Think Stats

What the book is

Think Stats: Exploratory Data Analysis in Python by Allen B. Downey describes itself as "an introduction to Probability and Statistics for Python programmers". Its thesis, kept since the first edition, is that many ideas that are hard to explain with maths are easier to explain with code. In the third edition Downey has "doubled down" on this, to the point where there is almost no mathematical notation left.

The third edition is free to read online and is published under a Creative Commons licence that lets you copy and adapt it for non-commercial use with attribution. A printed edition is also sold through book retailers.

What it covers

Fourteen chapters: exploratory data analysis, distributions, probability mass functions, cumulative distribution functions, modelling distributions, probability density functions, relationships between variables, estimation, hypothesis testing, least squares, multiple regression, time series analysis, survival analysis and analytic methods. Case studies use real datasets, including data from the National Institutes of Health, and extra worked examples tackle questions like whether a result could be luck.

Is it worth your time?

If you can already program in Python and statistics feels like a wall of symbols, this is one of the best ways in. Every idea is something you compute, plot and poke at, so you build intuition by running small experiments. It also teaches good exploratory habits: look at the distribution before you trust a mean, and question where the data came from.

Where it fits for a data science learner

Read it early, as your statistics foundation, before a statistical learning textbook. It gives you distributions, estimation, sampling error and hypothesis testing in working code, which makes later modelling and experiment analysis far less mysterious.

Prerequisites

Downey writes that he assumes you know core Python, including object-oriented features, and basic mathematics such as logarithms and summations. Knowing NumPy and pandas helps but is not necessary, and no statistics background is assumed.

How to run the code

Each chapter is a Jupyter notebook with a link to run it on Google Colab, which the author strongly recommends for at least the first few chapters because you install nothing. If you prefer your own machine, you need Python, Jupyter and libraries including NumPy, SciPy and StatsModels.

What you can do after it

You can explore a new dataset methodically, choose and fit a distribution, estimate a quantity with an honest measure of uncertainty, run and interpret a hypothesis test, and fit a regression with pandas and StatsModels.

Common questions

Is it free? The online version is free. Buying the print book is optional.

How long does it take? One chapter per sitting or two is a comfortable pace; do the exercises at the end of each notebook.