It is more complicated than I said

Two interpretations are possible

I apologise for doing a third numbers post in as many days - I promise the next one will be something different.

Having slept on it I think there is more to say on the correlations seen in mortality in subsequent halfs of the year after all. There are three unresolved questions and I am going to present arguments for each as I am still not certain.

1. Are deaths of despair together with vaccine deaths creating the correlation?

2. Was it an absence of variation that caused the correlations?

3. I got the baseline is wrong

1. Are deaths of despair together with vaccine deaths creating the correlation?

Matthew Crawford has kindly put some thought into this and published that some of my findings concur with his:

Clare Craig Analysis of U.S. Data: It's All Healthy User Bias

He points out that deaths of despair (drug, alcohol and suicide) cluster more in less wealthy/healthy counties. Such deaths have rocketed since lockdown and remain high in 2023.

Any vaccine benefit would need to be shown AFTER discounting these deaths which occur more in low vaccination areas.

The same applies for any calculation of vaccine caused deaths.

Pre 2020 excess mortality was dominated by seasonal spikes with returns to baseline between each spike. Post 2020 both deaths of despair and vaccine induced mortality are dominating.

He hypothesises that deaths of despair and vaccine induced deaths are together dominating the picture such that the correlation emerges over time.

2. Was it an absence of variation that caused the correlations?

What the correlations mean is that one six month period is a good predictor of the next six months. That could mean a new constant factor causing death but… it could also mean an absence of factors causing the variation - e.g. respiratory viruses or heatwaves etc. Fundamentally, after a huge number of the frail have died we should perhaps expect a period with little month to month variation.

To check if that was the cause I looked at the variation month to month for each period. If 2023 had oddly low variation we could put the correlations down to an absence of the usual factors.

For each 6 month period I took the total monthly deaths, calculated the mean and the standard deviation and then the coefficient of variance. This tells us how much the number of deaths by month varies compared to the average number of deaths for the 6 month period. A small CV means the deaths are fairly stable month-to-month, while a large CV means there’s more irregularity, possibly due to unusual events like a severe winter or a health crisis.

Here is the graph for all the states (except Hawaii- which is not in my spreadsheet as there is no vaccine data):

Here is the average value to make it clearer:

2023 had the usual amount of month to month variation suggesting all the usual seasonal factors were at play. That is therefore not the explanation.

3. I got the baseline is wrong

I still have concerns that I am overinterpreting a drifting baseline. Trying to predict how many deaths there ought to be is hard and gets more inaccurate with each year. If the “excess” is dominated by the error margin in the baseline calculation then we would see the correlations that were seen. The states with too high a baseline would have constant over estimates of excess and those with too low a baseline would have constant underestimates. The correlations would increase over time as the baseline and reality deviated from each other.

It is perhaps noteworthy that the biggest outliers, Alaska and Wyoming, both have low numbers of deaths so are the most susceptible to noise in the data and therefore the baselines might be the least accurate.

If anyone has any suggestions for improving the baseline I am open to them.