An illustration is given which suggests that important specification issues can arise in interactive random-coefficient regression models. These issues relate to the generality of the model, translation of results, and forecasting. A class of specifications is suggested which resolves some of these issues.