Bug in synthetic data set: mechanism 2

[jroessler]

Describe the bug
As you describe correctly here, the baseline main effect b* of the randomized trial (Setup B) in Nie and Wager (2020) is the following:

b*(X) = max{Xi1 +Xi2,Xi3,0}+max{Xi4 +Xi5,0}

However, in your implementation, you define b* as follows:
b = np.maximum(np.repeat(0.0, n), X[:, 0] + X[:, 1] + X[:, 2]) + np.maximum(np.repeat(0.0, n), X[:, 3] + X[:, 4])

Thus, your implementation of b*(X) is:
max{Xi1 +Xi2 + Xi3,0} + max{Xi4 +Xi5,0} but it should be
max{Xi1 +Xi2 , Xi3,0} + max{Xi4 +Xi5,0}

If that is indeed a bug, I would like to submit a merge request.

[jeongyoonlee]

Thanks, @jroessler, for filing the issue. It indeed looks like a bug. Please feel free to submit a PR.