Roll your own "When To Kill and Ad" Table (12)
Here's a 12-minute video on how to create a reference table to know when you have enough information to kill an non-performing ad.
It gives you decision criteria so you don't kill ads to quickly or let them run too long.
http://www.screencast.com/t/MeDUu88q
The link to the stats page is: http://statpages.org/confint.html
Holler if you have any questions.
That's awesome. Would be sweet to build something like that right into an ad management tool.
this is great stuff man.....much kudo's
Thank you very much, this is very interesting.
I like statistics, when I studied this matter at school, I was passionated about the topic. 
I've coded a tool too:
http://www.julienbagein.com/ConfianceStat.php
I wanted to blog it for month.
I think I'll never do it
It helps to determine confidence interval too, but with higher ratio than 80%.
It works both for calculating if you conversion rate or if your CTR is statically relevant.
You can compare the performance of 2 ads or 2 landing pages.
Basically, you insert the following informations:
1st column: name of the landing page or the ad you want to test
2nd column: number of positive actions (clicks or leads)
3rd column: (you don't have to insert data here)
4th column: number of the total impressions or visitors (impressions or clicks)
Then you choose your confidence ratio and submit.
If it's red, you don't have enough data to be sure that ad1 or LP1 is better than ad2 or LP2.
If it's green, you can be sure that for a ratio of {95% or 98% or 99%} ad1 or LP1 will always outperform ad2 or LP2.
I wanted to make a tutorial for everyone asking here "is 5 clicks enough on PPV to be sure that the offer doesn't convert?"
but my english is to bad to do it well 
It's just this tool: http://www.hutchon.net/ConfidOR.htm
that I've redesigned to be more 'visual'.
If some people find it interesting, I can translate this in english
Anyway, thank you very much for your great tutorial and the script!
Awesome, thanks!
got a question
using the same math formulas and trying to understand the meaning of this... i played with excel and got this:
with FB (shooting for 0.10% CTR) - kill ad:
if i ONLY have 0 clicks after 2000 imps (b/c I'm 80.00% confident that CTR won't be higher than 0.12%)
if i ONLY have 1 clicks after 3500 imps (b/c I'm 80.00% confident that CTR won't be higher than 0.11%)
if i ONLY have 2 clicks after 5000 imps (b/c I'm 80.00% confident that CTR won't be higher than 0.11%)
if i ONLY have 3 clicks after 6300 imps (b/c I'm 80.00% confident that CTR won't be higher than 0.11%)
if i ONLY have 4 clicks after 7600 imps (b/c I'm 80.00% confident that CTR won't be higher than 0.11%)
if i ONLY have 5 clicks after 8900 imps (b/c I'm 80.00% confident that CTR won't be higher than 0.10%)
if i ONLY have 6 clicks after 10200 imps (b/c I'm 80.00% confident that CTR won't be higher than 0.10%)
so can this be applied to ppv and killing landers?
assuming we are shooting for 5% CTR (10% is ideal) - kill lander:
if i ONLY have 0 clicks after 71 imps (b/c I'm 95.00% confident that CTR won't be higher than 5.06%)
if i ONLY have 1 clicks after 109 imps (b/c I'm 95.00% confident that CTR won't be higher than 5.01%)
if i ONLY have 2 clicks after 142 imps (b/c I'm 95.00% confident that CTR won't be higher than 5.00%)
if i ONLY have 3 clicks after 172 imps (b/c I'm 95.00% confident that CTR won't be higher than 5.01%)
if i ONLY have 4 clicks after 201 imps (b/c I'm 95.00% confident that CTR won't be higher than 5.02%)
if i ONLY have 5 clicks after 230 imps (b/c I'm 95.00% confident that CTR won't be higher than 5.00%)
if i ONLY have 6 clicks after 257 imps (b/c I'm 95.00% confident that CTR won't be higher than 5.01%)
if i ONLY have 7 clicks after 284 imps (b/c I'm 95.00% confident that CTR won't be higher than 5.01%)
--
0 clicks after 71 imps (b/c I'm 95.00% confident that CTR won't be higher than 5.06%) -> seems too early, even though it's statistically sound?
1 clicks after 109 imps (b/c I'm 95.00% confident that CTR won't be higher than 5.01%) -> is this too early as well?
THANKS for your insights!!!
Originally Posted by harrypotter
With that said, this assumes that you've taken a representative sample. If all those impressions came at the same time of day, it might not be representative. You need to be the judge of that. If your impressions have a good chance of being representative, then this math works.
Great freakin post man! Just one question, what's excel? J/K!!!
Originally Posted by inversion
representative sample is KEY and that was what got me really confused.
fact is, with PPV or any traffic source, even when we reach the impression count, it doesn't mean anything if we are only sampling a segment of the population!!!
AHHHHHHHHHHH
got another stats question that's related to this if you don't mind:
i was playing around with http://www.answersresearch.com/size.php and I ROUGHLY estimated the sample size needed for each confidence intervals
here is what i did... and found... just wanted to see if this is TOTALLY wrong or if these findings could be valid
lol... "effective" confidence, if there is such a thing in my definition = confidence level with maximum error
so example is if I want a confidence level of 85%, I would enter 99% confidence level with ± 14% error. so the worst case scenario would be 85% (effective confidence)
85% "effective" confidence?
sample size needed = 85 (99% confidence; ± 14% error)
sample size needed = 96 (95% confidence; ± 10% error)
sample size needed = 271 (90% confidence; ± 5% error)
90% "effective" confidence?
sample size needed = 205 (99% confidence; ± 9% error)
sample size needed = 384 (95% confidence; ± 5% error)
95% "effective" confidence?
sample size needed = 1037 (99% confidence; ± 4% error)
-> am i crazy? or does this make sense for finding sample size?
god damn I should hire inversion instead of letting him compete with me.
adpro - sorry if I'm taking all the monies. . . but, mainly just on mobile these days.
Home > Affiliate Marketing Forum >