# LMM for combined Bodner & Masson (1997) and Kinoshita (2006) data

# getwd()
# setwd( ... )

library(MASS)
library(lattice)
library(ggplot2)
library(lme4)

rm(list=ls())

load("MaskedPriming.rda")

# Helmert contrast for experiments
cmat <- matrix(c(-1,  1,  0,
	             -1, -1,  2),  3,  2, 
	             dimnames=list(c("BM1", "BM2", "SK"), 
	                              c(".BM1-2", ".BM-SK")))


# Baseline model
contrasts(d$e) <- cmat
print(m0 <- lmer(rrt ~ 1 + p*f*e + (1 | subjects) + (1 | words), data=d), cor=FALSE)

# Checking significance of correlations
# ... rrt  (replace "rrt" with "lrt" for log(RT) or "srt" for untransformed RT)
print(m1 <- lmer(rrt ~ 1 + p*f*e + (1 | subjects) + (0 + p | subjects) + (0 + f | subjects) + (1 | words), data=d), cor=FALSE)

print(m1.p <- lmer(rrt ~ 1 + p*f*e + (p | subjects) + (0+f | subjects) + (1 | words), data=d), cor=FALSE)

print(m1.f <- lmer(rrt ~ 1 + p*f*e + (f | subjects) + (0+p | subjects) + (1 | words), data=d), cor=FALSE)

print(m1.pf <- lmer(rrt ~ 1 + p*f*e + (1 | subjects) + (0+p+f | subjects) + (1 | words), data=d), cor=FALSE)

print(m2 <- lmer(rrt ~ 1 + p*f*e + (p + f | subjects) + (1 | words), data=d), cor=FALSE)

anova(m1, m1.p, m2)  # correlation of freq and priming with intercept improve fit; corr between freq and priming does not (.359)
anova(m1, m1.f, m2)  
anova(m1, m1.pf, m2)  

# For table only m0 vs. m2
anova(m0, m2)