st: different n's using "if"

["Sarah A. Mustillo" <smustillo@psych.mc.duke.edu>]

Hi -

This question is probably an easy one, but I am baffled...

I am trying to run GEE models for subsets of my sample separately: white 
girls, black girls, white boys, and black boys.  I have been using an "if" 
statement before the comma in the regression model, e.g., if 
sex==0&race==1, etc.  I should also mention that I am limiting the sample 
by time of observation as well, so what I really have is: if 
sex==0&race==1&period==1|2, for example.  I started getting suspicious that 
I wasn't doing what I wanted to do when my sample size stayed large.  So, I 
tried preserving the data set, the dropping male and black and running the 
model again, and the n was much smaller.

My question:

Shouldn't using "if" before the comma accomplish the same thing as dropping 
those people from the sample?  What am I missing?
Below are my examples:


xi: xtgee pul    per2Xlagccm period2 lagccm age if 
sex==0&racewh==1&period==1|2 [pweight=wt], robust corr(exch)

Iteration 1: tolerance = .00166807
Iteration 2: tolerance = 7.878e-11

GEE population-averaged model                   Number of obs      = 
2723
Group variable:                         id      Number of groups   = 
390
Link:                             identity      Obs per group: min = 
6
Family:                           Gaussian                     avg = 
7.0
Correlation:                  exchangeable                     max = 
7
                                               Wald chi2(4)       = 
150.68
Scale parameter:                  210.9225      Prob > chi2        = 
0.0000

                              (standard errors adjusted for clustering on 
id)
---------------------------------------------------------------------------
---
            |             Semi-robust
        pul |      Coef.   Std. Err.      z    P>|z|     [95% Conf. 
Interval]
-------------+-------------------------------------------------------------
---
per2Xlagccm |  -1.332656   1.009294    -1.32   0.187    -3.310835 
.645523
    period2 |  -5.228774   .6043163    -8.65   0.000    -6.413212 
-4.044336
     lagccm |  -.0626779   2.276359    -0.03   0.978     -4.52426 
4.398904
        age |  -.6751086   .3574791    -1.89   0.059    -1.375755 
.0255375
      _cons |   84.87411   1.566771    54.17   0.000     81.80329 
87.94492
---------------------------------------------------------------------------
---




versus:

.keep if sex==0
.keep if race==1

xi: xtgee pul    per2Xlagccm period2 lagccm age if period==1|2 
[pweight=wt], robust corr(exch)

Iteration 1: tolerance = .00331609
Iteration 2: tolerance = 1.041e-09

GEE population-averaged model                   Number of obs      = 
468
Group variable:                         id      Number of groups   = 
67
Link:                             identity      Obs per group: min = 
6
Family:                           Gaussian                     avg = 
7.0
Correlation:                  exchangeable                     max = 
7
                                               Wald chi2(4)       = 
87.58
Scale parameter:                  191.5843      Prob > chi2        = 
0.0000

                              (standard errors adjusted for clustering on 
id)
---------------------------------------------------------------------------
---
            |             Semi-robust
        pul |      Coef.   Std. Err.      z    P>|z|     [95% Conf. 
Interval]
-------------+-------------------------------------------------------------
---
per2Xlagccm |  -7.502376   2.768465    -2.71   0.007    -12.92847 
-2.076284
    period2 |  -6.324729   .8571098    -7.38   0.000    -8.004634 
-4.644825
     lagccm |  -.8279901   1.800555    -0.46   0.646    -4.357013 
2.701033
        age |  -.3743277   .6457195    -0.58   0.562    -1.639915 
.8912592
      _cons |   87.05734   3.265192    26.66   0.000     80.65768 
93.457
---------------------------------------------------------------------------
---

Thanks!

Sarah






Sarah A. Mustillo, Ph.D
Center for Developmental Epidemiology
Department of Psychiatry and Behavioral Sciences
Duke University School of Medicine
Box 3454
Durham NC 27710

919 687-4686 x234

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/