##########################################################################
# 
# perceptionResults.R
#
# copyright (c) 2007 by the Autumn Ling H286 class and Mary E. Beckman
# Department of Linguistics, Ohio State University
#
# Script for analyzing the results of the perception experiment that we  
# did in class together on October 10, 2007, for the term project in the
# Linguistics H286 "Analyzing the Sounds of Languages" course.

# Set the working directory to the directory where the data file is kept.
# The path will depend on how you have your computer organized.
setwd('c:/Lx286/projectParts')

# Read the data file in as a data frame (a type of table where you can 
# mix columns of categorical data and with columns of numerical data).
dat=read.table("Au2007perception.txt",header=T)

# Check to see that it read in correctly, by seeing how many rows and 
# columns you have, like this:
dim(dat)
# [1] 760   6
# If the above line of two numbers appears on your R console, the 
# data have been read in correctly.  There should be 6 columns and as 
# many rows as there are students (minus Chris) * 40.  

# Here are the names of the six columns that you should have.
names(dat)
# [1] "listenerDistinguishes" "listener"              "talkerDistinguishes"  
# [4] "talker"                "stimulus"              "response" 

# Add a seventh column to the data frame, where the value for every row 
# in this column is initially set to NA (for "not available").  Call 
# this column of data "correct" since each NA will eventually be set to 
# values indicating whether the response was a correct identification 
# of the intended word.  
dat$correct=NA
# Loop through the rows of the table, comparing the stimulus and 
# response and is 1 if they match, 0 if they don't
for (i in 1:dim(dat)[1]) {
	dat$correct[i]=
		as.numeric(dat[i,"stimulus"]==as.character(dat[i,"response"]))
}

# See how accurate listeners were overall.  (Since we have 1 for every 
# response that was correct and 0 for every other one, taking the mean 
# of this column gives us the proportion of correct responses, and 
# multiplying by 100 gives us the percent of correct responses.   The 
# round() command rounds up or down to the nearest integer value.  
round(mean(dat$correct)*100)
# [1] 62

# Now calculate accuracy for various subsets.  For example, here is how 
# to calculate the accuracy for the subset of tokens where the talker 
# distinguishes the two words and so does the listener:
YY=subset(dat, talkerDistinguishes=="yes" & listenerDistinguishes=="yes")
round(mean(YY$correct)*100)

# Here's how to calculate the accuracy for the subset of tokens where 
# the talker distinguishes the two words but the listener does not:
YN=subset(dat, talkerDistinguishes=="yes" & listenerDistinguishes=="no")
round(mean(YN$correct)*100)

# Here's how to calculate how many stimuli were identified as <hod> and 
# how many as <hawed>:
summary(dat$response)
# hawed   hod 
#   354   406

# Here's how to figure out which kind of error was more likely when 
# both the listener and the talker distinguish the two words.  

# First count the cases where the talker intended <hawed> but was heard 
# as saying <hod>:
dim(subset(YY, correct==0 & stimulus=="hawed" & response=="hod"))[1]
# and then count the cases where the talker said <hod> but was heard as 
# saying <hawed>:
dim(subset(YY, correct==0 & stimulus=="hod" & response=="hawed"))[1]

# Here's how to figure out which kind of error was more likely when 
# the listener distinguishes the two words but the talker does not.  
NY=subset(dat, talkerDistinguishes=="no" & listenerDistinguishes=="yes")
dim(subset(NY, correct==0 & stimulus=="hawed" & response=="hod"))[1]
dim(subset(NY, correct==0 & stimulus=="hod" & response=="hawed"))[1]

# Here's how to figure out which word the listener was more likely to 
# choose if the listener does not distinguish the two words.
summary(subset(dat, listenerDistinguishes=="no")$response)

# Here's how to figure out how accurate a specific classmate (such 
# as Alexa, in this case) was at identifying the intended word produced 
# by classmates who do and do not make the distinction.
Alexa=subset(dat, listener=="Alexa")
round(mean(subset(Alexa, talkerDistinguishes=="yes")$correct)*100)
round(mean(subset(Alexa, talkerDistinguishes=="no")$correct)*100)

# Here's how to figure out how accurate everyone was in identifying 
# a specific classmate's productions, illustrated here with Elyse as 
# the designated talker.
round(mean(subset(dat, listener=="Elyse")$correct)*100)

# Here's how to figure out how accurate people were at identifying 
# their own productions.
self=subset(dat, as.character(talker)==as.character(listener))
round(mean(self$correct)*100)

# Ask me if you want help in understanding the syntax of any of these 
# commands or if you want help writing R commands to answer other 
# questions you might have about the data.