Randomized experiments on social networks pose statistical challenges, due tothe possibility of interference between units. We propose new methods forestimating attributable treatment effects in such settings. The methods do notrequire partial interference, but instead require an identifying assumptionthat is similar to requiring nonnegative treatment effects. Network or spatialinformation can be used to customize the test statistic; in principle, this canincrease power without making assumptions on the data generating process.
展开▼
