Great video. I too initially got fooled by just how much lag there is in the data collected into the various CDC databases and reports. I thought I was being generous allowing for 6-8 weeks of lag, but especially in the case of accidental deaths, it takes 6 months to get a clear picture.
I started taking weekly snapshots of the CDC report "Weekly Provisional Counts of Deaths by State and Select Causes, 2020-2022"[1] starting in May of 2020 to analyze the rate of change week-to-week so I could forecast better, currently I have 104 consecutive snapshots of that data set - It seems to be the fastest way to keep tabs on All-Cause Mortality, Covid Mortality, and Accidental Deaths. Unfortunately the other causes tracked aren't that reliable (Cancer, Septicemia, Heart Disease, etc), so need to rely on the cumbersome Wonder for that drill down.
Long story short, having the snapshot data we can see just how long the accidental deaths take to be reported into the CDC.
Here is a table pulling the snapshots together for accidental deaths, for 2021.
Column A is the Week # of the year (1-52 for 2021), then each subsequent column shows the accidental deaths tracked to each snapshot - row 3 lists the date of the snapshot.
On 1/28/2021 the CDC got its first record for the year, week 1 they collected 900 accidental deaths. The next update (Column C) they added 133 more deaths to week 1 and made entry to week 2 of 457. This continues and continues as you move along to the right.
In rows 1 and 2 I added "% complete" for Week 1 and total Year, respectively.
For example, there would eventually be 5,872 deaths reported in the first week of 2021, so that first entry of 900 would only represent 15% of the total eventual deaths recorded.
There were at least 310,106 deaths for the total year so that first entry of 900 represents .3% of the eventual total.
As each snapshot adds more data, the % complete in rows 1 and 2 grow, but what is interesting is that it would 6 full months just to get 95% of the first week of 2021 (Column X). By the July 1st snapshot that year the CDC had tracked 5,557 deaths. It would take another 3 months just to get to 99% complete for that very first week of January 2021.
Even now, 18 months later, we are still tinkering (albeit only minor) with the final count. The 6/2/2022 update added another 6 deaths to the first week of January 2021.
Examining the full year total (tracked in row 2 % complete) note that it took us until April 10th 2022 to record 95% of the total accidental deaths (Column BH), getting to 295,167, but still 15,000 short of what looks to be the FINAL total, 310,106.
It might go higher than that though. The week-to-week increases were slowing down, but the most recent update on 6/2 added 1100 more deaths after a few weeks of being <+400. I had forecast initially 300,000 back in December of 2021, then revised to 310K by late January, now it seems it *might* nudge up another 2-5K.
It's notable that 310,000 accidental deaths breaks even 2020's record of 290,000 accidental deaths (previous 5 years ranged from 218K to 252K).
Yeah, that's like a claims triangle in insurance! We do that all the time -- I had to do that working in life reinsurance, and P&C actuaries have to do these claims triangles all the time.
Maybe I should do a video on how we do some of these reported vs. incurred triangles, because that's how we set up IBNR (incurred but not reported) reserves. It's how life actuaries make estimates of number of deaths that happened but we haven't heard about them yet; how P&C actuaries set up claims reserves, etc. We have a variety of techniques to complete these triangles, depending on sets of assumptions.
I didn't do that for COVID, etc., because I saw there were all sorts of problems... and the CDC's own estimates have been gross underestimates since the beginning.
Great video. I too initially got fooled by just how much lag there is in the data collected into the various CDC databases and reports. I thought I was being generous allowing for 6-8 weeks of lag, but especially in the case of accidental deaths, it takes 6 months to get a clear picture.
I started taking weekly snapshots of the CDC report "Weekly Provisional Counts of Deaths by State and Select Causes, 2020-2022"[1] starting in May of 2020 to analyze the rate of change week-to-week so I could forecast better, currently I have 104 consecutive snapshots of that data set - It seems to be the fastest way to keep tabs on All-Cause Mortality, Covid Mortality, and Accidental Deaths. Unfortunately the other causes tracked aren't that reliable (Cancer, Septicemia, Heart Disease, etc), so need to rely on the cumbersome Wonder for that drill down.
Long story short, having the snapshot data we can see just how long the accidental deaths take to be reported into the CDC.
Here is a table pulling the snapshots together for accidental deaths, for 2021.
Column A is the Week # of the year (1-52 for 2021), then each subsequent column shows the accidental deaths tracked to each snapshot - row 3 lists the date of the snapshot.
On 1/28/2021 the CDC got its first record for the year, week 1 they collected 900 accidental deaths. The next update (Column C) they added 133 more deaths to week 1 and made entry to week 2 of 457. This continues and continues as you move along to the right.
In rows 1 and 2 I added "% complete" for Week 1 and total Year, respectively.
For example, there would eventually be 5,872 deaths reported in the first week of 2021, so that first entry of 900 would only represent 15% of the total eventual deaths recorded.
There were at least 310,106 deaths for the total year so that first entry of 900 represents .3% of the eventual total.
As each snapshot adds more data, the % complete in rows 1 and 2 grow, but what is interesting is that it would 6 full months just to get 95% of the first week of 2021 (Column X). By the July 1st snapshot that year the CDC had tracked 5,557 deaths. It would take another 3 months just to get to 99% complete for that very first week of January 2021.
Even now, 18 months later, we are still tinkering (albeit only minor) with the final count. The 6/2/2022 update added another 6 deaths to the first week of January 2021.
Examining the full year total (tracked in row 2 % complete) note that it took us until April 10th 2022 to record 95% of the total accidental deaths (Column BH), getting to 295,167, but still 15,000 short of what looks to be the FINAL total, 310,106.
It might go higher than that though. The week-to-week increases were slowing down, but the most recent update on 6/2 added 1100 more deaths after a few weeks of being <+400. I had forecast initially 300,000 back in December of 2021, then revised to 310K by late January, now it seems it *might* nudge up another 2-5K.
It's notable that 310,000 accidental deaths breaks even 2020's record of 290,000 accidental deaths (previous 5 years ranged from 218K to 252K).
https://docs.google.com/spreadsheets/d/1U8XEbIaLnGYgRNzMGlnkFXrCJoL6TwYkTw57HF_Y75Y/edit?usp=sharing
[1] https://data.cdc.gov/NCHS/Weekly-Provisional-Counts-of-Deaths-by-State-and-S/muzy-jte6
Yeah, that's like a claims triangle in insurance! We do that all the time -- I had to do that working in life reinsurance, and P&C actuaries have to do these claims triangles all the time.
Maybe I should do a video on how we do some of these reported vs. incurred triangles, because that's how we set up IBNR (incurred but not reported) reserves. It's how life actuaries make estimates of number of deaths that happened but we haven't heard about them yet; how P&C actuaries set up claims reserves, etc. We have a variety of techniques to complete these triangles, depending on sets of assumptions.
I didn't do that for COVID, etc., because I saw there were all sorts of problems... and the CDC's own estimates have been gross underestimates since the beginning.