Logit/Probit fixed effects

[Chris_Ger]

Hello everybody,

currently I am modeling fixed effects for panel binary data based on dummy variables. The program works fine. Only for some datasets, Eviews displays the error message: "Quasi complete separations Dummy_18>0 perfectly fits binary responses in equation ..."

1.)
I checked the Eviews helpfile and it says: "This error occurs when one of the regressors contains a separating value for which all of the observations with values below the threshold are associated with a single binary response, and all of the values above the threshold are associated with the alternative response".
That means my dependent binary variable equals the dummy variable. The dummy variable identifies specific datasets, for example several responses from a respondent in a survey. So I would expect the error, when the dummy equals one and the binary variable also equals one for all observations marked with the dummy. But this is not the case as the binary data varies between 0 and 1 while the dummy is one for several datasets. Can anyone explain the error text, so that I can look for a solution?

2.)
For a specific binary model Eviews calculates the coefficients, but displays NAs for std.error, z-statistic and prob. What went wrong?

Thank you very much and best regards,

Chris

[EViews Glenn]

In the first case it's hard to say without looking at your dataset. You could post it along with the specification in question.

In the second case, it's probably a singularity. Does the equation output provide any indication that this is the case.