← Back to context

Comment by marginalia_nu

2 years ago

Some of these graphs are more convincing than others, this in particular seems to be fairly creative use of statistics.

https://cdn.substack.com/image/fetch/f_auto,q_auto:good,fl_p...

I think their paper they linked in the blog post is a bit like that too, maybe even more so, although I found both interesting.

I do agree some of the trends are more convincing than others, although they're generally consistent with the overall direction the author is suggesting.

It's notable that the lower error bar for the 1920s and the upper error bar for the 2020s only differ by about 2%. Aside from the fact the plot sure looks like random noise, even the trend line is consistent with no change.

I am a bit suspicious because the specific metric used for each type of media is different. That suggests some amount of cherry-picking.