Answer the question
In order to leave comments, you need to log in
How to a/b test a page element without creating a page duplicate?
How to a/b test a button located in multiple places on the visited site?
I know how to do such testing if it is only about one page for which a duplicate is created. But what if you need to test another interface element feed on a large running service?
The problem is how to properly divide users, what kind of sample to make and from whom? In order to disturb the majority of the community as little as possible and to get reliable data as much as possible.
Answer the question
In order to leave comments, you need to log in
Divide users into test and control groups using any of the sampling methods ( en.wikipedia.org/wiki/Category:Sampling_techniques). Then you show the corresponding elements to users from different groups (an extra if in the template code), pass information about which group the user belonged to to the analytics system used, and measure the conversion.
About the button, use GET parameters.
About the sample
1) Segment the sample by the frequency of using the service (there are more than N visits or more than N comments during the month). UserID or ClientID through custom dimension to help you
2) Analysis of variance. Data range and variation to understand sample deviation
3) Well, how to find the most stable groups, calculate the optimal size for the test from the expected effect
16*p*(1-p)/b2
p-current conversion (relative value)
b- expected effect (relative to value)
Well, it is logical to assume that since you serve the element differently when testing one page, then you do the same for a large service. So the answer is simple - climb into all the places where your button meets and replace it.
PS Apparently it's time to think about optimizing the output of this interface element.
Put a separate trigger to display this element and add an extra if in all the places you need. What is the problem?
Maxim, the test can be divided by time. In the first period, you collect statistics on the first interface option, and in the second - on the second one.
Of course not, this hp series is already hot, burn the mother I guarantee!
Didn't find what you were looking for?
Ask your questionAsk a Question
731 491 924 answers to any question