Abstract:
|
Multicollinearity is a situation often faced by practitioners when modeling a linear regression. For students of basic Statistics this phenomenon is a challenging concept to visualize and understand, especially its consequences on confidence intervals for the regression coefficients.Combining statistical simulations and game-based learning with R and the Shiny package, a web app was developed to smooth the way into teaching the influence of multicollinearity. The game is divided in two parts, allowing the student to explore the concept in a user-friendly interface.Different sample sizes can be chosen to be plotted with smoothed lines showing how the width, on the vertical axis, changes accordingly to the correlation coefficient, on the horizontal axis. Once the concept has been mastered, it gives freedom to try personalized examples. Four different interval widths can be generated, according to sample size, number of predictors and the correlation coefficient as inputs. The game can be used to challenge the students to arrange provided intervals and the answer can be verified in the app. It takes the concept beyond a simple explanation and helps to practice it and have fun learning
|