Definition of discrepancy
1 : the quality or state of disagreeing or being at variance
2 : an instance of disagreeing or being at variance
Discrepancy Definition & Meaning - Merriam-Webster
You were saying?
Sorry but you lose me here anyway. If something is within statistical noise then there is no discrepancy to be explained.
I’m not bothering to check any math at this point but when enough time has passed there are lots of factors that will lead to excess deaths and COVID-19 deaths to be different, without implying over or under reporting. Fewer MVAs, fewer respiratory deaths due to less particulate pollution, shifting of deaths to first half of year from second half, more suicides, more overdoses, more dying from care delayed out of fear to go in or appointments cancelled or screenings not done, and so on. In net will they result in more or less than reported and attributed COVID-19 deaths. I think way too soon to know.
I accept your apology.
In that case: whatever.
Many reasons. 1) They aren’t itemized separately - the CDC provides “All Causes” and “Natural Causes”, with “Natural” broken out into great ICD detail, and I like to be able to examine the detail; 2) I realized after a while that natural & “unnatural” deaths had different reporting patterns (natural arrives faster) , and I was already building 50 models…I didn’t want to separate them and build 100 models; and finally, 3) “unnatural” deaths historically only make up about 9% of total deaths, so they don’t move the dial that much.
All that said, I did look at the “unnatural” deaths for a while, and through early July-ish there wasn’t much of a change in pattern, maybe a little bit down. But hard to tell with the slow reporting & no model to adjust.
Thank you.
At what point would you think there was enough time passed that you would include “not natural” causes and no longer feel the need to model individual states?
I wouldn’t bother, honestly, the numbers are so small. That said, you can look for yourself - below I’ve put the deaths from natural causes side-by-side with ‘unnatural’ causes, with the same scaling for both. NOTE: as I’ve already said many times, please do NOT look at the more recent data, I’d say at least eight weeks worth…I just dumped it all in for your information. Apologies for the minimal Tableau skills.
On a side note, if the unnatural death graph is expandable, you’ll see a noticeable trend year over year…I assume that’s opioids. 
In weeks 1 through 14, 2020 is only the third highest in pneumonia deaths, and there is only a tiny variance between all of the years. Weeks 15 through 27, 2020 pneumonia deaths are double what they were in other years, and again, little variance between the rest of the years. Apparently, pneumonia deaths are highly predictable. Using that data, a shit ton of the underreported COVID deaths were attributed to pneumonia (which obviously could have been a factor IN ADDITION TO COVID-19).
So, while the OP was completely backwards in how he was interpreting the data, he did provide the necessary data to help find some of the missing deaths. Thanks, I guess.
Here’s a chart to help illustrate the point I was trying to make about there being no reason to necessarily expect the [average of past six years] to be a proper projection for the current year.
For whatever reason(s), those six-year averages (the red line) haven’t been good proxies for quite some time now. It is senseless to find a ‘discrepancy’ between two things that evidently aren’t expected to match in the first place.
In other words – yes, a very simple analysis like this would suggest that there is some sort of ‘discrepancy’ already baked into the model, if the model you have chosen is to say that you expect deaths in 2020 to line up (or else) with deaths from that particular baseline.
Perhaps it would improve the model a fair amount to use only the past three years. Or perhaps it would be somewhat enlightening to plot 2020 against the ‘max’ of the past several years. Or perhaps some of this goes away if we adjust for population changes – I have no idea. But I do know that you are going to have to live with a wide range of uncertainty if you define ‘excess deaths’ so simply.
Or maybe COVID has been killing Americans longer than we thought. There’s an interesting hypothesis.
We all understood the (irrelevant) point that you were trying to make.
Well, you didn’t seem to care, when you were so married to 54K as your target number.
54k (or thereabouts) is the target number.
Through 25 weeks of this year, there were 54k more deaths than an average year even after we factor in reported COVID deaths.
Repeatedly saying that we should not be making that comparison doesn’t change that fact. That’s the comparison that this thread is about.
Repeatedly saying that some of those deaths already appeared in the data before COVID doesn’t change the fact that those deaths exist. I showed that some of those deaths existed before COVID in the first dataviz I posted in this thread in post 7. You are not adding anything new to the discussion.
I can’t tell whether you are trying to change the topic of this thread or utterly failing to understand what the topic of this thread is.
The number of excess deaths from pneumonia this year is in that general range. What’s really telling is that when using a boxplot to show dispersion, anything further than 1.5 * IQR from the 25% or 75% quartile is considered an outlier. Metrics more than 3 * IQR are considered extreme outliers. Pneumonia deaths in 2020 are over 8.5 times the IQR. That’s just insane.
Remarkable.
It’s my understanding that MEs are testing every corpse (that doesn’t have a test on file) for COVID. Not necessarily for attribution of cause of death but simply for epidemiology purposes.
Where do you think that’s happening? Anything you can link to? Even if that is happening now, it certainly wasn’t a fairly short time ago.
That’s news to me. Medical examiners have always evaluated unexpected deaths for infectious disease, but I haven’t seen anything about testing every corpse.
I don’t think ME even see most bodies, do they?
That’s been declining, and around a decade ago it fell to ~ 5% to 8%, depending on what source I found. And a lot of places don’t have MEs, just coroners, so if suddenly everyone is getting a test - as this rumor proposes - they’re having to send the bodies elsewhere.
A quick Google search brought this up. Coronavirus Testing For The Dead? It Can Help Reveal The Scope Of The Pandemic : Shots - Health News : NPR
It does lean toward testing only the unexpected.