This paper discusses methods for model reduction of power system dynamics. Dynamical models for realistic power-systems can very easily contain several thousands of states. The dimensionality increases further when considering the dynamics of distributed energy resources; these systems are typically smaller in power rating, so many more are installed at the grid edge to scale capacity. Computationally efficient models that capture the dominant modes of the system are important for all aspects of power-system operation, control, and analysis. In this paper, we analyze two data-driven methods for model reduction of power systems: I) proper orthogonal decomposition, which is based on singular value decomposition, and ii) a constrained convex-optimization framework with stability guarantees. Advantages and disadvantages of both of these methods are discussed. Exhaustive numerical simulations for a low-inertia system with mixed synchronous generator and wind energy conversion system resources are provided to verify the accuracy of the model-reduction methods.