Punctuated Equilibrium Theory, Policy Change and Neural Networks: Can We Go Any Further Beyond Boundaries?

Punctuated Equilibrium Theory (PET) has already secured its place as one of the “classic” approaches in studying policy. One of its main features is describing policy process as a parallel to information process. In this way, policy process starts with inputs (sources), than it goes through throughputs (decisions made according to bounded rationality scheme and with noise factors) and it results with outputs (incremental and punctuated policy shifts). This process’s logic is also quite similar to neural networks: here one has input layer (data), hidden layer(s) (processing units) and output layer (results). Having in mind the above assumptions, the aim of the paper is threefold. First, it contributes to bridging the divide between two disciplines: political science and computer science. PET’s and neural networks’ theoretical similarities and empirical evidence (budget data from Policy Agendas Project) make them well suited for a more detailed examination. Second, the paper aims at addressing some issues already investigated in literature. One of the most critical points in any empirical study is delivering the right data. Here, it is suggested that using budget authority data instead of budget outlays is one of possible extensions to previous research. The rationale for this is that budget authority reflects policy priorities more accurately than outlays – the latter are spent with a time lag of two years because in the US the first budget plan is presented two years in advance. Third, empirical issues are addressed. Namely, based on a trained neural network, the question of policy change is investigated. Yearly budget changes are studied via cross-validation, confusion matrix and ROC analysis techniques. Thus, the paper contributes to respectively: theoretical, methodological and empirical research on policy agendas and policy change. Last but not least, the argument is supported by as little formal notations as possible.