I am running a large mixed model, 65k entries on 11 fixed effects and one
random. One of the fixed effects is "Course" a factor that takes on 14
different values>levels(Master.complete$Course)
[1] "B101" "B2A" "B2B" "B2C"
"C118A" "C118B" "C118C" "C2A"
"C2B"
[10] "C2C" "N101" "P7A" "P7B"
"P7C"
and another "Yfrm7A" that is continuous
>summary(Master.complete$Yfrm7A)
Min. 1st Qu. Median Mean 3rd Qu. Max.
-5.25000 -0.75000 0.00000 -0.07688 0.50000 6.00000
but all the values are 0(zero) when course==P7A>summary(Master.complete$Yfrm7A[Master.complete$Course=="P7A"])
Min. 1st Qu. Median Mean 3rd Qu. Max.
0 0 0 0 0 0
Thus when I run the following mixed model I get no errors
fit.full <- lme(fixed= zGrade ~ Rep + ISE
+Yfrm7A+Ufrm7A+Female+White+HSGPA+MATH+AP_TOTAL+Years+Course +
Course*Rep + Course*Female +Course*White,
random= ~1|SID, data = Master.complete,
na.action=na.exclude)
but if I add in a Course*Yfrm7A term
fit.full <- lme(fixed= zGrade ~ Rep + ISE
+Yfrm7A+Ufrm7A+Female+White+HSGPA+MATH+AP_TOTAL+Years+Course +
+ Course*Rep + Course*Female +Course*White+Course*Yfrm7A,
+ random= ~1|SID, data = Master.complete,
na.action=na.exclude)
I get
Error in MEEM(object, conLin, control$niterEM) :
Singularity in backsolve at level 0, block 1
I suspect I could solve this problem with ordering the levels of
"Course"
so that P7A was the first level and thus the one that others were compared
to but I am unclear on how to do so.
Robert
[[alternative HTML version deleted]]