@paulj said:

@Lemuria said:

There aren't any sample sizes, because they aren't any samples: These are populations. Statistical significance is typically a question of whether samples are large enough to make inferences *to* a population from which the sample was selected - but there's no such question here, because there's no such jump: The data include *all* known cases, and there are definitely cross-era differences.

Holy sh*t. I remember making this same argument years ago when making a comparison of teaching evaluations at the department level. You're right.

Your intuition is still relevant even though it isn't a sample. Outliers are more impactful to summary statistics in data sets with a small number of points.

If you look at every phish show in 2002 you can confirm that it didn't rain in any of the shows that year. That's an interesting pattern until you consider that there was only one show that year. When we see interesting patterns in small data it is reasonable to question whether the patterns are meaningful or simply a consequence of the data being limited.

I think this was paulj's observation.