Evaluating Recipes for Development Success

Evaluating Recipes for Development Success Avinash Dixit This article offers a provocative critique of the ability of research on the impact of institutions on growth to offer immediate and practical recommendations for reforming and redesigning institutions in developing countries and transition economies. The article suggests a Bayesian diagnostic procedure to identify the causes of economic failure in an individual country as a first step toward remedying the failure. The main purpose of the most scholarly research, both theoretical and empirical, is to improve our understanding of the phenomena and processes being studied. In the concluding section, I suggest a framework or methodology of research that combines general conceptual and empirical findings from academic research and the experience of practitioners to help narrow or identify the causes of failures in individual countries. Besley and Burgess (2002), using panel data from India, find that an informed and active electorate leads to effective incentives for governments to respond to economic problems and that mass media play an important part. Acemoglu (2003) argues that the lack of third-party enforcement in political contracts makes it harder to make credible commitments, and that this explains the absence of a Coase theorem ensuring efficient outcomes in political bargaining. Finally, the theoretical literature, using a repeated-game framework, shows how a partial improvement of an imperfect formal system, by providing a better outside alternative and thereby lessening the harmful consequences of breaking a relational contract, can worsen the outcomes of the informal system (Baker, Gibbons, and Murphy 1994; Dixit 2004). They find that a country's initial conditions are more important than policy changes in determining its economic performance during the first few years of transition; that is, whether the reforms are rapid or gradual is less important. Pmn Avinash Dixit 151 If we observe a particular effect, say E7, then the Bayesian posterior probability that a particular cause, say C5, is present becomes p P Pm 5 5;7 : i 1 pi Pi;7 If we want to be nearly certain whether a cause, say C5, is present, we need to find an outcome, say E7, which will more typically be a cluster of outcomes or symptoms and might be called a "syndrome," such that It is very unlikely to occur when the underlying cause is any other cause, that is, Pi7 is close to It is very likely to occur when C5 is present, that is, P5,7 is close to one, so the rest of the P5,j's are close to zero, and if some other effect is observed, the posterior probability of C5 becomes close to zero.