Robert Lynch <robert.b.lynch <at> gmail.com> writes: > I am modeling grade as a function of membership in > various cohorts. There > are four "cohorts". (NONE, ISE07,ISE08,ISE09) and two times of cohorts > coded as ISE = TRUE (ISE0#) or FALSE (NONE). There is clear co-linearity > but that is to be expected. > > running the following code > > CutOff <-0 > fit.base <- lme(fixed= zGrade ~ Rep + COHORT/ISE + P7APrior + Female + > White + HSGPA + MATH + AP_TOTAL + Years + EOP + Course, > random= ~1|SID, > data = share[share$GRADE >= CutOff,]) > > I get the following error > > Error in MEEM(object, conLin, control$niterEM) : > Singularity in backsolve at level 0, block 1 > > but if I take out the /ISE I get no error, simmilarly if I take out the > COHORT/. > > I want to test for the effects of the different cohorts within the ISE > subset and across ISE & NONE > > I can send the data (the whole is too large) if you wish.
Please send this to r-sig-mixed-mod...@r-project.org for more discussion. The short answer is that lme can't fit models with rank-deficient fixed effect model matrices -- in other words, there are redundant parameters in your model because COHORT and ISE between them use 6 parameters to model 4 independent quantities. http://stats.stackexchange.com/questions/35071/ what-is-rank-deficiency-and-how-to-deal-with-it ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
