Hi,
I got a computational problem with lme (nlme library R 1.9.1) using
corSymm(). Here is the data:
[,1] [,2] [,3] [,4] [,5] [,6]
[1,] 0.19639793 0.09127954 0.11733288 0.07598273 0.06545106 0.06211532
[2,] 0.22773467 0.10981912 0.16052847 0.38101187 0.18353474 0.24072918
[3,] 0.46743388 0.45733836 0.32191178 0.43356107 0.39159746 0.53984221
[4,] 0.29051642 0.32184177 0.24935224 0.31634000 0.16355260 0.12430647
[5,] 0.07353932 0.10695346 0.12149297 0.12052110 0.08682816 0.10002020
[6,] 0.18672701 0.09027081 0.07597009 0.23282325 0.14697649 0.12762290
[7,] 0.26289407 0.27713945 0.13372536 0.36492799 0.25986507 0.22709220
[8,] 0.10527164 0.04871462 0.06152876 0.14244950 0.08310103 0.08212435
[9,] 0.06620246 0.07877543 0.05525695 0.11684909 0.06401386 0.06903739
[10,] 0.12867578 0.07969712 0.09466616 0.12565510 0.10685046 0.11924971
[11,] 0.12426949 0.09819454 0.12543859 0.12321680 0.09282128 0.11261496
[12,] 0.14227878 0.07937399 0.08314904 0.12904447 0.08569169 0.08205721
[13,] 0.13509013 0.09850748 0.09838632 0.16405161 0.09319339 0.10619560
[14,] 0.44856255 0.25898742 0.15940404 0.56595476 0.30830903 0.40427274
[15,] 0.12786080 0.13699933 0.15702839 0.15100094 0.15037274 0.11427119
[16,] 0.15294429 0.12695011 0.11030015 0.13769743 0.09924456 0.08767253
[17,] 0.20091098 0.09847140 0.14685682 0.16175769 0.18205527 0.13926859
[18,] 0.11707101 0.13492758 0.07762873 0.13414458 0.11760926 0.13963532
As you can see, values in row [15,] and columns [,3] to [,5] are very
similar. To produce a vertical design matrix use:
datamat<-as.data.frame(test)
datamat<-stack(datamat)
datamat$subject<-as.factor(rep(c(1:18),times=6))
datamat$A<-as.factor(rep(c(1:3),each=18,times=2))
datamat$B<-as.factor(rep(c("one","two"),each=54))
Then fitting the model:
fm.datamat<-lme(values~A*B,data=datamat,random=~1|subject,correlation=corSym
m())
produced this error:
"Error in "coef<-.corNatural"(`*tmp*`, value = log((cStNatPar
+ 1)/(1 - :
NA/NaN/Inf in foreign function call (arg 1)"
If I skip observation subject 15 (row 15 above) the computation is
successfull. Is it because of the similar values ? Does it fail to estimate
a general variance structure ? Is there a better way than skipping row 15 ?
Thanks for some suggestions,
Stephan Moratti
-----------------------------
Dipl. Psych. Stephan Moratti
Dept. of Psychology
University of Konstanz
P.O Box D25
Phone: +40 (0)7531 882385
Fax: +49 (0)7531 884601
D-78457 Konstanz, Germany
e-mail: Stephan.Moratti at uni-konstanz.de
http://www.clinical-psychology.uni-konstanz.de/
