This study seeks to demonstrate a method for treatment effect estimation in a multisite observational study where the treatment is highly selective and the assignment mechanism varies across sites. The method is demonstrated by addressing three primary research questions about the effect of 8th grade algebra: (1) For students who take algebra in 8th grade, what is the average effect of taking algebra in 8th grade on algebra achievement by the end of 9th grade?; (2) Does the average effect vary across students with different levels of demonstrated 7th grade mathematics achievement and propensity for taking 8th grade algebra?; and (3) Does the average effect vary across classrooms and schools? Through these three research questions, the author focuses on preprocessing the data with propensity score matching and imputation of the counterfactual, and on the exploration of treatment effect heterogeneity with multilevel modeling. This paper recognizes, but does not directly address, the importance of sensitivity analysis. (Contains 3 tables and 2 figures.)