PCA with NAs produces mismatch between subset and data set

Question

I'm trying to PCA some columns in larger data set which contains NAs. When I remove the NAs it produces a mismatch in the number of items so I cannot use the data set for Label info. How do I fix this?

> ef <- sepData[c(4, 5, 6, 7, 8, 9, 10)]
> autoplot(prcomp(na.omit(ef)), data = sepData, colour = 'species', label = TRUE, label.size = 3)
Error in data.frame(..., check.names = FALSE) : 
arguments imply differing number of rows: 27, 24

sepData contains the sample names on each row. When I remove omit NAs I lose the order for some columns.

Related posts: [here](http://stackoverflow.com/questions/12078291/) and [here](http://stats.stackexchange.com/questions/35561) — zx8754, Oct 27 '16 at 19:35

score 0 · Answer 1 · answered Oct 27 '16 at 19:43

0

One very quick and easy way to get around this is by replacing missing values with column (variable) medians. The zoo package has a nice function na.aggregate() for this. Let's say your matrix is called mat.

library(zoo)
na.aggregate(mat, FUN = median)

Of course, you should really find out why your values are missing in order to impute them most appropriately, especially if there are many of them.

answered Oct 27 '16 at 19:43

Joe

8,073
1
52
58

This wouldn't bias a PCA? Heh, they're missing because the tissue was destroyed during dissection. :D – user974887 Oct 28 '16 at 14:49

PCA with NAs produces mismatch between subset and data set

1 Answers1