128

We have a data frame from a CSV file. The data frame DF has columns that contain observed values and a column (VaR2) that contains the date at which a measurement has been taken. If the date was not recorded, the CSV file contains the value NA, for missing data.

Var1  Var2 
10    2010/01/01
20    NA
30    2010/03/01

We would like to use the subset command to define a new data frame new_DF such that it only contains rows that have an NA' value from the column (VaR2). In the example given, only Row 2 will be contained in the new DF.

The command

new_DF<-subset(DF, DF$Var2=="NA")

does not work, the resulting data frame has no row entries.

If in the original CSV file the Value NA are exchanged with NULL, the same command produces the desired result:

new_DF <- subset(DF, DF$Var2=="NULL")

How can I get this method working, if for the character string the value NA is provided in the original CSV file?

Martin Gal
  • 16,640
  • 5
  • 21
  • 39
John
  • 1,401
  • 2
  • 11
  • 12

7 Answers7

188

Never use =='NA' to test for missing values. Use is.na() instead. This should do it:

new_DF <- DF[rowSums(is.na(DF)) > 0,]

or in case you want to check a particular column, you can also use

new_DF <- DF[is.na(DF$Var),]

In case you have NA character values, first run

Df[Df=='NA'] <- NA

to replace them with missing values.

JelenaČuklina
  • 3,574
  • 2
  • 22
  • 35
Joris Meys
  • 106,551
  • 31
  • 221
  • 263
  • 2
    Thanks for your fast answer (this was quick)! Indeed, due to the csv-delivery of the data, the 'NA' are character values and your second statement might be very useful. Can you also clarify your first statement? The use of rowSums() is not clear for me, since I will only check a particular column (there are plenty of columns). If that particular column (in the example it would be column Var2) has there an 'NA' character string (I will replace it with your second statement), then I would like to choose the entire row to be part of the new data frame. – John Nov 02 '11 at 14:29
  • @John : updated. THe point is to use is.na, I wrongly interpreted you wanted to check all variables. – Joris Meys Nov 02 '11 at 21:34
  • 3
    should that be `new_DF <- DF[is.na(DF$Var),]`, i.e. there appears to be an extra `(` bracket after `DF[` ? – PatrickT Jul 24 '13 at 19:01
62

complete.cases gives TRUE when all values in a row are not NA

DF[!complete.cases(DF), ]
user3226167
  • 3,131
  • 2
  • 30
  • 34
42

NA is a special value in R, do not mix up the NA value with the "NA" string. Depending on the way the data was imported, your "NA" and "NULL" cells may be of various type (the default behavior is to convert "NA" strings to NA values, and let "NULL" strings as is).

If using read.table() or read.csv(), you should consider the "na.strings" argument to do clean data import, and always work with real R NA values.

An example, working in both cases "NULL" and "NA" cells :

DF <- read.csv("file.csv", na.strings=c("NA", "NULL"))
new_DF <- subset(DF, is.na(DF$Var2))
maressyl
  • 983
  • 6
  • 11
  • 1
    Thanks for your answer. If I understand it correctly the first statement would do the same as Df[Df=='NA']<-NA in the example of Joris? The (small) difference then would be that it is done in your statment direcly in the beginning, when the data frame is created (this is a very clean programming method and I therefore like it). – John Nov 02 '11 at 14:42
  • Exactly. Joris suggested to replace "NA" strings by NA values manually, here i only suggest to use the "na.strings" feature of read.table() to achieve the same purpose. – maressyl Nov 02 '11 at 15:33
  • Joris' answer is actually the "preferred" way to accomplish this feat (if you are writing this in a script). See: http://stackoverflow.com/questions/9860090/in-r-why-is-better-than-subset – Jonathan Feb 22 '13 at 17:30
  • @Jonathan : Two distinct ideas here, the topic you cite says "[" should be prefered on "subset", but we were talking about the "na.strings" argument in read.table(), my subset was here only to visualize the effects. – maressyl Aug 01 '13 at 08:02
26
new_data <- data %>% filter_all(any_vars(is.na(.)))

This should create a new data frame (new_data) with only the missing values in it.

Works best to keep a track of values that you might later drop because they had some columns with missing observations (NA).

Dmitriy
  • 5,525
  • 12
  • 25
  • 38
Ronak Pol
  • 361
  • 3
  • 3
7

Try changing this:

new_DF<-dplyr::filter(DF,is.na(Var2))
csilk
  • 2,732
  • 3
  • 23
  • 40
drhnis
  • 113
  • 1
  • 2
  • 1
    Could you explain why this works, what this does etc.? – csilk Nov 22 '17 at 00:19
  • new_DF<-dplyr::filter(DF,is.na(Var2)) it basically use the filter function of dplyr package and filter out any observation in Var2 column which satisfy the condition is.na ie they pick all the observation with NA – drhnis Nov 28 '17 at 19:56
  • 1
    More nicely expressed as `DF %>% filter(is.na(Var2))` after `library(dplyr)`. – Joe Feb 22 '18 at 13:15
6

Since dplyrs filter_all has been superseded

Scoped verbs (_if, _at, _all) have been superseded by the use of across() in an existing verb.

and the usage of across() in filter() is deprecated, Ronak Pol's answer needs a small update. To find all rows with an NA anywhere, we could use

library(dplyr)

DF %>% 
  filter(if_any(everything(), is.na))

to get

# A tibble: 1 x 2
   Var1 Var2  
  <dbl> <date>
1    20 NA
Martin Gal
  • 16,640
  • 5
  • 21
  • 39
-1

Prints all the rows with NA data:

tmp <- data.frame(c(1,2,3),c(4,NA,5));
tmp[round(which(is.na(tmp))/ncol(tmp)),]
Manfred Radlwimmer
  • 13,257
  • 13
  • 53
  • 62
jstar
  • 847
  • 9
  • 9
  • @ZheyuanLi If you don't like the answer, simply down-vote it. Editing the answer to recommend flagging is NOT the appropiate action. Leave a comment if you feel the need to. – Manfred Radlwimmer May 29 '16 at 09:25