3 Teaching Pedagogy
3.1 Evidence-Based Approaches
This course is built on evidence-based pedagogical principles for teaching R and data science effectively.
3.2 Adopt a “Tidyverse-First” Approach
The “tidyverse” ecosystem (including packages like ggplot2, dplyr, tidyr, and readr) is designed to be “human-centered,” prioritizing the cognitive needs of the analyst over the computer’s convenience.
Consistency and Readability: The tidyverse shares a common high-level design philosophy and low-level grammar. This means that once a learner masters one package, that knowledge transfers easily to others.
Human-Readable Code: The syntax uses intuitive verbs (e.g., filter, mutate, select) and the pipe operator (|> or %>%) to structure code as a linear sequence of actions, making it easier to read and write. This readability helps learners focus on the analysis logic rather than obscure programming syntax.
“Batteries Included”: Installing the single tidyverse meta-package provides a comprehensive toolkit for the core data science tasks—import, tidy, transform, visualise, and program—reducing the friction of managing dependencies.
3.3 Prioritise Data Visualisation
Visual feedback is a powerful motivator for learners.
Start with Visualisation: Introducing ggplot2 early allows learners to create complex, compelling graphics with relatively little code. This provides immediate gratification and helps learners spot patterns or errors in their data quickly.
Layered Complexity: ggplot2 implements a “grammar of graphics,” allowing learners to build plots layer-by-layer. They can start simple and progressively add complexity (e.g., mapping additional variables to colours or shapes, or adding smooth lines) as their understanding grows.
3.4 Centre Learning on Real-World Case Studies
Traditional statistics courses often focus on theoretical properties using clean, toy datasets. Effective data science education brings applications to the forefront.
Problem-First Pedagogy: Organise learning around diverse case studies where the primary goal is answering a specific subject-matter question (e.g., “Is there gender bias in grant funding?” or “Who will win the election?”).
Real, Messy Data: Use real-world data rather than pre-cleaned examples. This exposes learners to the reality of “data wrangling” and the necessity of cleaning and structuring data before analysis.
The “Whole Game”: Structure exercises to mimic the full lifecycle of a data scientist’s work: formulating a question, gathering data, cleaning it, analyzing it, and communicating results. This teaches students to “connect” subject matter to statistical frameworks.
3.5 Integrate Literate Programming and Reproducibility
Learners gain confidence when they can document their thought process alongside their code.
Literate Programming: Tools like Quarto and R Markdown allow users to interleave narrative text, code, and output (plots/tables) in a single document. This supports rigorous thinking by encouraging learners to record why they performed an analysis, not just how.
Reproducible Workflows: Teaching reproducibility early—using R Projects, relative paths, and version control (Git/GitHub)—helps learners manage their environment and ensures their work can be verified and built upon.
Automated Reporting: Using tools like Quarto avoids the error-prone “copy-paste” cycle of moving results into Word or PowerPoint. This establishes a robust workflow where updating data or code automatically updates the final report.
3.6 Foster Active Learning and “Statistical Thinking”
Live Coding and Iteration: Instructors should write code live (using literate programming tools) to demonstrate the iterative process of analysis, including how to troubleshoot errors in real-time.
Three Key Skills: Effective instruction should aim to develop three core skills:
- Creating - Formulating questions and data
- Connecting - Linking questions to appropriate data and methods
- Computing - Using tools like R to execute the analysis
Minimizing Notation: When introducing statistical concepts, relying on computational approaches (like simulation and visualisation) rather than heavy mathematical notation can make abstract concepts more accessible.
3.7 Learning Science Principles Applied
3.7.1 Cognitive Load Management
- “Whole Game” first reduces extraneous load (see the forest before the trees)
- Consistent patterns (load → transform → visualise) reduce cognitive overhead
- Helper functions reduce syntax complexity
- Spaced practice via homework spreads learning over time
3.7.2 Active Learning
- Live coding with immediate practice (not passive lectures)
- Minimal lecture time (< 30% of each session)
- Real-world application from Session 1
3.7.3 Scaffolding
- Pre-installed software eliminates setup friction
- Helper functions abstract complexity initially
- Example datasets are clean initially, messier in later sessions
- Gradual release: instructor demo → guided practice → independent work
3.7.5 Metacognition
- Reflection time in each session wrap-up
- Debugging strategy explicitly taught
- Error messages explained (not feared)
3.7.6 Transfer of Learning
- All practice uses real WECA data/project structure
- No toy examples after Session 1
- Authentic task design (they build actual deliverables)
3.8 References
This pedagogy is informed by:
3.7.4 Social Learning