Confidence Interval of Difference of Means between two datasets - r

I'm working on two datasets, derrived fromm cats, an in-build R dataset.
> cats
Sex Bwt Hwt
1 F 2.0 7.0
2 F 2.0 7.4
3 F 2.0 9.5
4 F 2.1 7.2
5 F 2.1 7.3
6 F 2.1 7.6
7 F 2.1 8.1
8 F 2.1 8.2
9 F 2.1 8.3
10 F 2.1 8.5
11 F 2.1 8.7
12 F 2.1 9.8
...
137 M 3.6 13.3
138 M 3.6 14.8
139 M 3.6 15.0
140 M 3.7 11.0
141 M 3.8 14.8
142 M 3.8 16.8
143 M 3.9 14.4
144 M 3.9 20.5
I want to find the 99% Confidence Interval on the difference of means values between the Bwt of Male and Female specimens (Sex == M and Sex == F respectively)
I know that t.test does this, among other things, but if I break up cats to two datasets that contain the Bwt of Males and Females, t.test() complains that the two datasets are not of the same length, which is true. There's only 47 Females in cats, and 87 Males.
Is it doable some other way or am I misinterpreting data by breaking them up?
EDIT:
I have a function suggested to me by an Answerer on another Question that gets the CI of means on a dataset, may come in handy:
ci_func <- function(data, ALPHA){
c(
mean(data) - qnorm(1-ALPHA/2) * sd(data)/sqrt(length(data)),
mean(data) + qnorm(1-ALPHA/2) * sd(data)/sqrt(length(data))
)
}

You should apply the t.test with the formula interface:
t.test(Bwt ~ Sex, data=cats, conf.level=.99)

Alternatively to t.test, if you really only interested in the difference of means, you can use:
DescTools::MeanDiffCI(cats$Bwt, cats$Sex)
which gives something like
meandiff lwr.ci upr.ci
-23.71474 -71.30611 23.87662
This is calculated with 999 bootstrapped samples by default. If you want more, you can specify this in the R parameter:
DescTools::MeanDiffCI(cats$Bwt, cats$Sex, R = 1000)

Related

Loop or function over several columns in R [duplicate]

This question already has answers here:
Aggregate / summarize multiple variables per group (e.g. sum, mean)
(10 answers)
Multiple functions in aggregate
(5 answers)
Closed 2 months ago.
May I ask for help with R coding? I would like to calculate the simple mean and standard deviation as shown below. This is an example of calculating it for species richness. However, I would like to calculate it for several species, each one is in a separate column (species1, species2, species3, species4 etc).
How can I do it automatically (I guess using some loop or funcion) in R and get a nice overview table where calculations for each species come one by one in below?
mean1=tapply(edge2$species_richness, list (Management =edge2$Management), mean) sd1=tapply(edge2$species_richness, list (Management =edge2$Management), sd)
cbind (mean1, sd1)
Result for species richness:
mean1 sd1
AES 15.6250 5.875089
AES2 29.5000 9.570789
Control 6.9375 8.590450
Centre 16.3125 5.437141
I ask help with R coding
Using aggregate on mtcars (the aggregation variable is in the rows, as is yours, the aggregated variables are in the columns)
> aggregate(.~am,data=mtcars,function(x){c("mean"=mean(x),"sd"=sd(x))})
am mpg.mean mpg.sd cyl.mean cyl.sd disp.mean disp.sd hp.mean hp.sd drat.mean drat.sd wt.mean
1 0 17.1 3.8 6.9 1.5 290.4 110.2 160.3 53.9 3.3 0.4 3.8
2 1 24.4 6.2 5.1 1.6 143.5 87.2 126.8 84.1 4.0 0.4 2.4
wt.sd qsec.mean qsec.sd vs.mean vs.sd gear.mean gear.sd carb.mean carb.sd
1 0.8 18.2 1.8 0.4 0.5 3.2 0.4 2.7 1.1
2 0.6 17.4 1.8 0.5 0.5 4.4 0.5 2.9 2.2

Storing output from R multiple loops into a list

I'm trying to carry out the following action on the columns of a dataframe (df1):
term1+term2+term3*req_no
req_no is a range of numbers: 20:24
df1:
ID term1 term2 term3
X299 1.2 2.3 0.12
X300 1.4 0.6 2.4
X301 0.3 1.6 1.2
X302 0.9 0.6 0.4
X303 0.3 1.8 0.3
X304 1.3 0.3 2.1
I need help t get this output and here's my attempt:
Required output:
ID 20 21 22 23 24
X299 5.9 6.02 6.14 6.26 6.38
X300 50 52.4 54.8 57.2 59.6
X301 25.9 27.1 28.3 29.5 30.7
X302 9.5 9.9 10.3 10.7 11.1
X303 8.1 8.4 8.7 9 9.3
X304 43.6 45.7 47.8 49.9 52
Here's:
results <- list()
req_no <- 20:25
for(i in 1:nrow(df1){
for(j in rq_no){
res <- term1+term2+term3*j
results[j] <- res
}
results[[i]]
}
results2 <- do.call("rbind",result)
Help will be appreciated.
Here are a couple different approaches, though neither as succinct as Parfait's. Sample data:
df <- data.frame(ID=c("X299", "X300"),
term1=c(1.2, 1.4),
term2=c(2.3, 0.6),
term3=c(0.12, 2.4))
req_no <- 20:25
Loop approach
Your initial approach is headed in the right direction, but in the future, it would help to specify exactly what your error or problem is. For an iterated and perhaps easier-to-read approach, here's one answer:
results <- matrix(data=NA, nrow=nrow(df), ncol=length(req_no)) # Empty matrix to store our results
colnames(results) <- req_no # Optional; name columns based off of req_no values
for(i in 1:nrow(df)) {
# Do the calculation we want; returns a vector length 6
res <- df[i,]$term1 + df[i,]$term2 + (df[i,]$term3 * req_no)
# Save results for row i of df into row i of results matrix
results[i,] <- res
}
# Now bind the columns (named 20 through 25) to the respective rows of df
output <- cbind(df, results)
output
From your initial attempt, note:
We only do one loop, since it is easy to multiply by a vector in R
There are a few ways to subset data from a data frame in R. In this case, df[i,] gets everything in the i-th row, while $termX gets value in the column named termX
Using a results matrix instead of a list makes it very easy to copy the temporary computations (for each row) into rows of the matrix
Rather than rbind() (row bind), we want cbind() (column bind) to bind those results to new columns of the original rows.
Output:
ID term1 term2 term3 20 21 22 23 24 25
1 X299 1.2 2.3 0.12 5.9 6.02 6.14 6.26 6.38 6.5
2 X300 1.4 0.6 2.40 50.0 52.40 54.80 57.20 59.60 62.0
Dplyr/purrr functions
This could also be solved using tidy functions. In essence it's a pretty similar approach to Parfait's answer, but I've made the steps a bit more verbose to see what's going on.
# Use purrr's map functions to do the computation we want
nested_df <- df %>%
# Make new column holding term3 * req_no (stores a vector in each new cell)
mutate(term3r = map(term3, ~ .x * req_no)) %>%
# Make new column which sums the three columns of interest (stores a vector in each new cell)
mutate(sum = pmap(list(term1, term2, term3r), ~ ..1 + ..2 + ..3))
# "Unnest" those vectors which store our sums, and keep only those and ID
output <- nested_df %>%
# Creates six new columns (named ...1 to ...6) with the elements of each sum
unnest_wider(sum) %>%
# Keeps only the output data and IDs
select(ID, ...1:...6)
output
Output:
# A tibble: 2 x 7
ID ...1 ...2 ...3 ...4 ...5 ...6
<chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 X299 5.9 6.02 6.14 6.26 6.38 6.5
2 X300 50 52.4 54.8 57.2 59.6 62
Consider directly assigning new columns with sapply using your formula:
df[paste0(req_no)] <- sapply(req_no, function(r) with(df, term1 + term2 + term3 * r))
df
# ID term1 term2 term3 20 21 22 23 24
# 1 X299 1.2 2.3 0.12 5.9 6.02 6.14 6.26 6.38
# 2 X300 1.4 0.6 2.40 50.0 52.40 54.80 57.20 59.60
# 3 X301 0.3 1.6 1.20 25.9 27.10 28.30 29.50 30.70
# 4 X302 0.9 0.6 0.40 9.5 9.90 10.30 10.70 11.10
# 5 X303 0.3 1.8 0.30 8.1 8.40 8.70 9.00 9.30
# 6 X304 1.3 0.3 2.10 43.6 45.70 47.80 49.90 52.00

R - Create (Mutate) a new column as a function of past observations

Ok so i have a pretty large data set of around 500 observations and 3 variables. The first column refers to time.
For a test data set I am using:
dat=as.data.frame(matrix(c(1,2,3,4,5,6,7,8,9,10,
1,1.8,3.5,3.8,5.6,6.2,7.8,8.2,9.8,10.1,
2,4.8,6.5,8.8,10.6,12.2,14.8,16.2,18.8,20.1),10,3))
colnames(dat)=c("Time","Var1","Var2")
Time Var1 Var2
1 1 1.0 2.0
2 2 1.8 4.8
3 3 3.5 6.5
4 4 3.8 8.8
5 5 5.6 10.6
6 6 6.2 12.2
7 7 7.8 14.8
8 8 8.2 16.2
9 9 9.8 18.8
10 10 10.1 20.1
So what I need to do is create a new column that each observations is the slope respect to time of some past points. For example taking 3 past points it would be something like:
slopeVar1[i]=slope(Var1[i-2:i],Time[i-2:i]) #Not real code
slopeVar[i]=slope(Var2[i-2:i],Time[i-2:i]) #Not real code
Time Var1 Var2 slopeVar1 slopeVar2
1 1 1 2 NA NA
2 2 1.8 4.8 NA NA
3 3 3.5 6.5 1.25 2.25
4 4 3.8 8.8 1.00 2.00
5 5 5.6 10.6 1.05 2.05
6 6 6.2 12.2 1.20 1.70
7 7 7.8 14.8 1.10 2.10
8 8 8.2 16.2 1.00 2.00
9 9 9.8 18.8 1.00 2.00
10 10 10.1 20.1 0.95 1.95
I actually got as far as using a for() function, but for really large data sets (>100,000) it starts taking too long.
The for() argument that I used is shown bellow:
#CREATE DATA FRAME
rm(dat)
dat=as.data.frame(matrix(c(1,2,3,4,5,6,7,8,9,10,
1,1.8,3.333,3.8,5.6,6.2,7.8,8.2,9.8,10.1,
2,4.8,6.5,8.8,10.6,12.2,14.8,16.2,18.8,20.1),10,3))
colnames(dat)=c("Time","Var1","Var2")
dat
plot(dat)
#CALCULATE SLOPE OF n POINTS FROM i TO i-n.
#In this case I am taking just 3 points, but it should
#be possible to change the number of points taken.
attach(dat)
n=3 #number for points to take slope
l=dim(dat[1])[1] #number of iterations
y=0
x=0
slopeVar1=NA
slopeVar2=NA
for (i in 1:l) {
if (i<n) {slopeVar1[i]=NA} #For the rows where there are not enough previous observations, it outputs NA
if (i>=n) {
y1=Var1[(i-n+1):i] #y data sets for calculating slope of Var1
y2=Var2[(i-n+1):i]#y data sets for calculating slope of Var2
x=Time[(i-n+1):i] #x data sets for calculating slope of Var1&Var2
z1=lm(y1~x) #Temporal value of slope of Var1
z2=lm(y2~x) #Temporal value of slope of Var2
slope1=as.data.frame(z1[1]) #Temporal value of slope of Var1
slopeVar1[i]=slope1[2,1] #Populating string of slopeVar1
slope2=as.data.frame(z2[1])#Temporal value of slope of Var2
slopeVar2[i]=slope2[2,1] #Populating string of slopeVar2
}
}
slopeVar1 #Checking results.
slopeVar2
(result=cbind(dat,slopeVar1,slopeVar2)) #Binds original data with new calculated slopes.
This code actually outputs what I want; but again, for really large data sets is quite inefficient.
This quick rollapply implemenation seems to be speeding it up somewhat -
library("zoo")
slope_func = function(period) {
y1=period[,2] #y data sets for calculating slope of Var1
y2=period[,3] #y data sets for calculating slope of Var2
x=period[,1] #x data sets for calculating slope of Var1&Var2
z1=lm(y1~x) #Temporal value of slope of Var1
z2=lm(y2~x) #Temporal value of slope of Var2
slope1=as.data.frame(z1[1]) #Temporal value of slope of Var1
slopeVar1[i]=slope1[2,1] #Populating string of slopeVar1
slope2=as.data.frame(z1[1])#Temporal value of slope of Var2
slopeVar2[i]=slope2[2,1] #Populating string of slopeVar2
}
}
start = Sys.time()
rollapply(dat[1:3], FUN=slope_func, width=3, by.column=FALSE)
end=Sys.time()
print(end-start)
Time difference of 0.04980111 secs
OP's previous implementation was taking Time difference of 0.2666121 secs for the same

Extracting block of m rows at regular interval from large dataset

I have a small problem. I have a dataset with 8208 rows of data. It's a single column of data, I want to take every n rows as a block and add this to a new data frame.
So, for example:
newdf has column 1 to column 23.
column 1 is composed of rows 289:528 from the original dataset
column 2 is composed of rows 625:864 from the original dataset
And so on. The "block" size is 239 rows, the jump between blocks is every 336 rows.
I can do this manually, but it just becomes tedious. I have to repeat this entire procedure for another 11 sets of data so obviously a more automated approach would be preferable.
The trick here is to create an index of integers that refer to the row numbers you want to keep. This is simple enough with some use of rep, sequences and R's recycling rule.
Let me demonstrate using iris. Say you want to skip 25 rows, then return 3 rows:
skip <- 25
take <- 3
total <- nrow(iris)
reps <- total %/% (skip + take)
index <- rep(0:(reps-1), each=take) * (skip + take) + (1:take) + skip
The index now is:
index
[1] 26 27 28 54 55 56 82 83 84 110 111 112 138 139 140
And the rows of iris:
iris[index, ]
Sepal.Length Sepal.Width Petal.Length Petal.Width Species
26 5.0 3.0 1.6 0.2 setosa
27 5.0 3.4 1.6 0.4 setosa
28 5.2 3.5 1.5 0.2 setosa
54 5.5 2.3 4.0 1.3 versicolor
55 6.5 2.8 4.6 1.5 versicolor
56 5.7 2.8 4.5 1.3 versicolor
82 5.5 2.4 3.7 1.0 versicolor
83 5.8 2.7 3.9 1.2 versicolor
84 6.0 2.7 5.1 1.6 versicolor
110 7.2 3.6 6.1 2.5 virginica
111 6.5 3.2 5.1 2.0 virginica
112 6.4 2.7 5.3 1.9 virginica
138 6.4 3.1 5.5 1.8 virginica
139 6.0 3.0 4.8 1.8 virginica
140 6.9 3.1 5.4 2.1 virginica
Update
Note the OP states the block size is 239 elements but it is clear from the examples rows indicated that the block size is 240
> length(289:528)
[1] 240
I'll leave the example below at a block length of 239, but adjust if it is really 240.
It isn't clear from the Question, but assuming that you have something like this
df <- data.frame(A = runif(8208))
a data frame with 8208 rows.
First compute the indices of the elements of A that you need to keep. This is done via
want <- sapply(seq(289, nrow(df)-239, by = 336),
function(x) x + (seq_len(239) - 1))
Then we can use the fact that R fills matrices by columns and convert the required elements of A to a matrix with 239 rows
mat <- matrix(df$A[want], nrow = 239)
This works
> all.equal(mat[,1], df$A[289:527])
[1] TRUE
but do note that I have taken a block length of 239 here (289:527) not the indices the OP quotes as that is a block size of 240 (see Update above)
If you want this is a data frame, just add
df2 <- as.data.frame(mat)
Try this:
1) Create a list of indices
lapply(seq(1, 8208, 336), function(X) X:(X+239)) -> Indices
2) Select Data
Columns <- lapply(Indices, function(X) OldDF[X,])
3) Combine selected data in columns
NewDF <- do.call(cbind, Columns)
Why not just:
as.dataframe(matrix(orig, nrow=528 )[289:528 ,])
Since the 8028 is not an exactl multiple of the row count we need to determine the columns:
> 8208/528
[1] 15.54545 # so either 15 or 16
> 8208-15*528
[1] 288 # all in the to-be-discarded section
as.dataframe(matrix(orig, nrow=528, col=15 )[289:528 ,])
Or:
as.dataframe(matrix(orig, nrow=528, col=8208 %/% 528)[289:528 ,])

Apply formula for between species comparison

I have a data frame laid out in the follwing manner:
Species Trait.p Trait.y Trait.z
a 20.1 7.2 14.1
b 20.4 8.3 15.2
b 19.2 6.8 13.9
I would like to apply, for each species combination, (Xa) - (Xb) where is X is the trait value and the letter is the species and Xa > Xb. I.e has to be such that the larger value of each respective species combination has to come first, calculated for every trait
Would this be a multi-step process?
An example output could be
Combination Trait.p Trait.y Trait.z
a/b 0.3 1.1 1.1
I assumed you choose the largest value but David brings up a good point. I doubt this is the best approach but I think it gives you what you're after. Note I added a c as I'm sure your problem is a bit more complex that just a and b:
dat <- read.table(text="Species Trait.p Trait.y Trait.z
a 20.1 7.2 14.1
b 20.4 8.3 15.2
b 19.2 6.8 13.9
c 14.2 3.8 11.9", header=T)
li <- lapply(split(dat, dat$Species), function(x) apply(x[, -1], 2, max))
com <- expand.grid(names(li), names(li))
inds <- com[com[, 1] != com[, 2], ]
inds <- t(apply(inds, 1, sort))
inds <- inds[!duplicated(inds), ]
ans <- lapply(1:nrow(inds), function(i) {
abs(li[[inds[i, 1]]]-li[[inds[i, 2]]])
})
cbind(Combination = paste(inds[, 1], inds[, 2], sep="/"),
as.data.frame(do.call(rbind, ans)))
This gives us:
Combination Trait.p Trait.y Trait.z
1 a/b 0.3 1.1 1.1
2 a/c 5.9 3.4 2.2
3 b/c 6.2 4.5 3.3
Sorry for the lack of annotation but I'm heading to class.

Resources