As the COVID-19 situation is expected to get worse in the coming weeks in Florida, let’s take a few moments to go through a few common falsehoods you’ll encounter.
LIE # 1: Cases don’t matter. The only thing that matters is (INSERT EVER-CHANGING METRIC THAT DOWNPLAYS THE SEVERITY OF THE SITUATION).
This one’s easy. The number of cases is the fundamental measure of the state of things. Positive tests precede all other indicators and is, in general, the best, most reliable, easiest to follow measure of the spread of a disease. If you want to adjust for number of tests, or the proportion of positive tests – have at it, but it doesn’t matter that much. In every place since the start of the outbreak, as cases increase things get worse.
But who cares about all these new cases? They’re just a bunch of stupid college kids and dumb babies that are 100% immune!
Let’s take a look at the relationship between cases and hospitalizations:
Unless hospitalizations just happen to be changing by coincidence, the number of newly sick people (as measured in cases) likely determines the number of hospitalizations. Who knew!?
As the number of new cases in the previous two weeks increases, so do the number of currently hospitalized. For every increase in 1700 cases over the previous 14 days, the number of people hospitalized increases by 100 on average.
Note that hospitalization data was only released July 10 when cases among ‘young and healthy’ people were at their highest. Even when the record-breaking surge in cases in July was being downplayed as primarily young people, it made no difference on the number that were hospitalized (at least from a statistical modelling standpoint). Despite changes in percent positivity, number of tests, proportion of young over this period – the best single predictor is the volume of cases.
LIE # 2: The number of deaths reported is not AT ALL related to the state of things right now. The Media is intentionally reporting new deaths as if they occurred yesterday to scare people and stretch out the ‘Scamdemic.’
There are plenty of people that have taken up this crusade and staked their reputation (whatever that was worth) on this lie. Despite all reputable news sources clearly stating deaths as ‘Newly Reported,’ the Deaths-by-Date-of-Death Truthers cannot be deterred! This post describes the lie in more detail, but a short summary is that deaths reported on a day did not all occur on that day. And by the Truthers logic, all deaths are therefore suspect because there is a delay in reporting and processing deaths (and may or may not be part of a devious plot by George Soros and/or Bill Gates to form a New World Order and defeat the God-Emperor, Donald J. Trump).
While ‘deaths reported’ and ‘deaths by date of death’ are not the same, they are highly correlated. Depending on the rate at which deaths are reported (which varies and is increasingly unpredictable), the two should be close in value – and are, for the most part.
Consider the following graph, showing the relationship between deaths by date reported (x-axis) and deaths by date of death (y-axis). From March through August (data within the last 45 days is considered incomplete as deaths are still being added as far back as July), the correlation is unsurprisingly high. The number of deaths reported on a day may not be equal to deaths that occurred on the same day, but they are generally tracking on a monthly or even weekly scale.
Through September and October, with deaths on the decline, the correlation will decrease. It is expected that deaths reported in October will exceed the deaths that occurred in October because of the large backlog of deaths currently being reported.
The degree of correlation will depend on the proportion of recent vs backlog deaths that are reported but they are still correlated, nonetheless. In general, as true deaths increase so does the reported deaths (and vice-versa).
Any claim that deaths reported on a particular day is ‘unrelated’ or ‘not attached to reality’ to the deaths occurring on the day is blatantly false.
LIE #3: Deaths are decreasing because Deaths by Date of Death Graph shows a decrease. This is the actual number of people that died so it must be right. Facts over Feelings, Libtard!
This is related to Lie #2. When the number of deaths by date are plotted over time, we get the infamous ‘Deaths by Date of Death Graph,’ or DBDOD graphs. See some examples below from the Florida Department of Health COVID-19 dashboard from August 4 and October 5.
No matter when this data is shown, the graph will always show an artificial plateau or decrease in deaths due to the lag in reporting – even in periods of increasing record deaths. The left panel was displayed on August 4 when there were a record 236 deaths. The right panel, while deaths are truly decreasing, shows a steady decrease to zero. (If you want to see more examples, go to Covid Tracking Project’s website where they keep daily screenshots).
Admittedly, deaths by date is the gold standard and waiting months to identify trends is not ideal in a pandemic. Many projections of the true deaths by day rely on historical trends in death reporting and can become inaccurate with changes in reporting lag.
Some reasonable projections can be done based on this method, but the unfortunate truth is that the people spreading Lie #2 are not interested in projections of true deaths by date of death. Their sole agenda is to show decreasing deaths with no attempt to correct for the lag in reporting.
So how should this important data be presented in an honest and accurate way? One way is to remove the last several weeks or so entirely (previously DOH removed the prior three days’ worth of data, back when deaths were reported as they occurred, but DOH stopped doing that after they fired Rebekah), only showing data that is reliable and reasonably complete. What is lost in timeliness is made up for in accuracy. Even better is to add projections.
The graph above is the usual deaths by date of death graph (starting July 1) but with the projected number of deaths for the most recent 45 days added. Unlike ones that only show the reported deaths, this tells a different story – especially if deaths are increasing and a backlog is growing like what occurred over the summer in Florida. How are the projections made? Just as cases lead to hospitalizations, hospitalizations lead to deaths.
There is nothing too complicated or unexpected about how this is made. Recognizing that the number of hospitalizations on a given day is correlated with the number of deaths on the same day, we can build a simple but reasonable model using this relationship.
All data used is older than 45 days and there is some danger of extrapolation (the lowest number of hospitalized in the model is about 3500 while there are currently about 2000 hospitalized). But, unless the relationship between people hospitalized and people dying changes drastically or treatment has significantly improved in the past six weeks, it is reasonable to assume this relationship would remain stable.
Adding a second term for the cyclical nature of deaths within a week (peeking on Wednesday), we get what is shown in the graph.
It is a Poisson regression with three variables: hospitalizations, hospitalizations 7 days ago (to correct for autocorrelation) and day of week.
After finding the most likely value (red dots on the graph), we also plot 95% prediction limits to show the uncertainty in the predictions (red lines). A black dashed line is added to show the smoothed trend.
The estimated sum of outstanding deaths is shown in the bottom left corner. Based on the model, there are between 118 and 1,227 deaths (with 298 being the best estimate) that have already occurred that have yet to be reported.
Unlike the usual DBDOD graphs that portray deaths as always decreasing, a more accurate prediction shows deaths flattening, mirroring the flattening of hospitalizations. As cases and hospitalizations are expected to increase, remember that DBDOD graphs will be used to push an agenda as it was all summer even as deaths were peaking. We’ll update this as the situation changes but, in the meantime, follow the data and be vigilant. We’ve seen these lies before and we’re bound to see them again.