>1) I find numbers like “3-fold increased risk” a bit meaningless without knowing the baseline risk.
Yeah, I don't see this talked about enough. If it's three fold from 33% baseline to 99%, thats a big deal. If it's a three fold increase from 0.000000033% to 0.000000099%, even the new number is minuscule.
It's why, say, blue M&M's increasing risk of cancer by 75% isn't necessarily as big a problem as it sounds.
My father-in-law who died of pancreatic cancer in 2016 had excellent oral health. In fact, he never had a single cavity in his entire 76 years of life.
This is, among other things, a quirk of both statistics (models with relative effect measures are better behaved/easier than ones with absolute measures outside toy examples) and early causal inference where we underestimated just how bad smoking is for you.
Yeah, I don't see this talked about enough. If it's three fold from 33% baseline to 99%, thats a big deal. If it's a three fold increase from 0.000000033% to 0.000000099%, even the new number is minuscule.
It's why, say, blue M&M's increasing risk of cancer by 75% isn't necessarily as big a problem as it sounds.