Quite possibly it's just quirks in the trendlines caused by small sample size or something. I would imagine the original paper has more numbers and information

Quite possibly it's just quirks in the trendlines caused by small sample size or something. I would imagine the original paper has more numbers and information on methods, but sadly it looks like it's behind a paywall.