grouped data: generating a random error with zero group mean

I have a variable 'group' with 20 distinct values randomly assigned for 8000 observations. I need to generate a random error term,u, with zero mean in each group, and a specific covariance, say 10. how to do that?

I can do for an overall zero mean, as

u = random('Normal',0,10,length(group),1)

but not for zero group means.

Answer 1

[Tom Lane]

Your code generates each observation from a theoretical distribution that has zero mean. Consequently the theoretical mean is zero overall, and within each group. Can you describe more specifically what you need beyond this?