Simulating policy diffusion through learning: Reducing the risk of false positive conclusions