EViews.com

I would look for something more like a data error.

What exactly do you mean by data error?

Something like someone typed 10000001 instead of 1.0000001.

No, i did not find anything like that :(

Can I ask another question? I don't know who else to ask.

It is again something concerning dummy variables. When I only include the variable D1 in my equation this variable is highly signifikant. Then i want to perform an interaction with this dummy D1X1). I left the dummy variable in, as you recommend. Therefor i do have D1 and D1X1 in my equation. D1 is highly significant D1X1 is not. BUT if i leave D1 out of the regression and only leave D1X1 in then D1X1 is highly significant. How can this happen???
Everything was estimated using probit models.

That's about what you would expect if the interaction doesn't much matter. When you leave out D1, the interaction term tries to pick up both the interaction effect and the direct effect of D1 since the interaction is the only place D1 is allowed to enter. When both are in, the probit is allowed to properly attribute the effect of D1 rather than forcing it onto the interaction.

Thank you very much for all of your help!!!

I have another question regarding this subject.
I am looking at data from different schools and want to analyze when an incident occurs. Firtst i wanted to check, whether the size of the schools plays a role. The variable was negative and highly significant. Therefore this means: the smaller the school the bigger the probability of an occurence of the incident. Then i split the size of the school into 3 dummies: d1 small school, d2 middle, d3 big school. 1= incident, 0=no incident. I left d3 out as a reference category. The variable schoolsize is also left out.After running the regression d1 is negative and highly significant. D2 is not significant. This means, that small schools, in comparison to big schools, have a significant lower probability of having the incident. But how can thi.s two results be combined??? On the one hand the probability of the incident is higher the smaller the school and on the other hand the probability is bigger for big schools??

As no one seems to know an answer to my problem, let me ask a different question.

If I m looking at the school size and the estimation of the probit model gives me a coefficient of -2.7 and a p-value of 0.45.... does this mean, that the school size in general does not matter (meaning bigger schools as well as smaller schools have the same probability of an incident) or does it mean that smaller schools don't have a higher probability of an incident?