Applied Statistics

What it's about

Inferential statistics — hypothesis testing and confidence intervals — for a variety of situations, and the beginnings of multivariate statistics: simple and multiple linear regression. We learn each method by applying it to a real decision the pizzeria has to take, so that the statistics of a professional role — a data analyst's, say — is practised, not just heard.

Prerequisites

Courses on descriptive statistics and probability theory. Weeks 1–2 rebuild the descriptive part from scratch and Week 3 opens with the probability rules you need — so if the prerequisite is rusty, the first three weeks are where you fix it.

Use of AI

For the project you are encouraged to use ChatGPT and other AI tools freely to research, check and refine your work. The final exam is closed book: no AI, no computer, no phone. AI helps you learn and prepare; it will not be in the room when you are asked to demonstrate what you know.

What you'll be able to do

These are the course's official learning objectives. Each one is a thing you will be able to do by Week 16 — with the week that teaches it, so you always know where you are on the map.

• Apply the theory of sampling distributions. Explain why a sample mean from a sample of thirty behaves predictably even when the population does not — the idea every later method rests on. Week 5
• Estimate parameters and construct confidence intervals. Turn one sample into a statement of the form "the true value is between here and here, and I am 95% confident" — for a mean, a proportion and a variance. Week 6
• Perform statistical tests in a variety of situations. Take a claim, decide what evidence would refute it, compute the evidence, and reach a defensible conclusion — including what a p-value does and does not say. Week 7
• Compare two populations, and tell independent from paired samples. Decide whether a difference between two groups is real, and recognise which of the two designs you are looking at — the single most-marked judgement on the exam. Week 8
• Apply simple linear regression and compute its coefficients. Fit a line to two variables, interpret its slope in the units of the business, and test whether that slope is real. Week 9
• Apply multiple linear regression for model selection. Extend the line to several predictors and choose between competing models. Weeks 9 & 11
• Understand chi-square and F-tests. Handle categories and more than two groups — independence, goodness of fit, and one-way ANOVA. Week 10
• Understand machine learning and the main estimation methods. See where the statistics you have learned sits inside what industry calls machine learning, and what it adds. Week 11
• Use Python for data analysis. Reproduce the whole semester in pandas, SciPy and statsmodels — and read an output without panic. Week 12

60% · Final exam — Week 16

Individual, 90 minutes, graded to the half point. Closed book: the only authorised material is a simple non-programmable calculator and a personal summary of two sheets, both sides, hand-written. No computers, phones or communication devices.

Six exercises of ten points, one per block of the course — describe · probability and the normal · sampling and estimation · a test · two samples or regression · chi-square or ANOVA. Every one is a question type you have already met.

20% · Individual project — the pizzeria file

Every week ends with a workshop: a few questions that make you apply that week's method to your pizzeria, in your own words. Everything you type is saved automatically.

By Week 16 those entries are a pricing file you could take to a bank — a population defined, a market described, an interval estimated, a claim tested, a price defended. The project is that file, finished and argued. Graded to the half point.

Computer sessions prepare it: Week 11 uses DataTab, Week 12 is Python in Google Colab — nothing to install.

20% · Presence — pro rata

How it is recorded. In each session I open a short window, and the presence button on that week's page turns from grey to red. You press it once, on your own device, while you are in the room. Then I close the window.

How it is marked. Pro rata: your mark is the share of the sessions actually held that you were present for. Miss two of fifteen and you have 13/15 — there is no all-or-nothing cliff, and a session that never happened cannot count against you.

Presence is 20% because the room is where the theory gets argued out and the exercises get worked with someone to ask. That part is not recoverable from the site alone — which is exactly why it carries weight.

Homework · not graded, and still the most useful thing you do

A homework set is assigned every week — roughly two hours, on a different dataset from the one used in class, with hints and full worked solutions. It does not count toward your grade.

It is where you find out whether the lesson stuck. Do it by hand, with a calculator, the way the exam will be. Your scores reach my dashboard, which is how I know — before the next session — which idea the class has not yet landed, and what to open the session with.