Replace all missing values as "NA" (type mismatch?)

[Sarah Peters]

Hi - new to Stata.

I've run a script in Stata but now want to move my dataset into R. R is treating all Stata missing values as 0, so I want to recode them all in Stata as "NA". I've tried:

recode * (""=="NA")

But am getting the error code type mismatch. Is there a command to replace all missing variables, or to change the type of all missing variables? Or is there a better way to change variables before moving a dataset to R?

Thanks,
Sarah

[Carlo Lazzaro]

Sarah:
welcome to this forum.
You may want to try something along to the following lines:

replace A=.a if A==.

[Andrew Musau]

---Crossed with Carlo: It is easier to follow Carlo's suggestion and then export---

You are probably mixing up numeric variables with string variables. First, convert all variables to strings. Here is some method

sysuse auto, clear
*IDENTIFY NUMERIC VARIABLES AND STORE IN A LOCAL MACRO
local vars "price weight mpg"
foreach var in `vars'{
      gen s`var'= string(`var')
      drop `var'
      rename s`var' `var'
 }

*REPLACE MISSING WITH "NA"
foreach var in `vars'{
      replace `var'="NA" if `var'==""
 }

Note that you can use a Stata data set directly in R via the foreign package, and therefore you do not need to change the format. See this page for details.

[Martyn Sherriff]

Would 'save2rda-- Export data from Stata to R in RData(RDA2) format' from http://www.radyakin.org/transfer/save2rda/ be of any use?

[Nick Cox]

Just to explain that recode has nothing to do with string variables. That's in the help:

recode changes the values of numeric variables according to the rules specified.

I have to say -- sorry@Andrew Musau -- that I don't think converting all your data to strings is a good idea. Also, if #3 were a good idea, then the loop is a reinvention of tostring without any of the safety features of the latter.

Can you document R treating system missing as zero?

[Andrew Musau]

Nick Cox I had it in the back of my mind that there is a command that does the opposite of destring. I just never had to use it before.

[Jorrit Gosens]

It would be weird if R sets missings to zero. That's definitely not default behavior.
I just had a go at saving a copy of auto.dta, imported with the haven package in R, and that behaves as it should. Missings are missings.
Maybe you should share a snippet of your data, with dataex, here. See instructions on dataex in 12.2 of the FAQ: https://www.statalist.org/forums/help
It is possible, for examples, that your missings are numerical variables with text labels, which could cause problems.

[Marcos Almeida]

i fail to reproduce the problem mentioned in #1:

Below, an example of a "toy" data with "n.a.", then handled in Stata, finally imported in R:

. input str10 (treat value)

          treat       value
  1. DrugA 43
  2. DrugB 31
  3. DrugA 56
  4. DrugB "n.a."
  5. end

. replace value = ".a" if value == "n.a."

. destring value, replace

. labe define mylab .a "not available"

. label values value mylab

. list

     +-----------------------+
     | treat           value |
     |-----------------------|
  1. | DrugA              43 |
  2. | DrugB              31 |
  3. | DrugA              56 |
  4. | DrugB   not available |
     +-----------------------+

. sum value

    Variable |        Obs        Mean    Std. Dev.       Min        Max
-------------+---------------------------------------------------------
       value |          3    43.33333    12.50333         31         56

I saved this toy example as mytest.dta. Then, now in R, I imported it as a Stata file.

> mytest1 <- as.data.frame(mytest)
> mytest1
  treat value
1 DrugA    43
2 DrugB    31
3 DrugA    56
4 DrugB    NA
> summary(mytest1)
    treat               value      
 Length:4           Min.   :31.00  
 Class :character   1st Qu.:37.00  
 Mode  :character   Median :43.00  
                    Mean   :43.33  
                    3rd Qu.:49.50  
                    Max.   :56.00  
                    NA's   :1

Hopefully that helps.