Integrating Computational Thinking in High School Statistics Through Data Modeling With R

• In Event: 41.049 - Stories From the Field: Integrating Computational Thinking Across Curricular Domains

Abstract

Purpose. The CodeR4STATS project studies the integration of R, a programming-based statistical analysis tool used in professional science and mathematics, into high school statistics. Throughout the year, interspersed with more traditional lessons, students complete projects in which they collect or are provided with large datasets that can only be tractably analyzed using computational methods. Within this context, we are exploring the codevelopment of statistics knowledge, computational thinking, and students’ learning about the systems from which the data are collected.

Theoretical Framework. Data analysis and statistics are best learned through the pursuit of rich, meaningful problems (Cobb & McClain, 2004; Davies, Barnett & Marriott, 2010). Increasingly, such problems are captured in complex data (Nolan & Temple Land, 2010), which require skills like cleaning, visualization, and modeling to be productively explored (Finzer, 2013; Schutt and O'Neil, 2013). CodeR4STATS aims to help students develop such skills by using computational tools to manage, manipulate, and analyze scientific and social datasets relevant to their own lives. While the use of R does not dramatically change statistical learning objectives, it does introduce computational learning objectives. Learners must think about data and control structures, and engage in modeling and problem decomposition in order to productively apply the statistical tools they are learning about.

Methods. Over the course of the school year, two statistics classes in an urban public exam high school (33% free or reduced lunch) completed an introduction to R for statistical computing and three projects analyzing: simple bivariate datasets, distribution and forecasting with student-generated engineering datasets, and multivariate college admissions datasets. We collected video of student groups’ discussions and screen activities, work completed for each unit, and assessment items. For this poster, our analysis concentrates on the last session, during which students created models to predict what students would be admitted to colleges with different mission statements. Using frameworks from the science, math, and statistics education literature (Blum & Leiß, 2007; Pratt et al, 2008; Schwarz, et al, 2009), we analyze learners’ treatment of model and data in the presence and absence of computation.

Results. Our preliminary analyses of learners’ approaches to modeling and data reveals a tension: When learners were not involved in computational aspects of the project, they engaged in robust modeling processes, describing college competitiveness in ways that reflected relationships within and across student records. However, when working to build computational models, some groups abandoned or simplified those descriptions. Deeper examination of learners’ code and treatment of data, suggests this disconnect stems from difficulties creating and translating models to code. We will present two cases to illustrate this, one in which tensions between model and code are clear, and another in which learners transitioned coherently from model to code.

Significance. Introducing computation into topics such as statistics is a necessity; however these introductions are not automatically “integrative”. Students may isolate computational activities as different types of reasoning. Understanding how learners’ approaches to problem solving in a given domain map and can be translated to available tools is critical for supporting productive computational thinking in disciplinary domains.

Authors

• Michelle Hoda Wilkerson, University of California - Berkeley

• Elise Deitrick, Tufts University

• Eric Simoneau