A Note on The Relationship Between Covid-19 Death Occurrence and Reported Date

Article by the badger

Much has been written about the relationship between date of death and the date the death is reported by Florida DOH.

Despite it being a relatively trivial footnote in the Florida Coronavirus Saga, it has become somewhat of an obsession by some laymen.

While there is value in studying the distribution of time between date of death and the date that the death is reported, it has increasingly been used to push a conspiratorial narrative (i.e. ‘The Media’ is intentionally reporting deaths as increasing when they are actually decreasing). The basic argument – treated as some grand revelation worthy of a Nobel – is that the number of deaths on a day is not actually the number of people that have died that day; well, no shit. Would you like a cookie, Mr. Cole?

What Professor Cole and his like-minded disciples seem to ignore is the nature of arithmetic itself. The number of deaths reported must equal the number of deaths that have occurred. If some time periods are over-reported (reported deaths on a day greater than actual deaths) then, by definition, some days must be under-reported. Where is the outrage in the “misleading” reporting of that?Though you will never hear these self-proclaimed prophets mention this, the numbers do not lie. Let’s have a look.

The chart above shows the number of deaths reported (in blue) compared to the number of deaths that occurred (in red) on each day. Dates in the past 35 days are excluded as some deaths are likely to have been unreported thus far. The lines are added to smoothen the day-to-day noise (fitted with a generalized additive model.

We can see that in periods of increasing deaths (March 21 – April 10 and June 20 – July 20) that the actual deaths per date are greater than the reported deaths per date. This is not a conspiracy of “The Media” or Florida DOH; it is simply the nature of having to process hundreds of additional deaths. If we look at the first peek of deaths in mid-April, we observe a corresponding peek several weeks later in reported deaths – a ‘catch-up’ period (perhaps what we are seeing in reported deaths now). As deaths plateau and decline, the two are generally equal.

In theory, one could look at the past distribution of days between death and report to provide some inference into the ‘true peak’ of deaths. For example, if 90% of deaths are reported within two weeks, simply divide the count from two weeks ago by 0.9 to get an estimate of true deaths. While this may be a sound method assuming a stable distribution, changes in the distribution of lags makes this much more difficult in practice. And therein lies the rub.

The chart above shows that the distribution of lags has steadily been increasing over the past three weeks.

Deaths reported between August 4-10 had a median time of five days; that is, half the deaths reported occurred within the previous five days. Ninety-five percent occurred within 22 days. Compare this to the most recent week (August 18-24) with a 50th and 95th quantile of 8 and 33 days, respectively. Accounting for these unpredictable changes to ‘fill in’ reported deaths seem to be a fool’s errand rife with erroneous conclusions. All of this, of course, ignores the deaths that have yet to be counted. We know given a death is reported, the distribution of the date of death. It is increasingly difficult to know the reverse (which will never be truly known until long from now).

So what is the purpose in rushing to pin-point peaks of actual death other than to bolster a political ideal or agenda?

Florida officials have become infamous for premature celebration during this pandemic, so some measure of patience is needed when dealing with its data.

With the recent change in how medical examiner’s report deaths to the state, it’s unclear whether death information will be processed when DOH receives it, or if additional validation of COVID deaths will be required, further slowing down the rate of reporting, so all of this may be moot.

%d bloggers like this: