Identifying the effects of the New York State Experimental Prekindergarten Program on the development of children requires that a broad array of data be analyzed. In studying the effects of the program, the evaluation unit is interested in determining whether there is an overall effect, and whether children of different ages and different levels of initial development are affected in the same way. An illustration is given to clarify how the process of hypotheses testing reduces the number of assumptions one has to accept in drawing conclusions about the program effects. The analysis of covariance procedure illustrated involves 18 possible hypotheses; however, the branching nature of the analysis results in the testing of no more than six hypotheses. The result obtained in testing a given hypothesis leads to another hypothesis while simulataneously precluding other hypotheses. Thus, efficiency of design is achieved by considering a large number of hypotheses while testing only a limited number of them. At the same time, the design makes it possible to retain the most general hypothesis which adequately explains the data. (Author/MV)