Students
whose utility gain is somewhere between
the two extremes should be more likely to respond.
To control for such heterogeneity, we
estimate a conditional logit model with individual
fixed-effects. The average effect, therefore, is
not very representative and its estimation comes
with a large standard error.