Home > Paid Traffic Sources > Facebook & Instagram

How long to test / reaching statistical relevance (5)


02-16-2014 04:36 PM #1 bshimmer (Member)
How long to test / reaching statistical relevance

I see alot of people posting questions with really small sample sizes and they usually get a reply that there just isn't enough data to draw conclusions. Which makes total sense. But ... (sorry if this has been asked a alot , couldn't find it searching)

What would the number of impressions need to be at before you can start making conclusions about your ads?


02-16-2014 05:06 PM #2 iAmAttila (Veteran Member)

There's no general rule, but one thing is for certain, the more data you have the less gap between the min/max potential CVR.


02-16-2014 05:26 PM #3 bshimmer (Member)

Yeah , I know what my conversion rate is pretty solidly since I run my own offer I have alot of that info.

I'm just trying to get lower cpc and wondering how long I need to let ads run before I can make a conclusion. I'm reading bare min 10k impressions? so prob 20-25k impressions is a good place to start axing some ads?


02-16-2014 06:08 PM #4 waltermitty (Senior Member)

I have always had better luck targeting broad your cvr% dips but your cpc is much lower. Im talking super broad like US:Female:24+ which is over like 70 million on fb or some shit.
As far as cutting ads I just cut on cvr/cpc cost but everyone has there own method I guess. My issue with FB has always been getting your ads to a high enough CTR to bring cost down. That and also just getting FB to serve your ad in the first place for example 10 ads in the same campaign and only 3-4 get any impressions what so ever.

Bshimmer and me are in the same boat since we both run our own offers in house we actually have confidence in the data we generate, we do not have to deal with faulty data from shaving.


02-17-2014 01:27 PM #5 caurmen (Administrator)

I laid out my approach to statistical significance in sections 7, 10 and 11 of the Getting Started Guide: http://stmforum.com/forum/showthread...TION-AND-INDEX

In particular, using a Bayesian Inference calculator for split-tests rather than a conventional chi-squared approach or whatever is great for reducing the volume you need to make statistically significant inferences. I wrote about that more over here: http://stmforum.com/forum/showthread...Data-Read-This!


Home > Paid Traffic Sources > Facebook & Instagram