In arcuo/bachelors_meta: Bayesian Hierarchical Latent Mixture model meta analysis and plots.

knitr::opts_chunk$set(echo = TRUE)

Josh's model

Multibinomial model k = observed studies ~ dbinom(q, N) N = Total studies ~ multinom(n, A) n = conjugate prior to N ~ dirichlet() q = probability for publication <- R + p(1-R) R = probability for significant results ~ dbeta(.5, .5) rho = rate of unsignificant publications ~ dunif(0,1) (maybe set 0.1, 0.5 and 0.9)

Article:

N-k ~ dnbinom(k, q) = number failures it takes until we see k successes (finding a study) with Q probability. Q ~ dbeta(0.5, 0.5) = accounting for variability of prob of publication across fields. = R + rho(1-R): Assuming all significant studies are published and some proportion 0 <= rho <= 1 of non-significant studies are published. R = prob of producing significant results. Z ~ dbinom(k, r) = r = z/k

We need to assume that studies are of roughly the same sample size to have the same R across studies as larger power increases likelihood of significant results.

plot(dnbinom(seq(1,10), 7, 0.5))

arcuo/bachelors_meta documentation built on May 5, 2019, 2:41 a.m.