“Stay-at-home” behavior: A pretty graph but I have some questions

Or, should I say, a pretty graph and so have some questions. It’s a positive property of a graph that it makes you want to see more.

Clare Malone and Kyle Bourassa write:

Cuebiq, a private data company, assessed the movement of people via GPS-enabled mobile devices across the U.S. If you look at movement data in a cross-section of states President Trump won in the southeast in 2016 — Tennessee, Georgia, Louisiana, North Carolina, South Carolina and Kentucky — 23 percent of people were staying home on average during the first week of March. That proportion jumped to 47 percent a month later across these six states.

And then they display this graph by Julia Wolfe:

So here are my questions:

1. Why did they pick those particular states to focus on? If they’re focusing on the south, why leave out Mississippi and Alabama? If they’re focusing on Republican-voting states, why leave out Idaho and Wyoming?

2. I’m surprised that it says that the proportion of New Yorkers staying at home increased by only 30 percentage points compared to last year. I would’ve thought it was higher. Maybe it’s a data issue? People like me are not in their database at all!

3. It’s weird how all the states show a pink line—fewer people staying at home compared to last year—at the beginning of the time series (I can’t quite tell when that is, maybe early March?). I’m guessing this is an artifact of measurement, that the number of GPS-enabled mobile devices has been gradually increasing over time, so the company that gathered these data would by default show an increase in movement (an apparent “Fewer people stayed home”) even in the absence of any change in behavior.

I’m thinking it would make sense to shift the numbers, or the color scheme, accordingly. As it is, the graph shows a dramatic change at the zero point, but if this zero is artifactual, then this could be misleading.

I guess what I’d like to see is a longer time series. Show another month at the beginning of each series, and that will give us a baseline.

Again, it’s not a slam on this graph to say that it makes me want to learn more.

Stay-at-home orders

The above-linked article discusses the idea that people were already staying at home, before any official stay-at-home orders were issued. And, if you believe the graphs, it looks like stay-at-home behavior did not even increase following the orders. This raises the question of why issue stay-at-home orders at all, and it also raises statistical questions about estimating the effects of such orders.

An argument against stay-at-home or social-distancing orders is that, even in the absence of any government policies on social distancing, at some point people would’ve become so scared that they would’ve socially distanced themselves, canceling trips, no longer showing up to work and school, etc., so the orders are not necessary.

Conversely, an argument in favor of governmentally mandated social distancing is that it coordinates expectations. I remember in early March that we had a sense that there were big things going on but we weren’t sure what to do. If everyone is deciding on their own whether to go to work etc., things can be a mess. Yes, there is an argument in favor of decentralized decision making, but what do you do, for example, if schools are officially open but half the kids are too scared to show up?

P.S. In comments, Brent points out a problem with framing this based on “stay-at-home orders”:

In my [Hutto’s] state the order closing schools was on March 15. The “stay at home” order came on April 7.

As best as I can interpret the x-axis of the graphs, they have the April 7 order marked with the vertical line.

It’s no puzzle why mobility data showed more people staying at home three weeks earlier. Mobility became limited on Monday, March 16 when a million or so families suddenly had children to take care of at home instead of going off to school.

This also raises questions about estimates of the effects of interventions such as lockdowns and school closings. Closing schools induces some social distancing and staying home from work, even beyond students and school employees.

81 thoughts on ““Stay-at-home” behavior: A pretty graph but I have some questions

  1. In my state the order closing schools was on March 15. The “stay at home” order came on April 7.

    As best as I can interpret the x-axis of the graphs, they have the April 7 order marked with the vertical line.

    It’s no puzzle why mobility data showed more people staying at home three weeks earlier. Mobility became limited on Monday, March 16 when a million or so families suddenly had children to take care of at home instead of going off to school.

    Or maybe I misinterpret their time scales. At any rate, by the end of March everyone I know (admittedly, white collar desk workers with good jobs) was working from home. The order a week or so later had more to do with business closings than with actual mobility for the middle class office drones like myself.

  2. Lots of companies in Seattle area encouraged people to work from home several weeks before “shelter” orders were issued. Dunno exact dates, but but Microsoft has major facilities in several states and countries as do many other Seattle-area companies so its likely people especially in the tech industry were sheltering way ahead of orders.

    A friend of mine in Arizona recently tested positive for antibodies and was ill in Feb – roughly around the time of the first case in Seattle. So clearly this was far more widespread far sooner than is generally being touted. Probably thousands and maybe tens of thousands of cases went under the radar.

  3. > An argument against stay-at-home or social-distancing orders is that, even in the absence of any government policies on social distancing, at some point people would’ve become so scared that they would’ve socially distanced themselves, canceling trips, no longer showing up to work and school, etc., so the orders are not necessary.

    I don’t really think that’s a compelling argument. We ban murder even though most people still wouldn’t murder if that law didn’t exist. Arguably, laws should to some extent reflect what people would want to do anyway.

    > Conversely, an argument in favor of governmentally mandated social distancing is that it coordinates expectations. I remember in early March that we had a sense that there were big things going on but we weren’t sure what to do. If everyone is deciding on their own whether to go to work etc., things can be a mess. Yes, there is an argument in favor of decentralized decision making, but what do you do, for example, if schools are officially open but half the kids are too scared to show up?

    Yeah. To expand on this, it helps people maintain social distancing for a longer period, by protecting workers from reprisals from employers trying to force them to return to work. It also reduces the possibility of ending social isolation too early, if the public’s individual perception of risk is matched worse to the scientific evidence than the central guidance.

    I think these graphs give a good illustration of what I suggested previously – that attempts to quantify the effects of lockdowns by doing before/after type analyses are going to run into this sort of gradual onset issue.

    • Zhou:

      This is different from murder because the argument for social distancing has been about flattening the curve. Suppose there used to be 400 murders per year in a city and now there are 200. OK, that’s still 200 too many murders. But if the goal is for the network of interactions to be shrunk by at least a factor of X, and it is shrunk by that factor, then we’re ok. It doesn’t need to go to zero (except in some high-risk places such as eldercare facilities).

      • I’m not sure if that works. In fact it seems plausible that the opposite might be at play? Like, if a stay at home order merely halves the movement from normal, then it could turn out to be effectively useless if people are still moving around enough to spread the virus. Meanwhile, an order that gets rid of the last few super-spreaders on top of already reduced movement might not look like an impressive improvement from the POV of the GPS graph, but could make a massive difference to disease spread.

        If you play around with e.g.

        https://corona.katapult-magazin.de/

        Going from say, 25 to 5 makes a much larger difference than going from 200 to 100.

        • The only reason for the goal of 0 spreading of the virus would be to eradicate it. If everyone (or enough people to reach herd immunity) will eventually get it (because I highly doubt a vaccine is coming anytime soon), then taking the flattening of the curve mantra seriously, what we want is a manageable flow of cases. If hospital capacity (either current or quickly assembled temporary capacity) is sufficient to get most people the care they need, then there is no reason for people’s activity to remain at lockdown levels. The whole idea of getting the R0 below 1 so that there is no more spread doesn’t make sense unless you plan on eradicating the disease. It seems like the door is closed on that option.

        • The thing to remember though is that at R0 the thing spreads around doubling every 3 days, so we need to keep Reff well below R0 or we’re back to square one with nothing but lockdowns being ok. (people keep talking about “what is R0 now” but R0 is a fixed thing, it represents kind of “worst case” so when they say “what is R0 now” the appropriate terminology is “what is Reff now”, the current effective reproduction number

          The thing about this virus is it seems to be pretty darn contagious, not at the measles level, but contagious enough that you could imagine doubling every 5-7 days even if you have reasonable levels of caution while reopening things. At that rate it means you’re going to have to close after only a couple weeks or you’ll get a crush of patients again.

          The only thing that could potentially fix this is widespread contact tracing, so that cases can’t grow exponentially even if contact is resumed.

        • People may be using these terms differently then… R0 is “start of the epidemic” reproduction number, before anyone is immune, but it does vary by the amount of “baseline” social contacts (before interventions), by some descriptions. Now maybe that’s an inaccurate use of the term… but if it doesn’t incorporate that, it’s kind of useless.

          Maybe it would be useful to compare with different diseases, but not to predict anything.

          The social contacts (before anyone had heard of COVID) in Wyoming just have no resemblance to those in New York City. (Same for rural vs. urban areas of many states.)

          After all, some states never did stay-at-home orders, so they are probably about at “reasonable level of caution while reopening” already. And Arkansas, South Dakota, etc. did not see 5-7 day doubling times. (South Dakota’s curve is weird because it is basically entirely driven by one local outbreak, which seems to have passed its peak. Most of the state outside the Sioux Falls area has barely been touched at all.)

        • That’s absolutely correct, R0 is not a property of a virus, but rather a property of a virus in a particular region under the early stage conditions.

          R0 in Rural WY is inherently different from R0 in NYC, but more to the point, R0 in “long term care facilities in Idaho” is probably very similar to R0 in “long term care facilities in Nebraska” and both of them are probably different form “R0 in a medium size town in either ID or NE”

          R0 is inherently a mean-field averaged property of a region. So it depends on which region you average over.

          From Wikipedia: “Suppose that infectious individuals make an average of β infection-producing contacts per unit time, with a mean infectious period of τ. Then the basic reproduction number is:

          R_0 = β τ

          So obviously, in places where lots of people only leave their multi-acre property a couple times a month to get food, then inherently their R_0 is way way lower in that region than NYC where most people leave their apartment daily and then contact hundreds of people on their way to work on the subway or going up and down elevators or in crowded restaurants for lunch.

        • Yeah.

          So the critical question for reopening will be, is the R-effective with reduced (or eventually zero) social distancing measures, among the general population in the-US-outside-the-Northeast-and-Midwest, low enough to prevent really bad outcomes?

          It’s hard to tell so far, because in general the lower-density states have done less stringent measures, so there aren’t really good apples-to-apples comparisons between more stringent measures vs. less stringent measures; and because even the lowest-density states have put some fairly significant measures in place. (Even those without stay-at-home orders closed schools, hair salons, etc.)

          I tend to think that the consequences of reopening will not be as bad as seems to be generally expected, but that isn’t really based on anything but the fact that all the really bad outbreaks so far have been in places that are incredibly dense by US standards.

          Clearly certain kinds of high-density settings – ships, meatpacking plants, long-term care facilities, prisons – are dangerous everywhere.

        • Daniel, confused, et al.,

          Worth remembering that these models assume homogeneous infectivity after infection. (Correct me if I am mistaken.) But this often is not the case. Some infected people may never be capable of transmitting virus; others may be very capable of transmitting virus. The variance could be huge; or it could be small. I haven’t found data on this topic (yet).

          On a practical level, a major concern regarding SARS-CoV-2 is asymptomatic/presymptomatic transmission. In terms of virology, I can only find literature on asymptomatic/presymptomatic transmission that used data from PCR-confirmed SARS-CoV-2 presence in sputum or nasopharyngeal samples or Contact Tracing. If I’m missing any papers/data, please share them!

          For the PCR-confirmed samples, they do not perform viral culture to confirm viability of detected viral RNA in these samples — so it is unclear if the detected virus was viable (like this paper: https://www.nejm.org/doi/full/10.1056/NEJMc2001468; good discussion of this issue here: https://www.thelancet.com/pdfs/journals/lancet/PIIS0140-6736(20)30868-0.pdf). For example, many viruses like SARS-CoV-2, MERS, Ebola, and Measels can shed viral RNA for weeks after infectious virus clears.

          For Contact Tracing, results indicate potential asymptomatic/presymptomatic transmission, but IMO biological data must also corroborate data from Contact Tracing to have a clear, decisive argument for asymptomatic/presymptomatic transmission. (Not discounting the overall utility of Contact Tracing, though!)

          So as of now, we don’t know what the variance of transmissibility is for SARS-CoV-2 — which is a critical gap-in-knowledge.

          Also, it is worth noting that the 60-70% infected to herd immunity is not constant; it can be much lower if certain conditions occur. These two papers (https://arxiv.org/abs/2005.03085; https://www.medrxiv.org/content/10.1101/2020.04.27.20081893v2.full.pdf) provide a good explanation how this is possible: the more variable a population is with regards to individual susceptibility and exposure, the lower the threshold for herd immunity is (like 60% dropping to 10-20%).

          Not saying these papers are gospel, but they raise interesting points. Interested in hearing thoughts on them!

        • Twain –

          Re that 2nd link:

          https://twitter.com/caesoma/status/1257762721317224448

          Keep in mind, we have examples already where their projections of herd immunity threshold have not worked out (Chelsea,MA, Bergamo, Italy, Ecuador, prison, etc.).

          Yes, in a theoretical model the herd immunity threshold might prove lower – but what matters quite a bit are the particulars of real world context.

        • Also, this theory of lower HIT due to heterogeneity in susceptibility and exposure relies completely on an assumption of immunity and thsr immunity has a certain degree of durability.

          Could be right, but I would think we should be heading any bets in that regard.

        • Joshua,

          Thanks for sharing the link.

          Can you elaborate on your examples? I’m not familiar with the situations in Chelsea, MA, and Ecuador.

          Regarding prisons, I’m not sure they are representative of anything except maybe LTCFs — like them, prisons have poor sanitation, poor HVAC, etc., that increase susceptibility and some (depending on location) have demographics with many at-risk patients. But I have not looked at them in-depth, so I very well may be missing something.

          And yes, particulars matter. Which is what I found interesting about the articles — “infections for herd immunity” depends strongly on multiple “particulars” and is far from a constant value. (Again highlighting the urgent need for state-wide, city-wide, etc., seroprevalence studies!)

        • Twain –

          I’m going to take back Ecuador. But they supposedly hit something like 67% population infection rate in Bergamo. And that seroprevalence estimate study for Chelsea put them at 31% and that was weeks ago and it certainly has gone up since then.

          The point about prisons is that they blew past the supposedly lower HIT.

          As for the term of immunity – no problem with speculation but I think we should wait until the evidence provides us an answer. A lot of knowledgeable people recommend that practice. This virus has shown a lot of people wrong about a lot of things.

        • Joshua,

          Thank you for clarifying. And as I say above, I agree — we have evidence HIT could be 60% (or higher) from some seroprevalence studies.

          And yes, we want to obtain data on how long immunity lasts, or at least start obtaining it, ASAP. Why the doesn’t seem like a priority (at least publicly) of many states is confusing, given it’s importance. (And see my correction below; I meant to say “we [don’t] want to bet the house”.)

        • I think the argument about herd immunity thresholds is that they can vary radically depending on the social-contact network, which in some cases can make them much lower than the simple calculation based on R0 would suggest – not that they will always be low. So 60% infection somewhere doesn’t necessarily disprove a herd immunity threshold much lower than that somewhere else.

          Also, isn’t there an “overshoot” issue? If one superspreader infects everyone in the prison / on the ship on the same day, it seems like you could get arbitrarily high prevalence, up to 100%. (Not that you would ever get 100% in real life – just that it might have no connection to a calculated threshold.)

        • Of course those “rural” hotspots may be linked to meat processing plants, prisons, nusing homes or similar high-contact places. I agree that in many places avoiding contact with people is easier than in NYC and the “natural” spread of the infection may be slower and more clustered. I mostly wanted to put a link to that document which I found interesting.

        • Joshua,

          I would be surprised if someone infected with SARS-CoV-2 does not have immunity lasting for at least 1-2 years. It would be shocking, at least in terms of immunology (what I’ve learned, at least), if immunity fades after weeks or months; that just does not follow the purpose of the immune system (which is to remember as many past infections as possible and protect us from them in the future for as long as possible.)

          Of course, the above could be wrong and is something we want to test to be sure. We want to bet-the-house on immunity until we know it lasts.

          But as a modicum of hope (that helps me sleep at night), I don’t think immunity to SARS-CoV-2 will be any shorter than a year.

        • Twain said,
          “that just does not follow the purpose of the immune system (which is to remember as many past infections as possible and protect us from them in the future for as long as possible.)”

          Sorry to be critical, but this sounds like a very naive view of biology. I could agree with a statement such as, “Ideally, an immune system remembers as many past infections as possible and protects us from them in the future for as long as possible.” But “the purpose of the immune system” as used in the quote from Twain sounds like something from a creationist perspective, not from an empirically sound scientific perspective.

        • Martha,

          Poor choice of words on my part; I was trying not to be overly technical.

          A more accurate statement would be: “The probability of little to no immunological memory to SARS-CoV-2 after infection is low — in fact, it would represent a very uncommon occurrence in immunology if the case. Since infection from SARS, MERS, the the four common HCoVs, all produce immunity lasting ~1-2 years, I personally do not think this is likely; but, it is still something to test and be sure of moving forward.”

        • I would argue that you need to take into account severity of infection here, in particular symptomatic and asymptomatic infection.

          Something we have seen in other diseases is that the immune system’s response is proportional to exposure. If people with asymptomatic infection correspond to those exposed to very little viral load, then they are not necessarily immune – or at least not immune to the same extent as someone with symptomatic infection. This is part of why the WHO is cautioning against using antibody tests as “immunity passports”.

        • Zhuo,

          Good point. And sorry, I was unclear again: By “infection” above, I meant infection producing mild or worse symptoms (so an infection in the “traditional” sense that excludes asymtomatic ones).

        • We have gotten R0 below 1 though. If the number of cases is forced down sufficiently new options open up (e.g. contact tracing)

      • Andrew,

        My suspicion is that you and I have been taking the “flattening the curve” argument at face value when it was really more of a marketing meme than a decision rule for when a set of policies have served their purpose.

        • OOPS, CORRECTED VERSION

          Andrew,

          My suspicion is that people like you and I have been taking the “flattening the curve” argument at face value when it was really more of a marketing meme than a decision rule for when a set of policies have served their purpose.

        • Brent

          I agree with this. It was easy to explain that an aggressive response would save lives by avoiding the overwhelming of hospital resources with the “flatten the curve” graph and slogan. Easy to explain, easy to understand. For many the concept of exponential growth was new, or something they’d only thought about in an abstract sense back in high school or whatever. The contrasting graphs helped people develop an intuitive understanding. Authorities needed people to agree to take action FAST as the doubling rate at the time was around 3-4 days, there wasn’t a lot of time for education about the fact that far more might be accomplished by sheltering in place.

        • The full and proper concept was something closer to this:

          1) Hammer the growth rate down so that the world wasn’t overwhelmed with sick people all at once.

          2) Learn a bunch about how to treat this disease

          3) Ramp up testing capacity

          4) Hire an army of contact tracers

          5) Give people time to learn how to do social distancing properly

          6) work through the bolus of patients until cases per day decline to the level that the contact tracers can handle them.

          7) Reopen some relatively safe businesses while monitoring and contact tracing to see how well infection control is working while avoiding exponential growth.

          That’s a mouthful compared to “flatten the curve”. But as is common with “sloganeering” the slogan became the only goal and 2-7 were abandoned for the most part.

        • Daniel

          #2 I don’t think this has been abandoned. It’s a tough question.

          #3 and #4 California’s just begun building up tracing throughput. People are being hired to be trained. It will take some time. A friend who works for the state and has as part of her job tracing TB just got assigned to being on of the liasons to county-level public health departments. She’s now on a seven day a week schedule. Online training programs for contact tracers have been developed and are entering use. I realize that most states aren’t doing this, of course.

          #5 California again but two months should be sufficient, no? Unfortunately it appears to have been sufficient time for people to decide that “patriotism” means rebelling, refusing to distance, wear masks, etc. People in SoCal apparently can’t do without sunbathing on the beach for another few weeks. Sigh.

          #6 California’s new cases numbers aren’t declining as rapidly as hoped, this is an issue

          #7 California’s implementing very slight relaxing . The next step explicitly requires sufficient contact tracing for each county (with the state helping, as mentioned above), no fatalities for 2 weeks, new cases below 1/10,000 for two weeks.

          Again, I know a lot of states aren’t doing anything close to this, but quite a few states are. And the national level? Ha. CA/OR/WA/NV/CO pledged a month ago to agree on common criteria for re-opening, and are trying to essentially build a region of rational behavior.

        • Yes, I give CA props for trying to do the right thing. And if CA,OR,WA,NV,CO can build a regional plan that’s GREAT. Meantime though CA cases per day are rising linearly. I hope for widespread contact tracing in CA by July or so. If we have some seasonality, we’ll need a LOT of contact tracing in the fall.

          Meantime here in Pasadena, mid April someone held a large birthday party and one person super-spread to a cluster of something like 10 or 20 people (only 5 of whom have been willing to get tested, but there are many more with symptoms). That kind of thing can erase 10 or 20 households worth of work… it’s a hard problem.

        • Yeah, some states are doing more of this than others.

          And it does seem that some states *need* to do more than others. The more rural states that did more limited social distancing don’t seem to be doing clearly worse than other states per-capita. (Iowa being the potential exception; it’s still nowhere near New York/New Jersey levels of bad, but if it goes really exponential it could get there pretty fast.)

          I think it’s largely misleading to look at “the US” for this, except when evaluating federal policies in and of themselves. The US curve is a mishmash of state and local curves which are doing radically different things. Per-capita the US is fairly middle-of-the-road in severity compared to European countries, but NYC is one of the worst hit places on Earth (except maybe some parts of Lombardy), though now sharply declining; while places like Hawaii and many rural areas without meatpacking plants have seen very mild problems.

        • Brent –

          > My suspicion is that people like you and I have been taking the “flattening the curve” argument at face value when it was really more of a marketing meme than a decision rule for when a set of policies have served their purpose.

          This is perhaps misleadingly simplistic. “Flattening the curve” at face value had a lot of implications from a public health perspective. Yes, there was a “marketing” element for the tag line, but that doesn’t necessarily imply that it wasn’t a decision rule for when a set of policies served a purpose. The two aren’t mutually exclusive as you seem to suggest.

        • The decision rule for “flatten the curve” is based on hospital capacity, every other policy is a derivative of that metric. But hospital capacity (current or quickly-assembled temporary capacity) was never a metric that most people have focused on (though some politicians are using hospitalizations (not the same as capacity but at least it’s related) as a decision rule, e.g. Hogan from Maryland). The top line focus among most public health officials and public health guidelines for opening up has been the number of cases and the movement in that number. Where I live, there had been some temporary capacity set up at a couple of hospitals which was never used, and they started taking it down 2-3 weeks ago (and they were taking this excess capacity down as cases increased). Based on the logic of “flatten the curve”, people in my area should be allowed a little more freedom of movement, perhaps with mandatory masks and reduced occupant capacity at all places of business.

        • Exactly.

          Some officials in Texas are looking at hospital capacity closely, though.

          IF your goal is to avoid overwhelming hospitals, and you expect that a vaccine will not be available in time to be relevant (IE keeping policies in place until vaccine is totally infeasible), then most places other than the real hotspots should probably be loosening up more than they are: Texas for example could “afford” several doublings without reaching our capacity.

          The big question there is whether you expect treatments to improve significantly over the next few months (availability of remdesivir? Learning more about what supportive care to give when?), so getting infected 2-3 months from now would be meaningfully better than getting infected today. That could change the picture.

  4. The graphs make me curious as well – and the CUBEIQ website does not provide enough detail to provide answers. First, the CUBEIQ website provides county level data and compares it to an average of the prior 365 days while the posted graphic appears to compare each state to the weekday from the prior year. The virtually identical pattern for every state is very suspicious. I agree with the comments about many schools and businesses adopting stay at home orders even before states “officially” ordered stay at home, but that doesn’t really explain either why the graphs all start below zero nor why they peak with the official order. According to CUBEIQ, the data relies on voluntary opt-in to their platform. If so, then there is some unknown selection bias for all their data. As Andrew suggests, it is reasonable to think that more people have opted in compared with a year earlier, so perhaps this accounts for all states starting off below last year’s numbers (although it also raises the question of whether the measurement is an average of all opted-in people in one year compared with all opted-in people from the prior year, or is it the same people who opted-in for both years, a more apples-to-apples comparison). Did voluntary opt-in peak with the announcement of official stay at home orders? While plausible, I find that a bit hard to believe. So the pattern similarity seems too regular to believe, without more detail on exactly what is being measured.

    This type of data is fascinating and could be quite valuable. But where is the complete disclosure of what is exactly being measured? And, where is the data? (proprietary of course).

  5. @andrew: “I’m thinking it would make sense to shift the numbers, or the color scheme, accordingly. As it is, the graph shows a dramatic change at the zero point, … “.

    Once again, someone forgetting about color-blindness. The color change is not dramatic to someone (like me) with red-green color vision weakness. I can barely see that there is any difference. And what’s the point of using the colors anyway, when the no-change baseline is so strong?

  6. RE: “Yes, there is an argument in favor of decentralized decision making, but what do you do, for example, if schools are officially open but half the kids are too scared to show up?”

    The dimension you need to disentangle is autonomy.

    I knew people who had their kids staying home a full week before schools closed here. But that’s because they had stable, good-paying jobs which do not require them to physically show up to work every day.

    Once the schools closed, even the people without that option were forced to keep their kids home. Which in many cases entailed being laid off from their jobs because they could no longer show up to work.

    We like to talk about there being two COVID-19 experiences. One in places like NYC which were flattened like a hammer blow. And the other in places out in the hinterlands where it ended up (so far) causing no major, structural problems.

    But it’s also a different experience depending on whether you have a job that allows you autonomy versus one where you show up on demand, do the work that’s available or else you don’t get paid.

    • Brent –

      > Once the schools closed, even the people without that option were forced to keep their kids home. Which in many cases entailed being laid off from their jobs because they could no longer show up to work.

      You are looking at this from only one angle. There may have been many people who didn’t want to have to go into work, and would have been fired if they chose to be safe rather than run a cash register. The SIP gave them another option – stay at home to be safe and apply for unemployment.

      > We like to talk about there being two COVID-19 experiences. One in places like NYC which were flattened like a hammer blow. And the other in places out in the hinterlands where it ended up (so far) causing no major, structural problems.

      This works in another dimension as well. Whether in the hinterlands or in NYC, there is a differential impact from CV-19 on minorities, and somewhat as a artifact on that, essential workers. When people not from that category complain that their freedoms are being stolen, and we should open up, they are implicitly saying that they accept the differential impact from their preferred policy option.

      > But it’s also a different experience depending on whether you have a job that allows you autonomy versus one where you show up on demand, do the work that’s available or else you don’t get paid.

      I wonder if you have such a job? My belief is that there are many people who do, who much prefer to stay safe and collect unemployment than be forced to come into work under unsafe conditions. Or perhaps for their to be a structure in place to assist their employer to make the workplace safe before they do have to come back to work.

      • Joshua said,
        “You are looking at this from only one angle. There may have been many people who didn’t want to have to go into work, and would have been fired if they chose to be safe rather than run a cash register. The SIP gave them another option – stay at home to be safe and apply for unemployment.”

        Good point.

      • “There may have been many people who didn’t want to have to go into work, and would have been fired if they chose to be safe rather than run a cash register. The SIP gave them another option – stay at home to be safe and apply for unemployment.”

        I think *this* is the primary actual benefit of the orders. In my area and among people I know, most of the actual change in activity had happened well before the state put in a stay-at-home order; much of it even before the county/city orders.

        Combining that with how many people are ‘essential workers’, and how many of these jobs are high-social-contact (food industry, grocery stores, Walmart, etc.), it doesn’t seem particularly likely that the imposition of statewide orders had a particularly significant impact on the course of the epidemic.

        But they probably did have a real positive impact on people who would otherwise be “caught in the middle”.

        But I don’t know whether that positive impact outweighs the negative impact on people who are unemployed now, but wouldn’t have been unemployed due to voluntary changes in economic activity alone.

    • > We like to talk about there being two COVID-19 experiences. One in places like NYC which were flattened like a hammer blow.

      I’m curious, do you think that the lockdown was appropriate an appropriate response in NYC? You didn’t seem to think so six weeks ago and given your comments since then about numbers being inflated to suit somebody’s interests I’m not sure if your view has changed.

      • To my recollection my thinking pretty early on was that NYC had no choice but to do whatever they could to “flatten the curve”. But I can’t imagine any “lockdown” regardless of timing or strictness could have avoided a horrific death toll, due to the extreme population density and demographics there. That was tragic but I suspect it was inherent the nature of the virus and the nature of New York City.

        But I do also think the state and/or city of New York played political games with how and when they reported numbers under various revised definitions. Their political leadership was engaging in some sort of weird pissing contest with Donald Trump even as they fought to keep their medical system functioning from day to day.

        I also think that very few places in USA experienced anything within an order of magnitude of the extreme threat facing NYC. Many places would have come through the crisis just fine without over-reacting and causing needless disruption and hardship for those on the bottom half of the economic ladder.

        Finally, I think people like yourself and some of the other frequent commenters on this blog have staked out an entirely unreasonable position that anything short of the maximum possible lockdown, maintained indefinitely is unacceptable and must be shouted down by repeated cries of “Don’t you know how exponential growth works????” and the like.

        I am completely aware of how viruses spread, I mourn the human cost of this virus as much as anyone and I do not for a moment suggest ignoring it and just hoping it goes away. But I am also cognizant of the human cost of “lockdown” under various guises and view everything about the response to COVID-19 with a focus on costs vs benefits and a recognition of the huge geographical and demographic disparities of not only the virus’s damage but the economic damage entailed by countermeasures to the virus.

        • I don’t know, on March 26 you were saying things like “While an additional ~50,000 deaths per year would be tragic” which suggest that you were able to imagine a less horrific death toll even without lockdowns. If you already foresaw 20’000 deaths in NYC alone you didn’t share that with us even though I asked precisely what should be done there (unfortunately you never replied): “The population of NYC is similar (15% less) and have reported 366 deaths by now. Should they follow Lombardy on their lockdown policy for maybe a similar level of tragedy or should they keep business activity as usual even though the outcome is likely to be much more tragic? Both are valid policy choices, mind you. But I’m not sure what are you proposing precisely.”

        • > I also think that very few places in USA experienced anything within an order of magnitude of the extreme threat facing NYC. Many places would have come through the crisis just fine without over-reacting and causing needless disruption and hardship for those on the bottom half of the economic ladder.

          I don’t know if you meant “an order of magnitude” literally. Just in case: confirmed deaths in NYC are 1’800 per 1’000’000 and there are 11 states (plus Washington D.C. and not counting New York State), home to 75 million people, reporting over 200 deaths per 1’000’000. You’re of course right that many places have been much less affected, though.

        • While many advocate for “costs vs benefits” I have yet to see any quantitative model of anything that might inform such a cost/benefit analysis other than the epidemiological models that predict a rebound and continued exponential growth. There’s no mechanism other than perhaps seasonality which leads to a prediction like “we can just go back to fairly normal and things won’t get bad”

          I’m all for cost benefit analysis, so far the only people who have anything that could inform such an analysis are the people such as the UK group at Imperial College, or the https://www.covid19sim.org/ that Carlos pointed to previously who are willing to make predictions about the future outcomes under different scenarios. All of those models predict a rebound effect from more contact, and they all show cases growing fast enough that the duration of any relaxation would be at most in the vicinity of 2-4 weeks before we had an order of magnitude more cases per day (from ~1k/day in CA now, to about 10k/day by June 15 according to the covid 19 simulator of the scenario 1 more week of shelter followed by 8 weeks of “social distancing in public” but unrestricted commerce, peaking at 340k cases a day in mid Aug and 70k deaths a day at that point in CA alone)

          It’s fine to argue that that prediction is wrong, if you have an argument. You know, like some data that suggests that the real world has different parameters.

          And no, arguments that “people will just change their behavior like stay home if it gets bad” is not an argument that the parameters are wrong, it’s an argument that the parameters are right, and that level of risk is unacceptable, and so people will naturally do what is being recommended currently. We need an argument that if we keep up social distancing and let people go back to work, the cases *won’t* grow out of control. So far, no one has such an argument.

        • I’m not saying this is true — I honestly don’t know* — but I can see one possible mechanism besides seasonality that could mitigate things a lot in some areas (many areas in the US, a lot fewer in Europe).

          If the patterns of social contact, in the absence of interventions, are *so* different in rural America than in NYC, then the effective R value might be low enough that hospitals would not be overwhelmed even in the absence of social distancing measures.

          The really bad outbreaks so far have all been in places that are very dense by most-of-the-US standards (NYC/New Jersey, Lombardy, Wuhan, Madrid).

          *Even those states that didn’t do shelter-in-place/stay-at-home orders did mandate fairly significant social distancing measures (school closures, closing hair salons and sit-down restaurants, etc.)

        • Absolutely, R0 is a property of the virus and an environment. So R0 in Casper Wyoming is definitely different from R0 in NYC.

          Nevertheless, R0 in 300 person LTCF in Casper WY (or wherever, some rural medium size city) is probably very similar to R0 in a 300 person LTCF in Seattle or Atlanta.

          So, while it’s possible to relax things in regions where people don’t have very many daily contacts, only go to the store every couple weeks anyway, and many live on multi-acre properties, it’s not possible to let up at University Dorms or LTCFs or Sheraton Hotels even if they are in the middle of WY or SD or whatnot.

        • Yeah, LTCFs and some other high-density settings definitely still need to be protected. Really big mass events e.g. giant crowded concerts and sporting events are also a bad idea everywhere.

          It’s possible that in those more rural areas, life could *otherwise* go back to fairly normal, though.

          University dorms I’m not so sure of, since the risk of bad outcomes is so low among people in the 18-24 age group. We kept doing college as usual during the 2009-10 pandemic (I caught it there, but it was very mild for me). This disease is much more deadly *overall*, but since 2009 H1N1 was a pretty flat mortality curve by age, maybe slightly weighted toward younger, while COVID is hugely weighted toward the elderly, I’m not sure the picture is that different among the 18-24 demographic.

        • The thing about university dorms is that you’ll have LOTS of people sick in a very short time, also people will be exposed to high viral load, which will make the severity worse even among the young. Also the stuff about clots/strokes/kidney damage etc is bad even if 20 year olds are low fatality risk.

        • >>The thing about university dorms is that you’ll have LOTS of people sick in a very short time,

          This is true, but if the severity is very low (and the majority are asymptomatic) I’m not sure it matters. Unless the hospitals are overwhelmed (and hospitalization rates should be very low in this age group) it’s not clear why everyone getting sick in 2 weeks is worse than over 6 months. In some ways it’s better, as social disruptions don’t last as long.

          >> also people will be exposed to high viral load, which will make the severity worse even among the young.

          This may well be true to some degree. But was it not true in 2009-10 H1N1? We didn’t shut down dorms then.

          (And, from my experience being at college during 2009-10, I don’t know that it was the dorms that was the big problem, as compared to the dining hall with 1000+ people, computer centers with hundreds, etc.)

          >> Also the stuff about clots/strokes/kidney damage etc is bad even if 20 year olds are low fatality risk.

          True, if it happens in this age group. The things I’ve read about strokes in “young” people are still a significantly older group than college dorm inhabitants (the five young stroke patients reported from a New York hospital were ages 33-49).

          I am not saying there would be no bad outcomes or even no deaths. But I am sure that *some* of the 12,000+ US deaths from 2009-10 H1N1 were related to colleges, and I don’t remember any outcry that it was handled irresponsibly.

          (Yes, this pandemic is clearly much worse than 2009-10. But given the very different curves by age, I’m not sure it’s clearly much worse *in this particular age group*.)

        • Brent –

          > But I am also cognizant of the human cost of “lockdown” under various guises and view everything about the response to COVID-19 with a focus on costs vs benefits and a recognition of the huge geographical and demographic disparities of not only the virus’s damage but the economic damage entailed by countermeasures to the virus.

          Actually, you have little solid idea of what the differential of a “lockdown” is. People were sheltering in place prior to when the SIP orders were issued. With worse infection rates and greater mortality and morbidity and hospitalizations there may have well been greater fear and economic harm over the long term. There’s no way to actually know, yet we have many peple absolutely certain – as if economic harm and the SIPs were somehow coupled. They weren’t. SIPs might have made things worse or they might have made things better over the long term. In the areas where they have been lifted, things are not exactly going swimmingly, and we have no idea how they will be doing in a few weeks if infections spike because they were lifted.

          Sweden, as an example, is hurting economically without any “lockdown,” – in fact, if you are so sure about the differential effect, compare the economy in Sweden w/o a “lockdown” to the economy in Denmark with “lockdown.”. You will see there are many relevant factors in addition to the simple fact of a “lockdown.”

          Counterfactuals are hard. They require a high bar of evidence. You don’t have that. So why are you so sure?

        • The so-called “lockdown” measures in my area were late, piecemeal, short-lived and not infrequently ignored. Yet we never had any sort of overload of the health care system or other dire consequences.

          It’s not counterfactual reasoning to point out that a huge portion of the USA is experiencing a pandemic that’s one or more orders of magnitude less severe than NYC. It’s not “counter” at all, it is simply fact.

        • Brent –

          > It’s not counterfactual reasoning to point out that a huge portion of the USA is experiencing a pandemic that’s one or more orders of magnitude less severe than NYC.

          Of course. But you’re engaging in counterfactual reasoning about what would have been different, economically, if there were no lockdown but you were just in the middle of a pandemic with tons o’ uncertainty about what that means.

          In point of fact, you actually have very little idea about how it would have been different absent the SIP mandates. Again, that kind of counterfactual training is extremely complex and requires very high fidelity evidence (that you don’t have).

        • “if there were no lockdown but you were just in the middle of a pandemic with tons o’ uncertainty about what that means.”

          But isn’t that the actual state of things in several states (those which never put into place stay-at-home orders/lockdowns)?

          I think the uncertainty (which I agree is very high) is instead because we don’t know whether those states that didn’t have stay-at-home-orders and didn’t become horrible disasters (e.g. South Dakota and Arkansas) were “outliers”, inherently at much lower risk. (In the same way, but opposite direction, to how New York City seems to have been inherently at much higher risk, since other states that did stay-at-home orders with similar timing saw nothing similar occur.)

          So we don’t know whether, say, Texas and Colorado would have become horrible disasters if *those* states had never put in place stay-at-home orders.

        • confused –

          > So we don’t know whether, say, Texas and Colorado would have become horrible disasters if *those* states had never put in place stay-at-home orders.

          Of course not. Nor do we know that they would have been significantly better off economically has they not had SIP mandates, especially over the long term.

          All I’m saying is that people are fooling themselves if they’re highly certain about these counterfactual assumptions. In particular when those assumptions align with political orientation – which they almost always do.

        • Yes, I’ll agree with that.

          I’ll reveal my own biases here – I don’t identify with either (US) political party, and certainly not with what they have now become, though I am fairly ‘conservative’ in some of the ways relevant to this specific issue (concern about surveillance issues, for example – I am fairly skeptical about some of the technological contact-tracing solutions suggested).

          I do think that doing totally unprecedented public health measures with huge impacts probably ought to have called for a higher standard of evidence than was actually met.

          But I certainly couldn’t claim at this point that the evidence yet shows that the measures taken were wrong, or were right.

        • This is all true, but I also don’t think we can entirely disentangle “what people did” from “the government response” – especially the local government response that isn’t captured by “date of statewide order” on the graphs in the post. Here in Texas, school districts closed schools well before there was any statewide order, and that made a huge difference to people’s behavior. Counties and cities also had stay-at-home / business closure orders before the state did.

          And the media is another factor. If the media and the government reacted to this the way they did to the 1957 and 1968 flu pandemics, I don’t think there would have been nearly as much change in behavior by individuals, especially in the less-hard-hit states.

          I don’t know anyone who has been diagnosed with COVID, and the few friend-of-a-friend cases I’ve heard of were a) far away and b) not likely to make me personally concerned. So the fact that I and people I know are staying at home is largely driven by the government response and messaging, plus media reports of places that *were* hard hit.

          People in NYC or Detroit, on the other hand, are far more likely to personally know someone who has had a pretty bad case.

  7. There’s also this:

    > The Times looked at data from Google, Descartes Labs and Unacast, three companies that use location data to determine whether people are socially isolating.

    Florida Gov. Ron DeSantis issued a stay-at-home order that went into effect April 3. By comparison, California’s stay-at-home order began March 19, Illinois’ started March 21, New York’s on March 22 and Ohio’s on March 23.
    Despite Florida’s later action, the data showed that people sharply cut their travel well before their counties and the state issued stay-at-home orders.

    For example, in the five days before Miami-Dade County’s March 26 stay-at-home order, more than half the phones tracked by Descartes Labs never traveled more than a mile, according to the Times. That was a drop of more than 80% compared with data collected from mid-February to early March.

    https://www.cnn.com/2020/05/11/us/florida-coronavirus-travel/index.html

    • Joshua said,
      “For example, in the five days before Miami-Dade County’s March 26 stay-at-home order, more than half the phones tracked by Descartes Labs never traveled more than a mile, according to the Times. That was a drop of more than 80% compared with data collected from mid-February to early March.”

      Interesting — but not surprising (to me at least).

  8. The above-linked article discusses the idea that people were already staying at home, before any official stay-at-home orders were issued. And, if you believe the graphs, it looks like stay-at-home behavior did not even increase following the orders. This raises the question of why issue stay-at-home orders at all, and it also raises statistical questions about estimating the effects of such orders.

    Imagine that you don’t want to go in to work because there’s a raging pandemic. But your boss says if you don’t go into work, you’ll get fired. The the Gov. issues a SIP mandate. Your boss can’t fire you for not going in to work anymore.

    > An argument against stay-at-home or social-distancing orders is that, even in the absence of any government policies on social distancing, at some point people would’ve become so scared that they would’ve socially distanced themselves, canceling trips, no longer showing up to work and school, etc., so the orders are not necessary.

    See above. You’re assuming that the only effect of a SIP order is people’s mobility. There were plenty of other effects as well. That doesn’t necessarily assign a positive of negative effect from the SIP orders, of course.

  9. Good graphs. Makes the point very clearly. I don’t see much to quibble with. Any bias in the data would have to be severe to negate the conclusion.

    There are even some confirmatory results showing that more wide-open states like the Dakotas, Wyoming, Nebraska, and Utah stayed at home a bit less than other states, which I would expect.

    • Are you referring to the rise in SIP prior to the official date or the entire plots? I agree the former seems to confirm all the anecdotal evidence, but the early period below zero and the tapering off after the official policy still need explanation.

      • Remember all the panic runs on toilet paper? The period prior to the shelter in place orders had a lot of stocking up at stores so that people could fully shelter in place for several weeks without needing to purchase anything. All that activity shows as red on the graphs.

        • I don’t find that a convincing argument for an increase in travel behavior (just how many trips do you have to make to stock up? Unless you think they ran from store to store buying every roll of TP thay could get their hands on – and did that for weeks). We have repeatedly objected to the common practice in NHST of rejecting the null and using that as evidence in favor of a particular favored alternative. Similarly, I worry that we see an intuitively appealing part of these graphs and use that as evidence for the validity of the data measurements. Don’t get me wrong – data like this can be very useful. But I’m to skeptical to just accept that it is measuring what it claims to, when there are unanswered questions and the data is not made available.

        • No question that we shouldn’t just accept this measurement, but I do remember a lot of people talking about going from store to store trying to find stuff as the shelves emptied. I guess what I’m saying is that it’s not implausible that we had a flurry of activity before sheltering.

      • Agree on all your points.

        I was focused on the rise in SIP before the official date.

        I don’t feel a great need to explain the early period below zero. It could just be that the previous comparable period was low for some reason. Or it could be anticipatory: people traveling home from college or returning home from vacations and business trips.

        The decline after the initial rise I would guess is due to an inevitable loss of rigor: initially people stayed home rigorously, then began to venture out as needed or grew bored.

  10. Not sure these look super surprising, right? Without seeing any real data, I would imagine that there would be a lot of movement before any kind of shelter – travelers trying to get home, college students headed home, people out buying supplies, etc. I might imagine quite a bit of moving, even across country to get to a place to shelter if the thought was that it would be for a while (like leaving college and heading back to parents house).

  11. I think the metric they’re using – “percent of people who stayed at home on a given day” – isn’t really a good enough metric here. For instance, it may not distinguish between going outside for a walk by yourself, and going over to a friend’s home to socialize.

    I’d guess the biggest reason that people leave home on a day-to-day basis is to go to work. Many companies started working-from-home before the shutdowns, hence the increase in people staying home. The shelter-in-place order may not have changed the *amount* people stayed home, but it may have changed the amount of social interaction they have when they do leave home.

    • “isn’t really a good enough metric here.”

      Great point. Aside from the probable numerous assumptions about using aggregate cell phone data that we haven’t even begun to consider.

  12. In my state (Texas) the date of the statewide measures is very misleading, because much of the population is in the largest urban areas (Houston, Dallas-Fort Worth) and these cities and counties imposed measures before the state did.

  13. Don’t forget that there were very large global shocks: watching Italy (and then Spain) being swamped by the virus, and the crash in the stock market.

    Both these factors affected all states simultaneously. And over the first two weeks of March, we went from the virus being a faraway problem to it being obvious we were all going to be hit soon by it. The actual state shutdown orders were small potatoes when the pandemic boulder got rolling.

Leave a Reply to Terry Cancel reply

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