In this article, the author shares his comments on statistical analysis for multisite trials, and focuses on the contribution of Stephen Raudenbush, Sean Reardon, and Takako Nomi to future research. Raudenbush, Reardon, and Nomi provide a major contribution to future research on variation in program impacts by showing how to use multisite trials to estimate this variation. This is important because (a) variation in program impacts can have major consequences for policy, practice and research; (b) there is considerable current interest in studying this variation; (c) there are a growing number of multisite trials from which to do so; and (d) appropriate statistical methods for this work are lacking. The authors' main contribution is to extend the use of instrumental variables (IV) analysis with randomized trials or regression discontinuity designs to account for cross-site variation in both compliance with program assignment ([gamma][subscript s]) and the impact of program participation on the outcome of interest ([delta][subscript s]). This flexible extension of IV brings with it a new problem--bias from cross-site "covariance" between [gamma][subscript s] and [delta][subscript s]--which the authors call "compliance-effect covariance bias." To make their analysis framework operational, the authors present three approaches that vary in the way they examine the relationship between the effect of program assignment on outcomes ([beta][subscript s]) and compliance with program assignment ([gamma][subscript s]). (Contains 5 footnotes.)