Turn Rank Ordering Into Numerical Scores By Transposing A Data Frame
The big difference between the data sets is that many of your rows (16) have all missing values. None of mine do. If you run my data and yours, you will see that decast throws a warning "Aggregation function missing: defaulting to length" with your data but not with mine. As a result, instead of using the value of rank, dcast uses length(rank) which is always 1 except when there are multiple missing values when it is the number of missing values. This problem will occur whenever there is more than one missing value on a row. The simplest way to handle this is to create a function that returns the first value of a vector and use that with the fun.aggregate= argument:
first <- function(x) {x[1]}
d4<- dcast(d3, row~color, fun.aggregate=first, value.var="rank", fill=0)
The only drawback is that this will not warn you if a category was ranked twice except that the NA column will be zero and one of the other columns will be zero. The number of missing values is the number of zeroes in your category columns (not including row or NA) and the value in NA is the lowest rank that was missing.
David C
-----Original Message-----
From: Simon Kiss [mailto:sjkiss at gmail.com]
Sent: Friday, September 5, 2014 10:22 AM
To: David L Carlson
Cc: r-help at r-project.org
Subject: Re: [R] Turn Rank Ordering Into Numerical Scores By Transposing A Data Frame
HI, of course.
The a mini-version of my data-set is below, stored in d2. Then the code I'm working follows.
library(reshape2)
#Create d2
structure(list(row = 1:50, rank1 = structure(c(3L, 3L, 3L, 4L,
3L, 3L, NA, NA, 3L, NA, 3L, 3L, 1L, NA, 2L, NA, 3L, NA, 2L, 1L,
1L, 3L, NA, 6L, NA, 1L, NA, 3L, 1L, NA, 1L, NA, NA, 6L, 3L, NA,
1L, 3L, 3L, 4L, 1L, NA, 3L, 3L, 3L, NA, 3L, 3L, NA, 1L), .Label = c("accessible",
"alternatives", "information", "responsive", "social", "technical",
"trade"), class = "factor"), rank2 = structure(c(6L, 1L, 1L,
2L, 4L, 6L, NA, NA, 6L, NA, 6L, 4L, 2L, NA, 4L, NA, 6L, NA, 1L,
6L, 3L, 2L, NA, 3L, NA, 6L, NA, 6L, 6L, NA, 3L, NA, NA, 3L, 6L,
NA, 6L, 6L, 6L, 7L, 3L, NA, 1L, 6L, 6L, NA, 2L, 6L, NA, 2L), .Label = c("accessible",
"alternatives", "information", "responsive", "social", "technical",
"trade"), class = "factor"), rank3 = structure(c(1L, 6L, 4L,
3L, 2L, 4L, NA, NA, 4L, NA, 1L, 1L, 6L, NA, 1L, NA, 1L, NA, 7L,
3L, 6L, 1L, NA, 2L, NA, 4L, NA, 1L, 3L, NA, 6L, NA, NA, 4L, 2L,
NA, 7L, 1L, 1L, 6L, 7L, NA, 6L, 1L, 1L, NA, 4L, 1L, NA, 3L), .Label = c("accessible",
"alternatives", "information", "responsive", "social", "technical",
"trade"), class = "factor"), rank4 = structure(c(7L, 4L, 2L,
1L, 1L, 7L, NA, NA, 1L, NA, 7L, 2L, 7L, NA, 3L, NA, 2L, NA, 3L,
4L, 5L, 6L, NA, 4L, NA, 3L, NA, 4L, 4L, NA, 4L, NA, NA, 2L, 7L,
NA, 2L, 2L, 2L, 3L, 6L, NA, 2L, 5L, 4L, NA, 1L, 2L, NA, 4L), .Label = c("accessible",
"alternatives", "information", "responsive", "social", "technical",
"trade"), class = "factor"), rank5 = structure(c(2L, 7L, 6L,
7L, 7L, 2L, NA, NA, 2L, NA, 2L, 7L, 3L, NA, 6L, NA, 7L, NA, 6L,
7L, 4L, 7L, NA, 7L, NA, 7L, NA, 2L, 2L, NA, 2L, NA, NA, 7L, 1L,
NA, 3L, 7L, 4L, 2L, 2L, NA, 4L, 2L, 2L, NA, 6L, 4L, NA, 5L), .Label = c("accessible",
"alternatives", "information", "responsive", "social", "technical",
"trade"), class = "factor"), rank6 = structure(c(4L, 2L, 7L,
6L, 6L, 1L, NA, NA, 7L, NA, 4L, 5L, 4L, NA, 7L, NA, 4L, NA, 4L,
2L, 2L, 4L, NA, 1L, NA, 2L, NA, 7L, 7L, NA, 7L, NA, NA, 1L, 4L,
NA, 4L, 4L, 7L, 1L, 4L, NA, 7L, 7L, 7L, NA, 7L, 7L, NA, 7L), .Label = c("accessible",
"alternatives", "information", "responsive", "social", "technical",
"trade"), class = "factor"), rank7 = structure(c(5L, 5L, 5L,
5L, 5L, 5L, NA, NA, 5L, NA, 5L, 6L, 5L, NA, 5L, NA, 5L, NA, 5L,
5L, 7L, 5L, NA, 5L, NA, 5L, NA, 5L, 5L, NA, 5L, NA, NA, 5L, 5L,
NA, 5L, NA, 5L, 5L, 5L, NA, 5L, 4L, 5L, NA, 5L, 5L, NA, 6L), .Label = c("accessible",
"alternatives", "information", "responsive", "social", "technical",
"trade"), class = "factor")), .Names = c("row", "rank1", "rank2",
"rank3", "rank4", "rank5", "rank6", "rank7"), row.names = c(NA,
50L), class = "data.frame")
#This code is a replication of David Carlson's code (below) which works splendidly, but does not work on my data-set
#Melt d2: Note, I've used value.name='color' to maximize comparability with David's suggestion
d3 <- melt(d2, id.vars=1, measure.vars=2:8, variable.name="rank",value.name="color")
#Make Rank Variable Numeric
d3$rank<-as.numeric(d3$rank)
#Recast d3 into d4
d4<- dcast(d3, row~color,value.var="rank", fill=0)
#Note that d4 appears to provide a binary variable for one if a respondent checked the option, but does not provide information as to which rank they assigned each option, but also seems to summarize the number of missing values
#David Carlson's Code
mydf <- data.frame(t(replicate(100, sample(c("red", "blue", "green", "yellow", NA), 4))))
mydf <- data.frame(rows=1:100, mydf)
colnames(mydf) <- c("row", "rank1", "rank2", "rank3", "rank4")
mymelt <- melt(mydf, id.vars=1, measure.vars=2:5, variable.name="rank", value.name="color")
mymelt$rank <- as.numeric(mymelt$rank)
mycast <- dcast(mymelt, row~color, value.var="rank", fill=0)
#Compare
str(mydf)
str(d2)
head(mycast)
head(d4)
Again, I'm grateful for assistance. I can't understand what how my data-set differs from David's sample data-set.
Simon Kiss
On Sep 4, 2014, at 2:35 PM, David L Carlson <dcarlson at tamu.edu> wrote:
I think we would need enough of the data you are using to figure out how to modify the process. Can you use dput() to send a small data set that fails to work? David C -----Original Message----- From: Simon Kiss [mailto:sjkiss at gmail.com] Sent: Thursday, September 4, 2014 1:28 PM To: David L Carlson Cc: r-help at r-project.org Subject: Re: [R] Turn Rank Ordering Into Numerical Scores By Transposing A Data Frame Hi David and list: This is working, except at this command mycast <- dcast(mymelt, row~color, value.var="rank", fill=0) dcast is using "length" as the default aggregating function. This results in not accurate results. It tells me, for example how many choices were missing values and it tells me if a person selected any given option (value is reported as 1). When I try to run your reproducible research, it works great, but something with the aggregating function is not working properly with mine. Any other thoughts? Simon On Aug 18, 2014, at 10:44 AM, David L Carlson <dcarlson at tamu.edu> wrote:
Another approach using reshape2:
library(reshape2)
# Construct data/ add column of row numbers
set.seed(42)
mydf <- data.frame(t(replicate(100, sample(c("red", "blue",
+ "green", "yellow", NA), 4))))
mydf <- data.frame(rows=1:100, mydf)
colnames(mydf) <- c("row", "rank1", "rank2", "rank3", "rank4")
head(mydf)
row rank1 rank2 rank3 rank4 1 1 <NA> yellow red blue 2 2 yellow green <NA> red 3 3 yellow green blue <NA> 4 4 <NA> blue yellow green 5 5 <NA> red blue green 6 6 <NA> red green blue
# Reshape mymelt <- melt(mydf, id.vars=1, measure.vars=2:5,
+ variable.name="rank", value.name="color")
# Convert rank to numeric mymelt$rank <- as.numeric(mymelt$rank) mycast <- dcast(mymelt, row~color, value.var="rank", fill=0) head(mycast)
row blue green red yellow NA 1 1 4 0 3 2 1 2 2 0 2 4 1 3 3 3 3 2 0 1 4 4 4 2 4 0 3 1 5 5 3 4 2 0 1 6 6 4 3 2 0 1 David C -----Original Message----- From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On Behalf Of David L Carlson Sent: Sunday, August 17, 2014 6:32 PM To: Simon Kiss; r-help at r-project.org Subject: Re: [R] Turn Rank Ordering Into Numerical Scores By Transposing A Data Frame There is probably an easier way to do this, but
set.seed(42)
mydf <- data.frame(t(replicate(100, sample(c("red", "blue",
+ "green", "yellow", NA), 4))))
colnames(mydf) <- c("rank1", "rank2", "rank3", "rank4")
head(mydf)
rank1 rank2 rank3 rank4 1 <NA> yellow red blue 2 yellow green <NA> red 3 yellow green blue <NA> 4 <NA> blue yellow green 5 <NA> red blue green 6 <NA> red green blue
lvls <- levels(mydf$rank1) # convert color factors to numeric for (i in seq_along(mydf)) mydf[,i] <- as.numeric(mydf[,i]) # stack the columns mydf2 <- stack(mydf) # convert rank factor to numeric mydf2$ind <- as.numeric(mydf2$ind) # add row numbers mydf2 <- data.frame(rows=1:100, mydf2) # Create table mytbl <- xtabs(ind~rows+values, mydf2) # convert to data frame mydf3 <- data.frame(unclass(mytbl)) colnames(mydf3) <- lvls head(mydf3)
blue green red yellow 1 4 0 3 2 2 0 2 4 1 3 3 2 0 1 4 2 4 0 3 5 3 4 2 0 6 4 3 2 0 David C -----Original Message----- From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On Behalf Of Simon Kiss Sent: Friday, August 15, 2014 3:58 PM To: r-help at r-project.org Subject: Re: [R] Turn Rank Ordering Into Numerical Scores By Transposing A Data Frame Both the suggestions I got work very well, but what I didn't realize is that NA values would cause serious problems. Where there is a missing value, using the argument na.last=NA to order just returns the the order of the factor levels, but excludes the missing values, but I have no idea where those occur in the or rather which of those variables were actually missing. Have I explained this problem sufficiently? I didn't think it would cause such a problem so I didn't include it in the original problem definition. Yours, Simon On Jul 25, 2014, at 4:58 PM, David L Carlson <dcarlson at tamu.edu> wrote:
I think this gets what you want. But your data are not reproducible since they are randomly drawn without setting a seed and the two data sets have no relationship to one another.
set.seed(42)
mydf <- data.frame(t(replicate(100, sample(c("red", "blue",
+ "green", "yellow")))))
colnames(mydf) <- c("rank1", "rank2", "rank3", "rank4")
mydf2 <- data.frame(t(apply(mydf, 1, order)))
colnames(mydf2) <- levels(mydf$rank1)
head(mydf)
rank1 rank2 rank3 rank4 1 yellow green red blue 2 green blue yellow red 3 green yellow red blue 4 yellow red green blue 5 yellow red green blue 6 yellow red blue green
head(mydf2)
blue green red yellow
1 4 2 3 1
2 2 1 4 3
3 4 1 3 2
4 4 3 2 1
5 4 3 2 1
6 3 4 2 1
-------------------------------------
David L Carlson
Department of Anthropology
Texas A&M University
College Station, TX 77840-4352
-----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On Behalf Of Simon Kiss
Sent: Friday, July 25, 2014 2:34 PM
To: r-help at r-project.org
Subject: [R] Turn Rank Ordering Into Numerical Scores By Transposing A Data Frame
Hello:
I have data that looks like mydf, below. It is the results of a survey where participants were to put a number of statements (in this case colours) in their order of preference. In this case, the rank number is the variable, and the factor level for each respondent is which colour they assigned to that rank. I would like to find a way to effectively transpose the data frame so that it looks like mydf2, also below, where the colours the participants were able to choose are the variables and the variable score is what that person ranked that variable.
Ultimately what I would like to do is a factor analysis on these items, so I'd like to be able to see if people ranked red and yellow higher together but ranked green and blue together lower, that sort of thing.
I have played around with different variations of t(), melt(), ifelse() and if() but can't find a solution.
Thank you
Simon
#Reproducible code
mydf<-data.frame(rank1=sample(c('red', 'blue', 'green', 'yellow'), replace=TRUE, size=100), rank2=sample(c('red', 'blue', 'green', 'yellow'), replace=TRUE, size=100), rank3=sample(c('red', 'blue', 'green', 'yellow'), replace=TRUE, size=100), rank4=sample(c('red', 'blue', 'green', 'yellow'), replace=TRUE, size=100))
mydf2<-data.frame(red=sample(c(1,2,3,4), replace=TRUE,size=100),blue=sample(c(1,2,3,4), replace=TRUE,size=100),green=sample(c(1,2,3,4), replace=TRUE,size=100) ,yellow=sample(c(1,2,3,4), replace=TRUE,size=100))
*********************************
Simon J. Kiss, PhD
Assistant Professor, Wilfrid Laurier University
73 George Street
Brantford, Ontario, Canada
N3T 2C9
______________________________________________ R-help at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
********************************* Simon J. Kiss, PhD Assistant Professor, Wilfrid Laurier University 73 George Street Brantford, Ontario, Canada N3T 2C9 Cell: +1 905 746 7606
______________________________________________ R-help at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code. ______________________________________________ R-help at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
********************************* Simon J. Kiss, PhD Assistant Professor, Wilfrid Laurier University 73 George Street Brantford, Ontario, Canada N3T 2C9 Cell: +1 905 746 7606
********************************* Simon J. Kiss, PhD Assistant Professor, Wilfrid Laurier University 73 George Street Brantford, Ontario, Canada N3T 2C9 Cell: +1 905 746 7606