Hello! I have a simple question. If there are multiple experimental groups in sequential testing, how should N be set? especially, I'm curious about what to do when the ratios of each experimental group are different. Can I simply use the sum of all the experimental group sizes as N?

Hey! That's a good question. We should update our suggestion in our docs to be more clear.

I'm curious about what to do when the ratios of each experimental group are different.

You can only set one value for the tuning parameter for the full experiment, even if the group sizes differ. Hypothetically we could allow it to differ but that would be a lot of fine tuning that we aren't likely to build into the product right now.

Can I simply use the sum of all the experimental group sizes as N?

This is true for 2 arm experiments. Let me explain. The suggestion in our docs is to "set this parameter to the sample size you expect to get when you are most likely to make a decision on this experiment" since that is when the confidence intervals from sequential testing are least inflated relative to the classic standard intervals. This is based on the total sample size used in both arms in each comparison. So if you had a 3 way experiment where you want to have the tightest CIs at around 9000 total users (3k, 3k, 3k in each arm), each comparison would take 6k users, and so you would want to set the tuning parameter to

6000