difference value over subsamples

[Sa Fe]

Dear Statalist,

I have 2 variables, Y & X, where X is a categorical variable. I want the following value:

average_y if x==1 minus average_y if x==0.

I'd appreciate any insights. Thanks,

ps: I can do this manually by doing the following code:
sum y if x==0
sum y if x==1
then get their mean value and subtract by a calculator.
But since I have many variables and subgroups that I want to get the differences, I would want to get the differences more efficiently.

[Carlo Lazzaro]

Saber:
you may want to consider:

. set obs 10
Number of observations (_N) was 0, now 10.

. g y=runiform()

. g x=0 in 1/5
(5 missing values generated)

. replace x=1 if x==.
(5 real changes made)

. list

     +--------------+
     |        y   x |
     |--------------|
  1. | .3488717   0 |
  2. | .2668857   0 |
  3. | .1366463   0 |
  4. | .0285569   0 |
  5. | .8689333   0 |
     |--------------|
  6. | .3508549   1 |
  7. | .0711051   1 |
  8. |  .323368   1 |
  9. | .5551032   1 |
 10. |  .875991   1 |
     +--------------+

. g wanted=y if x==1
(5 missing values generated)

. replace wanted=y*(-1) if x==0
(5 real changes made)

. list

     +--------------------------+
     |        y   x      wanted |
     |--------------------------|
  1. | .3488717   0   -.3488717 |
  2. | .2668857   0   -.2668857 |
  3. | .1366463   0   -.1366463 |
  4. | .0285569   0   -.0285569 |
  5. | .8689333   0   -.8689333 |
     |--------------------------|
  6. | .3508549   1    .3508549 |
  7. | .0711051   1    .0711051 |
  8. |  .323368   1     .323368 |
  9. | .5551032   1    .5551032 |
 10. |  .875991   1     .875991 |
     +--------------------------+

.

[Nick Cox]

Consider some variation on

* Example generated by -dataex-. For more info, type help dataex
clear
input str1 group float(binary whatever)
"a" 0 10
"a" 1  1
"b" 0  9
"b" 1  2
"c" 0  8
"c" 1  3
"d" 0  7
"d" 1  4
"e" 0  6
"e" 1  5
end

bysort group : egen wanted = total((binary == 1) * whatever - (binary == 0) * whatever)

list, sepby(group)

     +------------------------------------+
     | group   binary   whatever   wanted |
     |------------------------------------|
  1. |     a        0         10       -9 |
  2. |     a        1          1       -9 |
     |------------------------------------|
  3. |     b        0          9       -7 |
  4. |     b        1          2       -7 |
     |------------------------------------|
  5. |     c        0          8       -5 |
  6. |     c        1          3       -5 |
     |------------------------------------|
  7. |     d        0          7       -3 |
  8. |     d        1          4       -3 |
     |------------------------------------|
  9. |     e        0          6       -1 |
 10. |     e        1          5       -1 |
     +------------------------------------+

The example just has one observation that is 1 and one observation that is 0 for each group, but the code isn't based on that being true.

[Sa Fe]

Saber:
. g wanted=y if x==1

. replace wanted=y*(-1) if x==0
[/CODE]

Thanks Carlo,
Actually I wanted to obtain the averages difference in y over subsamples of x. I edited my original post accordingly.
Thanks,

[Sa Fe]

Consider some variation on

* Example generated by -dataex-. For more info, type help dataex
clear
input str1 group float(binary whatever)
"a" 0 10
"a" 1 1
"b" 0 9
"b" 1 2
"c" 0 8
"c" 1 3
"d" 0 7
"d" 1 4
"e" 0 6
"e" 1 5
end

bysort group : egen wanted = total((binary == 1) * whatever - (binary == 0) * whatever)

list, sepby(group)

+------------------------------------+
| group binary whatever wanted |
|------------------------------------|
1. | a 0 10 -9 |
2. | a 1 1 -9 |
|------------------------------------|
3. | b 0 9 -7 |
4. | b 1 2 -7 |
|------------------------------------|
5. | c 0 8 -5 |
6. | c 1 3 -5 |
|------------------------------------|
7. | d 0 7 -3 |
8. | d 1 4 -3 |
|------------------------------------|
9. | e 0 6 -1 |
10. | e 1 5 -1 |
+------------------------------------+

The example just has one observation that is 1 and one observation that is 0 for each group, but the code isn't based on that being true.

Thanks Nick,
But what I want is to have only one number: average of y over x==1 - average of y over x==0.

Thanks,

[Nick Cox]

OK; start with

bysort group : egen mean1 = mean(cond(binary == 1, whatever, .))  

bysort group : egen mean0 = mean(cond(binary == 0, whatever, .))

[Sa Fe]

OK; start with

bysort group : egen mean1 = mean(cond(binary == 1, whatever, .))

bysort group : egen mean0 = mean(cond(binary == 0, whatever, .))

I get an error:
invalid syntax
r(198);

[Carlo Lazzaro]

Saber:
Nick's code works for me:

. bysort group : egen mean1 = mean(cond(binary == 1, whatever , .))

. bysort group : egen mean0 = mean(cond(binary == 0, whatever, .))

. list

     +-------------------------------------------+
     | group   binary   whatever   mean1   mean0 |
     |-------------------------------------------|
  1. |     a        0         10       1      10 |
  2. |     a        1          1       1      10 |
  3. |     b        0          9       2       9 |
  4. |     b        1          2       2       9 |
  5. |     c        0          8       3       8 |
     |-------------------------------------------|
  6. |     c        1          3       3       8 |
  7. |     d        0          7       4       7 |
  8. |     d        1          4       4       7 |
  9. |     e        0          6       5       6 |
 10. |     e        1          5       5       6 |
     +-------------------------------------------+

.

[Sa Fe]

Yeah, it worked now! Thanks,