2

I want to filter my dataset to keep cases with observations in a specific column. To illustrate: 

help <- data.frame(deid = c(5, 5, 5, 5, 5, 12, 12, 12, 12, 17, 17, 17),
               score.a = c(NA, 1, 1, 1, NA, NA, NA, NA, NA, NA, 1, NA))

Creates 

   deid score.a
1     5      NA
2     5       1
3     5       1
4     5       1
5     5      NA
6    12      NA
7    12      NA
8    12      NA
9    12      NA
10   17      NA
11   17       1
12   17      NA

And I want to tell dplyr to keep cases that have any observations in score.a, including the NA values. Thus, I want it to return:

  deid score.a
1     5      NA
2     5       1
3     5       1
4     5       1
5     5      NA
6    17      NA
7    17       1
8    17      NA

I ran the code help %>% group_by(deid) %>% filter(score.a > 0) however it pulls out the NAs as well. Thank you for any assistance.

Edit: A similar question was asked here How to remove groups of observation with dplyr::filter() However, in the answer they use the 'all' condition and this requires use of the 'any' condition.

Community
  • 1
  • 1
b222
  • 966
  • 1
  • 9
  • 19
  • 1
    possible duplicate of [How to remove groups of observation with dplyr::filter()](http://stackoverflow.com/questions/24583152/how-to-remove-groups-of-observation-with-dplyrfilter) – shadowtalker Jun 13 '15 at 16:08
  • Suppose if `help$score.a[3] <- -1`, what would be the result? – akrun Jun 13 '15 at 16:08
  • Not sure I'm following your question, @akrun ... sorry. When I run that the 3rd row of score.a is -1 – b222 Jun 13 '15 at 16:17
  • My question is that suppose one of the element in `score.a` for a particular group 'deid' is less than 0. Then would you select that group? – akrun Jun 13 '15 at 16:19
  • Yes. Correct. Essentially any observation in score.a and I want to keep that group. Although with the present data all observations are >= 0 – b222 Jun 13 '15 at 16:20
  • 1
    So, your condition is not `all`, but `any` – akrun Jun 13 '15 at 16:21
  • So, my first solution should work. I added the second solution as a bonus. – akrun Jun 13 '15 at 16:23
  • Yes! That worked... I ran the following code... `help %>% group_by(deid) %>% filter(score.a = any(!is.na(score.a)))` – b222 Jun 13 '15 at 16:23
  • 1
    But, that code is not checking the condition `score.a >0` Check my update – akrun Jun 13 '15 at 16:27
  • Correct. Got it. Thank you @akrun – b222 Jun 13 '15 at 16:27

2 Answers2

5

Try

library(dplyr)
help %>%
      group_by(deid) %>%
      filter(any(score.a >0 & !is.na(score.a)))
#    deid score.a
#1    5      NA
#2    5       1
#3    5       1
#4    5       1
#5    5      NA
#6   17      NA
#7   17       1
#8   17      NA

Or a similar approach with data.table

library(data.table)
setDT(help)[, if(any(score.a>0 & !is.na(score.a))) .SD , deid]
#    deid score.a
#1:    5      NA
#2:    5       1
#3:    5       1
#4:    5       1
#5:    5      NA
#6:   17      NA
#7:   17       1
#8:   17      NA

If the condition is to subset 'deid's with all the values in 'score.a' > 0, then the above code can be modified to,

setDT(help)[,  if(!all(is.na(score.a)) & 
         all(score.a[!is.na(score.a)]>0)) .SD , deid]
#   deid score.a
#1:    5      NA
#2:    5       1
#3:    5       1
#4:    5       1
#5:    5      NA
#6:   17      NA
#7:   17       1
#8:   17      NA

Suppose one of the 'score.a' in 'deid' group is less than 0, 

help$score.a[3] <- -1

the above code would return

 setDT(help)[,  if(!all(is.na(score.a)) & 
           all(score.a[!is.na(score.a)]>0, deid],
 #   deid score.a
 #1:   17      NA
 #2:   17       1
 #3:   17      NA
akrun
  • 874,273
  • 37
  • 540
  • 662
2
library(dplyr)
df%>%group_by(deid)%>%filter(sum(score.a,na.rm=T)>0)
Shenglin Chen
  • 4,504
  • 11
  • 11