Ignore missing values when generating new variable

Question

I want to create a new variable in Stata, that is a function of 3 different variables, X, Y and Z, like:

gen new_var = (((X)*3) + ((Y)*2) + ((Z)*4))/7

All observations have missing values for one or two of the variables.

When I run the aforementioned command, all it generates are missing values, because no observation has values for all 3 of the variables. I would like Stata to complete the function ignoring the missing variables.

I tried the following commands without success:

gen new_var= (cond(missing(X*3),., X) + cond(missing(Y*2),., Y))/7 
gen new_var= (!missing(X*3+Y*2+Z*4)/7)
gen new_var=  (max(X , Y, Z)/7) if missing(X , Y, Z) 

The egen command does not allow complicated functions; otherwise rowtotal() could work.

---

EDIT:

To clarify, "ignoring missing variables" means that even if any one of the component variables is not missing, then apply the function to only that variable and produce a value for the new variable. The new variable should have missing values only when all three component variables are missing.

Answer 1

I am going to guess that "ignoring missing values" means "treating them as zeros". If you have some other idea, you should make it explicit.

That could be

gen new_var = (cond(missing(X), 0, 3 * X) ///
+ cond(missing(Y), 0, 2 * Y) ///
+ cond(missing(Z), 0, 4 * Z)) / 7 

Let's look at your solutions and explain why they are all wrong either in general or usually.

(cond(missing(X*3),., X) + cond(missing(Y*2),., Y))/7 

It is sufficient is note that if it's true that X is missing, then cond() yields missing, as then X * 3 is missing too. The same kind of remark applies to terms involving Y and Z. So you're replacing any missing values by missing values, which is no gain.

!missing(X*3+Y*2+Z*4)/7

Given the information that at least one of X Y Z is always missing, then this always evaluates to 0/7 or 0. Even if X Y Z were all non-missing, then it would evaluate to 1/7. That is a long way from the sum you want. missing() always yields 1 or 0, and its negation thus 0 or 1.

(max(X, Y, Z)/7) if missing(X , Y, Z) 

The maximum of X, Y, Z will be the right answer if and only if one of the values is not missing and the other two are missing. max() ignores missings to the extent possible (even though in other contexts missings are treated as if arbitrarily large positive numbers).

Answer 2

If you just want to "ignore missing values" without "treating them as zeros", the following will work:

clear
set obs 10

generate X = rnormal(5, 2)
generate Y = rnormal(10, 5)
generate Z = rnormal(1, 10)

replace X = . in 2
replace Y = . in 5
replace Z = . in 9

generate new_var = (((X)*3) + ((Y)*2) + ((Z)*4)) / 7 if X != . | Y != . | Z != .

list

     +---------------------------------------------+
     |        X          Y           Z     new_var |
     |---------------------------------------------|
  1. | 3.651024    3.48609    -24.1695   -11.25039 |
  2. |        .   14.14995    8.232919           . |
  3. | 3.689442   9.812483    1.154064    5.044221 |
  4. | 2.500493   13.02909     5.25539    7.797317 |
  5. |  4.19431          .    6.584174           . |
  6. | 7.221717   13.92533    5.045283    9.956708 |
  7. | 5.746871   14.26329    3.828253    8.725744 |
  8. | 1.396223    16.2358    19.01479    16.10277 |
  9. | 4.633088   13.95751           .           . |
 10. | 2.521546   4.490258   -3.396854     .422534 |
     +---------------------------------------------+

Alternatively, you could also use the inlist() function:

generate new_var = (((X)*3) + ((Y)*2) + ((Z)*4)) / 7 if !inlist(., X, Y, Z)