Kit Clement (University of Illinois Urbana-Champaign), Jennifer Noll (TERC), Elen Huang (University of Illinois Urbana-Champaign), Luke Thorell (University of Illinois Urbana-Champaign)
Abstract
Background. Research has advocated for simulation-based inference to promote understanding statistical inference at a deeper conceptual level than the consensus curriculum (Chance et al., 2018, 2022; Tintle et al., 2012, 2014). While many simulation-based curricula leverage applet-based tools, there is some evidence that supporting students to build probability models to perform simulations may yield more success in students’ understanding of the purpose of simulation and its relationship to statistical inference (Hildreth et al., 2018). Other research has argued that modern statistics requires computational fluency (Çetinkaya-Rundel & Rundel, 2018). Yet, there is little research that investigates pedagogical approaches to scaffolding student learning from applet-based simulation tools or visual modeling and simulation tools (e.g., TinkerPlots or CODAP) to developing programming skills with widely-used computational tools (e.g., R). This study aims to develop a research base for understanding how to use applets or visual modeling tools to support learners' development of computational skills. The work compares students' understanding of simulation from one introductory statistics course that used applets and another that used model-based simulation. We focus on whether (and how) students can leverage their experience with simulation to read and understand R code that performs the same simulation.
Methods. Surveys were conducted on two classes of introductory statistics: one that used a modeling-based approach to simulation using TinkerPlots, and one that used applet-based simulation from Art of Stat. The surveys used part of the Simulation Understanding in Statistical Inference and Estimation instrument (Brown, 2021) to assess students' understanding of the purpose of randomization and simulation in a hypothesis test of two independent means. A task-based interview will be conducted with select students of comparable backgrounds and confidence in statistics/programming to determine their fluency with these simulations and interpreting R code that conducts a similar simulation.
Findings. Analysis of surveys is currently in-progress, and we are still identifying interview participants. However, we hope to have interesting results to show about how the two classes compared on the survey and how the interviewed groups of students from each class fared on the task-based interviews.
Implications For Teaching and For Research. If results from this study show positive outcomes for students that used TinkerPlots in their ability to read and understand R code for simulations, this may motivate the development of
R-based software that would allow students to build simulations without coding, as already done in TinkerPlots, but also provide side-by-side code that carries out their simulation. This could provide an environment for students to learn important coding structures like for-loops or conditional statements that are common to algorithms used in statistical simulations, and allow for students to move toward building more advanced simulations that go beyond a TinkerPlots-like UI.