What makes a good question?

Research Skills — Week 1

Research Skills for Geoscientists

Research Skills for Geoscientists

Welcome!

• The scientific method
• Testable or not?
• Falsifiability
• Meet R
• The biomass question
• Wrap-up

• Before starting, queue up these pages:
  • BlackBoard - for reading list
  • More or less - required listening
  • The Economist - required reading
  • PollEv - Q&A
  • Attendance

Who am I?

I’m Martin Smith, and I’m a palaeontologist.

Hallucigenia sparsa, Burgess Shale, ~508 million years old.
Fossil: NMNH 198658, Smithsonian; photo Martin Smith.
Reconstruction: Danielle Dufault

Youti yuanshi: internal organs reconstructed from X-ray scans.
Model: Emma Long

• [11:05–11:06] Who I am. Cambrian animals: Hallucigenia (famously reconstructed upside down and back to front) and Youti, whose internal organs we could trace in 3D from X-ray scans.
• The point of these slides: the skills in this module travel. They matter whether you stay in research or not.

Testing hypotheses needs statistics

I test hypotheses, and I have to communicate the results to convince others. So I write R packages:

New methods for research

Mapping tree space with TreeDist.
Smith 2022, Systematic Biology, CC BY 4.0

Data visualization

Contoured ternary plot: Ternary

• [11:06–11:07] Being a scientist means testing ideas against data and showing your working. That needs statistics and clear figures.
• Tree space: how different are the trees that an analysis produces? Are they in clusters? TreeDist maps them so you can see.
• Ternary: three-part data (e.g. sand/silt/clay) plotted on a triangle. Written because nothing else did what I needed.

Skills for all of life

My spin-out, StratoBayes:

Our subcontractors use the same tool suite you will learn this term.

© StratoBayes; illustration by Kilian Eichenseer

• [11:07–11:08]

• StratoBayes correlates rock layers between boreholes, with quantified uncertainty. The developers we hired work in the same toolkit: R, version control, reproducible reports.

• And my PhD student was hoovered by by Atom Bank because of their R skills - and they only worked with fossils!

• Project management, understanding data, statistical thinking and clear writing are what employers look for.

• Across life - understand fitness, health, supplements, sleep…

In demand in industry

Research Skills for Geoscientists

This module is about learning to design experiments, and to communicate what their results really mean. This, in a word, is: Science!

• [11:08–11:13] Welcome. Set the tone: warm, rigorous, honest.
• This module isn’t about memorizing tests — it’s about developing habits of thought. By Week 10 you’ll have a toolkit of reflexive questions (“Is that a big number?”, “What assumptions are we making?”) that will serve you in dissertations and careers.

Two strands:

1. Statistics — how to test claims with data
2. Communication — how to present evidence so it can be trusted

This term: Quantitative data

Next term: Spatial data

How to succeed

20 credits = 200 hours = 10 hours per week.

Each week: Content session (2 h) + Application session (2 h)

• Content: I’ll break it up with exercises. We work best when we learn with others. Work with others on your bench; we’ll feed back to the class.
• Application: Work through the exercises; engage with the extensions. Different working speeds - but always room to be pushing yourself (and having fun?)

• Do. Per ardua ad astra!

• My mantras: Writing is thinking; thinking is understanding. Good work takes time: it’s in struggle that we learn.

No. 50 Squadron Lancasters, 1943.
RAF official photographer / IWM TR 1158, public domain

• Read (and listen)

• Show students how to find the reading list, More or Less and log into The Economist

• Ask

• For course questions: Point out the QR code - we’ll come back to that later
• For technical questions (code & data): Allude to (don’t show yet) the GitHub discussion board

• Use AI intelligently (if at all)

• My AI conclusion: Half of the skill of using AI is knowing how to ask the right question. The other half is knowing when to trust the answer. That’s exactly what this module will teach you.
• AI output looks convincing - until you really look at it.

Two phases

Phase 1 (Weeks 1–4)

Scaffolded mini-project:

“Does UK biomass power help reach net zero, and should we keep paying for it?”

Real data & ambiguity.

Culminates in a peer-reviewed policy briefing.

Phase 2 (Weeks 5–10)

Your investigation:

Choose a sustainable energy question. Design the study. Analyse the data. Write a 2,000-word policy report.

This is your summative assessment (30%).

• Phase 1 is formative: low stakes, high learning. Everyone works with the same data so you can compare notes.
• Phase 2 is summative: apply the same skills to a question of your choice. Groups of 3–4 for analysis; individual write-up.

Building on what you have already learned

GEOL1151 taught you to code in Python, use Jupyter, and work with data.

This module uses R instead of Python.

Why? R is built for statistics. It uses the same ideas and concepts, but with a different syntax. Pre-term preparation has helped you to make the jump.

• Reassure: if you’ve done the pre-term preparation, you’re ready. If you haven’t, the application session is your safety net.
• R and Python have similar philosophies - they are distinct tools, best for different tasks. Knowing a little of both gives you the freedom to choose the best tool for a task.

Questions?

Submit questions:

PollEv.com/geol

text geol to 07480 781235

• Open the Poll Everywhere Q&A activity before the session and leave it active throughout.
• Questions appear in your Poll Everywhere presenter view; upvoted questions rise to the top.
• Address during transitions or at the end of concept blocks, not mid-explanation.
• Reset (archive) the activity between sessions so questions don’t carry over from the previous week.

The scientific method

• Research Skills for Geoscientists

The scientific method

🎓 Concept block 1

• Testable or not?
• Falsifiability
• Meet R
• The biomass question
• Wrap-up

What is science?

“All science is either physics or stamp collecting” (attrib. Rutherford)
Penarc, CC BY 3.0

• Not a body of facts

• A process for reducing uncertainty
• Nothing is ever “proven”
  • Instead, we accumulate and evaluate evidence

• [11:13–11:30] This is the intellectual foundation for the course.
• Start here: science isn’t what you know, it’s how you decide what to believe. Every claim in this module will be held to this standard.

The cycle

w01-ppdac-ring.R

Problem → Plan → Data → Analysis → Conclusion

Drax power station. © Drax

• PPDAC. This is the scientific method as practised with data, and it is the spine of the whole module.
• Each stage is itself iterative - its own mini-cycle
• Walk through each stage with the biomass question, our running case. Drax first started biomass conversion in 2013.
  • Problem: is “biomass is carbon-neutral” a testable claim?
  • Plan: (Hypothesis + prediction) what would we measure, and what result would disprove it? What could confound us?
  • Data: DUKES generation, pellet imports, emission factors.
  • Analysis: summarize, plot, compare scenarios (Weeks 2–3).
  • Conclusion: a policy briefing that someone can act on (Week 4).
• The cycle isn’t finished until the analysis becomes a decision: that is the communication strand.
• Phase 2 runs the same cycle again on your own question.
• Perhaps this is your mapping dissertation. What’d it look like?
• Perhaps it’s your non-mapping diss. Supervisor provides one box - then what?

Two directions of reasoning

Deductive

General → Specific

If plate tectonics is correct, then we predict matching fossils on separated continents.

We test the prediction.

Inductive

Specific → General

We observe matching fossils on separated continents. We infer the continents were once joined.

We build toward a theory.

USGS, public domain

• Both are legitimate but different. Deduction is stronger logically (if the premise is true, the conclusion must follow). Induction is how we actually discover most things.
• In practice, science uses both — we inductively form hypotheses, then deductively test them.

Wegener’s continental drift

• 1912: Wegener induces that continents move, from fossils, rock belts and glacial deposits that match across oceans
• He also proposes a mechanism: forces from Earth’s spin and tides
• Physicists deduce those forces are far too weak: mechanism refuted
• Most geologists throw out the drift along with the mechanism

The lesson: test each claim against its own evidence.

USGS, public domain

• The drift hypothesis came from induction (matching fossils and rocks). The mechanism was tested by deduction: forces not strong enough to move continents → hypothesis failed.
• Two hypotheses here. Mechanism was easy to test & dismiss. The drift hypothesis made predictions - and evidence kept piling up. Rejecting both hypotheses together was the error.
• Why did it happen? Wegener was a meteorologist, an outsider, and his mechanism made no physical sense. The establishment was blinded by the implausible cause and stopped looking at the evidence. The source of an idea doesn’t settle its validity.
• The mechanism arrived in the 1960s: seafloor spreading (Hess), magnetic stripes (Vine & Matthews), plates (Wilson; McKenzie & Parker). Roughly fifty years after Wegener.
• PPDAC link: a Conclusion can bundle several claims; each spins out its own subsequent cycle(s).

Testable or not?

• Research Skills for Geoscientists
• The scientific method

Testable or not?

✏️ Exercise 1

• Falsifiability
• Meet R
• The biomass question
• Wrap-up

Instructions

I’ll show you some statements.

For each one, decide: Is this a testable hypothesis?

• What predictions does the hypothesis make?
• What observations would / would not be consistent with the hypothesis?
• What data / evidence could be collected to test it?

Think for 30 seconds, then discuss as a desk.

• [11:30–11:40] Think, discuss as a desk, share. Show each statement, pause, then take a few responses.
• The biomass statement should produce disagreement — that’s the hook for the rest of the module.

“The Earth is approximately 4.5 billion years old”

Testable, and extensively tested (radiometric dating, meteorite & Lunar ages, astronomical evidence).

“Earthquakes are caused by divine punishment”

Not testable as stated — no observable prediction that distinguishes this from natural causes.

“Wind turbines cause headaches in nearby residents”

Testable — but tricky. How do you control for the nocebo effect? Awareness bias? The design of the study matters enormously.

• This foreshadows Week 5 (experimental design, confounding).
• Don’t resolve it — just flag the complexity.
• Introduce the “Dawkins Zone”: Dawkins would like every hypothesis to be scientifically testable. But some might be testable in principle, but not in practice - need impossibly large sample size to detect small effects; or effect may be swamped by confounders.

“CO₂ emissions from burning biomass should be counted as zero”

Testable? It depends.

• What timescale?
• What system boundary?

• “Carbon-neutral” is partly a scientific claim and partly a policy choice.

This question underpins our first four weeks.

• This statement should generate genuine disagreement. Some students will say “it depends on the trees regrowing” — good. Others may say “it’s a policy definition, not a scientific hypothesis” — also good.
• Don’t resolve it. Leave it open. Curiosity is the goal.

Falsifiability

• Research Skills for Geoscientists
• The scientific method
• Testable or not?

Falsifiability

🎓 Concept block 2

• Meet R
• The biomass question
• Wrap-up

Popper’s perspective

We can never prove a theory.

We can only fail to disprove it.

Every test that could have disproven our hypothesis but didn’t makes us a little more confident — but never certain.

Mute swan, Windermere
Michal Klajban, CC BY-SA 4.0

• [11:40–11:45] This is the philosophical core.
• Karl Popper (1934): the demarcation problem. What separates science from non-science? Falsifiability.
• You don’t need to go deep into philosophy of science. The practical lesson: seek evidence that could prove you wrong.

The Wason selection task

Each card has a letter on one side and a number on the other.

Rule: “If a card shows a vowel, then the other side has an even number.”

A

K

4

7

Which cards must you turn over to test the rule?

Vote at PollEv.com/geol

or text geol to 07480 781235

• [11:45–12:00] Let students think for 60 seconds, discuss as a desk, then poll via Poll Everywhere.
• Most students pick A and 4 (confirmation). The correct answer is A and 7 (seeking disconfirmation).
• Close the poll in Poll Everywhere before advancing to next slide (which shows the results).

Poll results

Wason selection task

Each card has a letter on one side and a number on the other.

Rule: “If a card shows a vowel, then the other side has an even number.”

A ✓

K ✗

4 ✗

7 ✓

A (vowel): if the other side is odd, the rule is broken

K (consonant): the rule says nothing about consonants

4 (even): Neither consonant-even nor vowel-even would disprove the rule

7 (odd): if the other side is a vowel, the rule is broken

• The key: 4 feels relevant because it confirms the rule. But confirmation doesn’t test anything. Only disconfirmation does.
• Link forward: “Hypothesis testing in statistics is a formalized version of this — we ask: if the null hypothesis were true, would we see data this extreme? We’re seeking disconfirmation.”

Confirmation bias

Bank holidays… always a washout?
Sea haar at Saltburn-by-the-Sea.
Colin Park / Geograph, CC BY-SA 2.0

Humans are wired to confirm, not disconfirm: we seek evidence for what we already believe.

• Give an everyday example: you think it always rains on bank holidays. You remember the rainy ones, forget the sunny ones. That’s confirmation bias.
• In data analysis: if you only look for patterns that support your hypothesis, you will always find them.

Useful hypothesis tests push back against this tendency.

In this module, you’ll learn to:

• Ask “what evidence would change my mind?”
• Seek comparisons, not just supporting cases
• Treat uncertainty as information, not failure

• This connects to every subsequent week. Week 3: “what assumptions are we making?” Week 6: “how plausible was this before we tested?” Week 8: “what is the model not capturing?”
• The habits are the real learning outcome. The specific tests and methods are tools in service of these habits.

Meet R

• Research Skills for Geoscientists
• The scientific method
• Testable or not?
• Falsifiability

Meet R

🎓💻 Concept block 3

• The biomass question
• Wrap-up

Why R?

You know Python (NumPy, Jupyter, pandas)

Same ideas. Different syntax.

R is built for statistics

• t.test(), lm(), aov() — one line each
• ggplot2 — the grammar of graphics
• Built-in data manipulation

You’ll see why this matters starting in Week 2.

• [12:00–12:15] This is NOT an R lesson. That was pre-work.
• The goal: show that R is working, connect it to what they know, and move on. Keep it to ≤10 minutes of live coding.

R demo

• elec <- read.csv("data/uk_electricity.csv") head(elec)
• Then build: elec$bioenergy_twh, bio <- elec$bioenergy_twh, summary(bio), plot(elec$year, bio).
• Narrate: “You already know how to do this. head() is like df.head() in pandas. plot() is like plt.plot(). The syntax is different but the logic is identical.”
• Make sure something goes wrong, to normalize errors: “This happens to everyone. R’s error messages are actually quite helpful once you learn to read them.” Typing elec$bio is a good one — R partial-matches, so it works, which is its own lesson.
• The editor swallows arrow keys and space, so slide navigation needs a click outside the cell first.
• ≤10 minutes. Then back to slides.

The biomass question

• Research Skills for Geoscientists
• The scientific method
• Testable or not?
• Falsifiability
• Meet R

The biomass question

✏️💬 Integrative exercise

• Wrap-up

w01-ppdac-ring.R

This section is an opportunity to work through the scientific process. Now you know what makes a good hypothesis: let’s try to formulate one. We’re going to start with observations & hypotheses today, then get testing over the coming weeks.

Context

The UK burns millions of tonnes of imported wood pellets for electricity.

Most come from forests in the southeastern United States.

w01-pellet-origins.R

• Little aside: I’m going to try to hold myself to the same standards of data presentation to which you should aspire. It’s your job to call out when I make a claim that’s not supported by the evidence I am presenting.
• Here’s a first (pedantic?) example: The chart records country of origin, not region: “southeastern” is from the sourcing literature, not from this data. The claim on the slide is finer-grained than the evidence behind it.
• “Millions of tonnes” is now a number they can read off. Plant the seed: is 9.3 Mt a big number? That question is next week’s.

Under international carbon accounting rules, the CO₂ released at the chimney is counted as zero.

Is that reasonable?

Discuss as a desk. You have 10 minutes.

1. Write one testable hypothesis about biomass energy
2. What data would you need to test it?
3. What evidence would disprove your hypothesis?

• [12:15–12:35] Circulate and listen. Collect a few examples on the board / shared doc.
• Good hypotheses might be: “Biomass electricity produces less CO₂ per kWh than coal” (testable, but depends on system boundary). “Transporting pellets from the US adds significant emissions” (testable with import data + transport factors).
• The disprove question is the hard one — and the most valuable. It forces students to think like Popper.

Attendance

Show the QR from https://attendance.durham.ac.uk/mytimetable, then ← back to the task.

What you came up with

• Type 3–4 hypotheses in as students call them out. The box is plain scratch space — nothing is saved, so screenshot it if you want it.
• In speaker view (press S), click into the box on the current-slide preview and type: it appears on the projected slide as you go. Esc or a click outside the box lets the arrow keys advance slides again.
• Without speaker view, click into the box on the slide itself: while it has focus, space and the arrow keys type rather than advance. Click outside (or press Esc) to get navigation back.
• Highlight: these are the questions that will drive the next four weeks. You’ll get the data in the application session.
• If anyone says “it depends on the timescale” or “it depends on what you include” — celebrate this. That’s exactly right, and it’s what Week 3 is about.
• End the session with the question feeling genuinely open. Resist giving the punchline.

Wrap-up

• Research Skills for Geoscientists
• The scientific method
• Testable or not?
• Falsifiability
• Meet R
• The biomass question

Wrap-up

Key points

1. Science is a process for reducing uncertainty, not a collection of facts
2. We test hypotheses by seeking disconfirmation, not confirmation
3. The biomass question — does it help reach net zero? — is genuinely open, and the answer depends on choices you’ll learn to identify

Any questions we missed?

Submit questions:

PollEv.com/geol

text geol to 07480 781235

• Continuation of the same Poll Everywhere Q&A activity opened at the start of the session — this is not a new poll, so don’t reset/archive it yet.
• Surface any questions that have come in since the early Questions slide, especially ones raised during today’s content.
• Once you’ve finished with this slide, archive (reset) the activity at the end of the session so questions don’t carry over.

Next time

Application session: “Meet the data (and Git)”

You’ll:

• Load real UK energy data in WebR
• Make your first plots
• Set up your GitHub repo and make your first commit

If you haven’t done the pre-term preparation, do it before the next session.

• End on a practical note. The application session is where the abstract ideas become concrete.
• Students who haven’t done the pre-work should not panic — the application session will support them — but they should try to catch up.