MemoHR fMRI Behavioral Data

This is the behavioral analysis to go along with the MemoHR fMRI project.

1 Load & process data

1.0.1 Options

subjects <- c('s04','s05','s06','s07','s08','s09','s10','s11',
              's13','s14','s16','s17','s18',
              's19','s20','s21','s22','s23','s25','s26','s27','s28') # subjects included in fMRI analysis
datadir <- '/Volumes/External/MemoHR/Data/Behavioral/'
outdir <- paste(datadir,'summary/',sep='')
readOriginal <- FALSE # read in original csv files; otherwise pull from existing behavior_unprocessedData.Rdata 
processData <- FALSE # recode variables etc; otherwise pull from existing behavior_processedData.Rdata 
writeFiles <- FALSE # save out clean datafile per subject 
dir.create(outdir)

'/Volumes/External/MemoHR/Data/Behavioral/summary' already exists

1.0.2 Read in original data

if (readOriginal) {
  encdata <- data.frame()
  recdata <- data.frame()
  for (sub in subjects) {
    sub.encdata <- read.csv(paste(datadir,sub,'/',sub,'_encoding.csv',sep=""))
    sub.recdata <- read.csv(paste(datadir,sub,'/',sub,'_recognition.csv',sep=""))
    
    sub.encdata$X <- NULL
    sub.recdata$X <- NULL
    encdata <- rbind(encdata,sub.encdata)
    recdata <- rbind(recdata,sub.recdata)
    save(encdata,recdata,outdir,subjects,file=paste(outdir,"behavior_unprocessedData.Rdata",sep=""))
  }
} else {
  load(paste(outdir,"behavior_unprocessedData.Rdata",sep=""))
}
table(encdata$participant) # number of rows per participant - should be 192 trials + 6 run onset rows (except s20, missing one run)

  4   5   6   7   8   9  10  11  13  14  16  17  18  19  20  21  22  23  25  26  27  28 
198 198 198 198 198 198 198 198 198 198 198 198 198 198 165 198 198 198 198 198 198 198 

1.0.3 Process data

if (processData) {
  # Recode run numbers and trial numbers (start from 1 not 0)
  encdata$runs.thisN <- encdata$runs.thisN + 1
  encdata$trials.thisN <- encdata$trials.thisN + 1
  recdata$trials.thisN <- recdata$trials.thisN + 1
  
  # Recode responses
  encdata$responses.enc <- encdata$response.keys
  encdata$responses.enc <- factor(encdata$responses.enc,levels=c("1","2","3","4"))
  recdata$responses.itemrec <- revalue(recdata$response_itemrec.keys,c("n"="new","b"="fam","v"="rem"))
  recdata$responses.itemrec <- factor(recdata$responses.itemrec,levels=c("new","fam","rem"))
  recdata$responses.sourcerec <- revalue(recdata$response_sourcerec.keys,c("v"="within SF?","b"="professional photo?"))
  recdata$responses.sourceconf <- revalue(recdata$response_sourceconf.keys,c("v"="sure","b"="unsure"))
  recdata$responses.sourceconf <- factor(recdata$responses.sourceconf,levels=c("sure","unsure"))
  # Score item memory responses
  oldmask <- which(recdata$TrialType=="Old")
  newmask <- which(recdata$TrialType=="New")
  Fresp <- which(recdata$responses.itemrec=="fam")
  Rresp <- which(recdata$responses.itemrec=="rem")
  Nresp <- which(recdata$responses.itemrec=="new")
  recdata$itemScore <- factor(NA,levels=c("Rec","Fam","Miss","R-FA","F-FA","CR"))
  recdata$itemScore[intersect(oldmask,Rresp)] <- "Rec"
  recdata$itemScore[intersect(oldmask,Fresp)] <- "Fam"
  recdata$itemScore[intersect(oldmask,Nresp)] <- "Miss"
  recdata$itemScore[intersect(newmask,Rresp)] <- "R-FA"
  recdata$itemScore[intersect(newmask,Fresp)] <- "F-FA"
  recdata$itemScore[intersect(newmask,Nresp)] <- "CR"
  table(recdata$participant,recdata$itemScore)
  
  # Score source memory responses
  encdata$ConLabel <- factor(encdata$ConLabel,levels=c("within SF?","professional photo?"))
  recdata$ConLabel <- factor(recdata$ConLabel,levels=c("within SF?","professional photo?"))
  recdata$responses.sourcerec <- factor(recdata$responses.sourcerec,levels=c("within SF?","professional photo?"))
  recdata$sourceScore <- factor(NA,levels=c("Correct","Incorrect"))
  recdata$sourceScore[recdata$ConLabel==recdata$responses.sourcerec & !is.na(recdata$responses.sourcerec)] <- "Correct"
  recdata$sourceScore[recdata$ConLabel!=recdata$responses.sourcerec & !is.na(recdata$responses.sourcerec) & recdata$TrialType=="Old"] <- "Incorrect"
  summary(recdata$sourceScore)
  recdata$sourceScore.sure <- factor(NA,levels=c("Correct","Incorrect"))
  recdata$sourceScore.sure[recdata$ConLabel==recdata$responses.sourcerec & !is.na(recdata$responses.sourcerec) & recdata$responses.sourceconf=="sure"] <- "Correct"
  recdata$sourceScore.sure[recdata$ConLabel!=recdata$responses.sourcerec & !is.na(recdata$responses.sourcerec) & recdata$TrialType=="Old"] <- "Incorrect"
  recdata$sourceScore.sure[recdata$ConLabel==recdata$responses.sourcerec & !is.na(recdata$responses.sourcerec) & recdata$responses.sourceconf=="unsure"] <- "Incorrect"
  summary(recdata$sourceScore.sure)
  table(interaction(recdata$ConLabel,recdata$participant),recdata$sourceScore)
  
  # Get rid of extra run onset rows in encoding
  encdata <- subset(encdata,!is.na(onsetTimeTrial))
  
  # Merge data
  alldata <- merge(encdata,recdata,by=c("participant","ID","EmotionType","TrialType",
                                     "Category","ConID","ImageFile","Category","condition",
                                     "ConList","ConLabel","CB"),all.x=TRUE,all.y=TRUE,suffixes=c('.E','.R'))
  alldata$EmotionType <- revalue(alldata$EmotionType,c("Negative"="Emo","Neutral"="Neu"))
  
  # Save cleaned up version
  outputdata <- alldata[,c("participant","ID","EmotionType","TrialType","Category","CB","ConLabel",
                        "runs.thisN","trials.thisN.E","onsetTimeCue","onsetTimeTrial","responses.enc","response.rt",
                        "trials.thisN.R","responses.itemrec","responses.sourcerec","responses.sourceconf",
                        "response_itemrec.rt","response_sourcerec.rt","response_sourceconf.rt","itemScore","sourceScore",
                        "sourceScore.sure")]
  
  save(outputdata,file=paste(outdir,"behavior_processedData.Rdata",sep=""))
} else {
  load(paste(outdir,"behavior_processedData.Rdata",sep=""))
}

1.0.4 Write clean datafiles

if (writeFiles) {
  for (sub in subjects) {
    # Write out a cleaned up csv of all of the data - used for generating model regressors later
    outputdata.cursub <- subset(outputdata,participant==as.numeric(substring(sub,2,3))) 
    dataFilename <- paste(datadir,sub,'/',sub,'_alldata.csv',sep="")
    write.csv(outputdata.cursub, file=dataFilename)
  }
}

2 Analyze memory data

2.1 Item recognition

2.1.1 Base response rates

# denominator
sumdata.itemrec <- filter(outputdata,!is.na(itemScore) & TrialType=="Old") %>%
  group_by(participant,EmotionType) %>%
  summarize(allFreq = length(EmotionType))
# numerator
curdata.itemrec <- filter(outputdata,!is.na(itemScore) & TrialType=="Old") %>%
  group_by(participant,EmotionType,itemScore) %>%
  summarize(respFreq = length(itemScore)) 
# rate
curdata.itemrec <- merge(curdata.itemrec,sumdata.itemrec,by=c("participant","EmotionType")) %>%
  mutate(respRate = respFreq/allFreq,
         x = interaction(EmotionType,itemScore))
# Plot mean+-SEM rates
meandata.itemrec <- filter(curdata.itemrec,!is.na(itemScore)) %>%
  group_by(itemScore,EmotionType) %>%
  summarize(meanval=mean(respRate),
            sdval=sd(respRate),
            sem=sdval/sqrt(length(respRate)),
            nsubj=length(respRate))
mybarplot(data=meandata.itemrec,xvar="EmotionType",fillvar="EmotionType",facets=c(".","itemScore"),errorbars=TRUE) + 
  ylab('Response Rate') + xlab('') + theme_minimal(24) # functions from memolabr

These are just the basic response rates. Now let’s estimate recollection and familiarity. We’ll correct for false alarms, and familiarity will be adjusted according to the independence assumption (i.e., familiarity reflects number of Fam responses corrected for the rate of Rec responses).

2.1.2 Corrected rates

2.1.2.1 Calculate corrected rates

# denominator
sumdata.itemrec.corr <- filter(outputdata,!is.na(itemScore)) %>%
  group_by(participant,EmotionType,TrialType) %>%
  summarize(allFreq = length(EmotionType))
# numerator
curdata.itemrec.corr <- filter(outputdata,!is.na(itemScore)) %>%
  group_by(participant,EmotionType,itemScore,TrialType) %>%
  summarize(respFreq = length(itemScore)) # trials of each type
# rates
curdata.itemrec.corr.allrates <- merge(curdata.itemrec.corr,sumdata.itemrec.corr,by=c("participant","EmotionType","TrialType")) %>%
  mutate(respRate = respFreq/allFreq) # calculate response rate
# Calculate rates - filling in zeroes for anyone with missing combination (e.g., no recollection FAs)
curdata.itemrec.corr.allrates.summary <- 
  group_by(curdata.itemrec.corr.allrates,participant,EmotionType) %>%
  mutate(tmp1=length(which(itemScore=="R-FA")),
         Rold=respRate[itemScore=="Rec"],
         Rnew=ifelse(tmp1>0,respRate[itemScore=="R-FA"],0),
         tmp2=length(which(itemScore=="F-FA")),
         Fold=respRate[itemScore=="Fam"],
         Fnew=ifelse(tmp2>0,respRate[itemScore=="F-FA"],0),
         tmp3=length(which(itemScore=="Miss")),
         missRate = ifelse(tmp3>0,respRate[itemScore=="Miss"],0),
         tmp4=length(which(itemScore=="CR")),
         CRrate = ifelse(tmp4>0,respRate[itemScore=="CR"],0)
  ) %>% 
  select(participant,EmotionType,Rold,Rnew,Fold,Fnew,missRate,CRrate) %>%
  unique() %>%
  as.data.frame()
# Calculate corrected rates - somewhat redundant with above for legacy reasons
curdata.itemrec.corr <- 
  group_by(curdata.itemrec.corr.allrates,participant,EmotionType) %>%
  mutate(tmp1=length(which(itemScore=="R-FA")),
         Rold=respRate[itemScore=="Rec"],
         Rnew=ifelse(tmp1>0,respRate[itemScore=="R-FA"],0),
         cRec=(Rold - Rnew)/(1 - Rnew),
         tmp2=length(which(itemScore=="F-FA")),
         Fold=respRate[itemScore=="Fam"],
         Fnew=ifelse(tmp2>0,respRate[itemScore=="F-FA"],0),
         cFam=(Fold/(1-Rold)) - (Fnew/(1-Rnew))
  ) %>% 
  select(participant,EmotionType,cRec,cFam) %>%
  unique() %>%
  gather("correctedRate","value",3:4) %>%
  as.data.frame()

2.1.2.2 Plot & stats for item recognition

# Plot mean+-SEM
curdata.itemrec.corr$correctedRate <- factor(curdata.itemrec.corr$correctedRate,levels=c("cRec","cFam"))
meandata.itemrec.corr <- filter(curdata.itemrec.corr,!is.na(correctedRate)) %>%
  group_by(correctedRate,EmotionType) %>%
  summarize(meanval=mean(value),
            sdval=sd(value),
            sem = sdval/sqrt(length(value)),
            nsubj=length(value)) 
mybarplot(data=meandata.itemrec.corr,xvar="EmotionType",fillvar="EmotionType",facets=c(".","correctedRate"),errorbars=TRUE) +
  ylab('Corrected Rate') + xlab('') + theme_minimal(24) #+ scale_y_continuous(expand=c(0.01,0.01))

# Stats
## Recollection
ezANOVA(data=subset(curdata.itemrec.corr,correctedRate=="cRec"),dv=value,wid=participant,within=c(EmotionType),type=3)

Converting "participant" to factor for ANOVA.You have removed one or more levels from variable "EmotionType". Refactoring for ANOVA.

$ANOVA

## Familiarity
ezANOVA(data=subset(curdata.itemrec.corr,correctedRate=="cFam"),dv=value,wid=participant,within=c(EmotionType),type=3)

Converting "participant" to factor for ANOVA.You have removed one or more levels from variable "EmotionType". Refactoring for ANOVA.

$ANOVA

# Store corrected recognition scores
itemrecdata <- spread(curdata.itemrec.corr,correctedRate,value)
itemrecdata %>%
  group_by(participant) %>%
  summarize(cRec = cRec[EmotionType=="Emo"] - cRec[EmotionType=="Neu"],
            cFam = cFam[EmotionType=="Emo"] - cFam[EmotionType=="Neu"],
            EmotionType = 'EmovNeu') -> tmp
itemrecdata %>%
  select(participant,cRec,cFam,EmotionType) %>%
  rbind(tmp) -> itemrecdata
head(itemrecdata)

There is a significant enhancing effect of emotion on recollection estimates but not on familiarity estimates.

2.2 Source memory

2.2.1 Both levels of confidence - Stats for source memory

# denominator
outputdata %>%
  filter(!is.na(sourceScore) & TrialType=="Old" & itemScore!="Miss") %>%
  group_by(participant,EmotionType) %>%
  summarize(allFreq = length(EmotionType)) -> sumdata.src
# numerator
outputdata %>%
  filter(TrialType=="Old" & itemScore!="Miss") %>%
  group_by(participant,EmotionType,sourceScore) %>%
  summarize(respFreq = length(sourceScore)) -> curdata.src
# rates
curdata.src <- merge(curdata.src,sumdata.src,by=c("participant","EmotionType")) %>%
  mutate(respRate = respFreq/allFreq) %>%
  filter(sourceScore=="Correct")
# Plot mean+-SE
curdata.src %>%
  group_by(EmotionType) %>%
  summarize(meanval=mean(respRate),
            sdval=sd(respRate),
            sem = sdval/sqrt(length(respRate)),
            nsubj=length(respRate)) -> meandata.src
mybarplot(data=meandata.src,xvar="EmotionType",fillvar="EmotionType",errorbars=TRUE) +
  ylab('Source Accuracy') + geom_hline(aes(yintercept=.5), linetype="dashed") + theme_minimal(24)

# Stats
## Source Accuracy
ezANOVA(data=as.data.frame(curdata.src),dv=respRate,wid=participant,within=c(EmotionType),type=3)

Converting "participant" to factor for ANOVA.You have removed one or more levels from variable "EmotionType". Refactoring for ANOVA.

$ANOVA

## Source Accuracy corrected for chance rate (for testing intercept)
curdata.src$respRate.corr <- curdata.src$respRate - .5
ezANOVA(data=as.data.frame(curdata.src),dv=respRate.corr,wid=participant,within=c(EmotionType),type=3,detailed=TRUE)

Converting "participant" to factor for ANOVA.You have removed one or more levels from variable "EmotionType". Refactoring for ANOVA.

$ANOVA

# #### Store source recognition scores
curdata.src %>%
  group_by(participant) %>%
  summarize(sourceRate = respRate[EmotionType=="Emo"] - respRate[EmotionType=="Neu"],
            EmotionType = 'EmovNeu') -> tmp
curdata.src %>%
  mutate(sourceRate = respRate) %>%
  select(participant,sourceRate,EmotionType) %>%
  rbind(tmp) -> sourcerecdata
head(sourcerecdata)

There is no significant effect of emotion on source accuracy (collapsing high and low confidence). What if we now consider high-confidence responses only?

2.2.2 High-confidence source memory

# ### Plot source memory distribution: Rate
# denominator
outputdata %>%
  filter(!is.na(sourceScore.sure) & TrialType=="Old" & itemScore!="Miss") %>%
  group_by(participant,EmotionType) %>%
  summarize(allFreq = length(EmotionType)) -> sumdata.src.sure
# numerator
outputdata %>%
  filter(TrialType=="Old" & itemScore!="Miss") %>%
  group_by(participant,EmotionType,sourceScore.sure) %>%
  summarize(respFreq = length(sourceScore.sure)) -> curdata.src.sure
# rate
curdata.src.sure <- merge(curdata.src.sure,sumdata.src.sure,by=c("participant","EmotionType")) %>%
  mutate(respRate = respFreq/allFreq) %>%
  filter(sourceScore.sure=="Correct")
# Plot mean+-SEM
curdata.src.sure %>%
  group_by(EmotionType) %>%
  summarize(meanval=mean(respRate),
            sdval=sd(respRate),
            sem = sdval/sqrt(length(respRate)),
            nsubj=length(respRate)) -> meandata.src.sure
mybarplot(data=meandata.src.sure,xvar="EmotionType",fillvar="EmotionType",errorbars=TRUE) +
  ylab('Source Accuracy') + theme_minimal(24)

# Stats
## Source Accuracy - High confidence
ezANOVA(data=as.data.frame(curdata.src.sure),dv=respRate,wid=participant,within=c(EmotionType),type=3)

Converting "participant" to factor for ANOVA.You have removed one or more levels from variable "EmotionType". Refactoring for ANOVA.

$ANOVA

# #### Store source recognition scores
curdata.src.sure %>%
  group_by(participant) %>%
  summarize(sourceRate = respRate[EmotionType=="Emo"] - respRate[EmotionType=="Neu"],
            EmotionType = 'EmovNeu') -> tmp
curdata.src.sure %>%
  mutate(sourceRate = respRate) %>%
  select(participant,sourceRate,EmotionType) %>%
  rbind(tmp) -> sourcerecdata.sure.all

There appears to be an impairing effect of emotion on high-confidence source accuracy. However, this estimate is not corrected for bias to use the high-confidence rating. So let’s fix that.

2.2.3 High-confidence source memory correcting for high-conf incorrect

# Relabel only "sure" incorrect responses as incorrect; unsure marked unsure
outputdata$sourceScore.sure.new <- factor(NA,levels=c("Correct","Incorrect","Unsure"))
outputdata$sourceScore.sure.new[outputdata$ConLabel==outputdata$responses.sourcerec & !is.na(outputdata$responses.sourcerec) & outputdata$responses.sourceconf=="sure"  & outputdata$TrialType=="Old"] <- "Correct"
outputdata$sourceScore.sure.new[outputdata$ConLabel!=outputdata$responses.sourcerec & !is.na(outputdata$responses.sourcerec) & outputdata$responses.sourceconf=="sure" & outputdata$TrialType=="Old"] <- "Incorrect"
outputdata$sourceScore.sure.new[!is.na(outputdata$responses.sourcerec) & outputdata$responses.sourceconf=="unsure" & outputdata$TrialType=="Old"] <- "Unsure"
# denominator
outputdata %>%
  filter(!is.na(sourceScore.sure.new) & TrialType=="Old" & itemScore!="Miss") %>%
  group_by(participant,EmotionType) %>%
  summarize(allFreq = length(EmotionType)) -> sumdata.src.corr
# numerator
outputdata %>%
  filter(TrialType=="Old" & itemScore!="Miss") %>%
  group_by(participant,EmotionType,sourceScore.sure.new) %>%
  summarize(respFreq = length(sourceScore.sure.new)) -> curdata.src.corr
# rate
curdata.src.corr <- merge(curdata.src.corr,sumdata.src.corr,by=c("participant","EmotionType")) %>%
  mutate(respRate = respFreq/allFreq)
# the above was the same as before - now correct for rate of high-confidence incorrect source responses
curdata.src.corr <- curdata.src.corr %>%
  group_by(participant,EmotionType) %>%
  mutate(nFA = length(respRate[sourceScore.sure.new=="Incorrect"]),
         respRate = ifelse(nFA>0,
                           respRate[sourceScore.sure.new=="Correct"] - respRate[sourceScore.sure.new=="Incorrect"],
                           respRate[sourceScore.sure.new=="Correct"])) %>%
  filter(sourceScore.sure.new=="Correct") # corrected for incorrect sure responses (old items only)
# Plot mean+-SEM
curdata.src.corr %>% 
  group_by(EmotionType) %>%
  summarize(meanval=mean(respRate),
            sdval=sd(respRate),
            sem = sdval/sqrt(length(respRate)),
            nsubj=length(respRate)) -> meandata.src.corr
mybarplot(data=meandata.src.corr,xvar="EmotionType",fillvar="EmotionType",errorbars=TRUE) +
  ylab('Source Accuracy') + theme_minimal(24)

#' #### Stats
#' Source Accuracy
ezANOVA(data=as.data.frame(curdata.src.corr),dv=respRate,wid=participant,within=c(EmotionType),type=3)

Converting "participant" to factor for ANOVA.You have removed one or more levels from variable "EmotionType". Refactoring for ANOVA.

$ANOVA

NA

After correcting the high-confidence source accuracy rates, there is no significant effect of emotion.

3 Summarize data

3.1 Save scores

# Merge together all summary measures (itemrec, sourcerec, sourcerec-sure)
sourcerecdata <- merge(sourcerecdata,sourcerecdata.sure.all,by=c("participant","EmotionType"),suffixes=c("",".sure"))
summarydata <- merge(itemrecdata,sourcerecdata,by=c("participant","EmotionType"),suffixes=c(".item",".source"))
# use subject IDs instead of just numbers
snamefun <- function(s) {
  if (s<10) {sname <- paste0('s0',as.character(s))}
  else { sname <- paste0('s',as.character(s))}
  return(sname)
}
summarydata$subject <- sapply(summarydata$participant,snamefun)
head(summarydata)

write.csv(summarydata,file=paste0(outdir,'behavioral_summary_scores.csv'))

3.2 Table of behavioral data

# separate for easy indexing in markdown-formatted table (below)
summarydata %>%
  filter(EmotionType=="Emo")  -> emoScores
summarydata %>%
  filter(EmotionType=="Neu")  -> neuScores
curdata.itemrec.corr.allrates.summary %>%
  filter(EmotionType=="Emo") -> emoScores.rates
curdata.itemrec.corr.allrates.summary %>%
  filter(EmotionType=="Neu") -> neuScores.rates
formatMSD <- function(df) {
  meanval <- mean(df)
  sdval <- sd(df)
  sprintf("%0.2f (%0.2f)",meanval,sdval)
}

3.2.1 Paper Table 1

Condition	Remember rate	Familiar rate	New Rate	Recollection estimate	Familiarity estimate	Source accuracy	Source accuracy - high confidence
Emotion - old	0.60 (0.16)	0.26 (0.10)	0.14 (0.10)	0.60 (0.15)	0.53 (0.19)	0.67 (0.06)	0.36 (0.14)
Neutral - old	0.53 (0.18)	0.33 (0.13)	0.14 (0.10)	0.51 (0.18)	0.53 (0.15)	0.68 (0.08)	0.40 (0.13)
Emotion - new	0.02 (0.04)	0.14 (0.16)	0.85 (0.20)	-	-	-	-
Neutral - new	0.04 (0.04)	0.17 (0.13)	0.79 (0.15)	-	-	-	-

3.3 Paper Figure 2 - Behavioral data plot

plotcolors1 <- c('#e9a3c9','#a1d76a','#e9a3c9','#a1d76a','#e9a3c9','#a1d76a')
plotcolors2 <- c('#d01c8b','#4dac26')
# Reformat dataframe for easier plotting
summarydata %>%
  filter(EmotionType=="Emo" | EmotionType=="Neu") %>%
  gather(measure,value,-participant,-EmotionType,-subject) -> summarydata.long
summarydata.long$measure <- factor(summarydata.long$measure,levels=c("cRec","cFam","sourceRate","sourceRate.sure"))
summarydata.long.rec <- filter(summarydata.long,measure=="cRec" | measure=="cFam" | measure=="sourceRate" )
summarydata.long.src <- filter(summarydata.long,measure=="sourceRate" | measure=="sourceRate.sure")
# Plot all behavioral measures
summarydata.long.rec %>%
  group_by(measure,EmotionType) %>%
  summarize(meanval=mean(value),
            sem=sd(value)/sqrt(length(value))) %>%
  ggplot(aes(x=EmotionType,y=meanval)) + 
  geom_bar(stat="identity",color="gray40",fill=plotcolors1) + theme_minimal(18) +
  facet_grid(.~measure) +
  geom_dotplot(data=summarydata.long.rec,aes(x=EmotionType,y=value,fill=EmotionType), binaxis = "y", stackdir = "center", binwidth=.02) +
  geom_errorbar(aes(x=EmotionType,ymin=meanval-sem,ymax=meanval+sem,width=.2)) +
  scale_fill_manual(values=plotcolors2) + ylab("Mean corrected rate") + xlab("")