Un bon mot s'il vous plait. I have coded an R routine to decompose messy data following Murray Aitkin and Brian Francis's NPMLE GLIM macros for the normal distribution only but extended to incorporate censoring and variance heterogeneity. Essentially it is a wrapper for nlm() in the M-part while the E-part re-estimates the weights in the same way as the GLIM macros. The big problem is that it is slow to run - even with homogeneous variance and no censoring it is substantially slow than the GLIM equivalent and when I incorporate censoring, variance heterogeneity, a few factors in the random part etc, the number of coefficients being estimated climbs very quickly and it becomes an overnight job. No SG Origin to hand, I am afraid, just good old Linux on an AMD 350 box. Being a simple chap, and not wanting to get my hands dirty with C, I have coded it using global variables (OK - back of the class) so that I can restart it if it hasn't converged within 500 iterations and I can see all the variables. Obviously the important part is the function call, which I will speed up, but other than that, do global variables slow things up? John -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html Send "info", "help", or "[un]subscribe" (in the "body", not the subject !) To: r-help-request at stat.math.ethz.ch _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._