expected_loss: Calculate the expected loss from assuming the wrong beta

In dgrtwo/splittestr: Bayesian and frequentist A/B split testing

Calculate the expected loss from assuming the wrong beta

Description

Find the expected loss from two Beta posteriors if one assumes (wrongly) that a / b > c / d. That is, E[max(Beta(c, d) - Beta(a, b), 0)]. This comes from https://www.chrisstucchio.com/blog/2014/bayesian_ab_decision_rule.html. Note that the approx = TRUE version is vectorized, but the exact version is not.

Usage

expected_loss(a, b, c, d, approx = FALSE, ...)

Arguments

a	alpha parameter for first Beta
b	beta parameter for second Beta
c	alpha parameter for first Beta
d	beta parameter for second Beta
approx	whether to use a normal approximation when calculating h
...	Extra arguments, ignored

dgrtwo/splittestr documentation built on May 15, 2022, 3:16 p.m.