The Wizard of Oz is a classic of American literature, and one of the most beloved movies of all time. The film, which is the most-watched in history, was based on the novel by L. Frank Baum. Baum's story carried a theme of being satisfied with what you have, and realizing how blessed you really are. It's a fanciful tale, and it set the bar for cinematography when it was made. The beginning and end are set in my native Kansas, and were filmed in black and white (likely a swipe at flyover country, but hey, what's new?). But the middle is filmed in glorious, vivid color, and the setting is a fanciful land watched over by a wise Wizard, who rules with benevolence and offers his subjects hope. The Great and Powerful Oz, as he's known, turns out to be a sham, nothing more than a huckster hiding behind a curtain, pulling levers and throwing switches to make things appear differently than they really are. But he truly is benevolent - he was thrust into power by the people of this magical land who revered this outsider. The power got to him, so he ran with it. But in a well-intentioned way.
Fast-forward to 2020 ("Do we have to?" you ask). There's a new Wizard here in the Land of Oz, one that is entirely different from Baum's Wizard - decidedly less benign, drunk on power, pulling mathematical levers and throwing graphical switches to make things appear differently than they are, not in a well-intentioned way, but to exert control. And rather than offering the subjects hope, this Wizard seeks to crush it.
The new Wizard of Oz is not one person, but rather an archetype. The new Wizard is Gov. Laura Kelly and her Kansas Department of Health and Environment (KDHE), led by Dr. Lee Norman.
My alter ego has posted several articles on social media outlining the statistical trickery plied by Dr. Norman to support his boss's policies regarding COVID. But the topic of this post warrants a deeper dive, hence we'll use the blog as our springboard. So we won't be talking about manipulating graph scales (which Norman has done), or mining data (which he's also done). We'll be talking about selectively changing metrics at the first sign of improvement in the old ones, so as to continue the exertion of control that the Wizard justifies by fomenting undue fear of the virus.
You may recall that, in the beginning, the metric was "flattening the curve." This meant not overwhelming the health care system. It had nothing to do with reducing cases. In fact, it had nothing to do with reducing deaths. The concept was that by stretching those cases and deaths out over a longer time period, we wouldn't overextend the capacity of our hospitals: beds, ICU beds, ventilators, and health care professionals.
Mission accomplished. We crushed that one. Kansas never came close to being overwhelmed. Total hospitalizations to date are a fraction of the total number of beds in the state, and remember that those hospitalizations have occurred over a period of six months. The number one location of clusters in the state was meatpacking plants - more than in correctional facilities and long-term care facilities. And whenever there was an outbreak at a meatpacking plant in western Kansas, there were ample doctors and nurses to dispatch from elsewhere in the state to deal with it. (I know, because I stayed in one of the extended stay hotels in one of those towns, and the desk clerk told me they had kept busy due to the health care workers staying there.)
So we flattened the curve - but wait. The Wizard was still reluctant to re-open the state economy. So gating criteria were established, based on rolling averages of new cases, hospitalizations, and deaths. Let's look at each of those in turn.
Overall, without looking at such metrics on a relative basis (cases and deaths per capita, hospitalizations vs. available resources), they are meaningless. Regarding deaths, first of all, the "flatten the curve" strategy was never about preventing deaths, remember? This is a virus. People are going to get sick. If they're already in poor health, they may die. But it ain't the plague. And second, daily new deaths in Kansas attributed (key word, with an asterisk after it, please) to COVID more than doubled from mid-August to mid-September, reaching 26 on September 16.
Twenty-six. That's .0009% of the Kansas population. Tragic for those lost and their families, but hardly sufficient criteria for determining whether businesses should be open or at what capacity, or whether someone can have a wedding with their full guest list. More people than that die each day in Kansas from cancer and heart disease, but the Wizard doesn't exert control over every Kansan's life to try to prevent those deaths. From the media, we hear, "Daily new deaths double in one month!" instead of "Daily new cases remain below .001% on peak day!"
As for hospitalizations, again, that was never an issue. And for many weeks, deaths were declining. So instead, the Wizard focused on new cases. Well, new cases did indeed rise after Kansas began re-opening. So hell-bent on maintaining control was the Governor that she even added a "Phase 1.5" to her phased re-opening plan. (Reminds me of the ineffective parenting ploy of telling your unruly kid that you're going to count to three to get them to stop doing something, and when you get to two, you start using fractions to allow the kid more time to misbehave before actually doing something about it.)
You know what else began increasing in Kansas about the time re-opening began?
Testing. Early on, Kansas was dead-last in testing per capita among all states. Today, it's "only" fifth from last. Regardless, increased testing is going to inevitably result in increased cases. For those who still don't believe this, consider that in places where testing has declined, so have cases. You accept that, so why can't you accept the converse? It doesn't mean that testing causes cases; let's not fall into the trap of confusing correlation with causation. What it does mean is that there have been more cases out there all along, since March, than we ever knew. Testing merely reveals them. And that means that this thing is a lot less deadly than all the malevolent Wizards would have you believe, because if the total deaths (known, but inflated by combining "died with COVID" with "died from COVID") were divided by the real number of total cases (which we'll never know), the mortality rate would be much, much lower than what's reported.
With much of the Kansas economy open (but with a mask mandate and remaining capacity limitations for some businesses), the Wizard needed a new target for exerting control. As Fall approached, that target became the schools. So a new phased plan was developed (Gov. Kelly loves her phases). And new criteria were established, chief among them being the "positivity rate." This is defined as the number of positive tests divided by the number of tests administered. Now, I could go after any of the other criteria, as they're all flawed. But let's focus on this one, because it's not just flawed, it's so statistically distorted that it means absolutely nothing.
First, while it's defined as the number of positive tests divided by the number of tests administered, that's not how the Wizard calculates it. It's being calculated as the total number of cases divided by the number of tests administered.
The first flaw in this methodology (besides the fact that it's not true to its definition) is that numerous people have been diagnosed on the basis of symptoms, and without a test being administered. They show up in the case count, but there's no positive test associated with it. (Don't believe me? I know one such person.) Is it possible to get such a diagnosis wrong? Could the symptoms actually be the flu?
Well, I had several of the COVID symptoms in early February, and I'd just spent a week on a cruise ship after spending the night in the port city, where a large festival was taking place with 100,000 attendees, so I was exposed to a lot of people. I might have thought later that I'd had COVID - except that, at the time, I tested positive for Influenza A. "Ah," you say, "but you're not a doctor. It's unlikely that a doctor would make such a mis-diagnosis."
Ya think? Let's set aside the fact that even the "experts" like Fauci still know precious little about this virus. My own doctor - who's been in practice for decades, and has treated infectious diseases both in the U.S. and at his clinic in Africa - also thought he had it in February. He had more of the symptoms than I did. He had an antibody test run (as did I, just to be sure) and it was negative (as was mine). So yes, mis-diagnoses do happen, especially for illnesses that have similar symptoms. It's why we test, whether for COVID or the flu.
The upshot of this is that total cases probably include some that were diagnosed on the basis of symptoms, and thus might not be actual COVID cases, thus they would tend to overstate the positivity rate.
A related issue that actually would result in a higher positivity rate is the fact that there are probably a lot of people walking around who have had COVID, dating back to March or even February, but never had symptoms and never got tested. "Aha!" you say, "that means the positivity rate is too high!"
High relative to what? What constitutes an acceptable positivity rate? The fact that there might be a large number of unreported cases out there (even the CDC has estimated that it may be as high as 2.5 times the number of reported cases) does speak to the infection rate. But, more importantly, it also shows that the hospitalization and death rates are far lower than what's reported. Thus this illness probably spreads more easily than we know, but it's far less deadly than we think, unless you're at high risk.
I could talk about the influence of false positives and false negatives in testing, but I won't waste time on that, because the impact would be relatively minor, and possibly offsetting.
But here's my real statistical issue with the positivity rate: it, like total cases, is a function of testing. There are two reasons for this. The first is that the lower the number of tests administered, the fewer cases you're likely to detect. But the second, and more important, is a function of who's being tested.
Have you been tested, even if you didn't have symptoms and hadn't been in contact with anyone who was infected? Would you be? Just for the sake of science, would you make an appointment or wait in line at a drive-thru testing site, and have a swab shoved halfway into your brain, if you felt perfectly fine? The vast majority of us would say, "No thanks."
So who's being tested?
People who have symptoms, or have been in contact with a person known to be infected.
Read it again, slowly. Those people will be more likely to have COVID than those of us who are just walking around, feeling fine, not having a lot of contact with people other than our families. So of course the positivity rate among those tested will be higher. Yet still, the Kansas rate is just 11.3%. Think of it this way: that's 88.7% of the population that aren't positive for COVID, even if they have symptoms (less likely) or have been in contact with a positive case (more likely).
In samples of people that were tested without symptoms or contact with an infected person, the numbers are starkly different. At the Tour de France recently, 785 tests were administered to riders and team personnel. The riders had been racing for two weeks in close quarters, without masks, spittle and mucus flying (gross, but true). Sure, they took precautions before and after each stage, but are you telling me that COVID can only spread when you're hanging out talking to reporters, or getting a massage, and not while you're racing?
How many positives? Zero. Zilch. Nada. Positivity rate: 0.0%.
At a Johnson County, Kansas hospital, 7,500 pre-op patients were tested (you have to get a test before having a procedure done in a hospital, including giving birth). These were people without symptoms and without contact with an infected person. The positive rate? 0.6%. So while there may be a lot more people who have had it back in the early months of the pandemic, there don't seem to be a lot of asymptomatic people with it.
Beyond that, how does the math play into this metric in terms of the relationship between number of tests and positivity rate? The table and graph that appear at the end of this post break it down, but I'll summarize here.
- Rhode Island has done the most testing per capita of any state, having administered over 637,000 tests per 1 million population (1M). Its positive rate is 3.5%, which is below the Kansas threshold for fully re-opening schools.
- Alaska is second, having administered more than 578,000 tests/1M. Its positive rate is 1.6%.
- New York has tested over 504,000/1M population. Despite having been the epicenter of the virus in the U.S., with more deaths than any other state, and more than Texas and Florida combined (in spite of having a smaller population than either of those states), New York's positive rate is 4.9%, again below the Kansas threshold to re-open the schools.
- As noted before, Kansas has administered the fifth-lowest tests/1M of any state, at less than 163,000/1M. That's just over half the average of all states. Thus its positivity rate is the aforementioned 11.3%, which is in the "Remote Only" range of the state's recommended guidelines (fortunately for the kids, most school districts aren't following those guidelines entirely - more on that later).
- To illustrate the influence of testing on the positive rate, consider this: the average positive rate of the top 25 states in terms of tests/1M is 5.4%. (If you look at a line of best fit on the graph below, it follows the top 18, which have an average rate of 5.2%.) The average rate of the bottom 25 is 8.3% (again, a line of best fit would follow the bottom 18, which have an average rate of 9.8% - nearly twice the top 18). The distortion is clearly evident.
Using that quite reasonable assumption, if Kansas had tested at the national average rate of tests/1M, its positive rate would be just 6.2%. And if it had tested at 500,000 tests/1M, as the top three states listed above did, its positive rate would be just 3.6%.
To further put the positive rate in its proper place (the dustbin of mathematics), consider that Kansas' rate ranks eighth among all states (only one state out of the top 33 in testing has a positive rate above 10%). Yet, Kansas ranks 29th in cases/1M, and 42nd in deaths/1M. Hardly a hotspot.
As an aside, but related, for anyone who still believes the debunked "study" that used disconnected extrapolation to claim that the Sturgis motorcycle rally resulted in 260,000 infections, I must ask: are you that completely innumerate? If so, you shouldn't be trusted with money, among other things.
About 460,000 people attended the Sturgis rally. If that resulted in 260,000 infections, we have an infection rate of 57%. "Well," you say, "it was a super-spreader event, after all." Riiiight. As noted previously, the primary cluster sites in Kansas were meatpacking plants. Those resulted in about 3,500 cases, among workers and within those communities. Now, a precise per capita figure isn't possible, as we can't define the "capita" in question precisely. But if we just look at the populations of the two counties where those plants are concentrated, it's about 70,000. So the infection rate per capita from those clusters is probably about 5%. If you actually believe that Sturgis could have produced an infection rate 11 times that of those clusters - well, let's just say you'd better protect your fingers and toes, because you probably need them to be able to count. (The same people who still believe this also believe that no cases resulted from protests or riots.)
Finally, a word about remote learning: it's bad for the kids, bad for parents, and bad for most teachers. The kids don't like it, the parents don't like it, and most teachers don't like it. Anecdotally, every kid I know who was given a choice between in-person and remote learning chose the former. And in every case I know of where the choice was the latter, the parent made that choice, not the kid.
In the Kansas City area, on the first day of remote learning this school year, several districts experienced widespread technology issues so severe that kids couldn't even connect, and thus didn't really start "school" until the next day. (This makes a very strong case for school choice. In the private sector, we test technology before putting it into production.)
The lesson in all of this is that anytime the Wizard begins to speak, it's a safe bet that levers are being pulled and switches are being thrown behind the curtain, all in an effort to distort reality. Let's see the Wizard for the sham s/he is.
No comments:
Post a Comment