lme with cyclic cubic regression splines
Gabriela Czanner via R-sig-mixed-models <r-sig-mixed-models at ...> writes:
Hello list, I am trying to fit linear mixed model with fixed effect being cyclic cubic regression splines, because my data are defined on a circle, while circle is divided into 24 directions. I defined a variable Direction which is numeric and has values 1,2,... 24. I receive this error message:
out.lme.8=lme(Y~ s(Direction,bs="cc",k=8),
+ random=~1|PatientID, + data=mydata,na.action=na.omit,method="ML") Error in model.frame.default(fixed, dataMix) : invalid type (list) for variable 's(Direction, bs = "cc", k = 8)'
I wonder if anyone has any suggestion, please?
As Alexandre Villers implicitly pointed out, specifying a smooth term via s() is restricted to the mgcv:gam(m) and gamm4:gamm functions. However, if you want to do this in lme (with spline order and knot positions pre-specified, rather than using penalized regression splines) it looks like you could use cSplineDes from the mgcv package to set up the splines yourself. However, I don't know how smoothly these will work with the built-in model matrix machinery -- using gamm(4) will probably be easier. Ben Bolker