The "Too Few Cases/Too Many Variables" Problem in Implementation Research

1F 5IRST GENERATION studies of implementation were, for the most part, detailed accounts of how a single authoritative decision was carried out, either at a single location (Derthick 1970; Pressman and Wildavsky 1973) or at multiple sites (Derthick 1972).1 Implementation was conceptualized as a complex and dynamic process, characterized by a variety of participants with a wide range of perspectives whose interpretation and operationalization of earlier authoritative decisions had multiple effects, some anticipated, others not. With few exceptions (e.g., Sapolsky 1972), most first generation authors were pessimistic about a program's chances of being implemented successfully.2 For example, in 1977, Eugene Bardach (1977: 5) observed that "even the most robust policy one that is well designed to survive the implementation process will tend to go awry. The classic symptoms of underperformance, delay, and escalating costs are bound to appear" (emphasis added). With the publication of additional empirical studies of implementation, however, the picture of implementation as inevitable failure began to fade.3 Whereas faulty execution may have been the rule, there emerged a more eclectic view of implementation, one that acknowledged its variability and stressed the political as well as managerial dimensions of implementation behavior.4 The lead, or orienting, questions of these second generation studies became, and still are, "Why was X so well implemented, particularly in comparison with the less successful implementation of Y?" (Mazmanian and Sabatier 1983: 3) or "Why was X so well implemented in state A, compared to state B?" (Abolafia 1979; Goggin forthcoming).