We analyze information diffusion by focusing on network structures. First, we propose a network growth model that produces networks with features required for analysis and perform a validation experiment using Twitter networks. The proposed model produces networks with features calculated from these real networks with high accuracy. Using this proposed model, we produce several networks that exhibit various features. We simulate information diffusion on these networks using an independent cascade (IC) model and calculate the Ability of Information Diffusion (AID). Second, we analyze how each feature affects information diffusion using this simulation. We found that the AID score was affected by the average shortest-path length L and variance of closeness centrality σι. We got a high AID score by a network with low L and σι.