AffDirst
Contributor
- Joined
- Jan 24, 2020
- Messages
- 29
Ive been hooked by this thread https://afflift.com/f/threads/when-...er-using-statistics-and-scary-math-stuff.3476 .
So Ive spent some time taking couple of courses about statistics and applied it with my programmer skills to automate testing. I just wanna share my research and maybe get some interesting ideas from you guys.
Most wanted aff. marketing questions:
1. When to exclude variable(zone/browser/os etc) on the optimization process (which has been covered on the link above)
2. How do we pick best offer/landing/ads etc, which refers to A/B testing.
This includes smaller questions like, how many data should I collect, how do I know if this conversion rate will be the same in time, how much money should I have for testing, etc.
BAYESIAN CONFIDENCE INTERVAL:
Avoiding all math stuff, and assuming we are talking about CR (Conversion rate) or CTR (click rate):
Traditional statistics says that true CR has fixed value and all data obtained comes according to the distribution that represented by this CR parameter(bernoulli distribution) but estimated CR is normally distributed. In the Bayesian statistics real CR has a distribution dependent of the given data.
By updating and calculating the distribution with the new set of data each time we can get CRs range
So Ive spent some time taking couple of courses about statistics and applied it with my programmer skills to automate testing. I just wanna share my research and maybe get some interesting ideas from you guys.
Most wanted aff. marketing questions:
1. When to exclude variable(zone/browser/os etc) on the optimization process (which has been covered on the link above)
2. How do we pick best offer/landing/ads etc, which refers to A/B testing.
This includes smaller questions like, how many data should I collect, how do I know if this conversion rate will be the same in time, how much money should I have for testing, etc.
BAYESIAN CONFIDENCE INTERVAL:
Avoiding all math stuff, and assuming we are talking about CR (Conversion rate) or CTR (click rate):
Traditional statistics says that true CR has fixed value and all data obtained comes according to the distribution that represented by this CR parameter(bernoulli distribution) but estimated CR is normally distributed. In the Bayesian statistics real CR has a distribution dependent of the given data.
By updating and calculating the distribution with the new set of data each time we can get CRs range
Attachments
Last edited:
