StatQuest with Josh Starmer

There's a moment in most statistics courses where the notation stops meaning anything and you start memorising. Josh Starmer built a channel to prevent that moment. Since 2016 he's been drawing the concept before he writes the formula — literal drawings, with a sung intro and a lot of shouting BAM when something clicks.

The silliness is load-bearing. Starmer's method is to take one idea, strip it to a tiny toy dataset you can hold in your head, walk through it slowly, and only then generalise. It works for p-values and it works for transformers, and the channel covers both ends of that range along with regression, gradient descent, decision trees, random forests, PCA, neural networks and back-propagation.

How to actually use it

Not as a course. This is a reference library organised by concept, and the right way in is to have a specific confusion and search for it. Playlists group related videos — the machine learning fundamentals sequence and the neural network series both work as ordered runs — but there's no curriculum, no exercises, no code to write, and nothing checking whether you understood.

Starmer sells books and study guides through his site, so some material is paid. The videos themselves are free.

Where it falls down

Two places. First, the pacing that makes it brilliant for beginners makes it slow for anyone who already half-knows the material — a ten-minute video often carries ninety seconds of new information, and the songs stop being charming around the fifteenth time. Playback speed is your friend.

Second, and more seriously: watching StatQuest is not doing statistics. The visual explanations are so satisfying that they produce a strong feeling of understanding that doesn't survive contact with a real dataset. You'll grasp what a p-value means and still not know what to do when your data violates the assumptions. There is no practice component at all, and the channel is not trying to provide one.

Who it's for — and who should skip it

Ideal alongside a hands-on course — pair it with Kaggle Learn or a university stats module and it fills exactly the intuition gap those leave. Also ideal if you've been using models you can't explain and it's started to bother you.

Skip it as your only resource. Skip it too if you want mathematical derivation and proof; Starmer trades rigour for clarity deliberately, and for some purposes you need the rigour.

Our take

The best conceptual explanations of statistics and machine learning on the internet, free, and structurally incapable of teaching you to apply them. Use it as the thing you watch when a concept won't sit still, not as the thing you watch instead of practising.