> … We compare two ways to elicit statistics such as means, fractiles, and confidence intervals. Of course you can ask for them directly. But we also tried eliciting the entire distribution with a graphical interface. Then you can compute any statistics from that. Doing so leads to greater accuracy, on both the individual and group level.

(The only violation is “directly”. Hemingway doesn’t like adverbs, I guess.)

Btw, I just added this to a draft of a similar post I’ve been nurturing for like a year: http://expost.padm.us/outloud

]]>