Demonstrator Guide
Quick reference for teaching assistants
Your role
You are a circulating helper during application sessions (and some content sessions). Your job is to unblock students, not to do their work. The goal is for students to leave each session having done something real — loaded data, made a plot, run a test, committed code.
Golden rule: Ask a question before giving an answer. “What have you tried?” and “What error are you getting?” are almost always the right first response.
Weekly summary
Phase 1: Biomass Mini-Project (Weeks 1–4)
| Week | Content session | Application session | Your focus |
|---|---|---|---|
| 1 | Scientific method, testability | Data exploration + Git setup | Git is the priority. Every student must leave with a working repo and one commit. Cut exercise time, not Git time. |
| 2 | Descriptive stats, ggplot2, “Is that a big number?” | Multi-fuel figures | Help with ggplot2 syntax. Common issues: wrong column names, forgetting library(ggplot2), wide-vs-long confusion. |
| 3 | Uncertainty, hypothesis logic, assumptions | Scenario analysis (transport, supply chain, payback) | Mostly arithmetic. Help students who are stuck on the R code, but the thinking matters more than the code. |
| 4 | Writing for policy, “How to lie with statistics”, exemplar reveal | Write briefing → peer review → traitor reveal | Do not reveal who the traitors are. Help with writing and GitHub Issues. Keep the energy up for the reveal. |
Phase 2: Group Projects (Weeks 5–10)
| Week | Content session | Application session | Your focus |
|---|---|---|---|
| 5 | Experimental design, confounders, sampling | Group formation, research plan, Git branching | Help with Git branching and PRs. Expect merge conflicts — walk groups through them. Push indecisive groups to choose a topic. |
| 6 | t-tests, assumptions, base rate fallacy | Groups apply t-tests to project data | Many groups will spend most of the time on data cleaning. That’s normal. Help them get to a testable state. |
| 7 | ANOVA, multiple comparisons, effect sizes | Deepen analysis, effect sizes | Not all groups need ANOVA. Help two-group teams deepen their t-test interpretation instead. |
| 8 | Linear regression, diagnostics, extrapolation | Fit and critique models | Check that groups are looking at diagnostic plots, not just R² and p-values. |
| 9 | AI, reproducibility, integrity | AI-assisted review + report writing | Students use AI to critique their own work. Your role: help them evaluate AI feedback critically, not just accept it. |
| 10 | Report structure, peer review, common pitfalls | Final draft → peer review → revise | Similar to Week 4 but higher stakes. Encourage honest, specific reviews. |
Common student problems
Git / GitHub Desktop
| Problem | Fix |
|---|---|
| “Repository not found” | Check they accepted the GitHub Classroom invite and are signed in to the right account. |
| Can’t push: “Authentication failed” | Sign out and back in via GitHub Desktop → File → Options → Accounts. |
| “Nothing to push” | They forgot to save the file before committing, or committed but didn’t push. |
| Merge conflict | Open the conflicting file, look for <<<<<<< markers, choose which version to keep, save, commit. |
| Accidentally committed to main (Phase 2) | Not a disaster. Help them create a branch retroactively if needed, or just move on. |
See also: Git troubleshooting guide on the course site.
WebR / Quarto Live
| Problem | Fix |
|---|---|
| Exercises not loading | Hard refresh (Ctrl+Shift+R). If still broken, try a different browser — Chrome and Edge work best. |
| “Start Over” doesn’t reset hints | Known issue. Refresh the page to fully reset. Code persists via localStorage. |
| Data not loading in exercises | Check the student is running the data-loading block at the top of the page first. |
R code
| Problem | Fix |
|---|---|
Error: object 'x' not found |
Variable wasn’t created — check spelling, check the assignment ran. |
Error: could not find function "ggplot" |
Forgot library(ggplot2). |
| ggplot is blank | Check aes() mappings. Most common: aes(x = year, y = coal_twh) not aes(year, coal_twh). |
pivot_longer confusion |
Wide → long: pivot_longer(cols = ..., names_to = "fuel", values_to = "twh"). Show them the before/after shapes. |
When to help vs. when to wait
Help immediately if:
- Git is broken and the student can’t commit (this blocks everything)
- A student is visibly frustrated and stuck on a technical issue
- An error message is genuinely confusing (not just “read the message”)
Let them work through it if:
- They haven’t read the hints yet
- The error message tells them what’s wrong (encourage them to read it)
- They’re trying to get the “right answer” without understanding why
- They’re comparing with a neighbour — that’s peer learning
Challenge them if:
- They’ve finished early: “What’s your effect size?” / “What assumptions are you making?” / “Can you make that figure tell a misleading story?”
- They’re reporting a p-value without context: “Is that a big number?”
- Their conclusion is stronger than their evidence warrants
HolmesCo scenarios
Several sessions use handouts from HolmesCo, a fictional geological consultancy with bad research habits. The handouts are designed to look like real consultancy reports. Student versions (without instructor notes) are on the course site. Staff versions with expected answers are in the Staff portal.
| Scenario | Week | What’s wrong |
|---|---|---|
| Site investigation | 5 | Selection bias — boreholes in valley, turbines on ridgetops |
| Gold assay | 6 | Base rate neglect — 50 “hits” are almost all false positives |
| Mineral survey | 7 | Multiple comparisons — 3 significant minerals from 400 tests |
| Report card | 7 | Conflates significance with importance; ignores effect sizes |
| Groundwater | 8 | Extrapolation failure; missing seasonal term in regression |
Key phrases to use
These are the course’s running refrains. Use them when circulating:
- “Is that a big number?” — whenever a student reports a number without context
- “Compared to what?” — whenever a comparison is missing
- “What assumptions are you making?” — whenever a test is run without checking
- “How plausible was this before you tested?” — after any hypothesis test
- “What is the model not capturing?” — when working with regression
- “What would HolmesCo do?” — shorthand for every bad practice
Session plans
For full session-by-session detail with timings and facilitation notes, see: