How to calculate mean of first 2 and last 2 non-NAs by row in R?

Question

I am calculating split-half reliability for certain behavioral items in my dataset and first need to grab the mean of the first 2 non-NA values per respondent followed by the last two non-NA values for each person (each row). I know there are ways to do this using packages runner , zoo and others by column, but I've yet to find a solution within rows.

For context, I designed a survey in which items were randomized in order to reduce item-level effects. Participants saw 1/2 of a random subset of items from a particular measurement scale at one point in the survey and the other 1/2 at a different point. Therefore, each participant will have the same number of non-NA as NA at each of the two-time points.

for instance, say I have 8 items total. Data for persons 1, 2, and 3 at time point 1 reads:

x1  x2  x3  x4  x5  x6  x7  x8
 1  NA  NA  2   NA  1   1   NA
 NA 4   3   3   NA  NA  4   NA
 3  2   1   NA  NA  NA  3   NA 

The resulting new variables (avg1 and avg2) should read:

x1  x2  x3  x4  x5  x6  x7  x8   avg1 avg2
 1  NA  NA  2   NA  1   1   NA   1.5   1
 NA 4   3   3   NA  NA  4   NA   3.5  3.5
 3  2   1   NA  NA  NA  3   NA   2.5   2

any help is appreciated, thanks!

Answer 1

Here is one potential solution:

m <- as.matrix(read.table(text = "x1  x2  x3  x4  x5  x6  x7  x8
 1  NA  NA  2   NA  1   1   NA
 NA 4   3   3   NA  NA  4   NA
 3  2   1   NA  NA  NA  3   NA ",
 header = TRUE))

# Only keep non-NA values
m2 <- t(apply(m,1,function(x) c(x[!is.na(x)])))

# Select the first two non-NA values
m3 <- m2[,1:2]

# Select the second-last and last non-NA values
m4 <- m2[,(ncol(m2)-1):(ncol(m2))]

# Bind the matrix to the mean of the first two and the mean of the last two non-NA values
cbind(m, "avg1" = rowMeans(m3), "avg2" = rowMeans(m4))
#>      x1 x2 x3 x4 x5 x6 x7 x8 avg1 avg2
#> [1,]  1 NA NA  2 NA  1  1 NA  1.5  1.0
#> [2,] NA  4  3  3 NA NA  4 NA  3.5  3.5
#> [3,]  3  2  1 NA NA NA  3 NA  2.5  2.0

^{Created on 2022-03-11 by the reprex package (v2.0.1)}