replace versus recode message

[alrocha2]

A student approached me and noticed that when I showed two different ways to generate a new dummy variable, the first yielded a message that made sense while the second had a message that didn't seem to make sense. Here are the frequencies for the variable we converted to a dummy.
veteran | Freq.
------------+----------
1 | 4
3 | 1
5 | 9
6 | 101
. | 20
------------+------------
Total | 135 100.00
Here are the two methods I demonstrated to create a dummy variable for veteran (1=yes/0=no). Both methods "worked" such that they yielded the dummy variable as intended.
* Method 1
ge vetcat1 = .
replace vetcat1 = 1 if veteran==1 | veteran==3 | veteran==5
replace vetcat1 = 0 if veteran==6 * Method 2
ge vetcat2 = veteran
recode vetcat2 1/5=1 6=0
The first method gave this output. . ge vetcat1 = .
(135 missing values generated)
. replace vetcat1 = 1 if veteran==1 | veteran==3 | veteran==5
(14 real changes made)
. replace vetcat1 = 0 if veteran==6
(101 real changes made)
The second method gave this output.
. ge vetcat2 = veteran
(20 missing values generated)
. recode vetcat2 1/5=1 6=0
(vetcat2: 111 changes made)

So the question is why does method 1 display 14 and 101 changes made (which makes sense) while method 2 displays 111 changes made (which is not the sum 14 + 101)?
I couldn't figure it out. Have you run across this?

[Roberto Ferrer]

Because there are four observations that do not need recoding. Namely, those that are already equal to 1. So 111 (recoded) + 4 (that are not recoded) + 20 (missing) = 135.

Please follow advice on formatting your posts. That is very difficult to read.

[ben earnhart]

Your syntax was a bit difficult to dis-entangle, since you re-arranged things and didn't use code blocks. But your difference is due to the fact that four cases *already* had a value of 1, so they weren't changed with your -recode- syntax
.

clear

set obs 135
gen veteran=.
replace veteran=1 if _n<=4
replace veteran=3 if _n==5
replace veteran=5 if _n>5 & _n<15
replace veteran=6 if _n>=15 & _n<116
tab veteran, mi 

* Method 1
ge vetcat1 = .
replace vetcat1 = 1 if veteran==1 | veteran==3 | veteran==5
replace vetcat1 = 0 if veteran==6
* Method 2
ge vetcat2 = veteran
recode vetcat2 1/5=1 6=0

[Sergiy Radyakin]

Well, if the value doesn't change, Stata doesn't count it into the values that change.
In method 1 all values were missing to begin with, and were assigned non-missing values. So number of values changed is _N.
In method 2 some values were preserved as original (ones remained ones), so they are not counted: 135 total = 111 changes + 20 missings unchanged + 4 ones unchanged.
Best, Sergiy Radyakin.