FYI - There is site that could take some of the hard work out of graphing things manually or just having a quick look at some “what if” scenarios. It’s called the Human Mortality Database. It has some nice features you can use to quickly do things like changing the base line years, split male female, age bands, raw counts and age adjusted rates etc. You can also includes different counties for comparison. However, it doesn’t break it up by state like you have.
Thanks for the site. You'll know that I'm a bit reluctant to step outside my current direction. I think the baseline is very critical and "changing the base line years" sounds like a feature of just adding up and averaging deaths of the past, like the ABS adds them up as a baseline. I think it needs something far more nuanced - something that reacts to these big changes in population like we're seeing in Australia. I added a pic of the crazy population movements in the student age group. Averaged deaths of the past can't possibly represent the current expectation - I think they underestimate the true expectation and overstate the pandemic event. I like current-time population and age-specific mortality rates.
I’m not suggesting you change your current direction. The app is just an interesting tool. I like it because you can see the difference between certain assumptions without having to crunch the numbers yourself. Their week specific trend option might interest you it uses a linear projection of the reference years to set the baseline (expectation) which is sort of similar to what your doing but I think they do a trend over say 5 years week on week rather than the overall trend over 5 years to project the expectation week by week into the future and it also accounts for the population base like you have (I think). I’m sure the death and population data they use comes directly from the ABS so it’s probably comparable. Looks like they even offer some different age bands not available in the ABS we get to see i.e. 0-14, 15-64, 65-74, 75-84, 85+
I was just playing with it to try and find something interesting to show you. Is this a Vax death signal I just accidentally found amongst the noise in 2021?
Not sure if you can access the link, but it should link to a draft post I just created, so you can see what I’m talking about.
So I went to the app link. Once there, I couldn't help playing with it all - love data :D
I reproduced their week by week projection - I used 2015-19 at first ( I'm familiar with it (I have the data)), and it's the immediate pre-pandemic time so it's most relevant. The problem with that method is that whatever trend is set up for that week, over 2015-19, is magnified into the future. I did projections on 2020, 2021, 2022 and 2023, and put them all on a graph together. They just amplify the varying trends of the week.
I should do a reference substack for how I do the Baselines. It's single year of age population (averaged across the 5 quarterly points of the year for a better representation of exposure) for the year of interest/1000, multiplied by a calculated single year of age-specific mortality rate (also for the year of interest). That age-specific mortality rate is derived from the trend (for that single age) over 2009 to 2019, and then projected forwards to 2023 and backwards to 2015, anchored at the actual average at 2017 (middle of 2015-2019). That is what I use to calculate the expected annual deaths. I could also find amplification using this method (so it was good to see what happened in the app), but using an 11 year term (2009 to 2019) is quite stable, and small distortions cancel each other out in the annual total. For the seasonal variation I use a proportional distribution derived from the actual weekly deaths, averaged over the 2015-2019 years (I tried weighting but it didn't have an impact), and do some detailed micro-adjustments (fiddly/annoying) for 52 and 53 week years. The fact that this fits well over 2015-2019, certainly at summer baseline, and only exaggerated by varying flu years, says it was a stable distribution over the pre-pandemic years, reasonable to assume it would've continued if not for the pandemic. I think it's a good base. After experimenting a lot today there is one further adjustment I might make.
I went back to the app to try a longer term for the week trends, but they only go back to 2015 (I haven't seen the ABS put out weekly data before 2015 - I was thinking of ordering some).
I looked at NZ on the app - our closest comparator. They have negative excess deaths in 2021, because, again, no flu. This is where there needs to be a change in the baseline, informed by the 2020 experience. I've been thinking about how do it mathematically for a long time. The fever & cough %, and shortfall in flu/pneumonia//covid attributions showed what was going on (went in through the back door on that one), but I think I know how to project a more realistic baseline forwards.
Just discovered they’ve got even more data if you register. Hold off paying for anything, they’ve got their own data sets they paid for from the ABS and made public.
It's fun, but it's not helpful for my purposes. There is no point matching reference years that are full of flu, to 2021 which had no flu, and looking for a clear vaccine signal. I think that couple of weeks that you see in the females in 2021 could also be seen in 2019 and 2018, partly because flu changed and partly because (imo but not shown yet) flu vaccines increased.
If I was going to match 2021 to any reference year to see the vaccine signal I'd use 2020, but there needs to be adjustments, AND a strong rationale for why you'd do it. That's why I needed all the flu data.
Using the weekly trend from 2015-2019 is chaotic, because the weeks are very chaotic. The trends go up and down from week to week. There is no rationale for using it for my purposes, though there is one aspect it can add to.
I was just reading the Actuaries Digital blog and found the paper below and thought it might be of interested to you. It outlines pretty much all their results and also includes their methods. It’s over 100 pages long and some of it is just stuff they have already published in their blog. The appendices go into detail about their model. Appendix C & D shows their modeled deaths (expected) vs actual. It’s a shame they don’t have the raw data in spreadsheets or csv format but the charts are pretty interesting (to me at least). Check out the flu looks like someone just flicked the switch to off 😀. I plan on reading it in detail to see if there is something I’ve previouslymissed in their blog but they look to have been pretty fair in how they have modeled the base line. I think it’s similar to your method. Although I don’t agree with some of their interpretations and findings, their number crunching looks pretty good to me. It’s a Shame they don’t have a cumulative graph showing the remarkable break in the trend as you have done. https://www.actuaries.asn.au/Library/Opinion/2023/REPORTV2COVID19.pdf
Try the link below. I wasn’t going to publish it (it was in my drafts). You used to be able to send link to drafts, looks like it doesn’t work anymore. Just added some stuff for others and published it (no email sent out).
We all know that statistics can be manipulated - intentionally or not. I've been following all the other analyses of the figures. I did have some training in stats back in my Uni days, but I'm not a mathematician by inclination. So I'm just following all this with a lot of interest & an open mind.
I'm a baby boomer myself, and in that excess death band (still healthy, unjabbed, and fully recovered from my one bout of covid). I can understand why they might want to be reducing our numbers in the demographics, but I'm not going to cooperate!
Whoop whoop! Don't cooperate :D So glad you're healthy after your one covid!
I'll have to wait and see what the data does (and hope it has integrity ;) ).
If that result was true, then maybe there's something about covid or the vax that is just bad (on average) for people at a particular stage of life. This doesn't of course address dodgy but beating hearts, and other ways that people could be struggling.
Another great post👍
FYI - There is site that could take some of the hard work out of graphing things manually or just having a quick look at some “what if” scenarios. It’s called the Human Mortality Database. It has some nice features you can use to quickly do things like changing the base line years, split male female, age bands, raw counts and age adjusted rates etc. You can also includes different counties for comparison. However, it doesn’t break it up by state like you have.
https://mpidr.shinyapps.io/stmortality/
Thanks so much Ivo.
Thanks for the site. You'll know that I'm a bit reluctant to step outside my current direction. I think the baseline is very critical and "changing the base line years" sounds like a feature of just adding up and averaging deaths of the past, like the ABS adds them up as a baseline. I think it needs something far more nuanced - something that reacts to these big changes in population like we're seeing in Australia. I added a pic of the crazy population movements in the student age group. Averaged deaths of the past can't possibly represent the current expectation - I think they underestimate the true expectation and overstate the pandemic event. I like current-time population and age-specific mortality rates.
I’m not suggesting you change your current direction. The app is just an interesting tool. I like it because you can see the difference between certain assumptions without having to crunch the numbers yourself. Their week specific trend option might interest you it uses a linear projection of the reference years to set the baseline (expectation) which is sort of similar to what your doing but I think they do a trend over say 5 years week on week rather than the overall trend over 5 years to project the expectation week by week into the future and it also accounts for the population base like you have (I think). I’m sure the death and population data they use comes directly from the ABS so it’s probably comparable. Looks like they even offer some different age bands not available in the ABS we get to see i.e. 0-14, 15-64, 65-74, 75-84, 85+
I was just playing with it to try and find something interesting to show you. Is this a Vax death signal I just accidentally found amongst the noise in 2021?
Not sure if you can access the link, but it should link to a draft post I just created, so you can see what I’m talking about.
https://krap.substack.com/publish/post/138857574
So I went to the app link. Once there, I couldn't help playing with it all - love data :D
I reproduced their week by week projection - I used 2015-19 at first ( I'm familiar with it (I have the data)), and it's the immediate pre-pandemic time so it's most relevant. The problem with that method is that whatever trend is set up for that week, over 2015-19, is magnified into the future. I did projections on 2020, 2021, 2022 and 2023, and put them all on a graph together. They just amplify the varying trends of the week.
I should do a reference substack for how I do the Baselines. It's single year of age population (averaged across the 5 quarterly points of the year for a better representation of exposure) for the year of interest/1000, multiplied by a calculated single year of age-specific mortality rate (also for the year of interest). That age-specific mortality rate is derived from the trend (for that single age) over 2009 to 2019, and then projected forwards to 2023 and backwards to 2015, anchored at the actual average at 2017 (middle of 2015-2019). That is what I use to calculate the expected annual deaths. I could also find amplification using this method (so it was good to see what happened in the app), but using an 11 year term (2009 to 2019) is quite stable, and small distortions cancel each other out in the annual total. For the seasonal variation I use a proportional distribution derived from the actual weekly deaths, averaged over the 2015-2019 years (I tried weighting but it didn't have an impact), and do some detailed micro-adjustments (fiddly/annoying) for 52 and 53 week years. The fact that this fits well over 2015-2019, certainly at summer baseline, and only exaggerated by varying flu years, says it was a stable distribution over the pre-pandemic years, reasonable to assume it would've continued if not for the pandemic. I think it's a good base. After experimenting a lot today there is one further adjustment I might make.
I went back to the app to try a longer term for the week trends, but they only go back to 2015 (I haven't seen the ABS put out weekly data before 2015 - I was thinking of ordering some).
I looked at NZ on the app - our closest comparator. They have negative excess deaths in 2021, because, again, no flu. This is where there needs to be a change in the baseline, informed by the 2020 experience. I've been thinking about how do it mathematically for a long time. The fever & cough %, and shortfall in flu/pneumonia//covid attributions showed what was going on (went in through the back door on that one), but I think I know how to project a more realistic baseline forwards.
I thought you’d like it 😀
Just discovered they’ve got even more data if you register. Hold off paying for anything, they’ve got their own data sets they paid for from the ABS and made public.
It's fun, but it's not helpful for my purposes. There is no point matching reference years that are full of flu, to 2021 which had no flu, and looking for a clear vaccine signal. I think that couple of weeks that you see in the females in 2021 could also be seen in 2019 and 2018, partly because flu changed and partly because (imo but not shown yet) flu vaccines increased.
If I was going to match 2021 to any reference year to see the vaccine signal I'd use 2020, but there needs to be adjustments, AND a strong rationale for why you'd do it. That's why I needed all the flu data.
Using the weekly trend from 2015-2019 is chaotic, because the weeks are very chaotic. The trends go up and down from week to week. There is no rationale for using it for my purposes, though there is one aspect it can add to.
I was just reading the Actuaries Digital blog and found the paper below and thought it might be of interested to you. It outlines pretty much all their results and also includes their methods. It’s over 100 pages long and some of it is just stuff they have already published in their blog. The appendices go into detail about their model. Appendix C & D shows their modeled deaths (expected) vs actual. It’s a shame they don’t have the raw data in spreadsheets or csv format but the charts are pretty interesting (to me at least). Check out the flu looks like someone just flicked the switch to off 😀. I plan on reading it in detail to see if there is something I’ve previouslymissed in their blog but they look to have been pretty fair in how they have modeled the base line. I think it’s similar to your method. Although I don’t agree with some of their interpretations and findings, their number crunching looks pretty good to me. It’s a Shame they don’t have a cumulative graph showing the remarkable break in the trend as you have done. https://www.actuaries.asn.au/Library/Opinion/2023/REPORTV2COVID19.pdf
I keep getting this on your link
"This page isn’t workingkrap.substack.com redirected you too many times.
Try deleting your cookies.
ERR_TOO_MANY_REDIRECTS"
I deleted all my cookies. Doesn't work.
What date was your post? I'll scroll down your list.
Try the link below. I wasn’t going to publish it (it was in my drafts). You used to be able to send link to drafts, looks like it doesn’t work anymore. Just added some stuff for others and published it (no email sent out).
https://krap.substack.com/p/vax-signal
Great analysis - thank you!
We all know that statistics can be manipulated - intentionally or not. I've been following all the other analyses of the figures. I did have some training in stats back in my Uni days, but I'm not a mathematician by inclination. So I'm just following all this with a lot of interest & an open mind.
I'm a baby boomer myself, and in that excess death band (still healthy, unjabbed, and fully recovered from my one bout of covid). I can understand why they might want to be reducing our numbers in the demographics, but I'm not going to cooperate!
Whoop whoop! Don't cooperate :D So glad you're healthy after your one covid!
I'll have to wait and see what the data does (and hope it has integrity ;) ).
If that result was true, then maybe there's something about covid or the vax that is just bad (on average) for people at a particular stage of life. This doesn't of course address dodgy but beating hearts, and other ways that people could be struggling.