A Bayesian Network for School Performance Massoud Moussavi * Causal Links, LLC Noel McGinn † Causal Links, LLC Abstract This paper describes a Bayesian Network model to diagnose the causes of low effec- tiveness of certain schools. Our aim is to build tools to assist policymakers in educa- tion to think through a policy, evaluate var- ious scenarios, and choose among competing policy options. These tools would help de- cision makers to make their tacit knowledge more explicit, and assimilate and systematize information from other sources. The model we describe has two potential uses: the ex- planation of learning outcomes in terms of conditions and processes within schools that are difficult to observe directly; and the esti- mation of the probability that a given inter- vention will affect those conditions and pro- cesses and hence learning outcomes. We be- lieve that models of this kind can be effective aids in making decisions, and in learning from them. 1 INTRODUCTION In the last two decades, a growing body of research has focused on understanding and developing causal as well as diagnostic models and their applications to policy analysis. (Pearl 2000; Hausmann, Klinger, and Wagner 2008). Our approach to development policy analysis is to build on these advances, develop the- oretical models, and link them to a specific body of development knowledge. Specifically, we aim to de- velop a diagnostic model of school effectiveness or qual- ity. That is, we focus on identifying major factors that cause an education system to be ineffective. The underlying framework for our approach is to build a * moussavi@causallinks.com † mcginn@causallinks.com model that represents the teaching and learning pro- cess as it takes place in schools. The work requires synthesizing results of empirical research and field ex- perience, consistent with advances in learning and or- ganizational theory. Bayesian modeling approach is particularly suited for integrating and synthesizing in- formation from a variety of sources. Using a Bayesian network approach model as our in- ference engine, we have built a software tool, called Policymakers Workbench. Through this tool, policy- makers can run unlimited iterations of various policy scenarios and connect existing knowledge to effective action. Policymakers Workbench has two main com- ponents: a “theoretical engine” and a computer user interface. The theoretical engine or the model is built upon a set of underlying relations or functions, such as “if attendance goes up, achievement goes up, but only if there is time on task”. This model is elaborated as a Bayesian belief network that specifies dependencies among the variables, which are based on the results of previous research, direct experience, and the tes- timony of experts and stakeholders in the particular domain or field. The computer interface enables users to evaluate the effect of changes in specific variables; perform diagnostics; and analyze the impact of vari- ous interventions or policies. A brief description of the user interface is given in Appendix 1. While much of the discussion in this paper is focused on education, the framework and the tool that we dis- cuss is general and could be employed in policy analy- sis in other development areas, such as health, the en- vironment, economic growth, migration, and so forth.