How to Use the Feynman Technique, With a Worked Example

The Feynman technique is a way of finding the gaps in your understanding by explaining a topic in plain words. This post covers where it comes from and what the evidence says, gives the steps, and then works through one topic from university statistics, including what goes wrong in a first attempt.

Where the technique comes from

Richard Feynman was a physicist who shared the 1965 Nobel Prize in Physics and was known for explaining difficult ideas in plain language. Two things he said show the idea behind the technique.

In a 1974 commencement address at Caltech about honesty in science, he told the graduating students that "you must not fool yourself—and you are the easiest person to fool" (Feynman, 1974).

His colleague David Goodstein told a story about the test Feynman applied to his own understanding. Goodstein asked him to explain why particles with spin one-half obey Fermi–Dirac statistics, and Feynman said he would prepare a first-year lecture on it. A few days later he came back and said he could not do it. He had not managed to bring it down to first-year level, and he took that to mean that physicists did not really understand it (Goodstein, 1989).

Feynman did not set out a numbered method himself. The steps that carry his name were written up by other people after his death in 1988. One widely read version is Scott Young's, from 2011.

What the evidence says

The idea behind it has reasonable support. When people are asked to write a step-by-step explanation of how something works, such as a zip or a flush toilet, their rating of their own understanding drops, because writing the explanation shows them gaps they did not know were there. The same research found this overconfidence was strongest for explanations of how things work, and weaker for facts, procedures and stories (Rozenblit and Keil, 2002).

Under its own name, the technique has only been tested in a few small studies, mostly in schools, such as Reyes et al., 2021. The closest method with a larger body of research, self-explanation, was rated moderately useful in a major review of study techniques: promising, but less tested in real classrooms than self-testing or spaced practice (Dunlosky et al., 2013).

How to do it

1. Pick one topic: an idea or a process, not a single definition.
2. Without notes, write an explanation of it for a classmate who has the same background as you but missed the lecture on it.
3. Wherever you get stuck, mark the problem. Look for three kinds: a claim you are not sure is true, a phrase you copied without understanding it, and a sentence you cannot finish.
4. Go back to your notes, textbook or lecture recording for the marked points only.
5. Write the explanation again from the start, without notes.
6. Repeat until there are no borrowed phrases and no unfinished sentences.

Popular versions tell you to explain the topic as if to a child. An imagined child will not ask about conditions or exceptions, so it puts no pressure on you to be precise. A classmate on the same course will not let you hide behind jargon, and does not need everything turned into an analogy.

A worked example: the central limit theorem

The central limit theorem is a standard topic in university statistics. Here is a first attempt at explaining it, written without notes.

First attempt

The central limit theorem says that if your sample is large enough, the data becomes normally distributed. That is why we can use normal-based tests on almost any data, as long as n is above about 30.

It reads fluently and is mostly wrong. Rereading the chapter would not show you that, because the chapter states it correctly and every sentence would look familiar.

What is wrong with it

1. A false claim. The data does not become normal. Sample a thousand household incomes and you have a thousand skewed incomes. The theorem is about the distribution of the sample mean across many repeated samples, not the shape of the data in one sample.
2. A borrowed phrase. "n above about 30" is a rule of thumb, not part of the theorem, and how large a sample needs to be depends on how skewed the population is. Treating 30 as the point where the theorem starts to work is a misunderstanding that statistics education researchers have reported in their students (Lunsford, Rowell and Goodson-Espy, 2006).
3. A sentence you cannot finish. Try to complete "this works as long as…" with the actual conditions, and you get stuck. The standard version of the theorem needs independent observations from a population with finite variance.

Each of these is now a specific thing to look up.

Going back for the gaps

Go back for these three points only: what the theorem is about, what conditions it needs, and where the 30 comes from. That is a few paragraphs of a textbook or a couple of minutes of a recorded lecture. Rereading the whole chapter would bury the three gaps under material you already know.

Finding those few paragraphs can be the slow part. In SymbioLearn, each Concept links to the page, section or timestamp in every Source that teaches it, so you can go straight to the paragraph, or to the right moment in a lecture video.

Second attempt

Take a sample of n independent observations from a population with finite variance, and work out its mean. Do that many times, and the means have a distribution of their own. As n grows, that distribution gets closer to a normal distribution, centred on the population mean, and it gets narrower. This is what lets us use normal-based methods for the mean even when the data is skewed. It says nothing about the shape of the data itself. How large n needs to be depends on the population: with mild skew a fairly small sample is enough, and with heavy skew or extreme outliers it can take far more than 30. If the variance is infinite, the theorem does not apply.

This version is longer, which is fine. What matters is that it has no borrowed phrases and no sentences you cannot finish.

Tips for using it on your own topics

• Write it or say it out loud. In your head it is easy to skip a gap without noticing, because a thought does not have to be a complete sentence.
• Keep your first attempt. It shows exactly where your understanding went wrong, which makes it useful to reread before the exam.
• Mark anything you wrote word for word from your notes, slides or textbook. A phrase you can only repeat is one you have not had to unpack yet.
• Spend fifteen to thirty minutes on one idea, and use the technique once per topic, not once per fact.

When to use something else

The technique is for ideas that have to be explained: why something happens, or how the parts of a process fit together. Terms, notation, formulas and lists are better learned by repeated recall, with flashcards or a short quiz. The evidence for recall over rereading is in active recall vs re-reading.

• Explaining why a price ceiling set below the market price leads to a shortage is worth fifteen minutes. The formula for price elasticity of demand is flashcard material.
• Explaining why a drug with a narrow therapeutic index often needs its blood levels monitored is worth doing. Remembering which drugs have a narrow therapeutic index is better done with flashcards.

What to do next

Pick the topic in your current module that you would least like to explain out loud. Write your explanation before you open any notes, and mark the three kinds of problem: a false claim, a borrowed phrase and a sentence you cannot finish. Those marks are your revision list for the week.