-
Notifications
You must be signed in to change notification settings - Fork 1
Description
Hi devs
I'm going through the Frontiers In article, HDDM: Hierarchical Bayesian estimation of the Drift-Diffusion Model in Python, particularly the 'Simulations' section. I'm working on something similar, but I don't notice a huge difference in t-test detecting a difference vs Bayesian estimation.
I went through the repository, and it looks like you've used two-sided t-tests instead of one-sided, and that's why the t-test fails to detect a difference when the Bayesian estimation does. The current scipy.stats docs list two-sided as the default alternative for ttest_rel and ttest_1samp.
Also, it uses a threshold of 0.025, which means p_value_2_sided < 0.025 iff p_value_2_sided / 2 < 0.025/2 iff p_value_1_sided < 0.0125, which should be compared against Bayesian probability of 0.9875.
Could you please confirm this? Thank you!