[R] data layout for crossed factors w/interaction in linear mix models

[Rafael Diaz]

Dear All,

I am trying to fit a simple linear mixed model (see below this paragraph) arising from a crossed factorial design with 2 factors and ubalanced number of replicates (from two to five) in each cell, but I keep getting an error message (see bottom of message).  The model is:

yijk = intercept + ai + bj + abij + ejik, where:

"intercept" is fixed, and the crosss factors, ai, i = 1,..,10, and bj, j= 1,..,10, are random.  I am interested in estimating the variance components of these factors AND their interaction.  I have tried:

fm1 <- lmer(formula = V1~1 + (1|V2) + (1|V3) + (1|V4), data = 'datos') using two types of data layout for "datos":

1) using a matrix with 3 columns:

y     intercept   ai's  bj's  abij's 
y111  1           1     1     1 (1x1)
y112  1           1     1     " 
y121  1           1     2     2 (1x2)
y122  1           1     2     "
y123  1           1     2     "
y131  1           1     3     3 (1x3)
.     .           .     .     .
.     .           .     .     .

2) using the design matrix from  Y = XBeta +Zb.  That is, using the same first two columns as above, but substituting 1020 columns (10 for ai's, 10 for bj's and 100 for abij's) for the last three columns.

I get the message: "Error in eval(predvars, data, env) : invalid envir argument"

Is my data layout mispecified? Do I need to input initial values for the random components in order to get the REML estimates?  I lmer valid for unbalanced designs?  Any help would be greatly appreciated.

Rafael Diaz
California State University Sacramento
Math and Stats