3 ms·
>Because otherwise a car crash in a small town would be equivalent to a major terror attack in NYC You can probably adjust for this using some sort of statisti
by gruez 5y ago
>Because otherwise a car crash in a small town would be equivalent to a major terror attack in NYC
You can probably adjust for this using some sort of statistical significance test. However, without actually doing the test I'll go on a limb and say that the world's population wasn't small enough in 1918 that this phenomena (ie. n being too small) would skew the results.
- rich_sasha 5y agoWorld population in 1920 was presumably somewhere between 1.6bn and 2bn: https://www.worldometers.info/world-population/world-population-by-year/ https://www.worldometers.info/world-population/world-populat... so 4-5x smaller than now.
- gruez 5y agoRight, but that's not small enough to make a difference. To go to the extreme, if your population size was 100, and some guy died of the flu, that's a 1% death rate! That doesn't actually mean the death rate of the flu is actually 1%, because you don't know whether that guy just happened to die because your group was unlucky, or that the death rate is 1%. As your population size gets bigger this becomes less of a problem. For instance, if your population was 1,000,000 and 10,000 died of the flu you can be fairly certain the death rate is actually 1%. With that in mind, you can intuit that the difference between between 1.6bn and 2bn (1920) and 7 billion (today) is negligible, although feel free to quantify this with actual calculations.
- rich_sasha 5y agoAh if it’s just about significance (I thought you were arguing about rate of death being negligibly different) then agreed, little difference. In this kind of scenario, errors decrease like 1/sqrt(sample size) so quadrupling the sample size leads to halving the error (which was tiny to start off with). If 1 out of 20 die of flu, the frequentist 1 sigma error is roughly 5% too, so same order of magnitude as the probability estimate. Different story with a hierarchical Bayesian model etc.