Dear statalisters,

I would like to ask you please some help interpreting margins

I use Stata 14. What I did is
Code:
  logit diabetes i.bmicatasia agegroup2 fobp2 sexr
diabetes is diabetes incidence in 2007 as my main outcome
bmicatasia is BMI category
agegroup2 is age-group
fobp2 is blood pressure category
and sexr is sex

HTML Code:
. logit diabetes i.bmicatasia agegroup2 fobp2    sexr

Iteration 0:   log likelihood = -565.88314  
Iteration 1:   log likelihood = -475.64377  
Iteration 2:   log likelihood = -411.65553  
Iteration 3:   log likelihood = -409.32254  
Iteration 4:   log likelihood = -409.30833  
Iteration 5:   log likelihood = -409.30833  

Logistic regression    Number of obs     =    1,904
    LR chi2(5)        =    313.15
    Prob > chi2       =    0.0000
Log likelihood = -409.30833    Pseudo R2         =    0.2767

        
diabetes             Coef.   Std. Err.      z    P>z     [95% Conf.    Interval]
        
bmicatasia 
       Over        1.112465   .2561249     4.34    0.000     .6104693    1.614461
      Obese      2.073727   .2269948     9.14    0.000     1.628826    2.518629
agegroup2   -.0525809    .198183    -0.27    0.791    -.4410124    .3358506
fobp2           1.288137   .1236135    10.42    0.000     1.045859    1.530415
sexr            -.4397445   .1989234    -2.21    0.027    -.8296272    -.0498618
_cons        -4.503091   .2703448   -16.66    0.000    -5.032957    -3.973225
after that that, i tried to calculate the probability of having diabetes incidence across the BMI category in normal Blood pressure (fobp2==0), female (sexr==0), and young age group (agegroup2==0), and i got the result

HTML Code:
.  margins bmicatasia, at(fobp2==0 sexr==0 agegroup2==0) post

Adjusted predictions                            Number of obs     =    1,904
Model VCE    : OIM

Expression   : Pr(diabetes), predict()
at           : agegroup2       =           0
fobp2           =           0
sexr            =           0

    
Delta-method
Margin   Std. Err.      z    P>z     [95% Conf.    Interval]
    
bmicatasia 
Normal     .0109534   .0029288     3.74   0.000     .0052131    .0166937
Over        .0325897   .0094497     3.45   0.001     .0140686    .0511108
Obese     .0809608   .0205576     3.94   0.000     .0406686    .121253
but then , i wanted to check the incidence by the BMI in that specific group (normal Blood pressure (fobp2==0), female (sexr==0), and young age group (agegroup2==0)) , and i got the confusing result

HTML Code:
.    tab diabetes    bmicatasia if    fobp2==1    & sexr==0    &    agegroup2==0

        BMI    Asia
    diabetes    Normal    Over    Obese        Total
                        
    0                   10          7            5        22 
    1                 0           0           1        1 
                        
    Total              10            7         6        23
You can see that in normal and overweight, nobody developed the diabetes but the -margins- command gave me the result


My question is:
1. Did i do the right thing to calculate probability?
2. Is that possible to have estimated probability from an empty cell?

I really appreciate your help on this, since my main goal is to develop the risk score using probability.

Thank you very much