How to calculate statistical significance for Cost of Sales

FK11

New Member
#1
Hi there, I work in ecommerce and we do a lot of AB testing.

I am testing 2 populations, and wanted to prove whether sample A has a significantly lower cost of sales vs. sample B

[note: cost of sales is calculated by cost / sales i.e. if you spend £20 and got £100 sales, then the cost of sales is 20/100 = 20%. The sample with a lower cost of sales wins]

What methodology should I use in order to calculate that?

Thanks!
 

noetsi

Fortran must die
#2
If you are testing two populations, not samples, and population 1 has lower cost than sample b than it does. There is nothing to test statistically.
 

FK11

New Member
#3
Hey thanks for the reply. Agree that if it is just pure cost we are looking then it doesn't require a test

However, what I want to see here is cost of sale. So the sales value is taken into account (i.e. Population A has a cost of sale of 20% vs. population B has a cost of sale of 10%). Any thought on how this can be tested?
 

Karabiner

TS Contributor
#4
How was your research design/your sampling scheme/your procedure for data collection, and how large were your samples?

With kind regards

Karabiner
 

noetsi

Fortran must die
#5
the question I raised was not one of how you measure something, but whether you have the population or a sample. If you have a population than the effect size you find, however its measured is the true effect size. No statistical test apply. You know they are different without a statistical test because you have the population. Statistical test are really determining if what you found in a sample is true in the larger population.

If population A has a cost of sales of X and Population B has a cost of sales of X + 1 (or X - 1) than you know they are substantively different. Statistical significance is meaningless in this case.