Accurate representation of wind power plants (WPP) in both offline and online power system stability studies has gained importance because of the rapid increase in installation of wind generation around the world. However, because increases in accuracy reduce computational efficiency, reduced WPP representation is preferred. To improve accuracy of the reduced WPP model, other authors have proposed changes in structure of the reduced model for large WPPs. Another alternative is model validation and calibration. In this paper, we compare accuracy improvements brought by changes in model structure with accuracy improvements brought by validation and calibration of the reduced WPP model with minimum changes in structure. We illustrate our findings using a 168-machine WPP connected to the IEEE 39-bus test system. The parameters of the reduced WPP model are either calculated with current equivalencing techniques or validated and calibrated against a more accurate model. The changes in structure of the reduced model are one-machine or two-machine reduced models connected with a one-line or two-line collector equivalent. We show that the most accurate response is obtained by calibrating the parameters of the reduced model with minimum changes in structure.