I am running an ordered logistic regression for my thesis. I am trying to test the relationship between counterterrorism aid and state repression levels in recipient countries. I ran my regression, and for my first model, my International War control variable has a coefficient-value of 38.007 and a very large standard error value (158,641,489.000). This seems very wrong to me, but I do not know what the problem is.
The independent variable of International War is binary 0 = the given country was not part of an international war in the given year, 1 = the given country was part of an international war in the given year.
the dependent variable is ordinal, ranges from 0 to 2. O meaning there were many state sponsored disappearances and 2 there were none.
Does anyone know how this is possible and what it means? Did I do something wrong with my data? 
Asked
Active
Viewed 34 times
0
EmStaLo
- 41
-
3Sounds like it might be perfect separation. – EdM May 18 '23 at 12:32
-
Why do you give three values (158,641,489.000) and what do they mean? – Christian Hennig May 18 '23 at 12:40
-
@ChristianHennig I think that means 159 million (or so). – Nick Cox May 18 '23 at 12:42
-
Rather than perfect separation I'd guess that the matrix of predictors may be near to ill-conditioned. It may be almost perfectly possible to reconstruct the International War variable from the other predictors, in which case the regression contribution of it will have a very large standard error and potentially a strange looking coefficient estimator. – Christian Hennig May 18 '23 at 12:43
-
@ChristianHennig what do you mean by the matrix of the predictors may be near to ill-conditioned? I am sorry, but I do not know what that means – EmStaLo May 18 '23 at 12:45
-
1I explain it in the then following sentences. – Christian Hennig May 18 '23 at 12:45
-
@ChristianHennig should I then leave the variable in the regression or best take it out? I just really do not know how I just go about it in my analysis part, I have never heard of perfect separation or ill-conditioned matrixes. – EmStaLo May 19 '23 at 06:55
-
Sorry, I can't make recommendations not knowing the data, aim of analysis, and further background. – Christian Hennig May 19 '23 at 09:31