# April 10, 2024

library(tidyverse)
library(ez)
library(lme4)


twofactorRM <- read.csv("http://whlevine.hosted.uark.edu/psyc5143/twoFactorRM.csv", stringsAsFactors = T)

# I want study time to be a factor
twofactorRM$studytime <- factor(twofactorRM$studytime)

# checking levels of things & default coding
levels(twofactorRM$studytime)
contrasts(twofactorRM$studytime)
contrasts(twofactorRM$wordtype)

            # a digression: study time does NOT need to be a factor; the numbers
            # in this case are meaningful with good interval qualities, so this
            # could be treated as a quantitative numerical predictor rather than
            # a factor ... but not today!

# single-sample t-tests instead of main effects

# find each person's abstract & concrete mean
wordContrast <- twofactorRM %>% group_by(Subject, wordtype) %>% summarise(M = mean(dv))

# what do we have
wordContrast

# rearrange the data a bit
wordContrast <- wordContrast %>% spread(wordtype, M)

# what do we have now?
wordContrast
# cool

# create difference scores
wordContrast <- wordContrast %>% 
	mutate(d = concrete - abstract)

# checking one more time
wordContrast

# do the t-test (vs H0: mu = 0)
t.test(wordContrast$d, mu = 0)
summary(lm(d ~ 1, wordContrast))

# single-sample t-tests for the simple effects of word type within each study
# time condition
twofactorRM %>% filter(studytime == "1") %>% t.test(dv ~ wordtype, data = .)
twofactorRM %>% filter(studytime == "2") %>% t.test(dv ~ wordtype, data = .)
twofactorRM %>% filter(studytime == "3") %>% t.test(dv ~ wordtype, data = .)


##### EXTRA STUFF (FYI)
#####
					# not discussed in class; here as an example

					# more-complex contrast (linear trend of time)
					linearTime <- twofactorRM %>% group_by(Subject, studytime) %>% summarise(M = mean(dv))
					
					# rearrange the data
					linearTime <- linearTime %>% spread(studytime, M)
					
					# rename the variables
					colnames(linearTime)[2:4] <- c("one", "two", "three")
					
					# create the contrast
					linearTime <- linearTime %>% 
						mutate(con = -1*one + 0*two + 1*three)
					
					# do the t-test (vs H0: mu = 0)
					t.test(linearTime$con, mu = 0)

#####					
					
# the semi-informative omnibus ANOVA
ezANOVA(data = twofactorRM,
				dv = dv,
				wid = Subject,
				within = .(studytime, wordtype),
				detailed = T)

# doing ALL the t-tests
pairwise.t.test(x = twofactorRM$dv,               # outcome
                g = twofactorRM$variable,         # factor
                paired = TRUE,                    # paired data = TRUE
                p.adjust.method = "bonferroni")   # attend to FWER

# via lmer
LMM <- lmer(dv ~ studytime*wordtype + (1|Subject), twofactorRM)
anova(LMM)     # no p-values AND F does NOT match the conventional ANOVA
summary(LMM)   # contrast slopes are what we're interested in, along with their t-tests