# st: "if" statement

 From Victor Mauricio Herrera <[email protected]> To statalist <[email protected]> Subject st: "if" statement Date Wed, 16 Sep 2009 15:43:58 -0500

```I'd appreciate if somebody could explain the following behavior of the "if" statement when used with "logistic" (I'm running STATA IC/10.1).

webuse nhanes2f
gen ageg=2 if age>=20 & age<30
replace ageg=3 if age>=30 & age<40
replace ageg=4 if age>=40 & age<50
replace ageg=5 if age>=50 & age<60
replace ageg=6 if age>=60 & age<70
replace ageg=7 if age>=70
replace sex=0 if sex==2

model 1 --> xi: logistic sex i.ageg
i.ageg            _Iageg_2-7          (naturally coded; _Iageg_2 omitted)

Logistic regression                               Number of obs   =      10337
LR chi2(5)      =       2.80
Prob > chi2     =     0.7302
Log likelihood =  -7150.626                       Pseudo R2       =     0.0002

------------------------------------------------------------------------------
sex | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
_Iageg_3 |   .9761655   .0632659    -0.37   0.710     .8597191    1.108384
_Iageg_4 |   .9971218   .0696639    -0.04   0.967     .8695188    1.143451
_Iageg_5 |   .9424283    .065592    -0.85   0.394     .8222534    1.080167
_Iageg_6 |   .9903392   .0554211    -0.17   0.862     .8874609    1.105144
_Iageg_7 |   .8963705   .0683978    -1.43   0.152     .7718562    1.040971
------------------------------------------------------------------------------

model 2 --> xi: logistic sex i.ageg if age>=30
i.ageg            _Iageg_2-7          (naturally coded; _Iageg_2 omitted)

note: _Iageg_4 dropped because of collinearity

Logistic regression                               Number of obs   =       8017
LR chi2(4)      =       2.35
Prob > chi2     =     0.6713
Log likelihood = -5544.1939                       Pseudo R2       =     0.0002

------------------------------------------------------------------------------
sex | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
_Iageg_3 |   .9789833   .0734452    -0.28   0.777     .8451164    1.134055
_Iageg_5 |   .9451487   .0748512    -0.71   0.476     .8092619    1.103853
_Iageg_6 |   .9931979   .0670654    -0.10   0.919     .8700789    1.133739
_Iageg_7 |   .8989579   .0765455    -1.25   0.211     .7607821     1.06223
------------------------------------------------------------------------------

Why is the age group 4 (40-49) dropped due to collinearity if there are 610 males and 660 females in this stratum? More worrisome, why is the age group 2 (20-29) still being used as reference when it should have been dropped as a consequence of the "if" statement (i.e. _Iage_3 should be the reference instead of _Iage_2)?

model 3 --> xi: logistic sex i.ageg if age<70
i.ageg            _Iageg_2-7          (naturally coded; _Iageg_2 omitted)

note: _Iageg_7 dropped because of collinearity

Logistic regression                               Number of obs   =       9352
LR chi2(4)      =       0.86
Prob > chi2     =     0.9303
Log likelihood = -6472.0855                       Pseudo R2       =     0.0001

------------------------------------------------------------------------------
sex | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
_Iageg_3 |   .9761655   .0632659    -0.37   0.710     .8597191    1.108384
_Iageg_4 |   .9971218   .0696639    -0.04   0.967     .8695188    1.143451
_Iageg_5 |   .9424283    .065592    -0.85   0.394     .8222534    1.080167
_Iageg_6 |   .9903392   .0554211    -0.17   0.862     .8874609    1.105144
------------------------------------------------------------------------------

Now the "if" statement seems to work fine, as subjects with age>=70 are excluded (i.e. the _Iage_7 group has been dropped!)

This also occurs if I run these models using STATA IC/9.2 or if one models another dichotomous variable using a different dataset.

Many thanks,

VICTOR M. HERRERA MD. MS.
Research Assistant
Population Health Sciences Department
University of Wisconsin
610 Walnut St. 626 WARF