Just paraphrasing for my own understanding: If you were making a decision based only on the expected ATE, you would do it. However, once you account for uncertainty in expected ATE, you wouldn't.
-
-
-
Bingo!
- Još 2 druga odgovora
Novi razgovor -
-
-
expected TE?
-
Treatment effect.
- Još 3 druga odgovora
Novi razgovor -
-
-
Rather than plotting posdens * loss it might be informative to do a change of variables to loss and show the density of that ("Need to integrate this function" will become "need to take expectation of this density"). The transform is monotonic so you can mark the effect size too.
-
Great idea. Will update.
- Još 1 odgovor
Novi razgovor -
-
-
statisticians compute the posterior, we need clients to tell us about their loss function, even getting them to think formally about loss over actions is a step forward. Life is sometimes hard.
-
Absolutely. My friends
@ericnovik and@djsyclik started a firm@GenerableHQ to do just this (mostly in Pharma, where data are almost as $$ as the decisions). - Još 3 druga odgovora
Novi razgovor -
-
-
(1/2) Interesting. Can I paraphrase to test I understand: Wrong way: Take a point estimate of TE and evaluate it using the loss function.
-
(2/2) Right way: Multiply the posterior by the loss function and then take the integral of the result to get the expected loss.
- Još 2 druga odgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.