I am analyzing a sample of two groups of 5,092 and 114,038 customers - and two very basic questions I assume.
Question 1: Do you think this is a problem in general AND for the usage of OLS, fixed-effects OLS, Cox proportional hazards and negative binomial regression in special?
Question 2: The mean of group 1 is 16.38 and of group 2 19.10; the mean of the whole sample is 16.50 - the result of the OLS with three control variables is 0.052, resulting in a percentage increase of 5.30 in y, when unit x is increased by one. Obviously, the difference between 16.38 and 19.10 is not 5.3 %. Is it possible to "backtransform" the percentage increase to the unit level?
Thanks in advance!