Analysis of cluster-randomized experiments: A comparison of alternative estimation approaches

Analysts of cluster-randomized field experiments have an array of estimation techniques to choose from. Using Monte Carlo simulation, we evaluate the properties of point estimates and standard errors (SEs) generated by ordinary least squares (OLS) as applied to both individual-level and cluster-level data. We also compare OLS to alternative random effects estimators, such as generalized least squares (GLS). Our simulations assess efficiency across a variety of scenarios involving varying sample sizes and numbers of clusters. Our results confirm that conventional OLS SEs are severely biased downward and that, for all estimators, gains in efficiency come mainly from increasing the number of clusters, not increasing the number of individuals within clusters. We find relatively minor differences across alternative estimation approaches, but GLS seems to enjoy a slight edge in terms of the efficiency of its point estimates and the accuracy of its SEs. We illustrate the application of alternative estimation approaches using a clustered experiment in which Rock the Vote TV advertisements were used to encourage young voters in 85 cable TV markets to vote in the 2004 presidential election.