* the program below (begins with the “program” command) 
* takes the desired sample size "n" as the sole argument 
clear all
program define regprg 
args n
drop _all
set obs `n'
* use a skewed version of the gamma distribution to
* generate an error variable – the 2 arguments to the rgamma() 
* function are the shape & scale parameters, respectively
gen r = rgamma(1,0.1)
* center the errors to have approximately zero mean
* (Note: the mean for the gamma = shape * scale)
replace r = r-0.1
* generate a binary predictor with approx. equal groups
gen x = rbinomial(1,0.5)
* generate the outcome using slope (intercept) of 2 (1)
gen y = 1 + x*2 + r
* fit model
reg y x
end