search

Richie-Lee / Bayesian_AB_Test

MSc_Thesis code repository
0 stars 0 forks source link

Bayes Factor division #5

Closed Richie-Lee closed 9 months ago

Richie-Lee commented 1 year ago

What are the marginal likelihoods that should be divided?

we have 6 options:

  1. Treatment + H0 prior (no effect)
  2. Treatment + H1 prior (effect)
  3. Control + H0 prior (no effect)
  4. Control + H1 prior (effect)
  5. Treatment: point estimate likelihood for H0 (regular)
  6. Control: point estimate likelihood for H0 (regular)

Intuitively I'd think either:

Sanity check:

Richie-Lee commented 1 year ago

recreated (flexible) new version & recreated old version. Available for side-by-side comparison

Richie-Lee commented 1 year ago

current method of choice: https://chat.openai.com/share/988257a8-f5c7-40f5-909f-c5c36ef4557f

BF = P[data_t, data_c | H1] / P[data_t + data_c | H0]

Richie-Lee commented 9 months ago

Fixed, implementation of Wan et al (deng et al)