Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Exponentiated coefficients after mi estimate: melogit

    Hello, I am looking to get exponentiated coefficients after estimating a multilevel, mixed effects logistic regression with multiply imputed data.

    My code is as follows:

    Code:
    mi estimate (_b[MED14_0]) ///
                (_b[MED14_0] + _b[2.WAVE#c.MED14_0]) ///
                (_b[MED14_0] + _b[3.WAVE#c.MED14_0]) /// 
                (_b[MED14_0] + _b[c.MED14_0#1.SEX]) ///
                (_b[MED14_0] + _b[2.WAVE#c.MED14_0] + _b[2.WAVE#c.MED14_0#1.SEX]) ///
                (_b[MED14_0] + _b[3.WAVE#c.MED14_0] + _b[3.WAVE#c.MED14_0#1.SEX]) ///
                , eform cmdok: ///
        melogit BINGE_ i.BINGE_base c.MED14_0##i.WAVE##i.SEX ///main terms + interaction
        DEP_0 ADHD_0 DB_0 i.FAMHIST_ALC i.FAMHIST_DRUG i.MJ_0 i.NIC_0 ///
            i.DRUGS_0 AGE_0 i.ETHNIC i.LUNCH ///covariates
        if MISSING_OUT==0 || SCHOOL: || SID: //extra stuff
    The output I get for this is:

    Code:
    Multiple-imputation estimates                   Imputations       =         10
    Mixed-effects logistic regression               Number of obs     =      4,218
                                                    Average RVI       =     0.0507
                                                    Largest FMI       =     0.1528
                                                    DF:     min       =     407.24
                                                            avg       = 282,605.14
    DF adjustment:   Large sample                           max       = 6308629.07
                                                    F(  27,      .)   =          .
    Within VCE type:          OIM                   Prob > F          =          .
    
    ---------------------------------------------------------------------------------------
                   BINGE_ |     exp(b)   Std. Err.      t    P>|t|     [95% Conf. Interval]
    ----------------------+----------------------------------------------------------------
             1.BINGE_base |   9.493373   1.882706    11.35   0.000     6.435928    14.00329
                  MED14_0 |   1.101308    .046136     2.30   0.021     1.014443    1.195611
                          |
                     WAVE |
                       2  |   1.253049   .3675333     0.77   0.442     .7045933    2.228425
                       3  |   2.643886   .7371351     3.49   0.000     1.530396    4.567533
                          |
           WAVE#c.MED14_0 |
                       2  |   1.003756   .0513321     0.07   0.942     .9077536    1.109912
                       3  |    .923678   .0449719    -1.63   0.103       .83954    1.016248
                          |
                      SEX |
                    Male  |   2.146329   .7539484     2.17   0.030     1.077666    4.274726
                          |
            SEX#c.MED14_0 |
                    Male  |   .8877788    .054471    -1.94   0.053     .7870215    1.001435
                          |
                 WAVE#SEX |
                  2#Male  |   .8968825   .3610686    -0.27   0.787     .4073536    1.974693
                  3#Male  |   .4687726   .1874344    -1.89   0.058     .2140023    1.026848
                          |
       WAVE#SEX#c.MED14_0 |
                  2#Male  |   1.033528   .0720354     0.47   0.636     .9014825    1.184915
                  3#Male  |   1.155732   .0809422     2.07   0.039     1.007326    1.326003
                          |
                    DEP_0 |   1.002952   .0061672     0.48   0.632     .9909373    1.015113
                   ADHD_0 |   1.235144   .1692749     1.54   0.123     .9441368    1.615847
                     DB_0 |   .9805965   .0124518    -1.54   0.123     .9564893    1.005311
                          |
              FAMHIST_ALC |
                     Yes  |   1.284036   .2242157     1.43   0.153      .911536     1.80876
                          |
             FAMHIST_DRUG |
                     Yes  |   1.117512   .2199697     0.56   0.573     .7593975    1.644504
                          |
                     MJ_0 |
                     Yes  |   1.700471   .3236107     2.79   0.005     1.171062    2.469213
                          |
                    NIC_0 |
                     Yes  |    1.66683   .3154218     2.70   0.007     1.150307    2.415286
                          |
                  DRUGS_0 |
                     Yes  |   1.532934   .3410133     1.92   0.055     .9912093    2.370726
                    AGE_0 |   .9446203   .1776188    -0.30   0.762     .6529157     1.36665
                          |
                   ETHNIC |
                   Black  |   .5275365   .2357775    -1.43   0.152     .2196909    1.266756
         Hispanic/Latino  |   .9414317   .2117789    -0.27   0.788      .605732    1.463178
                   Asian  |    .471768    .150416    -2.36   0.018     .2525062    .8814238
                   Other  |   .8965573   .2321105    -0.42   0.673     .5397553    1.489221
                          |
                    LUNCH |
                       1  |   .9288689   .2640768    -0.26   0.795     .5316108    1.622987
                       2  |   .7690034    .137333    -1.47   0.142     .5416204    1.091846
                          |
                    _cons |   .0751932   .2331758    -0.83   0.404     .0001705    33.16745
    ----------------------+----------------------------------------------------------------
        var(_cons[SCHOOL])|   .0116671   .0295897                      .0000809    1.681846
    var(_cons[SCHOOL>SID])|   3.039094   .3954284                         2.355    3.921907
    ---------------------------------------------------------------------------------------
    Note: Estimates are transformed only in the first equation.
    
    Transformations                                 Average RVI       =     0.1182
                                                    Largest FMI       =     0.1536
    DF adjustment:   Large sample                   DF:     min       =     403.10
                                                            avg       =   1,520.15
    Within VCE type:          OIM                           max       =   5,199.22
    
            _mi_1: _b[MED14_0]
            _mi_2: _b[MED14_0] + _b[2.WAVE#c.MED14_0]
            _mi_3: _b[MED14_0] + _b[3.WAVE#c.MED14_0]
            _mi_4: _b[MED14_0] + _b[c.MED14_0#1.SEX]
            _mi_5: _b[MED14_0] + _b[2.WAVE#c.MED14_0] + _b[2.WAVE#c.MED14_0#1.SEX]
            _mi_6: _b[MED14_0] + _b[3.WAVE#c.MED14_0] + _b[3.WAVE#c.MED14_0#1.SEX]
    
    ------------------------------------------------------------------------------
          BINGE_ |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
    -------------+----------------------------------------------------------------
           _mi_1 |   .0964984    .041892     2.30   0.021     .0143394    .1786573
           _mi_2 |   .1002475   .0425527     2.36   0.019     .0165946    .1839004
           _mi_3 |   .0171066   .0389594     0.44   0.661    -.0592702    .0934835
           _mi_4 |  -.0225343   .0448122    -0.50   0.615    -.1106248    .0655562
           _mi_5 |   .1332256    .067133     1.98   0.048     .0013654    .2650859
           _mi_6 |   .1618407   .0664786     2.43   0.015     .0312957    .2923857
    ------------------------------------------------------------------------------
    Note: Number of groups varies among imputations.
    Note: Number of observations per group varies among imputations.
    In the final table displayed, the coefficients are not exponentiated. Is there any way to display these as odds ratios, instead of log odds ratios? I include the eform option in my mi estimate command, but it appears to only apply to the main regression results, not the linear combinations that I generate. Thanks in advance for any help that can be provided.

  • #2
    You didn't get a quick answer. Following the FAQ on asking questions would improve your chances of a useful response. With your code and data, someone might have programmed how to do what you need. Note also that posting the simplest model the demonstrates your problem is helpful - the extra stuff just reduces the chances someone will puzzle through to see the real issue.

    I don't use this procedure, but you might look at dydx margins and see if it can do what you want. I think it allows transformations of things. Alternatively, you can access the coefficients themselves and do the calculations yourself. If after any Stata estimator (say regress) you enter regress,coefl Stata will tell you how to refer to each of the coefficients.

    Comment

    Working...
    X