On Thu, 7 Sep 2006, Arnaud Bervas wrote: Ah, I see. I think that what you're aiming at here could be done more simply. The following sticks closely to your design, but uses matrices. matrix SR = zeros(11, 1) matrix DL = zeros(11, 1) set echo off set messages off loop for i= 0..10 --quiet scalar j = i + 1 ols Y const X(0 to -$i) --quiet matrix SR[j,] = $ess # Is the following line right? Perhaps should be # just $nobs - $ncoeff (?) matrix DL[j,] = $nobs - $i - $ncoeff endloop matrix SR print matrix DL print loop for i= 1..10 --quiet scalar j = i + 1 scalar dfn = DL[i,] - DL[j,] scalar dfd = DL[j,] scalar Fval = ((SR[i,] - SR[j,]) / dfn) / (SR[j,] / dfd) scalar pv = pvalue(F, dfn, dfd, Fval) printf "Fisher: F(%d,%d) = %g (p-value = %.4f)\n", dfn, dfd, Fval, pv endloop Allin.