Dear Statalist

I'm trying to identify influential observations in a bivariate logistic regression using Peribon's delta beta seeing that this is the appropriate measure for logistic regression.
FYI: I'm using paneldata

Onset:
- 0= Peace
- 1= Civil war onset

v2x_regime_lag (lagged one year)
- 0= Closed autocracy
- 1= Electoral autocracy
- 2= Electoral democracy
- 3= Liberal democracy

Code:
logit onset i.v2x_regime_lag if estimationssample2==1
predict db_model1, dbeta
gen casenum=_n
scatter db_model1 year, ml(casenum)
Code:
. logit onset i.v2x_regime_lag if estimationssample2==1

Iteration 0:   log likelihood =  -999.0105  
Iteration 1:   log likelihood = -976.58866  
Iteration 2:   log likelihood = -973.41285  
Iteration 3:   log likelihood = -973.37284  
Iteration 4:   log likelihood = -973.37278  

Logistic regression                             Number of obs     =      6,724
                                                LR chi2(3)        =      51.28
                                                Prob > chi2       =     0.0000
Log likelihood = -973.37278                     Pseudo R2         =     0.0257

--------------------------------------------------------------------------------
         onset |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
---------------+----------------------------------------------------------------
v2x_regime_lag |
            1  |   .3910203   .1503304     2.60   0.009     .0963781    .6856626
            2  |  -.0252029   .2048356    -0.12   0.902    -.4266734    .3762676
            3  |  -1.505673   .3351286    -4.49   0.000    -2.162513   -.8488328
               |
         _cons |  -3.322635   .1072928   -30.97   0.000    -3.532925   -3.112345
--------------------------------------------------------------------------------
I have created a scatter with variable "db_model1" (Preibon's dbeta values) along the y axis and the variable "years" on the x axis, however, around 50% og the observations have values over 0.2. To be exact they have a value of 5.90e+19 which is way more than the cutoff on 0.2. Is is not possible to run a Peribon's delta beta with a bivariate model using af dichotomous and a categorical variable? I works perfectly fine when I add control variables I only have trouble with the bivariate model.

//Marco Liedecke