MAST20034 Critical Thinking with Data

How MAST20034 is assessed

What MAST20034 covers

Eleven reasoning weeks → one critique toolkit. Each links to its free chapter guide.

You are handed a graph, a study or a piece of statistical output and asked to name what is good, name what is wrong, and say how to fix it. The final exam is 60% of your grade, short-answer reasoning only — no calculator, no calculations, no software — sat with up to four sides of your own notes.

As the marking criteria put it, “explaining your reasoning and choices is typically more important than any answer”: every mark is a because.

This guide drills exactly that one move — name the concept → define it in a line → apply it to the scenario → state the consequence or fix — across the whole 11-week critique toolkit, from the PPDAC cycle to confounding to Bradford Hill.

Worked example · free

Naming a confounder — the signature short-answer, mark by mark

Key terms

PPDAC cycle

The investigation cycle that frames almost every critique in this subject: Problem → Plan → Data → Analysis → Conclusion. Most exam prompts are really a question about one node — was the Plan a sound design, were the Data well sampled, is the Conclusion licensed by the design? Walk it from memory and you have a checklist for any study.

Confounding

When a third variable drives both the exposure and the outcome, creating an association between them that is not causal. The classic answer to “these two things move together, so one causes the other”: name the confounder, explain how it inflates both, and control for it (stratify, match, restrict or adjust) before believing the link.

Internal vs external validity

Internal validity asks whether the study's own conclusion is sound for the people in it (did the design rule out confounding and bias?); external validity asks whether that conclusion generalises beyond them. A tightly controlled RCT can be high on internal validity yet low on external validity if its sample is narrow — the WEIRD-sample problem.

Sampling bias

A systematic error in how units are selected, so the sample is unrepresentative of the population in a direction a bigger sample cannot fix. Distinct from sampling error (random, shrinks with size). Includes selection, non-response, undercoverage and voluntary-response bias — and the “who's missing?” reflex is the fastest way to spot it.

Confidence interval

A range of plausible values for an unknown population parameter, built so that the procedure captures the true value a stated percentage (e.g. 95%) of the time over repeated samples. It is NOT a 95% probability that the parameter lies in this one interval, and it is NOT a range for individual data points — those two misreads are the most-penalised errors in the inference questions.

MAST20034 FAQ

Questions students ask in each MAST20034 chapter

What does ‘critical thinking with data’ mean if there's no calculation?

It means appraising how data were produced and what they can legitimately support — naming the design, the biases, the confounders and the limits of a conclusion. The skill is structured judgement in words, which is exactly why the exam has no calculator. Chapter 01 · Objectivity and the Critical Lens

How do I structure a graph-critique answer?

Follow the skeleton the markers reward: name two genuine good features, then one specific, named fix — citing the graphics principle each point relates to. Specificity (‘truncated y-axis’, not ‘misleading’) is what earns the mark. Chapter 02 · Honest Graphics and Visualisation

Why can only experiments show causation?

Because random assignment balances confounders across groups, so a difference in outcome can be pinned on the treatment. In observational studies the groups self-select and may differ in unmeasured ways, leaving association as the strongest legal claim. Chapter 03 · Study Design: Experimental vs Observational

How do I spot a confounder fast?

Ask for a third variable that plausibly causes BOTH the exposure and the outcome. If removing the exposure wouldn't change the outcome because the real driver is elsewhere, you've found it — name it and draw the Z→X, Z→Y, X–Y triangle. Chapter 04 · Observational Studies and Confounding

When should a study be qualitative rather than quantitative?

When the question is about meaning, experience, process or context — the ‘why’ and ‘how’ — or when you are exploring and generating hypotheses. Quantitative suits ‘how many’, ‘how much’ and testing pre-specified effects. Chapter 05 · Qualitative Methods

What does a 95% confidence interval actually mean?

That the method produces intervals which capture the true parameter 95% of the time across repeated samples. For your one interval, the parameter is fixed and either inside or not — so you say ‘95% confident’, not ‘95% probability’. Chapter 06 · Frameworks for Inference

What does ‘all models are wrong but some are useful’ mean for the exam?

That you judge a model by usefulness for its purpose, not truth. So a good answer interprets what the model usefully tells us and names where it simplifies or fails — never claims the model is ‘right’. Chapter 07 · Statistical Modelling

Why can't a big sample fix sampling bias?

Because bias is systematic — it shifts the estimate's centre away from the truth — while sample size only reduces random scatter around that centre. A huge biased sample is a precisely wrong answer; only an unbiased design fixes it. Chapter 08 · Sampling and WEIRD Bias

Why isn't one study enough?

Because any single study can be a fluke, a biased sample or a false positive. Confidence comes from replication and synthesis — consistent results across independent studies, pooled where appropriate, weighed for quality and publication bias. Chapter 09 · Accumulating Research

Does more data make an analysis more trustworthy?

Not by itself. Volume is not validity — a large dataset collected with bias is still biased, and at scale even trivial effects look ‘significant’. Big data needs the same design, sampling and ethics scrutiny as any other data, plus attention to effect size. Chapter 10 · Big Data and Context

How to study for the exam

Treat this as a critique-skills subject, not a content subject: the same handful of moves recur on fresh scenarios, so drill the moves, not facts. For every prompt, practise the four-part answer shape the markers reward — name the concept → define it in a line → apply it to the scenario → state the consequence or fix.

Build your four-side notes sheet as the course runs: definitions, taxonomies (the four observational designs, the sampling methods, the Bradford Hill criteria) and decision rules (the design tree, the graph-critique skeleton, the CI/P-value interpretation rules) — never arithmetic, because there is no calculator.

Study design and confounding (the heaviest-marked block) and the name-the-concept decoder are where the exam is won: rehearse spotting the confounder in any association and naming the legal conclusion for any design, until both are automatic.