Sell More Via the Web

3. outliers, because this non-parametric test doesn’t place undue emphasis on the occasional huge order.

This design works best for merchants who close a large portion of their sales on a customer’s first visit, rather than for merchants who typically experience long delays between visit and order.

For more statistical power, run the test longer. If you ran the test for four weeks, you’d still alternate day by day. After four weeks, you’d claim the alternative the winner if it beat the control on 12, 13 or 14 of the 14 day-pairs, and claim the alternative a loser if it beat the control only zero, one or two days of the 14.