Question

I have a data frame and I would like to impute missing values based on row mean instead of column mean.

id       Price1         Price2           Price3        Price4    HorizontalMean
004        NA             101              103           114            106
005       100             108               78            99             96.25
006        34              33               NA            78             48.333
...

I’ve looked at a few packages and can’t seem to find any that explicitly mention it. Any recommendations or do I need to do a transpose first (this could be a problem as the data is >100k lines).

Answer 1

Here is a nice little one-liner:

> df <- data.frame(Price1 = c(NA, 100, 34),
+                  Price2 = c(101, 108, 33),
+                  Price3 = c(103, 78, NA),
+                  Price4 = c(114, 99, 78))
> df
  Price1 Price2 Price3 Price4
1     NA    101    103    114
2    100    108     78     99
3     34     33     NA     78

> df <- ifelse(is.na(df), rowMeans(df, na.rm=TRUE), unlist(df))
> df
     Price1 Price2    Price3 Price4
[1,]    106    101 103.00000    114
[2,]    100    108  78.00000     99
[3,]     34     33  48.33333     78

EDIT: for @Charlie who asked how would you do to replace NAs with column means, you could use the same thing but replace rowMeans(…) with a vector repeating the column means:

df <- ifelse(is.na(df), rep(colMeans(df, na.rm=TRUE), rep(nrow(df), ncol(df))),
             unlist(df))

or apply ifelse to each column of the list:

df <- sapply(df, function(x)ifelse(is.na(x), mean(x, na.rm=TRUE), x))