#### Homework 10 Key ####

library(lmSupport)
library(car)
library(ggplot2)
library(psych)

#### 1.Read in, univariate and bivariate stats ####
d <- dfReadDat("HW10_Data.dat")
str(d)
head(d)
varDescribe(d)

# Univariate:
varPlot(d$Coordination)
varPlot(d$Score)
varPlot(d$Experience)

# Bivariate 
corr.test(d)
spm(d)


#### 2. Make a descriptives table ####
varDescribeBy(d$Score, IVList=list(Cond=d$Condition,Exper=d$Experience))

varDescribeBy(d$Coordination, IVList=list(Cond=d$Condition,Exper=d$Experience))

#### 3. Center variables ####
d$ExperienceC <- varRecode(d$Experience, c(0,1), c(-0.5,0.5))
d$ConditionC <- varRecode(d$Condition, c(0,1), c(-0.5,0.5))
d$CoordinationC <- d$Coordination-mean(d$Coordination)

#### 4. interaction experience and condition ####
#4a. 
m1 <- lm(Score ~ ConditionC * ExperienceC, data=d)
modelSummary(m1)
modelEffectSizes(m1)
confint(m1)

#4b.

# b0 represents the predicted score for someone who is neutral with respect to condition
# and experience.

# b1 represents the effect of condition for someone who is neutral with respect to experience
# Those in the alcohol condition, on average, trend toward doing worse than those in the
# placebo condition

# b2 represents the effect of experience for someone who is neutral with respect to condition (or at average
# of the conditions). There are no significant differences overall across experience groups

# b3 quantifies how the (simple) effect of condition differs between those with low experience and those
# with high experience. The interaction is significant.

#4c. Describe the experience effect for those in alcohol group
d$ConditionA <- varRecode(d$Condition, c(0,1), c(1,0))
m2 <- lm(Score ~ ConditionA * ExperienceC, data=d)
modelSummary(m2)
modelEffectSizes(m2)
confint(m2)
# There is a significant positive effect of experience for those in the Alcohol group, with
# experienced players scoring 22 points higher than novices, on average. 

#4d. Describe the experience efffect for those in placebo group 
d$ConditionP <- d$Condition #since it is already coded properly 
m3 <- lm(Score ~ ConditionP * ExperienceC, data=d)
modelSummary(m3)
modelEffectSizes(m3)
confint(m3)

# There is a marginal effect of experience for participants in the placebo group with experienced 
# placebo participants scoring 18.7 points lower than novices, on average

#4e.
#### GRAPHING ####

# Recode from numeric to actual names, make factors
d$ConditionStr <- varRecode(d$Condition, c(0, 1), c("Placebo","Alcohol"))
d$ExperienceStr <- varRecode(d$Experience, c(0, 1), c("Novice","Experienced"))
d$ConditionStr = as.factor(d$ConditionStr)
d$ExperienceStr = as.factor(d$ExperienceStr)

# Re-run the model with IVs as factors and make sure that everything looks fine
m4 <- lm(Score ~ ConditionStr + ExperienceStr + ConditionStr:ExperienceStr, data=d)
modelSummary(m4)

# Yhat predictions
YhatPlacebo <- data.frame(ConditionStr = 'Placebo', ExperienceStr = c('Novice','Experienced')) 
YhatPlacebo <- modelPredictions(m4, YhatPlacebo)

YhatAlcohol <- data.frame(ConditionStr = 'Alcohol', ExperienceStr = c('Novice','Experienced'))
YhatAlcohol <- modelPredictions(m4, YhatAlcohol) 

Yhats <- rbind(YhatPlacebo,YhatAlcohol) 

# Graph

plot4 <- ggplot(Yhats, aes(x = ConditionStr, y = Predicted, fill = ExperienceStr))+
  geom_bar(stat='identity', position=position_dodge(.9))+
  geom_errorbar(aes(ymax=CIHi, ymin=CILo), position=position_dodge(.9), width=0.25)+
  theme_bw()+
  theme(legend.title=element_blank(), panel.grid.major=element_blank(),
        panel.grid.minor=element_blank(), panel.border=element_blank(), axis.line=element_line(),
        text=element_text(family=''))+
  ylab('Game Performance') + 
  xlab('Intoxication Condition')
plot4


#### 5a. Interaction condition and coordination ####
m5 <- lm(Score ~ ConditionC * CoordinationC, data=d)
modelSummary(m5)
modelEffectSizes(m5)
confint(m5)

# 5b. 

# b0 represents the predicted score for someone who is average with respect to coordination
# and condition.

# b1 represents the condition effect for someone of average coordination. The effect is
# non-significant in this sample. 

# b2 represents the coordination effect for someone who is neutral with respect to coordination
# The effect is non-significant. 

# b3 indicates how much the coordination effect differs between conditions. This interaction
# effect approached significance. 

## 5c. 

#Describe the condition effect for someone with self-assessed coordination of 8

d$Coordination8 <- d$Coordination - 8
m6 <- lm(Score ~ ConditionC * Coordination8, data=d)
modelSummary(m6)
modelEffectSizes(m6)
confint(m6)

# Among those with a coordination score of 8, the effect of condition was significant, with
# those in the alcohol group doing significantly worse than those in the placebo group. 

# 5d. 
#### GRAPHING ####
# Already made factor variable for condition
# Model with factorial dichotomous variable
m7 <- lm(Score ~ ConditionStr * Coordination, data=d)
modelSummary(m7)

# predict data from the model separately by group
YhatPlaceboB <- data.frame(Coordination=seq(min(d$Coordination),max(d$Coordination)),ConditionStr='Placebo')
YhatPlaceboB <- modelPredictions(m7, YhatPlaceboB) 

YhatAlcoholB <- data.frame(Coordination=seq(min(d$Coordination),max(d$Coordination)),ConditionStr='Alcohol') 
YhatAlcoholB <- modelPredictions(m7, YhatAlcoholB)


plot5 <- ggplot(data=d, aes(x = Coordination, y = Score)) + 
  geom_point(aes(color=ConditionStr, shape=ConditionStr)) + 
  scale_colour_brewer(palette="Set1") +
  geom_smooth(aes(ymin = CILo, ymax = CIHi, x = Coordination, y = Predicted), # Placebo group
              data = YhatPlaceboB, stat = "identity", color="blue") + 
  geom_smooth(aes(ymin = CILo, ymax = CIHi, x = Coordination, y = Predicted), # Alcohol group
              data = YhatAlcoholB, stat = "identity", color="red") + 
  theme_bw(base_size = 14) + labs(x = 'Coordination Before Dart Game', y = 'Game Performance') +
  coord_cartesian(ylim = c(0,150), xlim = c(0,11)) +
  theme(legend.position = c(.92,.92), legend.background = element_blank(), legend.title = element_blank()) 
plot5