1: \begin{abstract}

2: We study the problem of estimating the average causal effect of treating every member of a population, as opposed to none, using an experiment that treats only some. This is the policy-relevant estimand when (for example) deciding whether to scale up an intervention based on the results of an RCT. But it differs from the usual average treatment effect in the presence of spillovers. We study both estimation and experimental design given a bound (parametrized by $\eta$) on the rate of decay of spillovers between units that are ``distant'' in a general sense, encompassing (for example) spatial settings, panel data, and some directed networks. We show that over all estimators linear in the outcomes and all cluster-randomized designs the optimal rate of convergence to the average global effect is $n^{-\frac{1}{2+\frac{1}{\eta}}}$, and provide a generalized ``Scaling Clusters'' design under which this rate can be achieved. Both these results are unchanged under the additional assumption (common in applied work) that potential outcomes are linear in population treatment assignments and when the estimator is OLS. We also provide methods to improve finite-sample performance, including a shrinkage estimator that takes advantage of additional information about the structure of the spillovers when linearity holds, and an optimized weighting approach when it does not.

3: 

4: \end{abstract}

5: