Home > Technical & Creative Skills > Design - Imagery, Banners & Landers

Design Split Testing (5)

11-19-2012 03:05 PM #1 adpro (Member)
Design Split Testing

Hello guys and gals! Ok, guys...

I was wondering how many views are enough to split test a change on a landing page? IE, If I change the header colour, how many views I have to send to both the old and new page each so I can wrap up the data and consider one the winner?

Thanks!

11-22-2012 07:31 AM #2 joejoechen (Member)

i usually do that in 100-300 clicks.. sometimes less if the data varies too much it's obvious within 100 clicks

11-22-2012 07:56 AM #3 deedsmedia (Member)

Originally Posted by adpro

There is no rule of thumb, use math to get a statistically significant result...or use this sheet like I do: http://www.wrightw.com/stat-testing.html

11-22-2012 01:14 PM #4 dconstrukt (Member)

you arent supposed to think about shit like this.

the software tells you when the test is done.

11-23-2012 10:11 AM #5 caurmen (Administrator)

If you're working with software that doesn't automate the statistical significance (Prosper, for example), then this blog post has a great, simple way to estimate statistical significance (in other words "when you should stop testing") for A/B tests: http://blog.asmartbear.com/easy-stat...-hamsters.html

If you fancy something a bit more advanced, http://statpages.org/confint.html is excellent for testing various statistical confidence levels.

Home > Technical & Creative Skills > Design - Imagery, Banners & Landers