using egen to generate average of specified cases

[Matthew Breckons]

As usual I feel that I'm missing something simple but haven't managed to work out what I need to do.

So I've generated a variable:

egen var_x = mean(var_y)

however I don't want the mean of the whole variable, I want to exclude some which meet a certain condition (which I have a separate variable for). var_z=1

I tried:

egen var_x = mean(var_y) if var_z ==1

however that generates the same mean value but only generates it for cases where var_z == 1 whereas I want to generate a var_x for all cases but taking the mean from a specific subgroup of var_y.

Am I overlooking something incredibly obvious?

[Marcos Almeida]

If you want to exclude according to a condition, you may use - != - after the if clause.

[Matthew Breckons]

Thanks for your response Marcos - I probably needed to provide further information - I still can't work out how I would need to phrase this - so supposing I had the following:

var_x	var_y	var_z
	14	1
	90	1
	12	1
	67	0
	95	1
	64	0

so I want a value in every row of var_x, I want this value to be the mean of var_y, but only the mean of var_y when var_z =1, i.e. I want var_x to equal the mean value of 14, 90, 12 and 95, ignoring the other two values (67 and 64) - does that make sense?.

Matt

[Carlo Lazzaro]

Matthew:
are you looking for something along the following lines?:

egen varx=mean( vary) if varz==1

[Matthew Breckons]

Thanks Carlo - this gives me the following:

var_x	var_y	var_z
57	14	1
57	90	1
57	12	1
.	67	0
57	95	1
.	64	0

which has been calculated from (14+90+12+67+95+64)/6 = 57. So var_x is the mean of var_y but if var_z is 0 then var_x is left missing.

but I would like to get to the following:

var_x	var_y	var_z
52.75	14	1
52.75	90	1
52.75	12	1
52.75	67	0
52.75	95	1
52.75	64	0

So when the mean of var_y is calculated, I would like values of var_y where var_z = 0 to be ignored in the calculation. So var_x = (14+90+12+95)/4 = 52.75

Does this make sense?

[Carlo Lazzaro]

Matthew:
the best work-around that springs to my mind is the following one:

. egen varx=mean( vary) if varz==1
(2 missing values generated)

. su varx

    Variable |        Obs        Mean    Std. Dev.       Min        Max
-------------+---------------------------------------------------------
        varx |          4       52.75           0      52.75      52.75

. replace varx=r(mean) if varx==.
(2 real changes made)

. list

     +---------------------+
     | vary   varz    varx |
     |---------------------|
  1. |   14      1   52.75 |
  2. |   90      1   52.75 |
  3. |   12      1   52.75 |
  4. |   67      0   52.75 |
  5. |   95      1   52.75 |
     |---------------------|
  6. |   64      0   52.75 |
     +---------------------+

.

[Matthew Breckons]

Thank you Carlo and Marcos - much appreciated. Matt

[Nick Cox]

Search this forum for mentions of dm0055. Sections 9 and 10 of the paper so identified discuss this problem systematically and give direct one-line solutions.