replace var = .c

[Helder Costa]

Hello,

Reading someone else's code, I have run into for the first time the expression "replace var = .c", instead of "replace var = .". At first, I assumed this was a mistake, but running the code myself, it works. My question is thus, what exactly does this do, and why can't I find any reference to it online or in the help files?

sysuse auto, clear
tab rep78, miss
replace rep78 = .c if rep78 == 5
tab rep78, miss

Running this, yields the following:

. sysuse auto, clear
(1978 automobile data)
r; t=0.00 9:26:42

. tab rep78, miss

Repair |
record 1978 | Freq. Percent Cum.
------------+-----------------------------------
1 | 2 2.70 2.70
2 | 8 10.81 13.51
3 | 30 40.54 54.05
4 | 18 24.32 78.38
5 | 11 14.86 93.24
. | 5 6.76 100.00
------------+-----------------------------------
Total | 74 100.00
r; t=0.00 9:26:42

. replace rep78 = .c if rep78 == 5
(11 real changes made, 11 to missing)
r; t=0.00 9:26:42

. tab rep78, miss

Repair |
record 1978 | Freq. Percent Cum.
------------+-----------------------------------
1 | 2 2.70 2.70
2 | 8 10.81 13.51
3 | 30 40.54 54.05
4 | 18 24.32 78.38
. | 5 6.76 85.14
.c | 11 14.86 100.00
------------+-----------------------------------
Total | 74 100.00
r; t=0.00 9:26:42

.
end of do-file

Thus, one purpose seems to be to differentiate missing values already in the data from those replaced by us. Is there any other purpose?

Thanks,
Best,
Hélder

[Andrew Musau]

Such values are referred to as extended missing values. They are useful in a number of situations, e.g., say you want to differentiate between "refused to answer" and "don't know" responses in a survey. See

help missing

[Helder Costa]

Thank you Andrew!

[daniel klein]

One place where the distinction between system missing values (i.e. the dot .) and extended missing values (.a, .b, etc.) often causes problems is multiple imputation (see [MI] Multiple Imputation). Within the mi framework, system missing values are regarded as "soft missing" while extended missing values are regarded as "hard missing". The latter are not imputed. I do not think this behavior is useful, but there you go.

[Nick Cox]

As you have found, that is perfectly legal, and as you have guessed, it is usually intended to flag some reason for being missing. In social research on individuals, researchers often want to distinguish did not answer, inapplicable, "Don't know", and so on. But in any field there can be specific reasons. Records lost, records illegible, flood over-topped gauge, test-tube dropped on floor, contamination of sample: the list goes on and on.

The scope for extended missing values is documented.

help missing 

leads to a version of [U] 12.2.1.

EDIT: I had this open while doing something else and so did not see the overlapping answers from Andrew Musau and daniel klein until later.

As the Bellman said in "The Hunting of the Snark", what we tell you three times is true.