Yer in luck. I have written a macro to properly devise contrasts in complex cases, like yours. To get a copy, send me an email.

Patrick wrote: > Greetings, all. > > This may be a bit basic compared to most of what goes by this > board, but I'm stuck. > > I'm running a 2x2x2x2 ANCOVA in GLM, and want to contrast two cells > at the 3-way interaction level. I set it up in what I thought was > the correct manner, but I'm getting 'CONTRAST is not estimable.' > > It may be that some of the cell sizes are too small (the smaller > ones are in the 10-12 range), but if anyone can spot something I'm > doing wrong, I'd appreciate it. I've never really gotten the hang > of CONTRAST. > > The code: > > proc glm data=ni; > class interv rprob aprob cohort male black site; > model read2=interv|rprob|aprob|cohort male black site iq; > contrast 'Two cells' interv*rprob*aprob 1 0 0 0 0 -1 0 0 /e; > run; > > The log and output are at http://www.duke.edu/ > ~malone/contrastouput.html . > > Appreciate any help, > Pat