A/B testing significance calculator (spreadsheet in Excel)

No. Here is a brief overview of the different tests.

The z-test depends on the assumption that the sum of random events is well approximated by a normal distribution. This happens with a large enough sample size. But is wrong for small sample sizes.

A further weakness with the z-test is that the same sample data is being used to estimate variance and averages. Depending on the specifics of your actual sample, both estimates can be off in a way that confuses the test. The t-test corrects for this possibility but assumes that the individual measurements you are making are normally distributed. It is quite sensitive to this assumption. Unfortunately in A/B testing the measurements are binary yes/no, 1/0 decisions, which are decidedly not normally distributed.

The g-test is designed to be a very good approximation to the exact problem that A/B testing tries to solve. Unfortunately it requires you to compute logs, which was hard before computers.

The chi-square test is designed as an easily computed good approximation to the g-test. It only needs you to square numbers and add them together. It therefore became the most widely used algorithm, and is the one that people have heard of.

Now that we have computers, the g-test is the test we should use. But nobody has heard of it. (However you're still free to use it! And should.)

If you get a large enough body of measurements then all tests converge to the z-test.

I've love to see a write-up on the statistics equations for A/B testing and guidelines on when to use which.

http://elem.com/~btilly/effective-ab-testing/ goes through this in some detail.

In case you need to use a web-based calculator (z-test), here is one on my site: http://visualwebsiteoptimizer.com/ab-split-significance-calc...