Hey Guys,
I'm in an odd position where I know that I shouldn't be spamming F5 endlessly, but at the same time I'm testing ads in a high volume environment and want to quickly see which test copies are performing well.
How often are you looking at your figures and when testing new ideas how long of a run do you give them before looking through numbers? I've heard 30 clicks but I can't imagine that's a reasonable sample size.
I've also found that I'm not very productive with any sort of stats tab open.
How do you guys manage this?
Thanks,
ATG
Not F5 that F5 often F5 I F5 Guess.... F5
30 is the number you need on the TOP of the equation for statistical significance. So, if you want to know what your CTR is, you should have 30 clicks out of whatever impressions it takes. If you want to know what your CR is, you need 30 conversions, etc.
So, if you estimate a 0.1% CTR, you'd have 10 clicks every 10k impressions. So . . . THAT's how often you should check if you don't want to spam the F5 button.
I wonder what it'd cost to get someone to code up an application for my phone that sends a message/email along the lines of "Your sample size is significant! Campaign: Name, Ad: Name". I imagine it could plug right into Prosper.
Edit: It'd be awesome to have a drop down menu to message for significance on Ad CTR, LP CTR, and overall CR %. Although an email on 30 clicks for Ad CTR would probably spam the hell out of me.
too often :-) but you're right 30 clicks is too less to get any significant data..100-200 would be more accurate..but that also depends on the offer..
there's no rule of thumb for any of this. If you want real statistical significance, use a Z or T score, but that will take more data than you want to get a say 90% confidence interval. My rule of thumb usually has to do with spend, and CPA of the offer (I usually let each new test spend at least 2-3x CPA before even looking).
Max
I hear this is a sick tool: http://stmforum.com/forum/showthread...More-Guesswork