All my various postings tracking the COVID-19 pandemic through the end of 2022.
The above shows the centers of ZIP codes with at least a 1% cumulative infection rate, as of data reported 5/13/2020.
Separately, here’s the cumulative count for the three Vienna zip codes, as of data reported 5/13/2020. I missed the 5/12/2020 update, so that data point is interpolated between 5/11 and 5/13. In the Town of Vienna ZIP (22180), we’re now up to 88 cases, with a steady 16% of those tested so far showing positive for COVID-19.
As of the 5/11/2020 data, there remains no material difference between these two groups of states in terms of their trend in COVID-19 cases. That said, it’s still probably too soon to expect to see any difference.
Detail follows. Continue reading Post #689: Statistical comparison of states re-opening and not, first revisit
Source: Harlem Bespoke (top image), Town of Winthrop, MA (bottom image).
I can with near-certainty predict that when people look back on this pandemic, praise is going to be reserved for those who got tough, early. Bear that in mind when you decide what to support.
So, we finally had a prominent local business institute a mandatory mask policy, CEO excepted.
That’s better late than never. It’s only been two months since Virginia schools were shuttered (on 3/13/2020). It’s only been a month and a half since the CDC recommended that we all wear masks in public (circa 4/3/2020). It’s only be a little over six weeks since the head of the Chinese CDC said that failure to require masks was “The big mistake” that the US and Europe were making.
I think people will figure out that they ought to be wearing masks, sooner or later. Some a lot sooner. Others a lot later.
Here in Vienna, Fresh Market had a mandatory mask policy at least a month ago. So that’s where I shop now. And it looks like Whole Foods finally did that about a week ago.
But Giant? Ah, looks like they’ve gone as far as maybe even encouraging employees to wear masks. Wow. Encourage. That’s a fail.
As an economist, all I can say is, how stupid do we have to be? Just ponder the economic cost of what we’re doing now in Northern Virginia. Consider that, as of today, it’s not working well. (Fairfax County racked up another 300 or so cases today.) Consider that a mandatory mask policy is virtually costless. Consider that the head of the Chinese CDC just flat-out told us that the lack of mask use is a huge mistake. Consider that the CDC changed its guidance — almost unheard of, mid-epidemic.
Masks are not some form of sure protection. But the odds are overwhelming that they help prevent spread of disease, to some degree. They ought to be mandatory in any setting outside the home where individuals share indoor space. Retail, workplace — it doesn’t matter. Anywhere.
And instead, we sit here with the same policies in place, no changes, no ramp-up, nothing new to try. We continue the beatings and hope morale improves.
It took having cases appear literally within the White House to get a mandatory mask policy for White House staff. If not for the President and Vice President themselves.
But at least they’re ahead of Giant Food.
COVID-19 Cases per 10,000 by Northern Virginia ZIP code, as of 5/11/2020. Vienna, VA (ZIP 22180) is the blue dot.
Sources: Case counts by ZIP code are from the Virginia Department of Health. Population by ZIP code is from https://simplemaps.com/data/us-zips . Map is via Bing Maps.
The measles chart above shows the density of diagnosed COVID-19 cases per 10,000, for all ZIP codes in Northern Virginia with at least 5,000 total population. (Which is, in fact, almost all of them). The larger the dot, the higher the density of cases. Vienna (ZIP 22180) is in blue.
This is a little hard to take in by eye, because the only thing that matters is the size of the dots, not the spacing. You want to focus on where the dots are big, not where they are close together.
By eye, the only thing that seems clear is the correlation between high density of cases per 10,000, and high population density. The heavily urbanized areas in the close-in DC suburbs have generally higher rates of COVID-19 cases than the less-densely populated areas.
But as a lifelong NoVA resident, I think I see “blue collar” in this chart as well.
Point 1: Look west on I-66, and you see a big red dot where the road curves. That’s in the Manassas area, in a part of town that was light industry and retail, but has added a large amount of medium-density housing (apartment buildings, townhouses) in the past decade or so. (E.g., there’s a big Salvation Army thrift store there, which tells you that it’s not an upscale neighborhood.) This is a view of the center of that ZIP code. The large gray blocks are a mix of apartments, condos, and townhouse developments.
Point 2: The big red dot next to that, I know well, because that’s where I grew up: That’s the Westgate/Sudley/Yorkshire areas of Manassas. When I lived there it was a middle-class blue-collar community, and it has gone downhill since it was built in the 1960s and early 1970s. Here’s a typical dwelling. No great shakes, and the owner probably isn’t a doctor, lawyer, or business executive.
Point 3: Springfield. Here, I’m less sure. Springfield was and is a significant commercial hub for this area. But in addition, they are the home of the single largest assisted-living/senior-living development in the area, Greenspring. So I don’t want to generalize there, because I don’t know whether the high rates there might be driven by having such a large concentrated senior-living facility in that area.
In any case, if you look at our own little patch of upscale living — Oakton, through Vienna, and then on up into McLean — you see nothing but small dots. I’d attribute that to a combination of low-density housing and high-income earners. Plausibly, we are rich in the sort of families that can effectively shelter in place with minimal outside contact, and that keeps our overall infection rate low.
My guess is that much of the spread outside of the home, at present, is spread in the workplace. (I mean, just look at the White House). It would be helpful if the Commonwealth’s epidemiologists could confirm that.
My conclusion is largely based on the age distribution of the known infections. The elderly are far more susceptible to serious (reportable) infection, and so you would think that the peak infection rate would be in the retiree population. But, in fact, aside from the very oldest old, the peak infection rate actually occurs in the working-age population. That, and the negative statistical association between ZIP-level infection rate and income (prior post), both point to the workplace as a likely vector for infection.
I hope that people eventually figure that out, take it to heart, and wear a mask at work.
The Commonwealth is now putting out a daily file with case counts by ZIP code. Vienna (22180) added 8 new cases yesterday, going from 61 to 69 persons that have tested positive for COVID-19.
That said, all three Vienna ZIP codes (22180, 22181, 22182) rank near the bottom of all Northern Virginia ZIP codes in terms of cases per 10,000 residents, having 25, 15, and 20 cases per 10,000, respectively, as of 5/9/2020. By contrast, the median for all NoVA ZIPs with 5000 population or higher was 34 cases per 10,000.
(I estimated a higher rate in yesterday’s post, not knowing that the 22180 population includes a lot of people not in the Town of Vienna.)
I’ve just started to look at it, but the first thing I wanted to see was the extent to which income was associated with risk of infection. It’s clear that there’s just a lot of seemingly-random variation in infection rates at the ZIP code level. That’s due both to generally small numbers, and (probably) to single-point events such as a cluster of cases in a nursing home, or within a large household.
That said, even taking a crude income measure (this is from the IRS statistics of income, 2017), it’s pretty clear that, on average, ZIP-level income and infection rate are negatively correlated (R = -0.47).
The two high outliers in the graph above are ZIP codes in Alexandria (22305) and Springfield (22150). That got me to thinking that, even within densely-populated Northern Virginia, you might see more spread in the more urbanized locations. And, to a degree that’s true — that what the second graph shows.
If I put those two factors into a simple linear regression, each has a statistically significant independent effect, and together they explain about a third of the ZIP-level variation in infection rates (adjusted R-squared 0.34).
My only point here is that the infection rates are not some total mystery, appearing at random. Even within a small area like Northern Virginia, household income and population density are systematically correlated with the infection rate. Unsurprisingly, ZIP codes where people can afford to stay home (or, as likely, have the sort of job that allows working from home), and can get around without having a lot of contact with others, tend to have systematically lower infection rates.
In brief: This is a contrast of the states that have and have not chosen to re-open their economies. On average, the states choosing to re-open now are much more rural, and have about a third as many COVID-19 cases per capita, compared to the states that are still waiting.
The question of the hour is whether or not this early re-opening will increase the spread of COVID-19.
Normally, evaluating the impact of a “natural experiment” like this would be difficult. That’s because health care decisions normally depend on … well, health. And that correlation messes up any simple-minded comparison of those who did and did not take some health-related course of action.
But in this case, luckily (?), it looks like the early-versus-late opening decision appears almost purely driven by politics, not public health. And since this is an equal-opportunity virus, that’s about as close as you can get to random assignment. It doesn’t care if you are R or D. It doesn’t care if you’re pro-Trump, anti-Trump, or have no opinion. The upshot is that trends in the growth of cases — for April, and for the last half of April, and for May so far — are virtually identical in the early-reopening versus later-reopening states.
And so, I’m putting my marker down for an acid test on this issue. (I.e, a quick and dirty test that likely reveals the correct result in a short amount of time.) I’ll revise the above table in a couple of weeks or so. If the numbers in that right-hand green column diverge, then we’ll know that early reopening cost us something in terms of health. If they don’t, then not.
Sometimes the only way to get the analysis done is to do it. From the raw data to the writeup. It’s what I used to do, before I retired.*
* So that I can now look in horror at the paper assets that my continued livelihood depends on, pondering which of them will survive this Black Swan event, and which shall go off to Money Heaven, never again to be seen on Earth. My financial advisor urges caution and stability. I respond that individuals who bought into stocks in 1929 were made whole again circa 1956. Despite the fact that I am an economist (Ph.D. — no joke, I actually have one!), I am a lousy investor, so let me just squelch this fear-fest and get back to statistics.)
What you see above is the Johns Hopkins case count data, matched to Census population and land area data, split by the New York Times classification of which states are re-opening their economies now, and which area not.
The use of the NY Times classification was a key simplifying step, because this is complex enough, and nebulous enough, that somebody has to study each state and make some sort of subjective judgment. I’m more than happy to accept what the Times has done in that regard.
For all you statistics fan(s) out there, I put this together using SAS, so there’s plenty of opportunity for (e.g.) incorporating additional explanatory data. But, honestly, I’m not quite seeing the need for that, at the moment.
To a health economist — and maybe to an old-school political economist — the results are kind of hilarious.
The states that are choosing to re-open now are far more rural, and have a far lower COVID-19 case load per capita, than those who are not choosing to re-open at this time. That kind-of makes sense, but I suspect that’s mostly just an artifact of which states have Republican governors, and the spread of COVID-19 having been concentrated in urbanized areas.
But the hilarious and interesting part is the trend data. As in, no-difference-in-trend. To a very close approximation, the two sets of states have had the same average COVID-19 growth rate for the month leading up to May 1, for the two weeks leading up to May 1, and for the week following May 1. So the idea that the early re-opening states have seen falling new case counts, and that makes them different from the other states — that’s just wrong.
But look at the interesting part. Look at how the growth rates are the same across the two groups, even though the cases/capita (yellow) averages three times higher in the non-reopening states.
Because, to me, that says we’re not all on the same path. More or less, on average, those early-reopening states look like they are going to avoid the worst of this. Most US states (with a few exceptions) imposed restrictions around the same time. And at that point, whatever density of cases you had, that carried forward. And the upshot is that both groups of states have followed the same growth curve. It’s just that, in absolute numbers, the rapid-reopening states never got hit as hard (on average), and, assuming they don’t screw it up, never will get hit as hard (on average), as the ones that are currently delaying economic re-opening.
Finally, all other details aside, the almost-complete-equivalence of those trend numbers provides a good quick-and-dirty test for whether or not re-opening costs lives. We need to come back and revisit that last column of numbers in a few weeks. If the early re-openers don’t diverge from the later-reopeners, given how well matched these two groups were historically (in terms of trends), I think that will be a fairly strong argument that re-opening early was harmless.
Soure: The New York Times.
This is my first followup to Post #673, talking about re-opening state economies. By that, I mean allowing a wider variety of business and social activity.
For this analysis, I went through various news organization summaries of state re-opening strategies. My original goal was to begin setting up a statistical analysis of the impact of re-opening. But it’s complicated enough that I have abandoned that for now.
Instead, here are my observations, from going through the re-openings with a fine-toothed comb.
Continue reading Post #684: State re-openings, and the key unanswered question
Above: Wuhan, China, daily new cases following total lockdown, from WHO report, as summarized in Post #551.
Above: Total new cases, Virginia, following executive orders of the Governor, starting with the closing of schools in mid-March, partial closure of businesses, and closure of all non-essential retail businesses.
Above: The Virginia graph in full perspective — blue is the time period of the prior graph, red is what has happened since then. Continue reading Post #683: One thing to be thankful for
This is a brief update to my various posts on aerosol spread of COVID-19 via singing, most recently, Post #679 on hymn singing.
Plus a little arithmetic.
Let me summarize the arithmetic. Given the likely prevalence of infectious (but asymptomatic) individuals in Fairfax County (Post #680), and the prevalence of aerosol super-emitters within the population (via this reference), it’s straightforward to estimate the risks incurred by holding church services.
So, how likely is it that a Fairfax County church service will have at least one infectious adult present, or, worse, one infectious adult that is a superemitter of aerosols? That’s the arithmetic below.
So, it’s odds-on that even a small gathering — 100 individuals — will have at least one infectious adult present. But for a small gathering, there’s only a 5 percent chance that the infectious adult is also an aerosol superemitter.
But as you ramp up the congregation size — or, equivalently, consider the odds for a large group of churches — the odds of having a service with some infectious adult present go up considerably. To the point where having 10 churches meet, each with 100 congregants present, pretty much guarantees that at least one church will have at least one infectious individual present.
So, at this point, I hope you can see why this is a serious issue. The odds are overwhelming that resumption of church services is going to increase people’s exposure to the disease. If churches choose to resume services, they need to take every reasonable precaution.
My wife is friends with a Methodist minister in Florida. Florida is one of those early re-opening states. By report, right now, the entire Methodist hierarchy there is trying to figure out how best to resume services without endangering their congregations.
My wife wanted to make sure that her minister friend was aware of the issues raised by singing. And what came back was an emphatic response. Oh, yeah, the Methodists are acutely aware that singing is a bad idea. They’re just not quite sure how to deal with that.
Congregations like to sing. Singing is a risk. And at present, it’s not clear how the Methodists are going to deal with that. But at least they are aware of the problem.
Based on that sample of one, I’d say it’s a fair bet that all the mainstream denominations are aware of this issue. Presumably, both the Lutherans and the Catholics have paid attention to developments in Germany, where singing in church services has been banned. (Part of Germany’s overall rational and organized response to COVID-19, starting with their approach to testing, as outlined in Post #605.)
As for the rest of mainstream religious doctrines, I’d have to guess that they all know about the issue at this point. As outlined in Post #679, there have been enough church-based super-spreader events that any religious organization with a world-wide view should be aware of the dangers. And if the Florida Methodists are struggling with the issue of hymn singing, it’s a fair bet that most large-scale religious organizations have gotten that message.
Whether or not smaller, independent churches and sects are aware of the issue, it’s tough to say. There has been enough crazily defiant behavior among that industry segment (see Darwin Test), that its a fair bet that at least some will ignore it.
That said, as an industry, churches are in a financial bind. A significant (possibly main) revenue source derives from a particularly risky line of business. Their main product is a weekly in-person audience-participation exercise involving group singing. Mass meetings are risky enough, but singing produces as much aerosol emissions as coughing does ((via this reference). It’s hard to conceive of another activity that the public might participate in that would put them at higher known risk.
As always, I’d like to quantify the risk if possible. I can’t tell whether or not infection will spread in some given circumstance. E.g., how far apart will people be? Will masks be required? And so on.
But determining whether or not an infectious adult is likely to be present at a mass meeting is a simple probability calculation. Suppose 1 person in 100 is infectious. Then there’s a 99% chance that any given individual is not infectious. And so the odds that neither of two individuals is infections would be the odds for the first, times the odds for the second, or 99% x 99%. And in general, the odds that N people are all not-infectious would be 99%^N. (Where ^ = to the power of).
That’s the simple math that underlies the table at the start of this. I’ve already cited my calculation for estimated prevalence of infectious-but-asymptomatic individuals in Fairfax County. For fraction that are aerosol superemitters, I take a stricter definition than was adopted in this basic research. They used a statistical definition (more than one standard deviation above the mean), I looked for outliers clearly separate from the mass of individuals. They had 2 in a sample of 30, so I took 1/15 (= 6.7%) as my estimated prevalence of aerosol superemitters in the population.
Oddly, the arithmetic for several churches together is just an extension for the arithmetic of one congregation. That’s not odd from the math standpoint — 1000 people is 1000 people, whether it’s one congregation of 1000, or 10 congregations of 100.
What’s odd is that the Commonwealth limits group sizes. I’m pretty sure that has more to do with ensuring distance between congregants than with the underlying arithmetic. If a mega-church wants to take its 1000-seat chapel and hold 10 services with 100 persons each, that doesn’t change the odds that at least one of those person will be asymptomatic-but-infectious. But it does increase the distance between congregants, and that should help reduce the spread of infection.
Source: Virginia Department of Health.
Given the ongoing growth in new cases in Virginia, I think I have to start asking this question: If the current situation holds, how long will it be before 70% of the … let’s say working-age adult population — has had coronavirus? That’s the point at which most experts expect this pandemic to die off naturally, due to “herd immunity”.
And, secondarily, is that 70% point likely to be passed before or after the date at which we can reasonably expect to have a vaccine? At which point, it’s moot: We’ll get immunity from the vaccine.
This is motivated by a couple of graphs, from the Virginia Department of Health, above. Realizing that death is not the sole outcome, and that death is a somewhat lagging indicator, that said, this disease:
Given this, stay-at-home orders that are uniform for all age groups are probably inefficient. I think there’s a case for letting the younger generation get on with their lives as they see fit. And, maybe, for using this as an opportunity for the older generations to get out of the way and hand off economic tasks to younger generations.
At some level, this is just arithmetic. I’ll do three scenarios.
A static scenario asks: At the current daily rate at which newly-diagnosed cases are appearing in Virginia, how long will it take is to reach 70% of the adult population? The answer this is, about two years. That’s well past the point where we have a vaccine under any plausible scenario.
A fixed-growth scenario asks: At the current rate at which those new case counts are growing (about 5% per day), if that growth continues, how long would it take? In that case, just about two months. But the case load toward the end of that would clearly overwhelm the Virginia hospital system. So that’s short, but not really feasible.
A modified-growth scenario: Let the current growth rate continue until we reach some plausible estimate of hospital capacity. I (gu)estimate that at around five times the current daily new case count. Under that scenario, it would take just under half a year to achieve herd immunity. Arguable, that’s well before we can reasonably expect a vaccine to be available in the US.
But haven’t you read that there’s no guarantee that you’re immune, once you’ve gotten over this? Yep, you’ve probably read that. My guess is, that’s exceptionally unlikely to occur. It did not occur in the 2003 SARS epidemic, and it probably isn’t occurring in this epidemic. Most of what you’ve read about people being “re-infected” comes from Korea, and my best guess is that they simply have not properly accounted for a small false negative rate in their testing regiment. See the extended small-type discussion in this post for details.
But there is an alternative, which is to let the younger generations get on with their lives. At least, those without direct household contact with the older generation.
Today (5/4/2020), the Commonwealth reports a cumulative total of about 6,000 diagnosed cases age 21-40, resulting in 318 total hospitalizations. Assuming that the un-diagnosed population is four times as large (see prior post), that would be a hospitalization rate, per infected person, of just about 1%. We have 2,250,000 adults of that age in the Commonwealth. Infecting 70% of them would generate 16,700 hospitalizations. But if we did that at an even pace, over the course of two months, that would be just 300 new hospitalizations per day, plausibly within the limits of the Virginia hospital system to absorb.
In other words, if we could pace the infection of the younger (21-40) population, we could achieve herd immunity within that population in just two months, plausibly without overwhelming the hospital system. (It’s not clear that the herd immunity concept really applies, unless they can be kept separate from the rest of the population.)
Based on the same calculation, you would expect a total of 315 deaths, to achieve “herd immunity in that population.” In a normal year, we would expect about 2800 deaths from all causes in that population.
