# tests for Quam and Swingley (submitted) word-learning study data
# Kyle Gorman <kgorman@ling.upenn.edu>

# get data
dat = read.csv('deebo.csv', header=1)

# some exploratory analysis
boxplot(LookPercentage ~ Treatment)
xtabs( ~ LookingPreference + Treatment)

# build models
m0 = lm(logit(LookPercentage) ~ Treatment, data=dat)
m1 = lmer(logit(LookPercentage) ~ Treatment + (1 | Subject), data=dat)
m2 = lmer(logit(LookPercentage) ~ Treatment + (1 | Item), data=dat)
m3 = lmer(logit(LookPercentage) ~ Treatment + (1 | Subject) + (1 | Item), data=dat)
m4 = lmer(logit(LookPercentage) ~ Treatment + (1 + Treatment | Subject), data=dat)
m5 = lmer(logit(LookPercentage) ~ Treatment + (1 + Treatment | Subject) + (1 | Item), data=dat)

# here, look at the ANOVA facts. you'll see m4 is by far the best

# outlier nalaysis
source('http://ling.upenn.edu/~kgorman/R/ranOutliers.R') # get this if necessary
ranOutliers(m4)