The Data That Lied

Before exploring the course, try this short activity. It takes about two minutes and requires no prior knowledge.

The claim
Stage 1 of 5 — The Claim

Which AI algorithm should this security team deploy?

A cybersecurity team has tested two AI threat-detection algorithms — AlgoA and AlgoB — on a sample of 100 network alerts. Here are the results.

Based on this data, which would you recommend for deployment?

Stage 2 of 5 — The Investigation

Let’s look more closely at the data.

The 100 test cases included two distinct categories of network threat. Explore both views before proceeding.

Stage 3 of 5 — The Revelation

The full picture is more surprising than it first appeared.

Here are all three views side by side.

Paradox
AlgoB outperforms AlgoA in every individual category — yet appears far worse in the aggregate figure. How is this possible?
Stage 4 of 5 — The Explanation

The two algorithms were not tested on the same cases.

The aggregate result is misleading because the test sets had very different compositions. AlgoA was tested mostly on simple threats; AlgoB mostly on complex ones.

Simpson’s Paradox
When groups are combined, the composition of the aggregate can reverse a trend that holds within every subgroup. AlgoA’s high overall rate reflects the easy cases it was mostly given — not superior performance. This is a well-documented statistical phenomenon with real consequences in clinical trials, hiring decisions, and system evaluations.
Stage 5 of 5 — The Connection

Data can mislead even when it is entirely accurate.

Every number in this activity was correct — no data was fabricated or manipulated. Yet the aggregate figure pointed firmly in the wrong direction. Spotting problems like this requires more than arithmetic: it requires knowing what questions to ask, what comparisons to make, and what the data cannot tell you.

This course develops exactly those skills. Units 8 and 9 focus directly on data analysis, visualisation, and statistical inference — covering how to represent findings honestly and interpret them critically. But rigorous analysis depends on everything that precedes it: a well-designed study, a sound methodology, ethical data collection, and a clear research question. The full course develops all of these together, across ten units taught by specialists in research design, literature review, generative AI, and academic communication.

Why research methods matters
How you design a study determines what your data can and cannot show. The goal of this course is to help you ask better questions, make more defensible decisions, and produce research you can stand behind.

Course Structure

Ten units spanning the full research process — from initial proposal through to peer review and professional communication.

Phase 1 — Proposal & Literature
Units 1–2

Framing a research project and writing a proposal. Finding, evaluating, and synthesising academic literature to identify research gaps.

Phase 2 — Design & Methodology
Units 3–4

Research paradigms and experimental design. Choosing and justifying a methodology to match your research question.

Phase 3 — Ethics & AI
Units 5–7

Research ethics and responsible practice. The role of generative AI in research — opportunities, limitations, and academic integrity.

Phase 4 — Data & Analysis
Units 8–9

Data analysis and visualisation. Statistical inference — from descriptive summaries to hypothesis testing and effect sizes.

Phase 5 — Capstone
Unit 10

Research proposal workshop and peer review — bringing all course skills together in a collaborative, formative context.

Unit 1
Introduction and Research Proposal
Unit 2
Finding Literature and Developing the Research Question
Unit 3
Project Taster/Showcase
Unit 4
Research Methodology and Experimental Design
Unit 6
Generative AI in the Research Process
Unit 7
Generative AI in the Research Process
Unit 10
Research Proposal Workshop and Peer Review
Go to Unit 5 — Research Ethics