We study geographic political representation and geographic distribution of local public goods within local jurisdictions with at-large proportional representation elections. We use detailed geo-coded data on politicians, the electorate and elementary schools. Descriptive analysis reveals that poorer neighborhoods are under-represented and that local politicians have a strong support base in their home neighbourhoods. Based on randomized election outcomes due to personal vote count ties, geographic representation has a causal effect on school closures. The probability of closure is cut in half when a candidate living close to the school is randomly elected. High-income residents react to closures by moving away from the neighborhood, thus reinforcing segregation.