A method proposed by Matsunaga [Phys. Rev. Lett. 99, 238103 (2007)] is applied to simple stochastic models and two model proteins composed of 46 amino beads with three different kinds of residues. The method, which is based on the combination of the principal component analysis and the finite size Lyapunov exponent, characterize the coarse-grained dynamics in different spatiotemporal hierarchies in protein dynamics. The application of the method to model proteins reveals that the low-indexed (large-variance) principal components carry less-divergent, regularized dynamics at the coarse-grained scales on a less-frustrated energy landscape, whereas this less-divergent nature is less pronounced for a protein model with a more frustrated energy landscape. It is also revealed that our technique can differentiate the collective motions on the projected principal component space inherent to the system and the apparent collective behavior which can appear even in high-dimensional stochastic systems.