Aggregation of NAs results in 0s instead of NAs

Question

I am trying to find the sum of three different variables in a data frame while grouping by another variable, but there are several NAs. The sum of the NAs is interpreted as zero instead of NA. Here is an example:

my_data <- data.frame(Month = c("1995-01-01", "1995-01-01", "1995-01-01",
                            "1995-02-01", "1995-02-01"),
                  Value_1 = c(1, NA, 2, NA, NA),
                  Value_2 = c(2, 2, 3, NA, 1),
                  Value_3 = c(NA, NA, NA, NA, NA))

#summing through dplyr
my_data %>%
  group_by(Month) %>%
  summarise_each(funs(sum(.,na.rm = TRUE)))

#summing through base R
my_vars = c("Value_1", "Value_2", "Value_3")
aggregate(x = my_data[my_vars], by = my_data["Month"], FUN = sum, 
na.rm = TRUE) 

For Value_3 in both months, for instance, I get that the sum is zero instead of NA. Any advice for how to sum NAs to get NA instead of zero would be greatly appreciated.

Answer 1

You can add an if/else to return NA if all the values in the variable is NA:

my_data %>% 
    group_by(Month) %>% 
    summarise_all(
        funs(if(all(is.na(.))) NA else sum(., na.rm = TRUE))
    )

# A tibble: 2 x 4
#       Month Value_1 Value_2 Value_3
#      <fctr>   <dbl>   <dbl>   <lgl>
#1 1995-01-01       3       7      NA
#2 1995-02-01      NA       1      NA

Answer 2

Base on you own approach, add ifelse

my_data %>%
    group_by(Month) %>%
    summarise_each(funs(ifelse(sum(is.na(.))==length(.),NA,sum(.,na.rm = TRUE))))

Answer 3

We can also do this using

library(data.table)
setDT(my_data)[, lapply(.SD, function(x)  sum(x, na.rm = TRUE) *NA^(all(is.na(x)))), Month]
#       Month Value_1 Value_2 Value_3
#1: 1995-01-01       3       7      NA
#2: 1995-02-01      NA       1      NA