search

dpuelz / BicliqueRT

Randomization Tests of Causal Effects under General Interference
https://arxiv.org/pdf/1910.10862.pdf
6 stars 4 forks source link

updates1011 #7

Closed szhuang1 closed 3 years ago

szhuang1 commented 3 years ago
  1. fix a bug in bisection method, which can lead to wrongly identify the position of true CI. Specifically, the original method starts to detect ub and lb at 0. If the true 95%CI is, say, [0.1, 0.2], then testing ub=0 will give a p-val < 0.05, hence it will decrease, and head all the way down to the initial lower bound (say initial lb=-1, initial ub=1), and we get something like [0.1, -1]. In the new version, I add a step to firstly give a rough location of CI, then start at the middle point of this rough CI. In this way, the initial trial for both ub and lb will have a p-val > alpha, so the updates in CIs will make sense.
  2. update NAMESPACE to allow some new functions to be called.
  3. change print-out message of "found greedy clique" to avoid printing too many lines of "found greedy clique xx"
dpuelz commented 3 years ago

Hey @szhuang1, let me know if I merged this pull request correctly. There was a conflict I had to resolve, and I want to confirm I did it correctly. Thanks!

szhuang1 commented 3 years ago

Hey @szhuang1, let me know if I merged this pull request correctly. There was a conflict I had to resolve, and I want to confirm I did it correctly. Thanks!

Hi David, yes I think it's correct.