Song,

I don't know IML, thus won't attempt to modify your code. But, that said, couldn't you just modify the same functions you are already using. I.e., use something like:

%let nummiss=.2; If ranUni(0) lt &nummiss then call missing(whatever_variable) else assign random value to whatever_variable

Art ------ On Tue, 3 Feb 2009 13:19:56 -0800, Song <sounpra@YAHOO.COM> wrote:

>Hi All - > >If I have generated a sample data as follows: > >/**********************************************************/ >%let simsize=10; %let seed=112; %let g1=25; >%let g2=25; %let p=5; >proc iml; > sig1=I(&p); > sig2=2*I(&p); > mu1=shape({0},&p,1); > mu2=t({2 0 0 0 0}); > m=j(&simsize*(&g1+&g2),&p+2,0); > > do j=1 to &simsize; > sim=j(&g1+&g2,1,j); > group1=j(&g1,1,1); > group2=j(&g2,1,2); > call vnormal(z1,mu1,sig1,&g1,j+&seed); > call vnormal(z2,mu2,sig2,&g2,j*&seed); > >z=z1//z2; >group=group1//group2; >m[((j-1)*(&g1+&g2))+1:(j*(&g1+&g2)),]=sim||group||z; >end; >create sample from m[colname={sim group x1 x2 x3 x4 x5}]; >append from m; >quit; >/ >************************************************************************/ > >How can I create a dataset containing missing at random data such that >approximately XX% of data are missing within each simulation? > >Thanks, > >Song