Beta-binomial negative deviance

[nehill197]

Beta-binomial deviance sometimes negative because saturated log-likelihood is smaller than the model log-likelihood.

dev_beta_binom(x = 1, size = 5, prob = 0.3, theta = 0.5)

[sebdalgarno]

perhaps worth seeing whether the other re-parameterization method has the same issues?

alpha = prob * theta and beta = (1 - prob) * theta

that one is used in the brms package

[joethorley]

done