Hi, I am analyzing different stages of conflict between states. Each episode has an attacking and an attacked state. My dependent variable (DV) is whether the attacking state gets what it wants (Success=1). I am running a heckprobit model to account for different escalation stages of conflict. In one version of my model, I added fixed effects (FEs) for the attacked state to account for its time-invariant characteristics.

My code looks like this:

Code:
heckprobit Success x y i.attackedstate, select(StageTwo= x y c.DurationToStageTwo##c.DurationToStageTwo##c.DurationToStageTwo i.attackedstate) vce(robust)

The dataset does not include successful and unsuccessful attacks for each attacked state. In other words, in several cases, there is no variation in the DV for the FE units. As a result, STATA 17.0 SE tells me:

Code:
note: 94.attackedtstate != 0 predicts failure perfectly;
      94.attackedstate omitted and 3 obs not used.

Even though STATA tells me that it omits and drops several observations when I add the FEs (which makes sense), the number of observations does not change. In other words, STATA tells me it drops observations but doesn't seem actually to do so. How come? I also noticed that STATA no longer computes the Wald chi2 value when I add the FEs. Does this suggest there may be a problem?

Thank you very much for your help!