One-sample T -tests of proportion of fixations for PComp and Unrl at 800ms bins?

mickey_tea_tasting

## T-test

library(plyr)
load(file="preprocessed.RData")

# Read data
trials <- tinfo4
responses <- dat4

# Change column names
names(trials)[4]<-"Start"
names(trials)[6]<-"End"

# Compute relative Onset
trials <- transform(trials, End = End - Start)
# Remove unneeded columns
trials$Start <- NULL
trials$Onset <- NULL
trials$Choice <- NULL
trials$X <- NULL
responses$X <- NULL
# Merge in the actual observations
all.data <- merge(trials, responses, by="RespID", all=TRUE)
# Compute time realtive to start
all.data <- transform(all.data, ms = End + ms)
# Group each session/trial and sort them interally by ms
all.data <- all.data[with(all.data, order(SessionID, RespID, ms)),]
# Find the ending time for each response (i.e. the starting time of the
# next response or the end of the trial).

all.data <- ddply(all.data, c(.(SessionID), .(RespID)), transform, to=c(tail(ms,-1), head(End,1)))

# Clean up 
all.data <- all.data[ ,c("SessionID", "RespID", "ms", "to", "AOI", "ItemID")]
# Compute amount of time in the interval 200-1000ms

window <- function(xs, from, to) pmax(from, pmin(to, xs))

all.data <- transform(all.data, in.interval = window(to, 200, 1000) - window(ms, 200, 1000))
# Extract the value with highest "in.interval" for each RespID, including none whenever
# there is no observation in the interval
all.data <- ddply( all.data, .(RespID), transform
                   , is.most = max(in.interval) > 0 & in.interval == max(in.interval) )
# Check whether you have to break ties, this will give an error if you do
stopifnot(all(!duplicated(subset(all.data, is.most, RespID))))

# Here we compute, seperately, the proportions for the 200-1000ms responses
# and the non 200-1000ms responses, for each participant
per.participant.props <- ddply( all.data, c(.(SessionID), .(is.most))
                                , function(xs) table(xs$AOI)/nrow(xs) )

# Filling missing data with NA
per.participant.props <-
  merge(expand.grid( SessionID=unique(per.participant.props$SessionID)
                     , is.most=c(TRUE, FALSE) ), per.participant.props, all=TRUE)

# comparing the different proportion informations to each other
with( subset(per.participant.props, is.most) # Extract the 200-1000ms responses
      , t.test(PComp, Unrl) )                  # Compare mean for PComp with mean for Unrl

t.test( paired=T
        # Proportion of PComp among PComp Unrl observations within 200-1000ms interval
        , with(subset(per.participant.props,  is.most), PComp/(PComp+Unrl))
        # Proportion of PComp among PComp Unrl observations outside 200-1000ms interval
        , with(subset(per.participant.props, !is.most), PComp/(PComp+Unrl)) )