Experiment Analysis

A common mistake is to attempt to generate a chart using only one variant.

The Experiment Results chart needs something to compare your control to in order to generate results. If you neglect to include both the control and at least one variant, your chart will not display anything.

When using a T-test, you will have to wait until your experiment reaches the specified sample size before Experiment Results will run the p-value and confidence interval computations.

With sequential testing, even with a large MDE, it can take some time to reach statistical significance if your experiment’s lift is small. A T-test will generally require fewer samples to detect the same lift.

Amplitude uses two parameters to calculate Return On for the Retention metric:

• The return event: The event you hope the user performs after the exposure event (aka, the starting event). The user is counted as retained if they trigger the return event.
• Return on the nth day/week/month: The number of days/weeks/months you want to see between a user performing the exposure event and the return event. This parameter is calculated in 24 hour increments and does not use calendar dates.

For example, a user performs an exposure event and a return event:

• Exposure event: Page Viewed
• Return event: Add to cart

With a return on the nth day value of seven (seven days), the user will be counted as retained if the Add to cart event is triggered between seven days and seven days plus 24 hours from performing the Page Viewed exposure event.

If the return on value shifts to an nth week value of one (one week or seven days), the user will be counted as retained if they perform Add to cart anytime between days seven (week one) and 14 (week two) since performing Page Viewed. 

For a return on nth month value of one (30 days), the user will be counted as retained if they perform Add to cart anytime between days 30 (month one) and 60 (month two) since performing Page Viewed.