Welcome to WebmasterWorld Guest from 54.159.246.164

Forum Moderators: open

Message Too Old, No Replies

is a 70 - 30 split valid?

valid split tests

   
9:39 pm on Apr 14, 2014 (gmt 0)



We are testing 2 landing pages sending 70% to the control and 30% to the challenger. After 3 weeks of testing, the challenger won. Is this statistically significant? Do split tests need to be a true 50 - 50? Why is this or is this not significant?

thanks,
4:21 pm on Apr 16, 2014 (gmt 0)

WebmasterWorld Senior Member 5+ Year Member



They do not need to be 50/50, but each sample needs to have sufficient size to increase statistical certainty to the level where decisions can be made.

You didn't say the total visits. :-) If it's 7 visits (control) and 3 visits (challenger), neither is large enough to prove anything.

Telling us 3 weeks also doesn't help us, it is the sample size, not the duration over which it's collected (though that can affect a test).
4:34 pm on Apr 16, 2014 (gmt 0)

WebmasterWorld Senior Member 5+ Year Member



This is the point where most people ask... how many items do I need in each sample to be meaningful?

Not so simple. It depends on the behavior of what you're measuring and how different the two sets behavior is.

An example:
~~~~~~
A = 1000 clicks, 3 conversions
B = 1100 clicks, 2 conversions
~~~~~~
Inconclusive, not enough conversions, A winning, but only 81% probability. Aim for >95% probability in your tests.

Same ratios of success (A to B), but each one's success rate is a larger set of data:
~~~~~~
100 clicks, 60 conversions
110 clicks, 40 conversions
~~~~~~
A is clear winner, 99%+ probability.

Here's a decent calculator:
[tools.seobook.com...]