Using nlcom after postestimation margins

[Itai Magodoro]

Dear forum,

Thanks in advance for any guidance you may give. I would like to estimate the ratio (95%CI) of two (2) predicted probabilities that were computed using postestimation margins after logistic regression. I thought I could use nlcom but I`m coming stuck.

logit dm i.sex 
margins sex, post

Logistic regression                                     Number of obs =  4,830
                                                        LR chi2(1)    =   7.04
                                                        Prob > chi2   = 0.0080
Log likelihood = -2596.4629                             Pseudo R2     = 0.0014

------------------------------------------------------------------------------
          dm | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         sex |
       Male  |  -.1912397   .0725017    -2.64   0.008    -.3333405    -.049139
       _cons |   -1.14623   .0420583   -27.25   0.000    -1.228663   -1.063797
------------------------------------------------------------------------------ 

.

margins sex, post

Expression: Pr(dm), predict()

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         sex |
     Female  |      0.146      0.012    12.62   0.000        0.124       0.169
       Male  |      0.117      0.014     8.17   0.000        0.089       0.146
------------------------------------------------------------------------------ 

logit dm i.lclass

Logistic regression                                     Number of obs =  3,338
                                                        LR chi2(4)    =  89.35
                                                        Prob > chi2   = 0.0000
Log likelihood = -1431.0752                             Pseudo R2     = 0.0303

------------------------------------------------------------------------------
          dm | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
      lclass |
          2  |   .1144973   .1324727     0.86   0.387    -.1451444     .374139
          3  |   1.012298    .181102     5.59   0.000     .6573448    1.367252
          4  |  -.6119494   .1525491    -4.01   0.000    -.9109402   -.3129587
          5  |   .3836895   .1778637     2.16   0.031     .0350831    .7322959
             |
       _cons |  -1.662886   .1079938   -15.40   0.000     -1.87455   -1.451222
------------------------------------------------------------------------------ 

. margins lclass, post

Expression: Pr(dm), predict()

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
      lclass |
          1  |      0.078      0.023     3.47   0.001        0.034       0.123
          2  |      0.044      0.010     4.46   0.000        0.025       0.064
          3  |      0.213      0.052     4.08   0.000        0.111       0.316
          4  |      0.051      0.039     1.30   0.196       -0.026       0.128
          5  |      0.296      0.104     2.84   0.005        0.091       0.501
------------------------------------------------------------------------------ 

 nlcom _b[Male.sex]\_b[1.lclass]

[Male.sex]\_b[1.lclass] not found 

Thanks,
Itai

[Bader Bin Adwan]

Try these in order:

1) gen ratio = exp(_b[Male.sex] + _b[1.lclass]) / exp(_b[1.lclass])

2) margins, at(Male.sex == 1 1.lclass == 1)

3) margins, at(Male.sex == 0 1.lclass == 1)

4) nlcom ratio_male_female: ratio / _b[/lnorm]