Teaching Statistics in the 21st Century
The late 1980s saw a wave of new undergraduate programs launched in computational physics, as the advent of affordable workstations and PCs made the power to compute and simulate more accessible. A decade later, though, many of those programs had drastically scaled back their ambitions or quietly wound down. The problem wasn’t the programming: the problem was that whenever a curriculum is designed as “X plus some Y”, it’s the Y that gets cut when time runs short, budgets are squeezed, or tough hiring decisions need to be made. “Computational physics” became “the physics we’ve always taught, but with assignments on computers” and then just “the physics we’ve always taught”.
That experience is part of why I’m so excited by things like Daniel Kaplan’s 2017 paper “Teaching stats for data science”, which is a great example of how some faculty are re-thinking pedagogical approaches from the ground up. Kaplan argues that much of what we currently teach in introductory stats courses is left over from a time when data was scarce and calculation was hard. In its place, he advocates a ten-step calculation-first approach:
- Data tables
- Data graphics
- Model functions
- Model training
- Effect size and covariates
- Displays of distributions
- Bootstrap replication
- Prediction error
- Comparing models
- Generalization and causality
UBC’s Stat 545 course is another great example of how people are not just putting old wine in new bottles, but approaching their subject from an entirely new angles. If you have any favorite examples, please add them to the comments–I’m sure our community would enjoy hearing about them.
Update: several people have pointed us at George Cobb’s The Introductory Statistics Course: a Ptolemaic Curriculum? and William Wood’s Innovations in Teaching Undergraduate Biology and Why We Need Them as other examples of curriculum being re-thought from the ground up. Please keep them coming.