[nicbos]

I have a small product that I'd like to do some testing on ans wondered the best way to go about it...

It seems there are a couple of ways to split test for figuring out the best layout and design for product sales, either run with scenario A for a few weeks then switch to scenario B, or use a split testing script that randomly displays A or B to each visitor...? Are there other ways?

Has anyone any experience of using a script, or does Google have split testing service built in somewhere?

Best, Nic.

Edit, sorry, I just noticed the split testing sub forum, should have posted there!

Edited by nicbos, 19 November 2011 - 10:46.

[melontrades]

I always wonder how accurate split testing is statistically.  I remember for statistics class to get true scientific data you need a sample size of at least 30.  So my thinking is that with a sample size of 30 this would help move out some external factors. So going with this theory.  With one site that alternating designs or layouts as compared to having 30 domains up you could in fact factor out some factors such as type of traffic hosting issues because it would all be the same.  The question I'm trying to raise here is how accurate is AB testing.  And how much traffic do you need to run an accurate test.  Maybe 30 unique visitors would be enough to come up with accurate data.  I could be over analyzing.

[irishfury]

melontrades, on 18 December 2011 - 09:35, said:

I always wonder how accurate split testing is statistically.  I remember for statistics class to get true scientific data you need a sample size of at least 30.  So my thinking is that with a sample size of 30 this would help move out some external factors. So going with this theory.  With one site that alternating designs or layouts as compared to having 30 domains up you could in fact factor out some factors such as type of traffic hosting issues because it would all be the same.  The question I'm trying to raise here is how accurate is AB testing.  And how much traffic do you need to run an accurate test.  Maybe 30 unique visitors would be enough to come up with accurate data.  I could be over analyzing.

Way to small of a sample size.  You need much larger sample size so oddities don't exist within the data.

[RobertR]

nicbos, on 19 November 2011 - 10:44, said:

I have a small product that I'd like to do some testing on ans wondered the best way to go about it...

It seems there are a couple of ways to split test for figuring out the best layout and design for product sales, either run with scenario A for a few weeks then switch to scenario B, or use a split testing script that randomly displays A or B to each visitor...? Are there other ways?

Has anyone any experience of using a script, or does Google have split testing service built in somewhere?

Best, Nic.

Edit, sorry, I just noticed the split testing sub forum, should have posted there!

You can't run A for a few weeks and then switch to B because there may be other factors such as quality of your traffic that may change at any moment and spoil the experiment.

That's why you definitely need to randomly display A or B to a visitor.

Google weboptimizer is what you need. It's free and it's not that dificult to use, it's a copy and paste javascript.

Edited by RobertR, 02 February 2012 - 15:34.

[nicbos]

RobertR, on 02 February 2012 - 15:34, said:

You can't run A for a few weeks and then switch to B because there may be other factors such as quality of your traffic that may change at any moment and spoil the experiment.

That's why you definitely need to randomly display A or B to a visitor.

Google weboptimizer is what you need. It's free and it's not that dificult to use, it's a copy and paste javascript.

OK thanks for the note, mate.

[dewaz]

so, any update result now?
I'll keep watching your methods and I'll adopt if it succeeded.

good post.

[irishfury]

dewaz, on 25 February 2012 - 23:57, said:

so, any update result now?
I'll keep watching your methods and I'll adopt if it succeeded.

good post.

It does work.  If you have any questions shoot me a pm.