That was my response to this:
Which was looking at excess deaths before Covid-19, in 2019.
I was saying those numbers are probably from a worse than average flu season, which was also pretty much over by the time Covid-19 was really taking off.
I also said pretty much what you did about the attribution issue with pneumonia as a cause of death.
I don’t want to really go into it in this forum (as it’s really not the right place for it), but let’s just say that I don’t necessarily see all posts in this thread. 
From a total-death standpoint, weeks 1 through 7 look not terribly bad, though still 738 per weeks on average higher than the average of the prior six years. The five weeks after that get pretty bad, though – and interestingly, the twelve weeks before look pretty terrible. You appear to have a relatively mild bit of excess – but still excess, no doubt at that, over the average, even if actually lower in number than three of those years – for the first seven weeks of 2020, with some pretty serious looking damage surrounding it.
Those weeks stand out so much, in fact, that one wonders if something changed in the accounting from week 52 of 2019 to week 1 of 2020. Like, it does make you wonder if there is something about the calendar year that could lead to delayed reporting – meaning we can expect those early week numbers to rise in a final accounting.
Just to highlight things for you, try to spot the trends in this column, if any:
1955
2772
3678
3174
2431
4150
3357
3756
2411
3198
2984
1746
2036
758
561
536
916
918
1704
1219
2010
2538
2393
2312
3567
8193
17487
24573
23343
20778
16037
13880
12080
8768
6433
4503
2856
Those are ‘excess deaths’ for six months worth of weeks spanning end 2019 to early 2020, using the average number of deaths per week from those same weeks the six prior years.
One thing to keep in mind is that there are several different pathways of information about COVID deaths, as there are about flu deaths and other epidemic/pandemic diseases.
The death certificates are the gold standard, representing the best and most complete information, but they also lag the longest. That’s what is used for the CSV files we’re bandying about here, compiled once the death certificates have been filed with the county or state health department, transmitted to the feds, and analyzed/coded/sliced/diced into categories. The physician has signed the death certificate with his/her best judgment as to cause of death, any necessary autopsies have been completed, the medical examiner has signed off if required, the family has provided information about the deceased’s age, occupation, marital status, military service, etc., and the funeral director has certified the method of disposition of the remains.
This is the most complete information about the decedent and their demise that will be available to the heath authorities, but it takes time to gather. If the physician is busy and hasn’t got time to review certificates, the unfiled certificates are sitting on his/her desk awaiting a signature. If Aunt May is in cold storage waiting for her family to decide where she’s going to be buried (or trying to come up with money to bury her), then the unfiled certificate is sitting on some funeral director’s desk. There is always some lag there, and this year, with some doctors and hospitals and morgues and funeral homes completely overwhelmed, the lag is greater than normal.
The death certificate system isn’t really designed for rapid surveillance of fast-moving disease outbreaks; there are other systems for that. For example, the CDC set up the National Healthcare Safety Network for facilities to report information about hospitalized patients, originally mostly for healthcare-associated infections but repurposed to include influenza and COVID-19 cases. (This is the system the Trump Administration sidelined in mid-July in favor of a new system at HHS.) Hospitals and other facilities are asked to supply information about number of patients, number of deaths, number on mechanical ventilators, number of admissions, quantity of PPE, etc., on a daily basis. This is the data driving daily decision-making, but it has some limitations. Notably, facilities are not necessarily required to participate, depending on facility type, state/jurisdiction, etc., but do so only on a voluntary basis, and participation varies. (By contrast, every state requires a death certificate.) Also, this is preliminary data, and there can be some fuzziness; for example, who exactly qualifies as a “suspected COVID-19 case”? The feds provide some definitions, but there is still room for interpretation and nobody is reviewing exactly how any given hospital chooses to interpret.
Allow me to jump in here, as someone who has downloaded every weekly version of that dataset back to early June, and modeled them to death. The short answer to the OP’s question is that he massively understates the delay in reporting when he says it “takes three weeks for the data to come in” - in fact, only the states with the best reporting come in that quickly. Some example states
Georgia - it takes about 7 weeks
Pennsylvania - amazingly, data from March is still updating now. I have no idea how long it takes.
Connecticut - six months? this is a reporting trainwreck.
Florida - 4-5 weeks.
Texas - 7-9
And so on. The CDC tries to adjust for that by only using ‘mature’ weeks; I adjust for it by trying to estimate reporting patterns by state. When I last stopped, I was at about 65K underreported deaths from natural causes (so I excluded murder, suicide, MVA, and I forget what else).
In any case, I’m not sure why the OP is looking at raw files when the CDC itself does the estimates (page down, toggle to Number of Excess Deaths).
My only complaint is that the low end of their range is estimated against the top of the confidence interval for historical run rate; but the high end of their range is against the historical midpoint estimate, not the lower end of the confidence interval. Being conservative I guess.
I missed the edit window, but wanted to add that I’ve been downloading the weekly, state-level version directly from the CDC, not the summary version he posted above. Just wanted to be clear.
The premise of the OP of this thread was that the number of reported COVID deaths was larger than the number excess of deaths in the first 25 weeks of 2020 compared to other years. This would be evidence that the number of COVID deaths had been over-reported if it was accurate, but there were a couple of flaws in that analysis. The main one being that he was comparing 25 weeks 2020 deaths by all causes to 36 weeks of COVID deaths.
When we compare similar time periods we see that there were over 50k more excess deaths in the first 25 weeks of 2020 compared to the average of the last six years than we account for by reported COVID deaths in that time period. This is evidence that the number of COVID deaths had been under-reported over that time period.
That’s it. That’s the whole argument. The OP said there were less excess deaths than could be accounted for by reported COVID deaths, the truth is there are more excess deaths than can be accounted for by reported COVID deaths.
The data linked in the OP does not give us reason to think COVID deaths have been over-reported. It gives us reason to believe COVID deaths have been under-reported.
i get it. Your argument is that if deaths in 2020 do not match average deaths from the last six years, the difference is either fully down to COVID or we have ourselves a discrepancy.
That’s really great information. Thanks very much for sharing all that. I consider myself enlightened!
You don’t get it or you wouldn’t be talking about ‘baseline excess deaths’ and misstating my argument.
The OP asked about a discrepancy, but that discrepancy has been explained by mismatched time periods. There is an actual discrepancy in the other direction than the OP posited when similar time periods are compared.
A discrepancy exists. Part, but not all, of the discrepancy is almost certainly due to under-reporting of COVID deaths. There’s no reason to believe that COVID deaths have been over-reported based on this data set.
I’m not talking about the OP’s observations. I’m talking about the chart you showed that identified a ‘discrepancy’ between ‘excess deaths’ and ‘COVID deaths’. I understand a discrepancy to happen when things that are supposed to match do not match. That’s what I’m talking about. And yes, I fully get what you mean. I may not see it exactly the same way you do, but I certainly understand your claim.
I was talking about the OP’s observations. This thread is about the OP’s observations.
Why are you putting quotes around those terms? There is, in fact, a discrepancy between excess deaths in the first 25 weeks of 2020 compared to the average of the first 25 weeks of the previous six years and reported COVID deaths.
Not a ‘discrepancy’, a discrepancy.
Okay, now maybe I’m no longer following you. You are saying that we should add:
and the sum will equal:
1+2) Average total deaths from all causes during same time frame over the last six years
I just don’t believe the claim that such an equation is necessarily true.
Are you fucking with me?
That’s what I’m saying, and what I’m saying is a straightforward and accurate representation of the data set linked in the OP.
Yes, your item #3 is certainly a straightforward and true representation of the data. Saying that there is a discrepancy between #1 and #2 does not necessarily follow. (There would, of course, also be differences between #1 and every other specific cause of death.)
If it did follow, though, to what should we attribute the same discrepancy that can be found if we look at the weeks running up to COVID?
Can you explain the exclusion of murder, suicide, MVA, etc.? They are potentially impacted by COVID-19 and its impacts as well.
(And again not only are the actual numbers delayed, the potential impacts are as well.)
It does.
There were 179k excess deaths in the first 25 weeks compared to the average of the previous six years over that time period. 125k can be attributed to reported COVID deaths in the first 25 weeks of 2020, but that still leaves 54k excess deaths. This kind of thing is called a discrepancy.
You seem to be arguing, poorly, that the discrepancy might not be due to COVID so it is not a discrepancy. I’m not really sure to be honest. Your baseline excess deaths musings seem to be an attempt to explain the discrepancy.
But we know we have excess COVID deaths in 2020 and no has put forth a reason that some other cause of death would be higher in 2020 than the previous six years. Furthermore, if there was some other cause of death that was higher in 2020 that would be an explanation for the discrepancy. It wouldn’t mean there wasn’t a discrepancy.
Fidget spinners, leprechauns, something other than COVID, double secret COVID. Go nuts.
If you make a good argument, it will be the reason for the discrepancy. It won’t mean that no discrepancy exists.
If the 54k excess can be explained by statistical noise, or flu, or an increasing year over year trend, that’s great. Discrepancy explained. It would not mean there never was a discrepancy.
Again, you are using the word discrepancy wrong. Just because there is a difference between two numbers doesn’t mean there is a discrepancy. Sometimes there is just a difference. By your argument, the measure of ‘excess deaths’ itself is a discrepancy, as it means that a measure in one year doesn’t match the average of the prior years. For that to be true, you would first have to establish that there is reason to expect them to necessarily match in the first place.
Let me try an analogy. Let’s say you think your company will have a good year and sell 50 more widgets than the average of the past six years. Turns out you’re right, but you in fact undershot it by a lot because an angel came along and bought another 150. So, now you have sold 200 more widgets this year than you did the last six years on average.
The number [average last six years plus 50] is not in discrepancy with the number [average last six years]. It’s just different. It reflects your good year. There is no common usage of discrepancy that would apply here.
And more to the point, the number [this year’s sales minus the lucky deal for 150] is not in discrepancy with the number [average last six years]. It’s different, yes. The difference is the increase you expected.
The data in the table clearly shows that the months leading into COVID were deadly months for the US by comparison to the average of the six years prior. You would call that a discrepancy that needs to be explained, I guess. I would just say that more people died than did the prior six years, of something. I’d see the difference, and wouldn’t call it a discrepancy. But one thing is for sure: if I did call it a discrepancy, I wouldn’t expect it to suddenly disappear when the next one came along.
You are wrong about this, but I don’t see the point of continuing this discussion.
Please substitute ‘unexplained difference’ for ‘discrepancy’ in my posts if that helps.