Template talk:COVID-19 pandemic data/United States daily deaths post emergency

Modeled Deaths

Since so many states are definitely not reporting deaths post emergency and more will probably join them, I took the number of deaths for the year ending 2023-03-31 and used that as a way of modeling how many deaths would be reported if we were still getting all the data. I chose that because 2020 had such localized outbreaks, 2020 and 2021 had so many differences in local restrictions as well as the speed and volume of vaccine uptake, whereas over the past year there've been pretty much no restrictions pretty much anywhere and vaccine levels within a given state have been much more level. The states I've marked with an "x" have explicitly said they're not reporting any more, but of course other states could have not said so yet or just aren't going to bother saying it outright, so I anticipate this being pretty fluid to start. Kinerd518 (talk) 15:09, 14 May 2023 (UTC)

Modeled Deaths Review

In the first two months of this project, there have been seven occasions where a state seemed to stop reporting only to resume later, resulting in a removal of x marks. If the model is perfect, the sum of these occasions should have resulted in an increase in reported deaths and no change at all in modeled deaths. What actually happened summing over all those occasions was an increase of 343 in reported deaths and an increase of 14 in modeled deaths. There is no evidence the model is flawed, and it may be very slightly conservative.

Using the modeled totals, there are two states on either end that are massive outliers from the last reported year, and both have shown remarkable irregularities in reporting per se.

Virginia would be expected to report 138 deaths, broadly similar to the 229 and 177 actually reported by similarly populated Michigan and Arizona respectively. Virginia in fact reported 24 deaths, comparable to the 29 and 22 actually reported by much smaller Utah and Vermont respectively. While it is relatively common for a state to report a large negative death result due to a change in methodology, Virginia multiple times in this period reported negative deaths with no explanation.

Kentucky would be expected to report 151 deaths, broadly similar to the 121 and 78 actually reported by similarly populated Minnesota and Louisiana respectively. Kentucky in fact reported 374, more than the 240 and 276 actually reported by much larger New York and Texas respectively. Kentucky like some other state websites does not display a total but requires mass selection to sum to a total, and like some other state websites defaults to a useless date range. Unlike any other state website, Kentucky defaults to two distinct useless date ranges (through 11 Feb 2023 on one section and through 26 Sep 2022 on another), offers different sums depending on which graphic is chosen(!), and changes the "updated on" date on that basis as well(!!!).

Obfuscation is common in COVID reporting, and it is unlikely based on prior experience that contacting these two states directly will offer any clarity, but with nothing to lose I am going to try to do so anyway. For now I will continue to include their claimed reports without modification. Kinerd518 (talk) 17:59, 17 July 2023 (UTC)

Virginia's response: "we are still doing our same process that we always have. Our numbers are mostly based on info we received from death certificates". This is insufficient on its face to explain the wild discrepancy between Virginia and other states, and noticeably does not explicitly state that Virginia's Department of Health is still receiving death certificates or any other data from all the same sources, and frankly that it took this long to get this non-answer is evidence in itself. Since they have chosen to not publicly quantify this, we have no way of modeling the actual number of COVID deaths in Virginia, and so I will continue to use their obviously incomplete data. I will also begin contacting Kentucky today. Kinerd518 (talk) 17:25, 28 July 2023 (UTC)

Kentucky's response: "The difference you noted occurred because the filters that the total deaths and the epidemic curve rely on are based on different variables. So any difference might be due to a field (ex: Report Week) not being completed for a record in the dataset. These types of issues are routinely monitored and addressed as part of our internal data quality processes, though on a different schedule than the website publication. For statewide total death counts, it's most straightforward to use the "Reporting Week" filter for the time period of interest." Also of note is that while the sums remain different, they are much closer now than they were on my first observation. Kinerd518 (talk) 18:42, 17 August 2023 (UTC)

Modeled Deaths Second Review

In the first six months of this project, there have been twelve occasions where a state stopped reporting only to resume later, resulting in a removal of x marks. If the model is perfect, the sum of these occasions should have resulted in an increase in reported deaths and no change at all in modeled deaths. What actually happened summing over all those occasions was an increase of 663 in reported deaths and a decrease of 42 in modeled deaths. There is no evidence the model is flawed, and it may be very slightly liberal.

Virginia remains the massive outlier of states still reporting in terms of underperforming the model, having reported 140 deaths when the model would expect it to report 500. Virginia has reported fewer deaths than New Mexico, West Virginia, Hawaii - clearly nonsense. I will contact them again, and do not expect any more response than last time. Kinerd518 (talk) 21:18, 12 November 2023 (UTC)

Virginia's response was that local data may not be as up to date as CDC data, and advised to contact CDC for data. Unfortunately they did not quantify how out of date the VDH data is, so neither can we, but at least we know now why Virginia reported deaths are such an outlier - they simply haven't reported them all, a coherent if deeply depressing explanation for the observed phenomenon. Although it is extremely unlikely that Virginia or anyone else will start doing a better job in the future, since it's merely a question of getting the data published, it is a possibility, and again with no way to quantify their shortfall I see no alternative but to keep using their source, no matter how starkly outdated it continues to be. — Preceding unsigned comment added by Kinerd518 (talk • contribs) 19:35, 15 November 2023 (UTC)

Florida 2023 Dec 18 Reporting Change

Florida increased their reported death total by 7,052 overnight. There is no explanation given whatsoever, including any indication of a definition change, and the change being extremely non uniform seems to me to rule that out: only 65 deaths were newly reported from 2020 to 2022, although some were in each year, leaving 6,987(!!!) newly reported deaths for 2023 alone, including 3,544 in the date range of this chart. Through 27 Oct 2023, Florida had been reporting on a roughly biweekly schedule, following which they reported daily, and their current resource offers data broken down by week, notably in a spreadsheet that is separate from and consistent with the yearly break down, which seems to rule out data entry error. I chose to re-enter all the data in this date period by week rather than put nearly 7,000 deaths as occurring today, and I increased the starting value by 3,508 to cover the 2020 through 12 May 2023 newly reported deaths. When I previously analyzed states for outliers relative to the previous year, Florida has not been particularly striking, and had if anything already been a little above expectations. I am at a complete loss to explain this, but I do feel the way I have charted it is the best we can do to reflect what they are currently claiming. Kinerd518 (talk) 18:55, 18 December 2023 (UTC)

Florida reverted this change today. No explanation. Kinerd518 (talk) 19:22, 19 December 2023 (UTC)

I analyzed the two versions Florida published. There is no conclusion but that they are purposefully lying about how many COVID deaths have occurred in Florida.

I compared the statewide totals week by week and saw they often approached but never quite exceeded double.

Florida COVID Death Obfuscation

Over 67 counties over 48 weeks, there were no results that exceeded double: 0 for 3,216.

1,391 county-weeks reported 0 in both versions.

0 county-weeks reported 0 in the lower version and nonzero in the higher version.

1 county-week reported 0 in the higher version and 1 in the lower version: Hernando 12/01/23.

Since the lower version is a couple of days newer, this is obviously a catchup posting.

Of the remaining 1,824 county-weeks:

55 are exactly the same, of which 48 were exactly 1 in both,

35 are between about 1.16x and 1.48x as much

379 are between exactly 1.5x and about 1.96x as much, and

1,355 are exactly 2x as much.

It is simply not possible for the higher version to be the result of accident or error. Over 95% of county-weeks have clearly gone through some version of "cut in half and round", and if we could compare versions from the same date that figure would only increase(!). Results being double recorded, for example in the county of death and the county of residence, would have caused many county-weeks to show up as zero in one version and nonzero in the other, since there are over 500 county-weeks with exactly 1 recorded death in the lower version, but this did not happen.

Explanations for why a few county-weeks haven't quite gone through "cut in half and round" are the same as for why this lie was revealed in the first place: Florida got sloppy and inadvertently posted the true numbers, which they had sloppily not lied about a few of in exactly the same way, or the dissenter(s) who posted Florida's true number has also been intermittently interfering with the lies Florida has been telling throughout.

I am attaching a link to what can only be described as the true COVID death numbers in Florida as of 2023 Dec 18. Because there is no reason to believe we will see true numbers again, I feel I have no choice but to continue to use the lies Florida is posting.

https://www.dropbox.com/scl/fi/ceem4ux992u8nbi4vnmd6/Covid19SurveillanceDeathsByWeek.xlsx?rlkey=c0eryqx13a4kjj3a4jlo1jafg&dl=0 Kinerd518 (talk) 19:10, 20 December 2023 (UTC)

Draft:Template:COVID-19 pandemic data/United States daily deaths post emergency

For this page, I found that the data is still available for Vermont (on the last page, see past surveillance for historical data) and Oregon (sum up all the numbers in the second section to get the total), and Kansas' data remained available until the end of last month (see here). It would be good if you could update the template! Horizon Sunset (talk) 13:07, 23 June 2023 (UTC)

this was great! the template has been fully updated, and i will use the existing sources going forwards. thanks :D Kinerd518 (talk) 17:04, 23 June 2023 (UTC)

More data

Hello Kinerd518, thank you for updating this page daily. I would like to know what you do when a state stops reporting cumulative fatality count but continue to report weekly count, such situation can get messy easily, as a state might be backfilling its data. And if possible, can you include Iowa also?

Iowa's weekly fatality count is available in its flu reports, but the cumulative count isn't reported, so I am not quite sure if it is possible to use it.

others

Mississippi's fatality count is recently updated here.

Arizona recently stopped reporting cumulative count, but the total death count I found by summing up the age group is slightly different from the counts reported.

Finally, for Kentucky, the weekly fatality counts on the new dashboard match almost perfectly with the previous dashboard, with the exception of the last week, which might be due to backfilling. --Horizon Sunset (talk) 04:59, 1 December 2023 (UTC)

ah thanks! :D what i do when a state stops reporting cumulatives is really on a case by case basis; for example CT stopped reporting in the "summer" and is reporting cumulative figures for this "season", so i marked them x for when they stopped and am using the cumulative figures since, but when TX stopped reporting cumulative they immediately started weekly reports covering the last two weeks, and since there are always higher deaths from two weeks ago i put those in too. as you point out, in whatever version of non cumulative reporting a state goes out of their way to start doing, information will be lost in both directions, but my feeling is posting deaths when they are reported is the best we can do given these versions of intentional obfuscation - we know deaths from two weeks ago happened two weeks and not one week ago, but we also know there will have been more deaths from three weeks ago that are never reported, so once one week ago right now is three weeks ago, it'll even out to some extent. it's an approach that frankly offends me with its sloppiness, but i see no better alternative

that MS source is excellent, and the IA source looks very promising! i will need to go through all of them after today's regular update, but certainly we'll be able to add some data. i'm cautiously optimistic the MS report might continue to be updated, since this would otherwise be such an odd time to issue a one-off report and they went months without updating otherwise, but either way, more data is better!

regarding AZ, my guess is that the sum over age being slightly smaller is due to deaths with age unknown. my experience with the various forms of binning is there pretty much always is an unknown category, even if it's not explicitly reported, and it explains this kind of phenomenon

and i've been very happy with how KY has improved their reporting, we can even get csv downloads now!

thanks for all your help :D Kinerd518 (talk) 17:37, 1 December 2023 (UTC)

@Horizon_Sunset okay, YIKES! the IA source is a disaster. it only has surviving links to weekly reports back to week 32 in early august, those reports have a cumulative total for "2022-23" with no indication of what specifically that means, then in week 40 in early october they switch to a truly calamitous graph with the same apparent visual heights for 4 2 and 9 and higher(!) for 3 and 5. even if i was confident in my ability to go pixel by pixel and recover data for past weeks, which i am not, it is literally impossible for the most recent graph to cohere with literally any prior reported weeks. i can't in good conscience remove their x, the model has proven far too reliable to substitute it with so much blatant nonsense. with that said, and although IA has made their reports objectively worse and worse, they are also changing them at a dizzying rate, so i'll keep an eye on the page in case they become responsible

in less depressing news, the MS source has been completely integrated into the table! and thanks again, this was all helpful feedback and i'm glad to have another set of eyes on this endeavor Kinerd518 (talk) 18:52, 1 December 2023 (UTC)

Thank you very much for integrating them into table. I would've done that myself if I knew how the modelled deaths are calculated.

There is a definition of 2022-23 total listed for IA, for example here, on the last page, it is defined as "cumulative from the start of the flu season (October 2, 2022 through the end of the current reporting week)". However, this figure doesn't seem to match perfectly with the historical data, probably due to backfilling. So you may choose to include it at your discretion.

You are right about AZ. Summing up the cases for the previous three seasons by age group would give 2,406,363 which is slightly smaller than actual count (2,406,967) (and summing up the three year average and multiply by 3 gives 2,406,975, which is really close to the correct value). Nonetheless, it appears that the sum by age group is greater than the sum recorded in the table. For example, summing up each age group gives 34,356, which is greater than the 33,967 recorded in the data. Summing up weeks and multiply by 3 doesn't work well as first week's data are apparently incomplete.

Based on the downloaded csv file for KY, there is a discrepancy between last three reporting weeks (8/27, 9/3, and 9/10) of the old dashboard and the same three weeks of the new dashboard , I think the counts from the new one should be used as it describes a more complete data. Horizon Sunset (talk) 06:07, 4 December 2023 (UTC)

@Horizon Sunset yeah i'm still strongly leaning against including anything from these IA reports, because by week 32 they had removed that blurb from the report, only to put it back in at some point before week 45... when they were no longer reporting cumulative season deaths, and *certainly* weren't reporting them from october *2022* to present D: everything about their reporting just makes me uncomfortable haha, there are just so many questionable decisions

i misunderstood what you were talking about with AZ before, i think i get what you mean now, and i appreciate that you went through the effort of summing over all the weeks. i think the play here is to adjust the starting point in this table for AZ, since it's the past data causing the discrepancy, and even with a little missing age data at least we'll be closer. if i'm misunderstanding you again please let me know though!

i also misunderstood what you were talking about with KY: strictly speaking i'm not using the values by week of either dashboard, but rather the values reported when they were reported. my feeling is that while editing backwards would be more precise, it's just not worth the effort, except in extreme cases like MS where we get months of data all at once. since we have to use such long periods for a rolling average anyway, redistributing a few deaths by a week or two usually doesn't even show up

to the larger question of how the model works, i describe it on the talk page of the chart, but briefly it just takes the % a state contributed to the national total for the year ending 2023-03-31, and scales each day's total upward by that % for whenever the state stopped reporting, as denoted by an x in the table. please feel free to give it a go if for example you feel strongly about tidying up KY, i'll be happy to look it over! Kinerd518 (talk) 19:35, 4 December 2023 (UTC)

I agree with your opinion on Iowa.

You didn't misunderstand me this time! For AZ, it wasn't until late September when the state modified the dashboard and remove the total count that this discrepancy happened. So I don't think anything before 9/27 should be changed. Since the first dashboard version after this is no longer available, I think the change can start at around 11/22, when the value was 34,336, which I recorded in my update at Template:COVID-19 pandemic data/United States medical cases by state, instead of 33,947. I think method of calculation should remain unchanged (meaning the value reported when they are reported should still be used after this point) beyond this to avoid amplifying the error of unknown ages.

@Horizon Sunset i see! that does seem the best path, i will review your source and push through that update for AZ in this table later today. thanks again :D Kinerd518 (talk) 17:15, 6 December 2023 (UTC)

Kentucky: I totally agree that editing backward isn't worth it, and thank you for explaining how the model works. But before further explaining my concern may I ask how did you get the value of 19,256 based on the last available total count of 19,183? Horizon Sunset (talk) 23:09, 5 December 2023 (UTC)

you bet! i started with the old dashboard value of 19,183 as of 7 october 2023, there have since been 7 backdated reports for the 2022-2023 season so that makes it 19,190, and the current season has 66 reports for a total of 19,256. unfortunately the new dashboard only has data back through 2022-2023, but i've only seen one instance of a state significantly changing reporting without comment for two plus years ago, so i'm reasonably confident in the value Kinerd518 (talk) 17:13, 6 December 2023 (UTC)

Kentucky:OK here is what I was thinking. I agree with the 66 from the current season. But the number of backfills is more than 7.

In old dashboard, the sum from the last several weeks are:

16+11+14=41

Compared to new dashboard:

17+15+20=52

So this is 11 counts (I checked the data from previous weeks and found little discrepancy between the dashboards prior to 8/27/2023)

Also note that the old dashboard only includes data up to 9/10. So the weeks of 9/17 and 9/24 aren't included in either the old dashboard or the current season of the new dashboard. According to the previous season of the new dashboard, the count should be 8+18=26.

So in total, there should be 37 backfills.

Even though it's not as likely, it's nice to always keep an eye on those values from previous season to see if they change. Horizon Sunset (talk) 18:15, 6 December 2023 (UTC)

@Horizon Sunset alright, we're all set! i could have sworn i'd been checking the 2022-2023 season sheets too, but that doesn't matter now. KY and AZ have been entered, thanks! Kinerd518 (talk) 19:25, 6 December 2023 (UTC)

OK thank you very much!!!! Since you have 56 jurisdictions to check, it's easy to miss out on a specific state. Again thank you for updating this chart daily!--Horizon Sunset (talk) 20:22, 6 December 2023 (UTC)

Virginia 2024 Jan 2 Reporting

VA reported 14 new deaths and 570 previously unreported deaths ranging from 1 Jan 2022 to 15 Dec 2023. They made no public comment whatsoever acknowledging this, and in fact the only tweet today from the VA Dept of Health contains a proud declaration that "we are dedicated to protecting and promoting the health of all Virginians". Although it would be possible to rebuild the numbers they posted all along at least in this template's date range, the sheer scope and shocking blatancy of the report gives me even less reason to trust their numbers now than when it was obvious they were intentionally holding back from reporting them; although there certainly weren't 570 deaths in the past couple of weeks, frankly I just don't believe the deaths happened when VA claims they did either. As before, this situation is deeply depressing and I feel that all I can do is record. Kinerd518 (talk) 19:21, 2 January 2024 (UTC)

it occurred to me to link the two most recent csvs from VA for posterity, here they are Kinerd518 (talk) 19:33, 4 January 2024 (UTC)

https://www.dropbox.com/scl/fi/wlxx4m61ge4b076pgioe6/VDH-COVID-19-PublicUseDataset-EventDate_20231227.csv?rlkey=05nonaoecoaa7gnb3r7v1o58e&dl=0

https://www.dropbox.com/scl/fi/7fzqe8k3td6ynqlexju8m/VDH-COVID-19-PublicUseDataset-EventDate_20240102.csv?rlkey=lvgb2a2cjc1zpwmk9hcoewl2u&dl=0 Kinerd518 (talk) 19:33, 4 January 2024 (UTC)

VA reported 4 new deaths and 331 previously unreported deaths ranging from 5 Jan 2022 to 1 Jan 2024, including 27 in the period from 16 Dec 2023 to 1 Jan 2024 that was the newly reported section of last week's major change in reporting. again, there was no public comment whatsoever, there is no reason to believe anything they say, and i'll keep writing down what they do. below is the link to the newest csv https://www.dropbox.com/scl/fi/fbmk7bz7icl6j085u5t94/VDH-COVID-19-PublicUseDataset-EventDate_20240109.csv?rlkey=5i74lhkry4zwqvebmojos6vnq&dl=0 Kinerd518 (talk) 19:45, 9 January 2024 (UTC)

Modeled Deaths First Annual Review

In the first year of this project, there have been 22 occasions where a state stopped reporting only to resume later, resulting in a removal of x marks. If the model is perfect, the sum of these occasions should have resulted in an increase in reported deaths and no change at all in modeled deaths. What actually happened summing over all those occasions was an increase of 4,568 in reported deaths and an increase of 1,019 in modeled deaths. There is evidence the model is conservative.

The underlying assumption of the model is that a state's proportion of national deaths in the last year of the declared emergency is equal to its proportion of national deaths post emergency, since there was extremely little change in the relative measures taken state to state - barely anyone was getting vaccinated any more so a state that was already way ahead in vaccination would be as far ahead as it was before, there were no requests to mask or distance anywhere, etc. I still believe this assumption is correct, but I have nevertheless come to believe that the model has a flaw, which is that we only have a state's *reported* proportion of national deaths in the last year of the declared emergency, a number that does not necessarily match a state's *actual* proportion of national deaths. By definition, this review only looks at a state that reported erratically during the past year, and I have finally come to believe that this is correlated with the same state reporting erratically prior, illustrated most dramatically by IL but also by MS and WI. Though outside the scope of this review, we also see this correlation in the above detailed adventures of FL and VA reporting.

There is no way around this, but since we know the model has demonstrated about 10,000 deaths were hidden in this first year, we can estimate that it should have seen about another 2,200 people lost. All failure is dispiriting, and one that couldn't see the intentional obfuscation of thousands of deaths is especially so. There is no way around this. We must continue watching, and recording. Kinerd518 (talk) 20:08, 13 May 2024 (UTC)