A / B Poll Statistics

I am trying to do some statistical analysis of various A / B tests to see which alternative is better, and found conflicting information about this.

Firstly, I’m interested in a couple of different things:

  • Tests that measure success by counting events, such as conversions or sent emails
  • Tests measuring success by counting income
  • Tests that have only two alternatives (control and new)
  • Tests that have several alternatives (management and several new ones)

I was hoping to find a simple set of formulas or rules for this analysis, but found more questions than answers.

This site says you cannot compare multi-element tests; you can only make paired comparisons and analyze a chi-square to see if the whole test is statistically significant or not.

This site offers a way to test A / B / C / D (starts with slide 74) by analyzing the results using the G-Test (which he says is related to the chi-square), but is unclear in the details of using the fiction factor. It also suggests that you can use the A / B / C / D approach to eliminate alternatives until you get a clear winner in the A / B comparison.

A/B/C/D ( ) , . , , ( ).

, , , , . , / . , .

+3
1

-. p, . X N, p Beta (X + 1, N-X + 1).

, P (pA > pB), pA pB - -. paper.

E [pA-pB], .

+1

Source: https://habr.com/ru/post/1725747/


All Articles