Reply to post: Statistics are tricky

Statistics are tricky

Excess mortality is, indeed, the only parameter that makes sense in the context. Anything else will be affected by differences in testing, counting, transparency, etc., including "of Covid" vs. "with Covid" and what not.

Having said that, the material in the link pulls my eyebrow up. Annual data from 2015-2019? No way you can get significance from a sample of 5. (Aside: their "P-score" is not the p-value, which is the probability that the observed effect could happen by chance. You cannot get low p-value, and thus a high confidence that there is an effect, from a sample of 5) Besides, why use death counts? Population changes, too - use death rates (per 100K, per million, whatever - I found this normalization important, see below).

Next comment: "worst mortality for 8 years" means either "worst mortality over a very short period of time" or "nothing very special" (if the time series is actually a lot longer).

I did a comparison of 2020 mortality for the country I live in (hit by Covid as everyone else, several lockdowns, 2020 saw no vaccinations anywhere in the world, etc.), based exclusively on published data from our equivalent of ONS. I checked normalized (by population) mortality for 2000-2019 against 2020 (had to wait till the final figures for the latter were published).

To my surprise, the 2020 mortality was low against the previous 20 years - lower than either the mean or the median of the control sample. I struggle to see how lockdowns could save many lives to offset presumed Covid-related excess deaths. E.g., annual deaths in traffic accidents are low here (maybe ~20 times lower than deaths officially attributed to Covid in 2020). 20 years seems to be a short enough period to discount advances in medical care as a significant confusion factor (the country has advanced medical care in general). That was, in fact, a factor in my decision to limit the time series.

2020 mortality turned out relatively high against the previous 10 years, but not the highest. I have no idea what caused, e.g., 2012 to be so bad here. Even if 2020 mortality is high compared to the previous 5 years (I have not checked) the result can't be significant

The exercise was done purely for curiousity's sake. My conclusion: mortality in 2020 was nothing particularly special in this particular neighbourhood in this millennium.

I assume anyone can repeat the exercise for, say, UK after trawling ONS for some more data.

Disclaimer: the above should not be construed to imply that Covid is not a very serious disease. It obviously is, for some groups of people at least. I offer no firm explanation for the (surprising, for me) result. One may hypothesize that the fact that the life expectancy of the patients most likely to die of Covid (old and vulnerable) was something like 3 months might have something to do with it, but that's no proof. I didn't do any age-adjusted analysis.