AB split test graphical Bayesian calculator

This calculator tells you, given the split test data you have, how likely is it that one version is better than another.
A Bayesian approach avoids many of the issues with z tests and G tests, and can often enable you to draw meaningful conclusions, even where conversion rates and sample sizes are low.
The probabilities calculated go directly to the business question of which version is best.

Visit durations normally distributed? Not likely.

Time on site is an incredibly useful metric, but fraught with difficulties of collection and interpretation. For example Google’s presentation in the behaviour/engagement reports, at first glance, makes visit duration look normally distributed, but they aren’t.
Using the analytics API in a spreadsheet enables a much closer look at the distribution of visit lengths.