negative probabilities post GLM

[Daniel Sutcliffe]

Hello,

I ran a logistic regression in Stata as follows:

glm cost_ind ${sexagespl} $morbs $comorbs *_D2 if s1==1 ,family(binomial) link(logit) eform

After the model had ran I tried to obtain predicted probabilities using:

predict double chat if s1==1, p

However, I am getting negative probabilities. I thought maybe that because I used 'eform' in the command line, and maybe I need to log the results - but that would make no sense for negative numbers.

I have tried searching for what the issue might be but have not been able to find an answer.

[Nick Cox]

help glm postestimation##predict explains that predict, p after glm gives you Pearson residuals, which will often be negative and even larger in magnitude than 1 on occasion.

You just need to drop that option and use the default.

. sysuse auto, clear
(1978 automobile data)

. glm foreign weight, link(logit) f(binomial)

Iteration 0:   log likelihood = -30.546142  
Iteration 1:   log likelihood = -29.056683  
Iteration 2:   log likelihood = -29.054002  
Iteration 3:   log likelihood = -29.054002  

Generalized linear models                         Number of obs   =         74
Optimization     : ML                             Residual df     =         72
                                                  Scale parameter =          1
Deviance         =  58.10800384                   (1/df) Deviance =   .8070556
Pearson          =  61.02398411                   (1/df) Pearson  =   .8475553

Variance function: V(u) = u*(1-u)                 [Bernoulli]
Link function    : g(u) = ln(u/(1-u))             [Logit]

                                                  AIC             =   .8392973
Log likelihood   = -29.05400192                   BIC             =  -251.7847

------------------------------------------------------------------------------
             |                 OIM
     foreign | Coefficient  std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
      weight |  -.0025874   .0006094    -4.25   0.000    -.0037817    -.001393
       _cons |   6.282599   1.603967     3.92   0.000     3.138882    9.426316
------------------------------------------------------------------------------

. predict predicted
(option mu assumed; predicted mean foreign)

. su predicted

    Variable |        Obs        Mean    Std. dev.       Min        Max
-------------+---------------------------------------------------------
   predicted |         74    .2972973    .2906982   .0019454   .8492585

.

[Daniel Sutcliffe]

Thank you