This paper investigates how local jurisdictions in a federal system influence each other in the adoption of policy innovations. We look at school districts in Michigan and their participation in a public school choice program launched in 1996. Districts' participation decisions are modelled as simultaneous discrete choice decisions using a spatial latent variable model. Strong effects are found saying that lagged adoptions of neighbors positively affect the current probability of participation. This finding is robust to various changes in specification. The results suggest that in federal systems the diffusion of policy innovations is stimulated by horizontal interactions between jurisdictions.