In this paper, variations in the impedance parameter of the conventional spatial interaction models are examined in a quasi-empirical manner. In a previous paper, a spatial demand model based on the translog indirect utility function was simulated to provide interaction data. If the inputs to the translog demand model are varied systematically, the impedance parameter estimates for both models also vary systematically. In general, the impedance parameter becomes less negative as the set of destinations becomes more dispersed, as relative size among the origins and the destinations becomes more varied, and, in particular, as fewer destinations become substitutable or the degree of substitutability decreases. Meanwhile, model performance also deteriorates. The need for skepticism regarding model transferability and parameter interpretation is confirmed, and the possibility of ever correctly specifying the models is questioned.