it is difficult to move from one level of aggregation to another (for example, to infer the effects of a large-scale program from results of a demonstration program in a few schools). Christopher Elias expressed similar doubts about the generalizability of results of impact evaluations, since so much depends on how the programs were implemented.
Amy Tsui summarized questions that had recurred in the discussion of seemingly disparate topics: What are the behaviors that are of most importance for improving reproductive health? What are the effective interventions, and how feasible and affordable are they? And how confident should we be that we know the answers to these questions? Tsui suggested that it would be important to get more information on sexual behavior as an antecedent to risky behaviors. Research on all types of services is needed to help define the terms of the trade-offs among quality, effectiveness, and coverage. She concluded the meeting on an optimistic note. The contraceptive revolution has shown that large-scale changes in ideas and behavior can be achieved in a matter of a few decades, much shorter periods than anyone dared hope in the 1960s. Achieving many reproductive health goals may require a more complex mix of services than was the case with early family planning programs, greater efforts at generating demand, and the ability to sustain services over long periods, but there are no grounds for thinking these changes infeasible.