Computational Infrastructures for School Improvement: How to Move Forward R. Benjamin Shapiro, Hisham Petry, and Louis M. Gomez {rbs, h-petry, l-gomez}@northwestern.edu Learning Sciences, Northwestern University Abstract. The instructional practices common in today's schools reveal a disconnect between instruction and evidence of the effects of that instruction on student learning. In this paper, we propose the creation of computational infrastructures that will help teachers make more informed decisions in their practice. These infrastructures formalize student and teacher routines to facilitate data collection and mining, in order to create actionable information. We then show an instance of such a computational infrastructure and describe its potential for improving instruction. 1 Introduction Studies of today’s (American, and many European) schools reveal a disconnect between instruction and evidence of the effects of that instruction on student learning. New constructivist understandings of learning, knowledge, and effective instruction [11] challenge school reformers to instantiate formative assessment and instruction by reconstituting the daily routines of teaching into ones "where a teacher's day-to-day decision-making is instrumentally constructed based on the interaction of detailed observations about students' work in the classroom (and the personal background students bring to their work) and the aims in view for subsequent instruction" [2]. Transitioning the existing American system of education into a system of practice where this occurs could be made more likely by the development of new Computational Infrastructures for School Improvement (CISIs) that help teachers to use, and reflect on their use of, evidence to make decisions. This paper describes a rationale and framework for such architectures, gives a concrete illustration of the implementation of such an infrastructure, and shows how it provides an opportunity to apply Educational Data Mining (EDM) techniques to some of the most challenging problems that teachers face. This is not fundamentally an empirical paper. Rather, it is an argument for how we could organize the EDM field and the technologies it produces to have higher cumulatively and greater impact on educational practice and research. It is also not a claim about a specific technology; some of the ideas here about feedback and interactive support are manifested in some form in existing work, such as Interactive Tutoring Systems. Instead, we are trying to make a broader claim about how attending to the daily practices of teachers, learners, and school leaders could guide the design of interconnected technologies and practices, allowing EDM to have a much broader impact than it otherwise might. 2 Problem of Practice Teaching is a complex, difficult, and uncertainty-ridden job. As Higgens notes: One starting point for inquiry into the moral phenomenology of teaching is Philip Jackson’s famous observation that teachers make over 200 decisions per hour. If Donald Schon is right, all practices require ‘reflection in action’, but teaching takes the demand for improvisation to new levels. It is in large part this radical unpredictability of teaching which shapes its phenomenology. It does not seem to matter how many times one has taught. Each time one begins a class, there is that unique blend of excitement and dread