contrast command after mi estimate

[Markos Valsamis]

I am running a logistic model on multiply-imputed data, and one of the covariates I include in the model has multiple categories. The Stata output after running the model produces coefficients and significance values for each category, but I would like to obtain a p value for the covariate as a whole. I have previously used the contrast command which has worked well in non-mi data, but this appears not to work on mi data, as with the code below. Any advice on how I can do this?

mi estimate: logit outcome variable_a variable_b
contrast variable_b

Error code is produced: "requested action not valid after most recent estimation command"


Many thanks in advance

[daniel klein]

If you want a joint test, use mi test. Here is an example

. // example data
. webuse mheart1s20
(Fictional heart attack data; BMI missing)

. 
. // create categorical predictor
. generate age_cat = irecode(age, 35, 45, 55, 65)

. 
. // logit model
. mi estimate : logit attack i.age_cat bmi i.female

Multiple-imputation estimates                   Imputations       =         20
Logistic regression                             Number of obs     =        154
                                                Average RVI       =     0.0260
                                                Largest FMI       =     0.1212
DF adjustment:   Large sample                   DF:     min       =   1,321.47
                                                        avg       = 106,533.13
                                                        max       = 509,383.50
Model F test:       Equal FMI                   F(   6,131614.8)  =       1.82
Within VCE type:          OIM                   Prob > F          =     0.0908

------------------------------------------------------------------------------
      attack | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
     age_cat |
          1  |   .3141378   1.345188     0.23   0.815     -2.32244    2.950715
          2  |   .8507624   1.296884     0.66   0.512    -1.691131    3.392656
          3  |   .7430001   1.319771     0.56   0.573    -1.843772    3.329772
          4  |   1.631275   1.311714     1.24   0.214    -.9396832    4.202233
             |
         bmi |   .0960174   .0462712     2.08   0.038     .0052444    .1867905
    1.female |  -.0860744   .3979733    -0.22   0.829    -.8660895    .6939408
       _cons |  -3.568647   1.868242    -1.91   0.056    -7.231645    .0943514
------------------------------------------------------------------------------

. 
. // joint test
. mi test 1.age_cat 2.age_cat 3.age_cat 4.age_cat 
note: assuming equal fractions of missing information.

 ( 1)  [attack]1.age_cat = 0
 ( 2)  [attack]2.age_cat = 0
 ( 3)  [attack]3.age_cat = 0
 ( 4)  [attack]4.age_cat = 0

       F(  4,745085.1) =    1.83
            Prob > F =    0.1195