Serious question: if people treat p values -- continuous variables from 0 to 1 -- dichotomosly, what would stop them from treating CIs, which literally dichotomize the parameter space into "inside" and "outside", dichotomously? Telling them not to?
Conversation
Replying to
Great question: I propose to use x±y/2, with y the width of the 95%CI.
Makes it clear how ridiculous it is to try to make a difference between 0.1±0.099 and 0.1±0.1.
One is significant, the other not, but they bring the same info.
See here: economistjourney.blogspot.com/2018/06/why-i-
1
2
Clever but works only for symmetric CIs.
I think any CI for bounded effects (correlations, proportions) is asymmetric by construction.
1
1
Ok, good point, but most of our effects are Effect Sizes or Odds Ratios, which are unbounded by definition.