I'm running an experiment where we are taking 1500 of a certain type of product and changing the price of half of them to see if this influences sales. There are 5 categories on which I will be comparing the data that are exponentially distributed (i.e. sales volume over time for different regions).
I need to put the selected products into Groups A and Group B where Group A will be the products with the price change.
I also need to decide how long to run this test to make it valid.
A few questions that arise...
- Before starting with any statistics... Is it proper to just take all the products with an odd index number and place those in Group A and place all the even indexed products in Group B? My issue with doing this is that the two groups will not be similar with respect to average sales volume for the different regions, but does that really matter? It seems to me that even if the values were different I'm just testing to see at what rate Group A or Group B changed; the similarity for the two starting groups is meaningless.
- However, if it is important that the two groups are similar before starting the test, is it possible to "finagle" the data into equal groups with me selecting the products individually to go into Group A or Group B? Someone mentioned that this is improper since one you select items to get awesome p-values for a t-test you "lose independence". I don't really know what that means though.
- I've been looking on the boards and saw some stuff on the "equivalence" package in R, however, even after using that a bit it mentions in the documentation that your data must be normal. How would I tweak this for exponential data?
- Let's say I figure out a method to divide the two groups. How long should I have the experiment run?
Any help would be much appreciated.