# Updated March 4, 2006

# R code for running the analyses in
# "An Experiment in Minimalist Experimental Syntax"
# using GLMM (generalized linear mixed modeling).
# This code can simply be copied and pasted into R
# ("#" marks comment lines ignored by R).

#         **********************
#         **    IMPORTANT!    **
#         **********************
# Start by using "Change dir..." in R's "File" menu
# to change the directory to where you saved the data file!
# Otherwise you'll have to type in the whole path, which is a pain.

# The GLMM analysis uses R's lme4 package (written by Douglas Bates).
# If this is your first time using the lme4 package, you need to
# install it. Use "Install package(s)..." in R's "Packages" menu
# to find the source nearest you, or simply run the following command: 

install.packages("lme4")

# After installing it, you need to load it.
# You can do this by using "Load package..." in R's "Packages" menu,
# or you can run the following command. Note that the lme4 package (actually,
# the prerequisite package "Matrix") is very large, so it takes a little
# while to load (about 30 seconds on my machine). So don't panic if your
# computer seems to freeze at this point! This preinstallation helps the
# package run faster when actually used.

library(lme4)

# Note that you need to load lme4 this every time you start R
# (assuming you want to use lme4 again).

# R is an open-source project, so it is frequently updated.
# To check if the package has been updated since the last time you
# installed it, use "Update packages..." in R's "Packages" menu, or 
# the run the command "update.packages()".

# The lme4 command used here to conduct GLMM is called "lmer",
# for "linear mixed-effect ... something." Maybe that's an agentive suffix?
# To see R's help page on lmer, you can use "R functions (text)..."
# in R's "Help" menu, or enter the following command (without "#"):

#   help("lmer")

# The syntax of lmer is like this:
# lmer(Y ~ X... + (1|S), data = D, family = "F", method = "M")
# Y = output variable
# X... = formula for input variables
#   (For example, if there are two input variables X1 and X2
#    and you want to check their interaction, you would use   X1*X2
#    but if you don't care about their interaction, you would use  X1+X2.)
# S = subject. The (1|S) syntax makes R group data by subjects, and treats
#    subjects as random. By contrast, the X... variables are "fixed".
#    GLMM uses both random and fixed variables so it's "mixed".
# D = name of data source (see below)
# F = name of statistical distribution family.
#    (Since our data are binary, we need the "binomial" family.)
# M = method for estimating GLMM, since GLMM can't be computed exactly.
#    (Note that the analyses here use the "Laplace" method, which is the
#     best method currently available with the lmer function,
#     as of March 4, 2006. If you don't specify the method, lmer uses
#     the slightly faster but somewhat less accurate "PQL" method by default.)

# OK, let's start!

# Load the data file and give it a name
# (any name is OK if you're consistent):

minexp = read.table("minexp.txt",T)

# Allow R to identify columns by their headers:

attach(minexp)

# First analysis (ignoring order):

lmer(Judgment ~ Topic*Adjunct + (1|Speaker), data = minexp,
family = "binomial", method = "Laplace")

# Second analysis (testing satiation in main factors):

lmer(Judgment ~ Order*Topic + Order*Adjunct + (1|Speaker),
data = minexp, family = "binomial", method = "Laplace")

#Third analysis (removing nonsignificant interactions):

lmer(Judgment ~ Topic + Order*Adjunct + (1|Speaker),
data = minexp, family = "binomial", method = "Laplace")

# The above output, the key results are listed in the "Fixed effects"
# tables. The sign of each "Estimate" shows the direction of the effect,
# and the p-value ("Pr(>|z|)") indicates significance. The stars help
# to make the significance patterns visually clear.

# If you want to know how I did the reanalyses using subsets of speakers,
# contact me; the code I used is too ugly to distribute yet!