Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Interpreting interactions in logistic


From   Filipa de Castro <filipa.castro@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   st: Interpreting interactions in logistic
Date   Sun, 6 Sep 2009 19:25:32 -0500

Dear statalisters,

I am writing to ask for some help in interpreting the result of a
model I am running with my data.
I have already asked some colleagues and the responses have been quite
contradictory.
So. My model is:

Y= suicidal ideation 0=no 1=yes
X1= male 0=female  1=male
X2= active_sexlife  0=no  1=yes

Y= X1  X2   X1*X2

the output for this is:

Survey: Logistic regression
Number of strata   =         3                     Number of obs
=     12424
Number of PSUs     =       217                  Population size    =    992834
                                                             Design df
         =       214
                                                              F(   3,
  212)    =    131.24
                                                             Prob > F
         =    0.0000
------------------------------------------------------------------------------
             |                              Linearized
ideacion_cat        | Odds Ratio   Std. Err.      t    P>|t|     [95%
Conf. Interval]
-------------+----------------------------------------------------------------
_Iactive_sexlife |   1.828036   .1347743     8.18   0.000     1.580781
   2.113964
       Male          |   .4996468   .0253318   -13.69   0.000
.4521289    .5521587
    _IactXmale    |   .7020524   .0743205    -3.34   0.001
.5698326    .8649516
------------------------------------------------------------------------------

Now. If I do the same model but instead of Male I use Female where
0=male and 1=female  I get this:

Survey: Logistic regression
Number of strata   =         3                  Number of obs      =     12424
Number of PSUs     =       217                  Population size    =    992834
                                                Design df          =       214
                                                F(   3,    212)    =    131.24
                                                Prob > F           =    0.0000
------------------------------------------------------------------------------
             |                   Linearized
ideacion_cat |     Odds Ratio   Std. Err.      t    P>|t|     [95%
Conf. Interval]
-------------+----------------------------------------------------------------
_Iactive_sexlife |   1.283377    .110368     2.90   0.004     1.083268
    1.52045
        female      |   2.001414   .1014703    13.69   0.000
1.811074    2.211759
    _IactXfemale |   1.424395    .150789     3.34   0.001     1.156134
   1.754901
------------------------------------------------------------------------------

My questions are:
Why does the model change just by inverting the dummy for sex?
How can I know the effect for _active_sexlife for man and woman from
just looking at one of the outputs?
If I look at model 1 I see that odds for having suicide ideation for
males is 49%, but when I look at model 2 I see that odds for women is
200% ? I am really puzzled with this result as well as with the
different OR for the vars in 2 models.
If the interaction is significant how come I cannot detect it
graphically with postgr3?? neither by a tabulation of sex x
active_sexlife  with prtab ?


--------------
best wishes and thanks
Filipa de Castro

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index