02-27-2012 12:43 PM
I am simulating multivariate data using proc simnormal. I have known parameters rather than generating them using proc corr, so I am creating a dataset that looks like the one proc corr would have output. Proc simnormal seems to require that this dataset includes means standard deviations correlation coefficents and N. I can't see how this N is being used. It seems that unless N=1 the value N takes doesn't really matter, but if N=1 then the simulated values all take the mean values that are input (which makes some sense). Bas I have no known value for N as an input parameter is my assumption that it has no effect so I can choose any value (except N=1) correct?
08-31-2012 02:46 AM
I raised this question to SAS and the best way to apprach it appears to be to create a COV dataset rather than a CORR dataset when simulating from known parameters but with no knowledge of the N that they are based on. Using this dataset does not encounter the same problems as the CORR dataset when no N is provided.