Question

Why doesn’t a model matrix necessarily have the same number of rows as the data frame?

mergem = model.matrix(as.formula(paste(response, '~ .')), data=mergef)
dim(mergef)
# [1] 115562     71
dim(mergem)
# [1] 66786   973

I tried looking for hints in the documentation but couldn’t find anything. Thanks in advance.

Answer 1

Well, if a row has NAs in it, that row is (by default) removed:

d <- data.frame(x=c(1,1,2), y=c(2,2,4), z=c(4,NA,8))
m <- model.matrix(x ~ ., data=d)

nrow(d) # 3
nrow(m) # 2

This behavior is controlled by the option “na.action”:

options(na.action="na.fail")
m <- model.matrix(x ~ ., data=d) # Error: missing values in object