st: xi: and if

[Jonathan Sterne <Jonathan.Sterne@bristol.ac.uk>]

A colleague and I were recently caught by the way that xi: interacts with 
if clauses. In the following, the whole dataset contains a value 0 for 
variable  e1swabtype, but the subset that we wish to deal with does not. 
Because xi: creates indicator variables corresponding to every value in the 
dataset, not every value to be processed by the logistic command, logistic 
has to pick an indicator to be dropped, which it does automatically. 
Unfortunately for us, it dropped the indicator for level 2, leaving the 
results of the logistic regression the opposite way round to the way that 
we would have expected (i.e. the way that we got when we dropped the 
unwanted observations, then ran a logistic regression.)

Most of the time, we think of these two options (run the command on a 
subgroup using -if-  or drop the observations not in the subgroup) as 
interchangeable. Is there a good reason that xi: does not respect the if 
clause?

Jonathan Sterne


. tab e1swabtype

e1swab type |      Freq.     Percent        Cum.
------------+-----------------------------------
         0 |      9,966       51.01       51.01
         1 |        337        1.73       52.74
         2 |      9,233       47.26      100.00
------------+-----------------------------------
     Total |     19,536      100.00

. tab e1swabtype if e0gppracticeid=="L81022" & e0sex==2

e1swab type |      Freq.     Percent        Cum.
------------+-----------------------------------
         1 |        337       49.93       49.93
         2 |        338       50.07      100.00
------------+-----------------------------------
     Total |        675      100.00

. tab e1swabtype responder1 if e0gppracticeid=="L81022" & e0sex==2

   e1swab |      responder1
     type |         0          1 |     Total
-----------+----------------------+----------
        1 |       223        114 |       337
        2 |       239         99 |       338
-----------+----------------------+----------
    Total |       462        213 |       675


. xi: logistic responder1 i.e1swabtype [pw=sampweight] ///

>     if e0gppracticeid=="L81022" & e0sex==2
i.e1swabtype      _Ie1swabtyp_0-2     (naturally coded; _Ie1swabtyp_0 
omitted)

note: _Ie1swabtyp_2 dropped due to collinearity

Logistic regression                               Number of obs   = 
675
                                                 Wald chi2(1)    = 
4.98
                                                 Prob > chi2     = 
0.0257
Log pseudo-likelihood = -419.26431                Pseudo R2       = 
0.0091

---------------------------------------------------------------------------
---
            |               Robust
 responder1 | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf. 
Interval]
-------------+-------------------------------------------------------------
---
_Ie1swabty~1 |   1.583557   .3262572     2.23   0.026     1.057456 
2.371402
---------------------------------------------------------------------------
---

.
. keep if e0gppracticeid=="L81022" & e0sex==2
(19098 observations deleted)

. xi: logistic responder1 i.e1swabtype [pw=sampweight] ///

>     if e0gppracticeid=="L81022" & e0sex==2
i.e1swabtype      _Ie1swabtyp_1-2     (naturally coded; _Ie1swabtyp_1 
omitted)

Logistic regression                               Number of obs   = 
675
                                                 Wald chi2(1)    = 
4.98
                                                 Prob > chi2     = 
0.0257
Log pseudo-likelihood = -419.26431                Pseudo R2       = 
0.0091

---------------------------------------------------------------------------
---
            |               Robust
 responder1 | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf. 
Interval]
-------------+-------------------------------------------------------------
---
_Ie1swabty~2 |   .6314897   .1301046    -2.23   0.026     .4216915 
.9456657
---------------------------------------------------------------------------
---




----------------------

Jonathan Sterne

Department of Social Medicine
University of Bristol
Canynge Hall
Whiteladies Road
Bristol BS8 2PR
UK

E-mail: jonathan.sterne@bristol.ac.uk
web:    www.epi.bris.ac.uk

*
*   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/