.

It's incredibly weird to read that in the same article that also points out how many countries did in fact stop the spread. There are many other similar examples in the text.

There are only two countries (China, South Korea) and two city-states (Hong Kong, Singapore) that have managed to stop the spread, I wouldn't call that "many". For comparison all of Europe plus the US and Canada in North America are navigating in the dark, with no visible end in sight.

Now you're being sensationalist - Those places that stopped their local epidemics were A) the first ones to have the epidemics, and B) took effective action to stop it.

Saying there's "no visible end in sight" is meaningless in this context. If you model the epidemic, as many researchers have, you will see when the end of the epidemic is.

> If you model the epidemic, as many researchers have, you will see when the end of the epidemic is.

Do you have a link to a study of said researchers saying when the epidemic will end? I'm asking because I've been following this whole epidemic thing since just before Wuhan was put into lockdown, I currently live in Europe and I don't see us getting out of the dark with the current measures being taken.

It is generally accepted that the pandemic will end once just about everybody has caught it, because even those cases where people are thought to have re-caught it, if you assume they really happened, are a small fraction of the population.

> It is generally accepted that the pandemic will end once just about everybody has caught it

Yeah, that's probably how the Black Death ended in the mid-1300s, everything "ends" at some point, I was inquiring of the possibility of this "ending" without the obituary pages in local newspapers growing ten-fold in a matter of 7 or 8 days.

You didn't link to these many models, so I'll ask directly: when they model to the end, do 500M people die? Because that would be worthy of sensationalism. If instead, it's 1M people globally, I guess I agree with you.

Social distancing actually stops the spread but the Western countries will see their confirmed cases rate number drop only after 10-15 days, because they only account for symptomatic cases that underwent a test, not for real-time scans of the full population. That’s why the ‘true cases’ in the OP document are estimated at 10x-30x the confirmed cases, and governments’ lockdowns now actually aim to quarantine all of them.

I know all of that, but the first numbers that will either confirm of contradict the lockdown measures will come this Tuesday-Wednesday (for Italy), but you've still got Spain where countless madrilenos ran for the seaside this weekend (no matter the state of emergency already being instituted the day before), France actually holding elections (?!) and countless bars and restaurants still being full over the weekend, Switzerland ignoring the situation almost completely and Germany most probably fudging the numbers. The UK is a special case which I'll not get into because it's too complicated.

Which is to say that yes, nobody knows where and when the light at the end of the tunnel will show up. This positivist "attitude" of first "it's just a flu" and now of "we'll get over it in no time" has really started getting onto my nerves.

What is the UK situation?

The government saying a couple of days ago that they're going to do almost nothing and rely on what they called "herd immunity". They've since slightly changed course and today they've announced (through unofficial channels) that they're thinking of isolating the over-70 year old people (either at home or at other designated places, it's not sure for now).

I think some countries have controlled it, but I wouldn't say they've stopped it. The moment business and society goes back to normal behavior new cases will skyrocket until we can get new cases down near 0 per day.

For example South Korea diagnosed 851 cases on March 3 and 107 cases on March 14. That's a massive achievement and will undoubtedly save many lives. But there are still 100+ new cases being diagnosed every day there, and every one of those are potential vectors for future infection.

I can't load the Docs page, but the author of this post was criticized for being a non-expert and making many statistical mistakes. Here's a model from a viral dynamics researcher: https://alhill.shinyapps.io/COVID19seir/

HTTP 503 error. Any mirror?

I've seen that article a few times now. Who is he, and what are his qualifications? There's nothing linked off the medium article that I can find.

I believe this model was created by Tomas Pueyo. His full article is well worth a read: https://medium.com/@tomaspueyo/coronavirus-act-today-or-peop...

Man, I'm trying to just copy the model and keep getting the error "The file might be unavailable to heavy traffic". Never seen that before.

Yeah, Google Docs isn't really intended to handle much more than 100 or so concurrent viewers. They've optimized it for "1 million simultaneous users looking at 300k documents", not "1 million simultaneous users looking at 1 document".

Can’t seem to access it at all, it says don’t have access to it

I made a copy a couple days ago (if it's the same sheet I think it is; I can't see it now either) and I just made some additional copies to share here.

Here are five copies: (Fittingly, if other people also each make five copies, then we can cover the need in just a few generations...)

https://docs.google.com/spreadsheets/d/1qLKWavPg_c9G-RGKOAn2...

https://docs.google.com/spreadsheets/d/18wnbmV4TiaLF3mXLbD8k...

https://docs.google.com/spreadsheets/d/1y85X64AlDE93HYJqRMpm...

https://docs.google.com/spreadsheets/d/1ieS6YF0lpdvAPwliP5wX...

https://docs.google.com/spreadsheets/d/159tPZV31fl04DyzFR8ey...

a no spreadsheet model: https://staythefuckhome.com/

Preventing hospitals from being overwhelmed with this disease is like rapidly accumulating compound interest, and the dividends are people's lives. Early investment is monumentally better than late investment.

As someone specialized in the moral aspects of decision theory I unfortunately find this dubious. Any decision model makes value judgments and trade-offs between them, and normally you would have to elicit those judgments from the decision maker.

I don't understand how this could work without taking the individual value judgments into account, especially since the long-term consequences of home work differ from business to business.

The author provide a "direct link to copy" at:https://docs.google.com/spreadsheets/u/1/d/17YyCmjb2Z2QwMiRR...

However, its currently not working due to heavy traffic.

Mildly related, DRACO[0] needs funding.

[0]: https://en.wikipedia.org/wiki/DRACO

Reply on news.ycombinator.com