If your sample sizes for each cell are large, then the t-test would be approximately correct. I'm not sure what your data context might be, but one thing I would think about is if your users can interact with each other in some way. That would potentially hurt your the independent, identically distributed assumption used by the t-test.
Thanks so much. In typical AB testing done online, can we assume it's like a two sample t test?
If your sample sizes for each cell are large, then the t-test would be approximately correct. I'm not sure what your data context might be, but one thing I would think about is if your users can interact with each other in some way. That would potentially hurt your the independent, identically distributed assumption used by the t-test.