Mixed-Effects Modeling

• Preprint, 40 pages [PDF].
• Simulated normally-distributed data, as used in Models (3) to (7) using a nested (multilevel) design, as well as in Models (8) and (9) using a crossed design [ASCII].
  This data set consists of 4 columns (variables) and 864 rows (observations). The variables represent: (1) subject ID, (2) item ID, (3) treatment condition, (4) response score, respectively. The decimal symbol is a point (not comma). Footnote 1 in the manuscript gives the variance-covariance matrices used in the simulation.
  Using the nested design, these data are analyzed as 12 independent observations per condition per participant (i.e., ignoring the item codes), for 24 participants and 3 conditions per participant.
  Using the crossed design, these data are analyzed as the same random sample of 36 test items per participant, for 24 participants, with items rotated and balanced over 3 conditions within each participant, according to a latin square.
• Annotated log of example analysis, of mixed-effects modeling with crossed random effects, performed with lmer in R.
• Annotated log of example analysis, of mixed-effects modeling with crossed random effects, performed with MLwiN.
• Binomial data, as used in Models (13) to (15) [ASCII].
  This data set consists of 4 columns (variables) and 864 rows (observations). The variables represent: (1) subject ID, (2) item ID, (3) treatment condition, (4) "hit" or binary response, respectively. There are 24 subjects and 36 items, rotated over conditions, as with the corresponding normally-distributed data set above.
  Using R, this binomial data set was derived from the corresponding normally-distributed data set, which was already available in data frame x24, with the following commands:
  R> h24 <- x24 # copy data frame x24 to h24
R> h24$resp <- as.integer(x24$resp>1) # 0=false=miss, 1=true=hit
R> table(h24$resp) # tabulate values of resp in h24, to check
R> write.table(h24,file="h24.txt",col.names=F,row.names=F)
• Annotated log of example analysis, of GLMM (mixed-effects logistic modeling) with crossed random effects, performed with MLwiN.