New York coronavirus antibody study: Why I had nothing to say to the press on this one.

The following came in the email:

I’m a reporter for **, and am looking for comment on the stats Gov Cuomo just released. Would you be available for a 10-minute phone conversation? Please let me know.

Thanks so much, and here’s the info:

Here is the relevant part:

In New York City, about 21 percent, or one of every five residents, tested positive for coronavirus antibodies during the state survey. The rate was 16.7 percent in Long Island, 11.7 percent in Westchester and Rockland Counties, and 3.6 percent in the rest of the state.

Almost 14 percent of those tested in New York were positive, according to preliminary results from the state survey, which sampled approximately 3,000 people over two days at grocery and big-box stores.

I spoke with the reporter for a few minutes and gave my quick take:

– Those California studies estimating 2% or 4% infection rate were hard to assess because of the false-positive problem: if a test has a false positive rate of 1% and you observe 1.5% positive tests, your estimate’s gonna be super noisy. But if 20% of the tests you observe are positive, then the false-positive rate is less of a big deal.

– That said, I know nothing about how the tests were done. I have no idea what exactly it’s measuring or what its error rates are.

– Also, is 21% the raw proportion of positive tests, or has the number been adjusted for false positive and negative rates?

– In any case, the 20% number seems reasonable. It’s hard for me to imagine it’s a lot higher, and, given the number of deaths we’ve seen already, I guess it can’t be much lower either.

– I can’t figure out if doing a survey at grocery and big-box stores will overestimate or underestimate the infection rate. On one hand, people who are going out to the store could be at higher risk for exposure. On the other hand, sick people would be staying at home, right? But, then again, think of all the uninfected people holing up and never going out . . . And where exactly were these stores? I keep going in circles on this one.

– In any case, this sort of survey is a good thing. We keep doing surveys in different ways, each survey has its own bias, put them all together and we’ll learn some things.

After 10 minutes I’d spewed out enough material for as many quotes as the reporter could possibly have wanted from me.

But then I got off the phone and started thinking . . . Do I want to be quoted as saying this estimate is pretty good? Or that it’s flawed? I don’t know. There’s no data, no report. Based on the news report, the study seems to have been conducted by the state health department, and that sounds like a good sign. In general I’ll have more trust in a study from the state health department than from some Stanford professors. I’m not joking here: the health department are professionals and they don’t have the same incentives that academics have to hype their research. But, still, I have no idea what’s going on.

So I sent an email to the reporter:

Hi—just thinking more about this, I’d prefer if you not quote me in this article. I make this request because I don’t think I have anything particular to add here, and it would be better for you to quote an expert, or to not have a quote at all. I have not seen the data or even any report undelying the claims from this New York State study, so I can’t really make any reasonable statements.
Thanks for understanding.

There was more to say about those California studies because (a) the people who did those studies released some partial information, (b) the thing about the false-positive rate mentioned above, and (c) the people involved made a bunch of loud and weakly-supported claims. This new study is more of an official number, so there’s not much that can be said from a statistical perspective until the results are unpacked in some way.

69 thoughts on “New York coronavirus antibody study: Why I had nothing to say to the press on this one.

    • Yes, and here is another story for a class on Communicating Statistics…

      Perhaps it would help the public and elected officials understand what has happened.

      In this fable, we have a candidate for governor who has hired an expert in campaign modeling to advise about his campaign’s progress and potential success. The expert does some testing (survey of voter support by phone calls) and records the number of positive answers (#yes answers) over the key month of the campaign, as follows:

      Day 00 — with 100 phone calls and 10 yes answers.
      Day 03 — with 200 phone calls and 20 yes answers.
      Day 06 — with 400 phone calls and 40 yes answers.
      Day 09 — with 800 phone calls and 80 yes answers.
      Day 12 — with 1600 phone calls and 160 yes answers.

      Now, you can see that the candidate’s support is staying about a constant 10%.

      But the expert does not want to show that relevant metric,
      and instead only shows the results and graph for the increasing number of positive responses.
      Then the expert determines from the graph that the number of positive responses is
      doubling every 3 days, and so using that doubling rate his model projects that by the election day,
      the candidate’s support will have increased to well over 50% of the voters.

      Do you see anything wrong with this analysis and model result?

      This is about the same thing that has happened with the Covid modeling, and with the
      media’s presenting the scare curve of increasing cases (positive tests) which was mainly
      an artifact of the increasing rate of testing, instead of showing the relevant metric.

      • The facts don’t bear that out. Positive test rate in Germany has ranged from 3.1% to 9.0% and is sinking now.
        https://www.rki.de/DE/Content/InfAZ/N/Neuartiges_Coronavirus/Testzahl.html

        The death rate is not dependent on testing, and it qualitatively tracks the case rate. (The deaths/completed cases ratio by age does some weird things I don’t fully understand.) It’s a lot over the long-term average and peaks higher than flu seasons in many regions.

        Your model is too simplistic.

      • What’s missing from this analysis is that there’s a causal relationship going the other way where symptomatic cases drive testing. The difference between random testing (like this one) and non-random testing-on-request is instructive here. Until you have a proper model of what is driving testing, making the assumption that testing rate is an independent variable can be deeply deceptive.

        • I mean, consider how radically different your analogy would be if instead of the expert calling people to ask their opinion, people instead phoned into the office to offer their support…

      • That’s a pretty strong claim that “about the same thing had happened with Covid modeling”. You’ll need to back that up with some examples of modeling that ignored changes in test prevalence as one of the driver of changes in case count.

        I don’t doubt that that has happened in the media. Of course, the media has also been all over the inability to get adequate testing.

        And you are of course wrong about the “curve of increasing cases” being mainly an artifact. It’s not like God said in the 8th day they’re will be 200,000 confirmed positives in NYC. We had to get they’re from zero somehow. A true-infection “scare curve” had a somewhat different shape, but i’s

    • Anonymous

      Re: ‘Proximity gives you an *idea* of who was infected. But it misses a lot. What did they touch in their proximity? Before touching these things, did they happen to cough, sneeze, tough their face, tough their mask, etc., that could have transmitted virus? Who then touched a said surface that was also in this proximity during the next hour? 2 hours? 6 hours? 1 day?’

      Very good point. This is a complicated situation.

      • I believe surface transmission has never been proven to happen in the real world?
        And it’s easily preventable, don’t touch your face unless you’ve washed your hands before.

        What you need to track is distance and time, under 1.5m/6ft and over 15 minutes constitutes high risk, with a 5% attack rate (11/217) in the Munich study (“Outbreak of Covid-19 in Germany”, Lancet pre-print) for non-household contacts.

  1. I had an idea on how to correct for the sampling bias and I wondering whether anyone has any comments on it:

    One of the key numbers we are trying to estimate is the IFR and in order to do that we need a decent estimate of the number of exposed and the number who die from that exposure. This study gives us a good estimate for the prevalence of COVID-19 in people who go to stores frequently, with people getting sampled roughly according to the frequency with which they go to stores.

    It seems like the natural way to correct for this is to determine the supermarket attending frequency for everyone who has died. We could then re-weight the deaths to get an estimate for the amount of deaths in an equally biased sample.

    That should allow us to get a decent estimate of the IFR without too many other assumptions.

    • “This study gives us a good estimate for the prevalence of COVID-19 in people who go to stores frequently” — you need to support this. If 70 people go once a week and 10 people go daily, the survey catches 20, both groups at equal rates.

      The interesting question is, how likely are people who have survived an infection to go out shopping for others?

    • Besides Medel’s point about varying sampling rates, there’s also the *current* illness rate of people who shop, which would presumably also be correlated with frequency of shopping. Currently symptomatic people would be much less likely to be currently shopping so they’d also be excluded from the sampling population, not just those who have died.

  2. > In any case, the 20% number seems reasonable. It’s hard for me to imagine it’s a lot higher, and, given the number of deaths we’ve seen already, I guess it can’t be much lower either.

    Does this mean you could have guessed ~20% before seeing those stats?

    • Guy:

      I don’t think I’m on record with that one, but, yes, 20% was kinda what I was guessing based on everything I’ve been reading. But this was not any sort of informed technical opinion; it was more of me taking a mental average of everything the blog commenters had been saying. If a well-conducted study reported an infection rate of 10% or 40%, that would not shock me either. If it were 5% or 80%, then I’d be pretty surprised, but, again, my understanding of these numbers is shallow and is pretty much just the product of some combination of introspection and shallow reading of the news. As I wrote in my post a few days ago, it’s hard tor me to imagine a really high rate, given that I haven’t heard of anybody I know having the disease, just about nobody at Columbia had it back in March when they shut the place down, etc. But a really low rate doesn’t seem possible because of all the deaths they’ve had. But even I could’ve guessed 20% or something like it before seeing the latest numbers, that wouldn’t make such a study useless. First, there’s a big difference between 10% and 20% or between 20% and 40%. Second, at some point it’s good to know things and not just be guessing.

      • If you live in New York and haven’t heard of anyone you know having the disease, shouldn’t your estimate be a lot lower than 20%? Or do you think it’s more plausible that the people you know are very unrepresentative of the larger population?

        • That’s true. Anything between 5% and 30% seems possible to me based on the number of deaths. It does look like the death rate for this illness can get pretty high in vulnerable and unhealthy populations. That’s why 5% doesn’t seem that crazy to me in NYC. But who knows.

        • I assume that when Andrew says he doesn’t know anyone who has had the disease, he means he doesn’t know anyone whom he knows has had the disease (what else could he mean?). Apparently quite a few people can have no symptoms or very mild symptoms and may therefore be infected without knowing it. And I’m guessing Andrew hasn’t kept in touch with everyone he knows, so if some of his acquaintances had the disease he might not know it.

          And of course, Andrew’s acquaintances may be less likely to get infected: more able to avoid going out (e.g. not working in ‘essential jobs’) that sort of thing.

          Basically I agree with Andrew’s math: It doesn’t seem like the infection rate for New Yorkers as a whole could be as high as, say, 50%, because even asymptomatic infections and selection effects etc. don’t seem like enough for Andrew not to have known a few people who were known to have been infected. On the other hand, single-digit percentages seem unlikely because of all the deaths. 20% seems possible. I’d probably take the under but I wouldn’t bet the farm.

        • Reflective of the attributes of our society and our healthcare system, I’d be willing to bet that pretty much anyone who lives and works in poor, minority communities knows people who have gotten a symptomatic case, and likely knows someone who has died.

      • > I haven’t heard of anybody I know having the disease

        Adjust that for the fact that Andrew doesn’t follow social media. I don’t either, but …

        I know of two families in NYC who got coronavirus. One’s a neighbor and one’s a pretty close friend I saw regularly before quarantine. I’ve met people who have coronavirus, like one of my father’s best friends. I’ve met people who’ve died of coronavirus, like Floyd Cardoz, who I photographed back in the 90s. One of my childhood friend’s mothers died of coronavirus in the Detroit area.

        • I haven’t heard of anybody that I know well having corona virus — but I have heard of someone I’ve met having it (and dying of it): John Conway.

      • I live far, far from NYC so maybe it’s not unexpected but I don’tt have first-hand or second-hand (friend of a friend) connection with anyone who knows they’ve had it.

        I know of two people (one a relative of mine and the other the daughter of a friend) who would have bet anything they had it about a month ago. Really sick, high fever, headache, dry cough, pretty much all the symptoms. They both tested negative. It makes me wonder what the false-negative rate is, in practice, for those nasal swab.

    • Do we know which antibodies they were testing for? Apparently it is a test designed by NY state.

      Given that it would take 7-14 days for antibodies to show up in your blood, you’re actually looking detected people who were infected a week or two ago.

      Or did they try to adjust for that?

  3. Would the 3.6% in the rest of the state represent an upper bound for the false positive rate for this test – or am I just too tired to think this through coherently?

    • If the test has cross-reactivity to the common cold, then where are you more likely to have caught one? NYC or Albany? or Southern California?
      There are still a lot of known unknowns.

  4. “I can’t figure out if doing a survey at grocery and big-box stores will overestimate or underestimate the infection rate. ”

    That’s great news. So now instead of figuring it out from poor data with a bunch of unverifiable assumptions, people will have to go out and actually measure something useful.

    Just guessing though, it’s not the type of store – everyone needs groceries – it’s the time of day and the neighborhood. Measuring at what are normally peak hours, you’re probably measuring people with the highest risk tolerance and thus the people most likely to be infected. A rep sample might be measuring every 100th customer from open to close at several comparable grocery chains in several different parts of the city selected for capturing the income distribution seems like a starting point.

  5. Various articles such as the New York Times article mentioned in the post claim that the shoppers were picked randomly. There are clues though that this is not correct in at least some cases. Example from https://www.timesunion.com/news/article/Antibody-testing-at-Capital-Region-grocery-stores-15212590.php:

    “Now, more than a month later, Battista still doesn’t feel quite right. She’s coughing more than usual and feels short of breath at times. She works as a nurse in the neonatal intensive care unit at St. Peter’s Hospital in Albany, and said she worries about bringing something home to her three kids.

    That’s why, when she got a text on Monday from a work friend who said testing for coronavirus antibodies was being conducted at the Price Chopper in Albany’s Westgate Plaza, she dropped what she was doing, donned a mask, and left to get tested.”

    • 1. For the record, this was intended to be a random sample of shoppers. From http://www.cnn.com/TRANSCRIPTS/2004/19/cnr.14.html:
      “CUOMO: The antibody first thing that we’re doing now this week is a random sample. So it’s not like testing where somebody can ask to be tested. It has to be a random sample. That is conducted throughout the state. Thousands of people get tested.

      So we know this percentage of the population had the antibodies. That is not a test where a person can call up and say I want to be tested or go to a place. That has to be done on a random sample basis.”

      2. Another purported anecdote: from https://twitter.com/biancazumpano/status/1253359632439029761: “at my local store a lot of people were going specifically to take the test and lining up instead of it being randomized, so prob plenty of people who thought they had it before or had been around someone but had no symptoms so couldn’t get a normal test.” From https://twitter.com/biancazumpano/status/1253506830090854402: “cuomo announced they were going to start antibody tests a bit ago. they were randomly asking at the store for a bit and then word spread pretty fast around town that it was going on and people started going to the store specifically to get the test.”

    • It doesn’t matter that shoppers were picked randomly. The population of shoppers is not a random sample of the NY State population at any moment in time and especially not during a pandemic with stay-at-home law and tons of people ordering delivery. Any of these studies has to explain why in a world of rationed testing (triage, whatever you call it), that they don’t selectively attract people who have symptoms and reason to believe they have the virus, and was denied at test or could not afford a test before they learned about the free test. Release that study and let the scientists speak!

  6. Had the quote been from a researcher involved in the study, I think it would be fine for you to express caution about their conclusions to the reporter, despite not seeing a write-up. At the very least, it’s not a good sign for the quality of a study when the researchers rush to publicize their findings instead of rushing to properly write them up. But the quote was from the governor, who is a non-scientist and the researchers’ “boss,” and his discussion of the findings is surely beyond their control. This is in contrast to quoted in previous posts from authors themselves.

  7. Roughly speaking, if you consider New York’s study reasonably reliable, and if we consider 3.6% as a lower bound on prevalence, then is the Santa Clara study’s estimated prevalence in the range of 2.49% to 4.16% all that unreasonable? Is there a good reason to expect, based on whatever information we have about upstate New York’s demographics, population density, currently reported cases and deaths, and known earliest cases, that this an apples-to-oranges comparison? (And if so, what direction would the difference be in?)

    I’m just doing a back-of-the-envelope calculation here, but assuming 3.6% prevalence in the state of California, using a population of 39.51 million and current case count (from https://www.worldometers.info/coronavirus/country/us/) of 39,555, would give a ratio of true cases to reported of 35.96. We know that reported cases will continue to rise which will lower this number, but their estimate of 50x to 85x sounds potentially reasonable, given that estimate was for early April.

  8. A problematic claim in the New York Times article is “the death rate in New York from Covid-19 would most likely be far lower than previously believed, possibly 0.5 percent of those infected.”
    This divides confirmed (15,000) or suspected (15,000 + 5,000) deaths through estimated total infections, which might miss many deaths.
    For getting the actual death rate one also has to estimate the number of deaths based on excess mortality.

    • I don’t see where you get the claim that the death rate is lower than estimated. .5 is in line with most estimates I have seen, although on the lower end. If you believe 21% of NYC is infected, then you get a death rate of .6 to .9 depending on how how count probable deaths. .5 would be consistent with the .21 being an undercount.

      • I am not arguing at all whether the death rate matches expectations, or about the 21% estimated infections.

        My point is that it is important to also double-check by how much the other side of the division, the numbers of confirmed and suspected deaths, might be an undercount.

        We do not know how many “natural deaths” in retirement homes might be caused by COVID-19.

        COVID-19 is known to often cause blood clotting, and there are anecdotal reports of healthy people in their 30s having a stroke caused by asymptomatic COVID-19. For how many current deaths attributed to a stroke or heart attack is the root cause COVID-19?

        Looking at excess mortality could indicate whether or not there might be a huge amount of hidden mortality.

  9. There really isn’t anything quantitative to comment on if they don’t release their data. 21% of New York City infected ~.2% of the total population there has died, feels about right, but feelings aren’t worth much.

    I’m more focused on calibrating my Bleach/Lysol injections, and determining which orifice to insert my UV lamp into.

  10. “In any case, the 20% number seems reasonable. It’s hard for me to imagine it’s a lot higher, and, given the number of deaths we’ve seen already, I guess it can’t be much lower either.”

    Typical Bayesian ;-)

  11. Andrew: I dunno, I think your quick take is still valuable, even if your overall takeaway is “I dunno”. Because there will be less scrupulous ‘experts’ out there with more of a tendency to make pronouncements with false confidence, and it boosts publication bias if only confident quotes get published in the news media.

    • Yeah, since this is about being picky about what we say, I don’t think the response is that good either. It’s like good in the way a (-infinity, infinity) prediction has good coverage.

      > and it would be better for you to quote an expert

      > or to not have a quote at all.

      Like Zhou said, this is practically just telling the person to talk to someone else until they get something interesting. That’s not good. Also you are an expert. If you think there’s someone else out there who is more of an expert, say who that is.

      > I have not seen the data or even any report undelying the claims from this New York State study, so I can’t really make any reasonable statements.

      Well then say, “Well I don’t know if I should trust these numbers. Here are the things I want to know to evaluate that”.

      You could also say, “I need to take a pass through the article before it is published to approve how I am quoted”.

      I don’t know if that is possible, but it sounds like you aren’t comfortable being quoted by this reporter and that’s not good for anyone. If you care about such things, maybe start asking for more editorial control (maybe even post-publishing control) until you are comfortable (presumably the outlet can do something — they took the time to contact you for this).

      • “this is practically just telling the person to talk to someone else until they get something interesting. That’s not good. ”

        Nothin’ sayin’ they wouldn’t do that anyway.

        I think Andrew made the right call. He’s demonstrating to others the proper way for a person to assess a study that they know nothing about.

        • > Nothin’ sayin’ they wouldn’t do that anyway.

          Yeah but the message for the Stanford people was, “take 5 seconds to talk to a statistician!”

          It’s a different situation here, but the message to the journalist here seems like, “statisticians know nothing.”

          In fact, Andrew knows a lot! The problem with interpreting these numbers is that all the information behind them wasn’t published. If Andrew had requested this in a very public way, then maybe that would help it come out! What is that information that we need? No reason the journalist should know — they’re not statisticians! That’s why they asked the expert!

          And this is important information. Like, the journalist probably doesn’t know whether or not there was some public research behind the report. The answer is there isn’t, as far as we know. There’s not some secret Ivory-tower advising network here — that’s probably half the reason this reporter e-mailed Andrew!

          He can, of course, send an e-mail to an anonymous NY state government e-mail address and hope it bounces around to the right people so they know the correct things to report next time. That seems rather unlikely to work.

          Missed opportunity, honestly. We should expect experiments like this to come with testing protocol info and such. One line in a random internet article isn’t gonna change this, but may as well put it out there. But I assume if it happened once, it’ll happen again. And it’s totally up to Andrew whether or not he wants to talk to journalists on this stuff anyway. But he definitely knows things.

        • ‘ the message to the journalist here seems like, “statisticians know nothing.”’

          ?????

          I don’t know where you get that. Andrew hasn’t seen the specifics of the paper. How can he comment knowledgably on it? The fact that he’s withholding comment until he sees the details of the paper is hardly a message that “statisticians know nothing”!!

  12. “(a) the people who did those studies released some partial information, (b) the thing about the false-positive rate mentioned above, and (c) the people involved made a bunch of loud and weakly-supported claims. This new study is more of an official number, so there’s not much that can be said from a statistical perspective until the results are unpacked in some way.”

    I don’t think there is that much difference in how all of these studies have been reported/released or the incentives faced by the researchers. At least in the CA studies, there was enough info released that you could actually critique the studies. And who knows what kind of pressure the government workers were under to release this. Is it good news or bad news? Seems that this will most likely be taken as good news (Covid is more widespread and less deadly than we thought), and I assume Cuomo (the researchers’ boss) would probably want good news out in the open quicker than bad news. My perspective is that of a government statistician, so I’m not knocking them; I just think the incentives are not that different.

    As an aside, I have been baffled as to the reason I can download a daily updated CSV on the number of cases New York state has, but the deaths I can only find on the health department’s dashboard display.

    • > an aside, I have been baffled as to the reason I can download a daily updated CSV on the number of cases New York state has, but the deaths I can only find on the health department’s dashboard display.

      +1 here. And same for NJ.

      It is 2020. How is it not possible to de-identify the raw patient data and post it? My understanding is that charts should all be digital now, too. What is the hangup?

      Why post it to some dashboard already stratified? Is that not a form a censorship? For example, NYC posts deaths age 18-45. How many of those deaths were age 40-45? How many age 44-45? Why lump all underlying conditions together? I understand it helps with reporting to general public, but that doesn’t mean you cannot or should not make the raw data available as well.

      Agh!

  13. Public health academics/researchers do rely on state health statistics from my reviewing article citations. It’s that nearly every statistic comes with assumptions that are not featured clearly.

    Incidentally, what would be the potential incentives operating with respect to Santa Clara and LA studies?

  14. The numbers mean NYC deaths now generally agree with the death rates I see in the MA official numbers. That is, when you take them apart by age. That suggests the problem in NYC was that more people were sickened, and that this pushed their hospital system much closer to breaking. The hospital capacity in Boston was never truly challenged. In that regard, when the state finally started pushing out data about hospital utililization – which I think included the extra beds they arranged (which have mostly been unused) – I was startled to see we had over ⅓ available. That’s now approaching 50% and the state is trying to convince people who have other problems to get treated; as you must know, the number of stroke, heart attack, etc. patients has dropped by anywhere from 40-60% and now the medical system is realizing they’ve got to address that harm.

    Just to mention, the Globe looked at excess deaths. They asked the state for death records for March of this year and March of years back to 1999. They then made an elementary error: they calculated the average from 1999. Problem: the state’s population has grown substantially and has become noticeably older. This is reflected in the 5 prior year average, which reduced the excess deaths from a scary number (500+) to something understandable (around 280), assuming the excess represents the same really old group of 80+ people.

    • My off-the-wall hypothesis for fewer stroke and heart attack admissions is that the stress of driving in traffic causes these, and stress in being face to face with the boss, and other workplace-related stresses, and that we’re seeing a health benefit of stay-at-home. Would be nice if it turned out true!

      • I think it’s pretty well established that air quality affects strokes and heart attacks. And air quality around the world is dramatically improved. Also diet, and I suspect people are eating healthier, fewer big macs and fried chicken and soforth. I can think of a number of reasons why at the moment we might have much less heart attacks and strokes.

        But I don’t know the magnitude really. So although those things may be acting, they may not be making as big a difference as is being seen at hospitals. I don’t know.

        • Excellent point about the air quality!

          “The estimated odds ratio of ischemic stroke onset was 1.34 (95% confidence interval (CI): 1.13, 1.58; p<0.001) following a 24-hour period classified as “moderate” (PM2.5 15–40 μg/m3) by the US Environmental Protection Agency’s (EPA) Air Quality Index compared to a 24-hour period classified as “good” (≤15 μg/m3). Considering PM2.5 as a continuous variable, the estimated odds ratio of ischemic stroke onset was 1.11 (95% CI: 1.03, 1.20; p=0.006) per interquartile range increase in PM2.5 (6.4 μg/m3). The increase in risk was greatest within 12–14 hours of exposure to PM2.5 and was most strongly associated with markers of traffic-related pollution."
          https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3639313/

          This means we can expect a drop in strokes by 25% if the air quality goes from "moderate" to "good".
          If it's traffic-associated, and traffic has dropped even below "good" levels, who knows?

  15. Andrew: Bravo on this decision. I was so glad I read to the end to know it. Our only comment should be that it is strange and irresponsible for the Governor to dictate this result to us without releasing any data or even a proper report. I’m more than a little disturbed by this front-running. Why aren’t these reporters pressing for the scientific paper or the data? They should be doing that instead of asking us to comment on someone’s soundbite.

    • > Our only comment should be that it is strange and irresponsible for the Governor to dictate this result to us without releasing any data or even a proper report.

      This!

      This frustrates me with Andrew’s response. Andrew should have outlined for the reporter:

      1. Who exactly is the other expert that the reporter should be seeking out
      2. What information that needs to be included in the scientific paper/tech report/whatever that would make these numbers believable

      Sure Andrew didn’t tell him something he didn’t know, but he didn’t tell him the things that he did know! But small mistake or whatever. It is better to say nothing than just rolling out edgy statements.

      > Why aren’t these reporters pressing for the scientific paper or the data? They should be doing that instead of asking us to comment on someone’s soundbite.

      Honestly the reporter found Andrew, so I’ll give them points for that. Of the state, Andrew, and the reporter, I don’t think the reporter is the one to critique here.

      • Ben:

        I think you’re right. What happened was, I spoke with the reporter and said a bunch of things, then when I got off the phone I realized that some of the things I said could be misleading, so I backed out for reasons of damage control. If I’d been quick enough on my feet, I would’ve given a quote such as suggested in this comment thread. Or just pointed the reporter to Cass Sunstein.

        • Reporters could misquote intentionally or unintentionally. I think best practice for an expert should be to insist on providing written comment and proofreading the final quotes.

  16. Hi,

    A wide range of infection percentages are consistent with the fatality and hospitalization rate in NYC. At 11800 deaths and assume that there is about half delayed mortality (consistent with South Korea) – say 23600 total deaths if no new infections. At 1% IFR (Infection Fatality Rate) would be 2,360,000, with NYC population of 8.4 million that is 28% of the population can be expected to be infected. If 3% IFR would be 787,000 and thus 9.4% of the population can be expected to be infected. So I think 20% could be in the ‘ballpark’ but is probably a bit high – I’d bet 15%.

    • > At 11800 deaths

      That’s the number reported by the state for NYC (11817). The NYC health department reports 16270 (10961 confirmed and 5309 probable).

      “Data Collection Differences

      The State Department of Health reports data on deaths from:

      The State Hospital Emergency Response Data System

      Daily calls to hospitals and other facilities that are caring for patients, such as nursing homes

      The NYC Health Department reports data that reflect both:

      Positive tests for COVID-19 confirmed by laboratories

      Confirmations of a person’s death from the City’s Office of the Chief Medical Examiner and our Bureau of Vital Statistics, which is responsible for the registration, analysis and reporting of all deaths in the city.
      Due to the time required by the City to confirm that a death was due to COVID-19, the City’s reported total for any given day is usually lower than the State’s number.”

  17. 3.6% for the entire state of NY and 21% for NY City isn’t unreasonable, that effectively makes the true number of cases 2-3x higher than the confirmed count for the entire state and 10x for NY City. That being said, I haven’t seen any error bounds on those estimates. Presumably a reasonable bound on the proportion is going to be +/- 1 to 2% for the entire state. That still makes the overall multiplier well below 10x for the entire state. That just requires 90% of cases to be asymptomatic or unavailable for testing. That’s reasonably consistent with most estimates of 60-80% unsymptomatic and the fact that only 20-30% of tests come back positive.

    • NYC hasn’t been able to test all of its *symptomatic* cases for a long time.
      20%-30% test positive rate is *a lot* and indicative of a large undercount.
      Get that rate under 10% and I’ll believe that you’re counting most symptomatic cases.

  18. Dude, just admit you were wrong on this. Get off now before the data just overwhelmingly indicates you were way off base, caught up in the liberal status quo and piled on. It’s OK, happens to everyone, but anybody in statistics could have seen for some time where this was going. More and more data is going to come out on this, and you’re going to stick to making some esoteric point on this specific study and end up looking like a fool for suggesting the outcome was wrong, when almost daily people are releasing not only initial studies but with Cuomo’s comments today, substantiating studies, finding the same thing. The reason they’re emailing you is because you were so prideful and erudite to request an apology form an MD/PHD at Stanford that history will judge to have been right. Your unwillingness to concede the point is exactly why people buy into Trump.

    • Blopes:

      Is there a specific thing that you think is in error in the above post? I’m happy to take criticism, but you have to be specific. I’m honestly not quite sure what is the point you want me to concede!

    • Cuomo said “0.5%” in the press conference today, more than twice the IFR that Santa Clara study showed, and 20 times the lethality of the flu.
      And the numbers are still wonky, the statewide percentage of the older people above 65 who make up the overwhelming majority of deaths dropped compared to last week, by 2-3 percentage point or about 20%.

      If you want to talk studies, name them, but if you cite Cuomo as the basis for conceding, then you ought to acknowledge that 0.5% has been in the WHO established range of 0.3%-1% since February, but is higher than the 0.12-0.2% that the Santa Clara people said it was going to be, so why don’t you ask them to concede?

Leave a Reply to yyw Cancel reply

Your email address will not be published. Required fields are marked *