How should we understand the OLS estimates when some regressor
is binary? When we have more than one dummy variable as regressors,
how should we interpret the coefficient associated with their
interac