Yesterday, I said I’m no longer calling a peak in the case count in Florida. I’d said I was ready to call it based on merely looking at the rate of change in the weekly average number of detected positives in Florida. These are shown with the greenish filled circles in the figure below:
Notice that in this figure, the total number of cases detected in Florida declined in the past week. Based on the ratio of the final change to the second to last week, cases had declined about 8%.
Then, I looked more carefully and saw that positive test ratio increased and the number of tests in Florida declined last week.
I knew (from thinking about Illinois statistics) that when the number of tests varies from week to week, estimating weekly average number of actual infected in the population based on the weekly average number of detected positives requires a correction for the change in the number of tests administered. For Illinois, over the past few weeks, detected case counts briskly had risen, but the positive/test ratio was fairly flat. This is because in Illinois, numbers of tests had been rising pretty dramatically (this may have stopped this week.)
But I knew it wasn’t right for me to use to correction in Illinois to say things were less bad than they seemed, but then ignore that in Florida where the correction I’m applying to Illinois has the opposite effect and makes Florida’s situation a bit worse.
So, I did a little math to quantify– this is a bit crude and it’s an estimate. However, based on the correction I find, I’m not calling a peak in the rise in the Florida’s new infections rate yet.
The math
Suppose a state administers Ntest tests in a given period (e.g. week) and find the number of positive tests Ncount. If the tests samples are selected randomly in from the total population, and the test is perfect, in the limit where we take a very large number of tests,
(1) Npos = cpos* Ntest
where cpos is the true rate of infected members of the population. In the current epidemic, cpos is unknown and it cannot be measured directly. We can measure Npos out of some number of tests, Ntest.
To call a “peak”, I don’t necessarily need to know cpos, but I would like an estimate of Δcpos and in particular, I want to know if it is positive of negative.
The hard way
The following will discuss a somewhat complicated way to estimate Δcpos/cpos based on measurements of the number of positive samples counted, and the number of tests administered. One of the purposes in doing it the complicated way is to relate the estimate for the change in infection rate to the information we bet from the graph of ‘new positive tests’ only.
The analysis will be subject to the assumptions stated above. (The discussion of the shortcomings of that assumption are deferred.)
We’ll now do a little calculus and algebra.
If (1) is true, we can estimate the relationship between the change in the count ΔNpos over a period of time as:
(2) ΔNpos = Δcpos* Ntest + cpos* ΔNtest .
Here Δcpos and ΔNtest represent change of infection rate (i.e. concentration of infected) at constant Ntest, and the change in Ntest at constant infection rate. These are essentially the first two terms in a Taylor series expansion for function of more than two variables. Higher order derivatives are neglected.
The first term on the right hand side of (2) is the term most people expect based on understanding what happens if the number of tests is held constant. It says that if we take the same number of samples in all time periods, the number of positive tests will increase linearly in proportion to the increase in the infection rate of the population.
The second term tells us that if the rate of infection in the population were constant and we increase the number of samples drawn, then, quite naturally the number of detected positive tests increases.
The full formula merely tells us that the number of positives detected varies for both reasons: It can increase because the actual rate of infection in the population increases or it can increase because the number of samples increase. When estimating, we must account for both effects.
With algebra we can show
(3) Δcpos/cpos = (ΔNpos/cpos – ΔNtest ) /Ntest
We eliminate cpos which is unknown and not subject to direct measurement using (1),
(4) Δcpos/cpos = ΔNpos/Npos – ΔNtest/Ntest
Equation (4) allows us to estimate Δcpos/cpos, a value of greater interest than the individual measurements. When this is positive, the rate of underlying infection is increasing; when negative, it is decreasing.
It is useful to consider both terms on the right hand side of the equation. When the number of tests administered during measurement periods does not change, ΔNtest=0. Equation (4) indicates the fractional rise in infected individual is exactly equal to the fractional change in the number of positives detected.
This first term on the right hand side is what one mentally estimate when looking at graphs of the number of cases over time only without considering the rate of change in number of cases tested over the month. It is in effect what one mentally estimates when looking at information on the rate of increase in number of positive cases only without considering the rate of change in total testing. So basically, it’s one things of as important when “eyeballing” figures like the one at the top of this post.
It is also the value most easily estimated from breathless news report which generally discuss number of positive cases, but omit information on things like the ratio of positive tests to total tests or rate of change in testing.
However, the second term indicates that we need to correct the first term when the number of tests administered during a adjacent time period is changing. In particular, if the number of tests is increasing during a period, then the fractional increase in infections in the population is less than the fractional increase in the population in the tested group.
Everything on the right hand side of (4) can be computed from data available at github, so we can use (4) as an estimate of the rate change in the infection rate. I’ll do that in a moment. First I’ll discuss the big caveat.
The big caveat
Note that the entire analysis is predicated several assumptions. The largest that even though the number of tests changes over time, it is assumed the tests samples are selected randomly in from the total population. This is certainly not true. In reality, people elect to get tested and all sorts of things affect whether they undergo a test.
Some of the factors affecting the distribution of those who get tested relative to the distribution of infected in the population are related to the infection rate itself. For example: some people decided to get tested because they exhibit at least some symptoms of Covid; others because they discover they have been exposed to a person who tested positive. Since these people have a higher than average likelihood of actually being infected, some of the change in number of tests, ΔNtest overall, is proportional to the change in actual infection rate.
But recall, in the Taylor series which resulted in the approximation that leads to equation (2) ΔNtest represents the change in Ntest holding ΔN constant. So in some sense, neglecting these in consistent with the approximation. (The problem could be reformulated– but in that case we need to have access to data that I think are unavailable.)
Some of the factors affecting the change in the number of tests administered are definitely not affected by the infection rate. Some people need tests to report for work, some to schedule elective surgery, some because they are hypochondriacs and so on. Of those who wish to be tested, some actually get tested; news reports suggest that, at least in Florida, some people see long lines and just go home. Even of those who believe they were exposed or who have symptoms, some will have some other ailment (e.g. flu) and some will be mis-informed about having been exposed.
To the extent that these affect the number of both uninfected and infected people taking tests ΔNtest can vary independently of the actual infection rate. One difficulty is that we (or at least I) don’t have a good way to get a numerical estimate of the effect of increases in the true infection rate on the increase in the number of people being tested.
So my view is: It seems to me given the data we have, the best estimate for the change in infection rate falls somewhere between two values. When ΔNtest is negative, the range is:
(5) ΔNpos/Npos ≤ Δcpos/cpos ≤ ΔNpos/Npos – ΔNtest/Ntest.
So (5) is what I will use as bounds. (Note: all the math uses large numbers in any case. So even these estimate are likely biased in some way.)
Shoving in numbers
Using the two bounds, and estimating the number of tests, and positive outcomes from data for the past two weeks, I currently get
-8.7% < Δcpos/cpos < 2.3%.
( The eagle eyed will notice the figure has -8% for the left hand side while above I have -8.7%. The difference is using lagging vs. centered values. )
So, using these two methods, the upper estimate for the change this week could still be positive.
The easier way.
There is an easier way to estimate the change in Δcpos. Recall
(1) Npos = cpos* Ntest.
What this suggest is that at any given time, in the limit that we take a large number of tests and with the same assumptions as previously,
cpos = Npos/Ntest
So Δcpos = [Npos/Ntest]this week ] -[Npos/Ntest]last week ]. This means that merely comparing the change in the positive ratio is a quick way to estimate whether the number of cases in the population increased or decreased. For the most recent week, the ratio ratio rose Δcpos/Δcpos) increased about 3% which is comparable to the value estimated the other way.
(Note once again, these are estimates, and the numerical value of the estimate are affected by choices to use one of the other end point or a mid point when computing a numerator or denominator. No particular choice is “right”. )
My verdict
If so, my view is we can’t call the peak in new cases in Florida. Mind you: there’s lots of other noise in this so even if both were negative, we might want to wait a bit for confirmation. But if I ignore that other noise, I’m still not calling a peak. I think cases in in Florida appear to have risen this past week though very slightly.
You may, of course, call the peak using your own judgement. None of us here are the Governor, and our call won’t affect subsequent deaths or transmissions. So it really doesn’t matter, but I like to consider the same factors when examining Florida, the US and Illinois, not just pick the one “I like”. I mostly did this for that reason. 🙂
I look at these two recently:
https://experience.arcgis.com/experience/96dd742462124fa0b38ddedb9b25e429
https://www.nytimes.com/interactive/2020/us/florida-coronavirus-cases.html
.
The difference being in how deaths are counted (by day, or date of death). Using the averaged trends the cases look to be at peak, assuming all things being equal, which of course they aren’t, but no real evidence of any significant changes in counting techniques or population behavior. One thing for sure is FL is out of the fast exponential rise phase of late June. What happens from here is quite the unknown. My guess is it will stay above 5K cases a day for the next month.
.
Deaths are near peak, but still climbing a bit, given the trend of the cases I would expect them to be peaked in the next 7 days.
.
The mystery of 12x the cases but only 3x the deaths is the most compelling question. I think maybe it is that many more people are getting tested voluntarily (thus decreasing the hidden case count) and perhaps people are getting lower initial doses of infection due to behavior changes, the most vulnerable are already dead, and the elderly are being better isolated. Nobody in the MSM seem to be wondering about this very important difference. This is very important to understand.
.
A trend of population antibody percentages would be useful to sort out this question. I don’t think anybody is doing this.
Lucia,
Yes, there is some uncertainty in the peak in cases, even in a single state…. there are too many potentially distorting factors which could, potentially, or in fact, lead to misleading counts.
.
Not so with deaths. If you look at case/death ratios from different locations (domestic and international) they are all over the place…. so cases are not terribly informative. But dead is dead; almost everywhere (and certainly in Florida or Illinois) someone in the hospital in serious condition will have been confirmed positive by the time they die. Deaths are the only measure I pay much attention to.
I like deaths better as a measure, but it’s potentially lagging. Death’s in Illinois are dropping. But some people around here always want me to discuss the fact that cases are rising. (Some people means my little sister. )
.
On my graph above, I have per-capita deaths rose about 18% last week. That’s slower than it did before, but still rose. (I know this is not the data that pins the deaths to the right day. But the one that does is like Swedens– “yesterdays” death’s will increase “tomorrow”, so you need to wait 10 days to 2 weeks to know the deaths for a particular day. So…. )
.
Around here, detected cases are rising because testing has been exploding.
Yes. I think you can’t really compare one place to another. The population demographics differ too much. (Yes, I know the differ over time in a state too. But perhaps not too much over the course of a few weeks in a single state.
.
The fact is: nothing is perfect.
Tom Scharf
Yes. No matter which way I look at it, on the semi-log graphs, the rate of rise in infections has at least slowed. It looks near the peak. Some ways of looking look past… but I think more points to just before the peak.
.
I can’t really say someone is absolutely wrong if they think it’s peaked. I just thing… nah. Not quite yet.
lucia,
Isn’t delta_N_test actually the partial derivative of N_test when C_pos is held constant? I think it is. But that is not a quantity that you can get from the data if C_pos is changing.
.
Also, I think that you are assuming a homogeneous population, or at least a population in which the sample is drawn equally from all groups. But I don’t think that is valid. I’ll try thinking that through for specific subgroups.
lucia,
One subgroup would be people who get tested because they have had contact with someone who has tested positive. The number of such people would be Nc*N_pos, where Nc is the number of contacts per subject. The number of positive tests would be R*N_pos where R is the reproduction number. The positive test rate in that group would be R/Nc. What if that is lower than the average test rate? Then fewer positive tests overall would lead to fewer people in this group being tested and a higher positive test rate overall. So falling test numbers with a rising positive rate *could* be a result of a falling number of cases.
.
I am not saying that is happening. I am just pointing out that we are dealing with a system with feedbacks that can confound simple analysis.
lucia,
Another subgroup would be people who get tested because they have symptoms. So if there is a summer cold going around, a lot of people might be getting negative tests. Then when the wave of summer colds passes, fewer people are getting tested but a higher fraction test positive. That tells you nothing about the ratio of positive tests to actual number of people who get tested.
MIkeN,
A) Yes, that’s what it is.
B) You estimate it from the tangent to the curve.
That’s a pretty conventional way to estimate it from data when you use the the first two terms in the Taylor series. I’ve officially neglected the terms proporitonal to Δc ^2, ΔN^2 and Δc *ΔN in the expansion when I do (2), and the term your discussing is order Δc *ΔN .) So it exist, but is neglected.
The practical problem is that if I really wanted to consider the effect of Δc on ΔN, I need to write:
Ntest = {p(test|covid) * cpos + p(test|no covid) * (1- cpos ) } Npopulatin. Then differentiate that-. But then we’d have many more terms in the equation– none would be Ntest, and we’d replace things we can’t measure for things we can measure.
We might get some insight. I have to go dance or I’d discuss this a bit more. But you can chew on it a bit and you’ll see that the effect on
Saying that CA and FL now have more cases than NY is comparing apples to oranges or maybe to rocks. Do I remember correctly that the positive test ratio in NY at the peak was on the order of 50%, not that it was likely you could get tested unless you were on your way to the ER.
As a first approximation, let’s assume that the IFR in FL and NY is actually the same. That would mean that cases should be measured by deaths, not tests. That would put the NY case total at at least 2 million on the same basis as cases in hot spots now.
DeWitt,
Yes, the comparison of case numbers in Florida and New York is as batty as the NYT editorial page. However, it is unlikely the IFR is the same in both states. The people who died in NY were mainly elderly, thanks in no small measure to Cuomo’s ‘enlightened, progressive’ nursing home policies. The confirmed cases in Florida have a younger average age, so even if Florida actually had more confirmed cases than NY (and we both know that hypothesis is pure rubbish) it would continue to have far lower deaths.
As of today, 19.6% of all ICU beds in Florida are empty. There are a handful of counties with tiny populations and very few ICU bed (eg 10 total) that are full, but in each case there are adjacent counties with plenty of ICU beds. The bigger counties (Miami-Dade, Broward, Orange (Orlando), etc) all have plenty of ICU beds. Funny how this very good news goes completely unreported by the MSM…. it was just 10 days ago that they were blaring constantly how the UCU beds would soon disappear due to the “frightening surge†in confirmed cases. The hospitals are not going to be overwhelmed, and Florida will soon see both confirmed cases and deaths fall dramatically, just as they did in New York….. but at a much lower death toll.
SteveF
Well… testing is more widely available. NY probably failed to count a lot of people. (In fact we know they did because … didn’t Cuomo have serology tests that showed “surprising” numbers of people had been been infected w/o symptoms?
.
The only reasonable comparison at the end of everything will be deaths. Even that’s not entirely fair– but still better than case counts given the critical shortage in test early on,..
lucia (Comment #188319) “The only reasonable comparison at the end of everything will be deaths.” Shouldn’t that be the change in deaths versus a seasonal average — NY and Florida had peaks at different times of the year so we would expect different death rates.
Andrew Kennett,
I’m not sure how you would seasonally average fairly. We don’t know the seasonal average for Covid-19. We also don’t know whether the seasonal average “should” be the same in cold climates vs. warm ones.
I thought NYC was at fault for not shutting down subway with its lockdown. Now I think it might have helped NYC recover faster.
Why do you think shutting down the subway might have helped NYC recover faster?
I have been thinking about spread of covid-19. The standard models consider a population to be initially homogeneous, both in susceptibility and in space. More sophisticated models can include a range of susceptibility, as Nic Lewis has been writing about for a while. But I think another important factor is spacial inhomogeneity of the illness. The best description of extreme spacial inhomogeneity is how a grass fire moves across a burning front, with nothing much happening before (unburned) or after (char) the burning front passes. Seems to me analyzing the spacial spread might help to understand how the number of illnesses and deaths evolves over time. In addition, different places certainly have different Ro values, so the local speed of spread, and the fraction of local population infected before the pandemic dies out, will both vary spatially. Even within a single state there could be a wide range of local Ro values which influence how the cases and deaths evolve over time.
I think that MikeN is suggesting that by spreading the virus, the subway helped NYC reach herd immunity faster. Perhaps. But NYC is one place where they got to herd immunity too fast.
.
Shutting down the subway was not an option. Doing so would have caused the city and its health care system to collapse since the subway is how essential workers get to work. But if they really wanted to reduce the risk from the subway, they would not have cut back service so severely that they made it even more crowded than usual.
MikeM,
I agree shutting down the subway is not a remotely attractive option in NYC.
SteveF
For sure. Luckily for us in Illinois. Pritzker switched to more fine scale divisions of the state allowing locals to make more tailored rules. Lightfoot can enact rules without them applying to extremely rural Grundy county! ( In fact, Cook county is divided into “Chicago” and “Not Chicago/Cook”. )
.
The Ro in Grundy county is likely naturally pretty low. In some other more rural counties, the reasonable rules should focus on meat packing, nursing homes, food processing and any congregate living facilities they might have. (Kankakee gets outbursts– it’s those things.) But in Chicago, they likely do need to have actual rules for bars, restaurants and so on. It’s crowded. Without rules, too many people collect with too many strangers and near strangers too often, and Ro will be high.
.
(Sadly, even with rules Chicago is going to have too many people collecting with too many other people too often.)
SteveF (Comment #188339): “I have been thinking about spread of covid-19. The standard models consider a population to be initially homogeneous, both in susceptibility and in space.”
.
The complex models constructed by educated idiots like Professor Panic over in England actually do incorporate a great deal of supposed heterogeneity; they have a huge number of subgroups both geographically and within regions and. But they assume that transmission is pretty much the same in all groups. So the complexity does not really matter much, except to create the illusion of knowing much more than they do.
.
That might not be too bad an assumption for something like the flu that is spread largely in schools.
———-
SteveF: “In addition, different places certainly have different Ro values, so the local speed of spread, and the fraction of local population infected before the pandemic dies out, will both vary spatially.”
.
I have been saying that for months. An important consequence is that exponential growth is the exception not the rule since R0 will naturally decrease with time.
——–
SteveF: “Even within a single state there could be a wide range of local Ro values which influence how the cases and deaths evolve over time.”
.
Not could be. That is so even within a community. For instance, a nursing home or a meat packing plant.
.
There are hundreds of thousands of subgroups within the country. They have a huge range of R0 values, from spreading like wildfire to quickly dying out because R0 << 1. Transmission within the groups are largely random. The groups are too small for the law of large numbers to apply within them. Transfers between the groups are also essentially random.
.
That sounds to me like a recipe for chaos. Not in the everyday sense, but in the mathematical sense. Largely random bursts of new cases scattered across the map. On a large enough scale, you get enough averaging that you can see overall patterns, or what appear to be patterns. But there is no reason those patterns should behave at all like a simple model.
.
To use lucia's phrase, there aren't one or two waves, there are a multitude of wavelets.
Mike M. (Comment #188340)
July 28th, 2020 at 6:45 am
But did they really get to herd immunity at all? They are still required to wear a face covering when on public transport or when in a situation where social distancing is not possible. They also lived through the worst of its peak and would probably be more cautious in general than people living in states which never experienced anything as bad as they did. I would think that social changes are more responsible for the reduction in transmission than any perceived herd immunity.
Lucia,
“Sadly, even with rules Chicago is going to have too many people collecting with too many other people too often.”
.
Depends on who the people are. Herd immunity is the most likely outcome in most places, and who gets the virus is probably more important than how many get it if the objective is to reduce total deaths by the time the pandemic subsides. NYC is long past herd immunity, but at a much higher price in deaths than was necessary. I really think the objective of public policies should be to limit the total cost in lives and economic damage on the way to herd immunity.
I know this is a little bit off-topic but… Don’t ya hate it when a state changes its reporting methods and a whole bunch of extra deaths are added to the total. This just happened with Texas.
.
skeptical,
We won’t know for sure until Cuomo and di Blasio allow NYC to open back up again. And herd immunity would likely only apply to NYC and the collar counties on Long Island and north of the city. The rest of the state will likely still be vulnerable. New cases haven’t gone to zero and likely won’t for a long time.
skeptical,
“I would think that social changes are more responsible for the reduction in transmission than any perceived herd immunity.”
.
And I think you would likely be mistaken. As DeWitt points out, the only way to know for sure is to relax the rules and see if there is a rapid increase in cases. You can see this in Florida, California, Texas, Louisiana, and many other places. The initial social distancing rules did in many places reduce Reff below 1, and infections fell over time in those places, but as soon as rules were relaxed (and/or people tired of the social distancing) the cases rose rapidly. The striking difference in Florida is that the deaths have not increased at anywhere near the same pace as the rise in cases, which suggest to me that the age profile of cases has changed, with more younger people contracting the virus. The ratio of deaths to confirmed cases in Florida has dropped by a factor of ~3 since late May.
skeptical
Yes. It makes it difficult to understand what’s going on. Georgia subtracted a whole bunch of cases about 2 (or so) months ago. States like GA or TX doing it in one fell swoop does help remind people that it’s difficult to compare data from one region to another.
.
Adding or subtracting a chunk in one fell swoop also makes it fairly obvious some policy changed which helps people better interpret data. In some states, they just change policy and the definition of “Covid case” before and after some date are different. It can make someone casually looking at data think there was a change in spread of disease on that day which is mistaken.
.
Overall, I prefer states to add and subtract in one fell swoop compared to practices that “hide” the change in policy.
skeptikal (Comment #188343): “But did they really get to herd immunity at all? … I would think that social changes are more responsible for the reduction in transmission than any perceived herd immunity.”
.
The big behavior changes happened in mid-March but new cases in New York did not peak until the second week in April. If that was solely due to behavior changes, the effect would have shown up sooner. Also, behavior has been returning toward normal and new cases have not gone up. Those facts agree with herd immunity as the major factor, not behavior.
.
Behavior changes probably have at least some effect on R. Smaller R reduces the herd immunity level. Then as behavior relaxes, R goes up, and more infections are needed to maintain herd immunity.
The behavior changes in March probably somewhat slowed the spread and lowered the herd immunity threshold, so it likely helped. But it was not the controlling factor.
By the same logic, once new cases come down to a level that the medical system can cope with, then restrictions should have been gradually lifted to allow herd immunity to be reached with minimum damage. In most places, no restrictions at all would have been needed; voluntary behavior changes would have sufficed.
——–
SteveF (Comment #188345): ” I really think the objective of public policies should be to limit the total cost in lives and economic damage on the way to herd immunity.”
.
Exactly right.
lucia (Comment #188349): “Overall, I prefer states to add and subtract in one fell swoop compared to practices that “hide†the change in policy.”
.
It is good to not hide changes in policy. But it is also good to not confound what is going on. What should be done is to have different categories: Reported on death certificates, deaths with positive test, suspected deaths, etc. Then a change in policy might result in a new category, while maintaining the old ones.
.
There is precedent for that. According to the CDC, there are about 175K pneumonia deaths a year in the U.S. According to the CDC, there are about 60K pneumonia deaths a year in the U.S. Two different ways of counting. Unfortunately, if you find one number, there is no hint as to the existence of the other.
MikeM,
“In most places, no restrictions at all would have been needed; voluntary behavior changes would have sufficed.”
.
I completely agree, but I think we are in a small minority in the States. I think in most states, a majority of people are convinced that the only acceptable policy is forcing everyone to live by the rules they prefer.
.
If you do your best to explain to those who are at risk how to minimize their risk, and make sure those who can’t reduce their risk are protected from poor practices (like people in nursing homes), then I think people should be able to make their own choices. But alas, that is not going to happen in most places.
MIkeM,
Sure. It would be better to have more refined categories. As a practical matter it also requires having staff to re-organize the data bases, reclassify and so on.
.
There probably is some “hint” about the differences on pneumonia. We just don’t happen to know them. Ordinarily, you just don’t have a shitwad of “newbies” trying to look at and process a whole bunch of data.
.
With Covid– we usually don’t have a disease where tons of people suddenly have an avid interest.
WSJ editorial advocating for defunding parts of academia unless they adhere to free speech and intellectual diversity.
https://www.wsj.com/articles/higher-ed-and-the-fragmentation-of-america-11595865575
.
“Legislators control the purse strings, and they can fight back by withholding funds from schools that enforce unconstitutional speech codes. They can go a step further in strengthening academic freedom by tying federal financial support to a university’s willingness to adopt some version of the Chicago Principles, which protect students and professors with divergent views.”
“Alumni and donors also play a role. They can push for change by targeting certain programs with their gifts and by using contractual requirements to demand a detailed accounting of how their money is being used. They can also make their contributions contingent on a university’s commitment to cultivating viewpoint diversity and upholding First Amendment freedoms.”
“If you understand how those institutions could arrive at a stable, no-conservatives equilibrium even without overt hostile action, then you understand part of the social dynamics behind systemic racism. The way small decisions cascade into major social forces is how Americans who profess no racial hatred — and declare their implacable hatred for racism in all forms — could nonetheless end up contributing to patterns of residential, educational and employment segregation that left the average black American with fewer opportunities for well-paid office work than the average white person.”
.
Personally I’d defund the social sciences in a heartbeat because they have become intellectually poisonous in my mind. I don’t think most people understand how bad this rot has become. The taxpayer should not be required to fund this stuff. At the very least academia needs a shot over the bow in a language they can all understand, funding. Intellectual outliers in academia need to know that the taxpayer has their back and demand non-penalized representative argumentation inside the ivory tower.
lucia (Comment #188353): “It would be better to have more refined categories. As a practical matter it also requires having staff to re-organize the data bases, reclassify and so on.”
.
Indeed. But the people and systems were not in place. In other words, the CDC screwed up big time. The CDC has become a corrupt and almost useless entity.
.
They KNEW this was coming. They did not know what virus it would be or when it would happen, but they KNEW it would be something, sometime. It was the CDC’s job to be ready; that is why the CDC exists. And they weren’t ready. Not even remotely.
MikeM,
“They KNEW this was coming.”
.
They are bureaucrats; you can’t have very high expectations.
.
Tom Scharf,
Universities should be mostly defunded by stopping student loans and all “research grants” for anything but actual research in STEM. Were that to ever happen, most of the foolish nonsense would wither and disappear. I don’t see how it is going to happen in today’s political environment, which makes the McCarthy era seem quaint by comparison, and can best be described as uninformed racist Jacobinism.
SteveF (Comment #188357): “They are bureaucrats; you can’t have very high expectations.”
.
Indeed. The CDC used to be an outstanding organization, capable of living up to high expectations. But it is now a bureaucracy of which competence is too much to expect.
The CDC tarnished their image badly in this pandemic. The initial testing debacle was unforgivable although one wonders if it would have made a real difference at this point. Perhaps NYC would be different. On the other hand their website is simple, clear, and concise.
MikeM
Before Covid: I don’t think they “knew” there would be a need for separating out “virology positives” from “serology” positives for a novel new disease. I don’t think they “knew” they would need all sorts of fine grained diagnoses of “death with”, “main cause of death”, “secondary cause of death”. Currently, the lack of fined grained binning on this has become a big political issue. But these bins aren’t that important from the point of view of public health.
.
We have enough data to know that the infection has been spreading. It’s cause lots of deaths. We would benefit if we found treatments and if we got a vaccine. Lots of this “extra” stuff amounts to pissing contests. Even if we had these expanded databased we’d still have the pissing contests because that’s what happens in politically charged situations like epidemics.
.
We’ve already created all sorts of “bins” for data bases: We know race, age, zip code, blah, blah, blah… complaining that somehow they don’t have some bin you’d like is really just complaining. Yes: some knowledge is going to be imperfect. That’s always going to be true.
FYI: Local story on positivity rate:
What is the positivity rate in coronavirus data and why is it important?
https://www.tampabay.com/news/health/2020/07/28/what-is-the-positivity-rate-in-coronavirus-data-and-why-is-it-important/
“Even if we had these expanded databased we’d still have the pissing contests because that’s what happens in politically charged situations like epidemics.”
.
What does the creationist say when the missing link is found? Aha, now there are two missing links!
Tom Scharf,
“What is the positivity rate in coronavirus data and why is it important?”
.
The author of the article appears to understand nothing. Of course, the CDC guidance is less than helpful in gaining understanding.
lucia (Comment #188361): “Before Covid: I don’t think they “knew†there would be a need for separating out “virology positives†from “serology†positives for a novel new disease. I don’t think they “knew†they would need all sorts of fine grained diagnoses of “death withâ€, “main cause of deathâ€, “secondary cause of deathâ€.”
.
Only if they didn’t bother to do their jobs.
The need to distinguish between primary cause of death, contributing factors, and non-contributing factors is totally obvious. Also obvious is the need to distinguish between suspected cases and confirmed cases as well as between different types of test results. Also obvious was the need for accurate real time data on hospital usage and capacity. That might not have been obvious to people like you or me who never thought about it. But it was their JOB to think about such things very carefully. They didn’t do their job.
I’d also like to be able to get stats on length of time in hospital for each case. Are the current cases staying less time as I suspect?
I don’t know. I know a huge file is available at github.
https://github.com/open-covid-19/data
The “main file” is ginormous. It takes a long time to download. You can have a look at the heading files and see.
.
Individual countries may have other data. Those headings are the ones the people at github are accumulating in one big file.
School is still requiring all immunizations for enrollment.
I guess whooping cough can spread thru wifi.