Hi:
This is an abridged version of the reply I sent privately to the OP.
#### Generate an artificial data frame
# function to randomly generate one of the Q* columns with length 1000
mysamp <- function() sample(c(-1, 0, 1, NA), 1000, prob = c(0.35, 0.2, 0.4,
0.05), replace = TRUE)
# use above function to randomly generate 10 questions and assign them names
in the workspace
for(i in 1:10) assign(paste('Q', i, sep = ''), mysamp())
# create a data frame from the generate questions
C <- data.frame(time = rep(1:4, each = 250),
sector = sample(LETTERS[1:6], 1000, replace = TRUE),
Q1, Q2, Q3, Q4, Q5, Q6, Q7, Q8, Q9, Q10)
####
# A function to generate the scores from the combined questions
# for an arbitrary input data frame d:
scorefun <- function(d) {
dm <- matrix(unlist(apply(d, 2, table)[-(1:2)]), nrow = 3)
tsums <- cbind(rowSums(dm[, 1:3]), dm[, 4],
rowSums(dm[, 5:6]), rowSums(dm[, 7:8]),
rowSums(dm[, 9:10]) )
dprop <- function(x) (x[3] - x[1])/sum(x)
100 * (1 + apply(tsums, 2, dprop))
}
library(plyr)
# Apply scorefun() to each sub-data frame corresponding to time-sector
combinations
ddply(C, .(time, sector), scorefun)
Dennis
On Sat, Jan 8, 2011 at 10:19 PM, Kari Manninen <kari at econadvisor.com> wrote:
This is my first post to R-help and I look forward receiving some advice
for a novice like me...
I?ve got a simple repeated (4 periods so far) 10-question survey data that
is very easy to work on Excel. However, I?d like to move the compilation to
R but I?m having some trouble operating on count list data in a neat way.
The data C
'data.frame': 551 obs. of 13 variables:
$ TIME : int 1 1 1 1 1 1 1 1 1 1 ...
$ Sector : Factor w/ 6 levels "D","F","G","H",..: 1 1 1 1 1 1 1 1 1 1 ...
$ COMP : Factor w/ 196 levels " (_____ __ _____) ",..: 73 133 128 109
153 147 56 26 142 34 ...
$ Q1 : int 0 0 1 1 0 -1 -1 1 1 -1 ...
$ Q2 : int 0 0 0 -1 0 -1 0 0 1 -1 ...
$ Q3 : int 0 0 0 1 0 -1 -1 1 1 -1 ...
$ Q4 : int -1 0 0 0 0 -1 0 -1 0 -1 ...
$ Q5 : int 0 0 0 -1 0 -1 0 -1 0 0 ...
$ Q6 : int 0 0 0 1 0 -1 0 -1 0 0 ...
$ Q7 : int 0 1 1 0 0 0 1 0 1 1 ...
$ Q8 : int 0 0 0 0 0 -1 0 0 1 0 ...
$ Q9 : int 0 1 0 0 0 -1 0 -1 1 -1 ...
$ Q10 : int 0 0 0 0 -1 -1 0 -1 0 0 ...
summary(C)
TIME Sector COMP Q1 Q2
Min. :1.000 D:130 A: 4 Min. :-1.000 Min. :-1.0000
1st Qu.:2.000 F:126 B: 4 1st Qu.: 0.000 1st Qu.: 0.0000
Median :3.000 G:158 C: 4 Median : 1.000 Median : 0.0000
Mean :2.684 H: 26 D: 4 Mean : 0.446 Mean : 0.2178
3rd Qu.:4.000 I: 20 E: 4 3rd Qu.: 1.000 3rd Qu.: 1.0000
Max. :4.000 J: 91 F: 4 Max. : 1.000 Max. : 1.0000
(Other):527 NA's :60.000 NA's :69.0000
The aim is to produce balance scores between positive and negative answers?
shares in the data. First counts of -1, 0 and 1 (negative, neutral,
positive) and missing NA (it would be som much simple without the missing
values) for each question Q1-Q10 for each period (TIME) in 6 Sectors:
b<-apply(C[,4:13], 2, function (x) tapply(x,C[,1:2], count))
I know that b is a list of data.frames dim(4x6) for each question, where
each ?cell? is a count list.
For example, for Question 1, Time period 2, Sector 1:
List of 1
$ :?data.frame?: 4 obs. of 2 variables:
..$ x : int [1:4] -1 0 1 NA
..$ freq : int [1:4] 3 9 12 2
Now I would like to group questions (C[, 4:6], C[, 7], C[8:9], C[10:11]
and C[, 12:13]) and sum counts (-1, 0, 1) for these groups and present
them in percentage terms. I don?t know how to this efficiently for the whole
data. I would not like to go through each cell separately
Then I?d give each group a balance score based on something like:
Score = 100 + 100*[ pos% - neg%] for each group by TIME, Sector, while
excluding the missing observations.
### This is not working
Score <- 100 + 100*[sum(count( =="1")/sum(count(list( "-1", "0","1") -
sum(count( =="-1")/sum(count(list( "-1", "0","1")] for each 5 groups
defined above and by TIME, Sector
I would greatly appreciate your help on this.
Regards,
- Kari Manninen