[mgraczyk]

The relevant statistic is not how many people die, it's how many people die in a short time window. The probability of a particular person dying in a 5 minute interval after being beheaded is much larger than the probability of a particular person dying in an interval after they were not beheaded.

[thaumasiotes]

That just shows that beheadings are targeted at people who were already about to die.

[mgraczyk]

Yeah I think it really reveals a breakdown in our ability to quantitively model the intuitive notion of "causality".

A targeted public health intervention that prevented beheadings just at the time of beheading would clearly break the link in the causal change. This result is insensitive to any imaginable confounders. However, the executioner would find another strategy, say a gunshot or noose, which did not appear anywhere in our dataset.

Really makes you question the validity of observational studies. At least when acts of government are concerned.