What is Occam's razor? Why simpler explanations win

Occam's razor is a rule of thumb for choosing between explanations: when two explanations account for the facts equally well, prefer the one that needs fewer assumptions. It is named after William of Ockham, a 14th-century English philosopher. It says what is the better bet, not what is true, and it applies only when the explanations fit the evidence equally well.

In short

• Occam's razor says not to add assumptions an explanation doesn't need. It "shaves off" the extras.
• It works because every extra assumption is another way to be wrong, so simpler explanations usually start out more likely.
• It is a guide, not a law of nature: sometimes the truth is complicated, and the razor never overrides evidence.

What does Occam's razor actually say?

The popular wording is "entities should not be multiplied beyond necessity." According to the Stanford Encyclopedia of Philosophy, that exact sentence is nowhere in Ockham's writings; it was attached to him later. What he did write, in several forms, is that one should not posit more things than needed, unless reason, experience or the authority of scripture requires it. Ockham, born around 1287 and a Franciscan friar, used it mostly to cut down the number of abstract entities philosophers of his time believed in.

The idea is older than Ockham. As the encyclopedia's entry on simplicity notes, Aristotle preferred demonstrations from fewer starting points, and Thomas Aquinas held that if one means suffices, using several is superfluous. Isaac Newton made it the first of his rules of reasoning in 1687: "We are to admit no more causes of natural things than such as are both true and sufficient to explain their appearances."

A common misreading is "the simplest explanation is usually right." The razor is about assumptions, not about how short or easy an explanation sounds. An explanation such as "a ghost did it" is short, but it assumes a great deal that nothing else supports.

Why do fewer assumptions make an explanation more likely?

Each assumption an explanation needs is a separate thing that has to be true. If they are independent, their chances multiply, and multiplying probabilities below 1 always makes the total smaller. So an explanation that needs two unsupported assumptions starts out less likely than one that needs only one of them.

Worked example: the car that won't start

Your car won't start on a cold morning. Two explanations both fit that one fact.

• Explanation A: the battery is dead. Suppose that on a cold morning, a battery like yours fails about 1 time in 100.
• Explanation B: the starter motor and the fuel pump both failed on the same morning, independently. Suppose each fails about 1 time in 1,000 on a given day.

The chance of B is 1/1,000 × 1/1,000 = 1 in 1,000,000. The chance of A is 1 in 100. So A starts out about 10,000 times more likely, purely because it needs one failure instead of two coincidences. The numbers are illustrative, but the structure holds for any rates.

Now add evidence. You turn the key and the headlights shine brightly and the dashboard lights up. That evidence is hard to square with a dead battery, so A's chances drop sharply. The razor ranked the explanations before the evidence came in; once evidence arrives, evidence wins. That is the right order: use the razor to decide what to check first, not to stop checking.

How do scientists use Occam's razor?

In science the razor works as a tiebreaker and as a guide to what to test. A history of two planets that were predicted the same way shows both its power and its limits.

	Uranus's wobble, 1840s	Mercury's drift, 1850s to 1915
The puzzle	Uranus strayed from the path Newton's laws predicted	Mercury's orbit shifted slightly more each century than predicted
The economical fix	Keep Newton's laws; add one unseen planet pulling on it	Keep Newton's laws; add one unseen planet, called Vulcan, near the Sun
What happened	Neptune was found in 1846, close to the predicted spot	Vulcan was never found
The real answer	One extra planet	A new theory of gravity, Einstein's general relativity, in 1915

Both times, adding one planet was the move with the fewest new assumptions, and testing it was right. Once Vulcan failed to show up, the simple fix had to go, and the deeper, more complicated theory explained the facts.

Statistics builds the razor in. A model with many adjustable settings can fit almost any data, including noise, so it predicts new data badly. Methods for comparing models charge a penalty for each extra parameter, a formal version of not adding assumptions you don't need.

When does Occam's razor fail?

When the world really is complicated. Around 1900, some leading scientists, the physicist Ernst Mach among them, doubted that atoms were real, treating them as an assumption physics didn't need. Experiments by Jean Perrin on the jiggling of tiny particles settled the matter, and he won the Nobel Prize in Physics in 1926 "for his work on the discontinuous structure of matter".

Medicine has a counter-rule, Hickam's dictum: a patient can have several diseases at once. Doctors are taught to look for one diagnosis that explains all the symptoms, but in older patients with many conditions, several common problems are often likelier than one rare one. A 2023 study by Donald Redelmeier and Eldar Shafir found that when a hypothetical patient already had one diagnosis, such as flu, both members of the public and health professionals rated the chance of a second illness, COVID, far lower, cutting the odds they gave by roughly half or more. Properly read, the razor doesn't forbid that; three common illnesses can need fewer unlikely assumptions than one rare syndrome.

The razor also gets misused in arguments. "The simplest explanation" is often just the one the speaker already believes, or the one that leaves out inconvenient facts. An explanation that ignores evidence isn't simpler, only incomplete. A checklist before reaching for the razor:

• Do both explanations fit all the evidence? If not, the razor doesn't apply yet.
• Count assumptions, not words. What does each explanation need to be true that nothing else supports?
• Ask what each predicts. The better explanation makes a test you can run.
• Look at the missing data. Sometimes the simple story only looks simple because the cases that contradict it never reached you, which is survivorship bias.

Still argued

Philosophers disagree about why simplicity should track truth at all. Probabilistic accounts say simpler hypotheses fit fewer possible outcomes, so they are rewarded more when their predictions come true. Others say the preference is a practical one, about which theories are easier to test, with no guarantee that nature is simple. The Stanford Encyclopedia's entry on simplicity also notes that judgments of simplicity can depend on how a theory is described, so two people can disagree about which theory is simpler.

Where does the razor fit?

Occam's razor is one of the oldest tools in philosophy, and it is part of everyday critical thinking: a reason to ask what a claim is quietly assuming. In science, it guides which hypothesis to test first, and the test decides.

Check yourself

• Without looking back: state Occam's razor in one sentence, and say why the popular version "the simplest explanation is usually right" gets it slightly wrong.
• Using the car example, explain why multiplying the chances of independent assumptions favors the explanation with fewer of them.
• Why did the razor point to an extra planet for both Uranus and Mercury, and what does the Vulcan case teach?

Sources

• William of Ockham — Stanford Encyclopedia of Philosophy
• Simplicity — Stanford Encyclopedia of Philosophy
• The fallacy of a single diagnosis — Redelmeier and Shafir, Medical Decision Making, 2023
• The Nobel Prize in Physics 1926 — Nobel Prize Outreach