Occam's Razor: When the Simplest Explanation Wins, and When It Loses

The razor is a rule about choosing between explanations, and it has a precondition that is easy to forget: it applies only when the explanations fit the evidence equally well. Given that tie, it says to prefer the one that posits fewer things, fewer causes, fewer agents, fewer coincidences. The justification is partly probabilistic. An explanation that requires five independent things to be true is less probable than one that requires one of them, so long as each of the five is less than certain. It is partly practical, since simpler hypotheses are easier to test and harder to bend to fit any outcome.

The bias arrives when simplicity stops being a tie-breaker and starts to outweigh the evidence. Tania Lombrozo showed in 2007 that people prefer an explanation with one cause over an explanation with two, even when the two-cause story is more probable, and that the pull of the simpler story weakened when the probability information was made explicit and easy to use. People also tend to treat a simple explanation as more probable than it is when no probabilities are given. Simplicity is doing the work of evidence.

Philosophers have argued for a long time about whether the razor has any general justification, and Elliott Sober's 2015 book concluded that it does not. Parsimony earns its place only locally, where some background fact ties simplicity to truth or to predictive accuracy: in model selection, for example, where a simpler model that fits the data almost as well tends to predict new data better because it has not fitted the noise. Where no such fact holds, the razor is a preference rather than a principle. That is the position this page takes: the razor is a good default and a bad argument.

Its failures run in both directions. The first is the one the demo shows, where a simple story survives facts that should have killed it. The second is treating the world as if it had to be simple. In medicine this is the tension between Occam's razor and Hickam's dictum, the saying that patients can have as many diseases as they please; an older patient with several complaints more often has several conditions than one rare condition that explains everything. In science, the simplest theory consistent with the data has often been wrong, and the next round of data showed why.

There is no original experiment for a principle that is seven centuries old, so this page's anchor is the history of the principle and the research on how people actually use it. William of Ockham was an English Franciscan philosopher and theologian who taught at Oxford in the early fourteenth century, and his writings use parsimony as a working tool: plurality is not to be posited without necessity, and nothing should be assumed without a reason given for it. He used it to argue against entities his contemporaries had posited, not as a theory of nature. He did not claim that nature was simple, only that an argument should not take on more than it needs.

The famous formulation, that entities must not be multiplied beyond necessity, is not his. W. M. Thorburn searched Ockham's works for it in 1918 and reported in the journal Mind that it appears nowhere in them; the earliest use he found was in a 1639 commentary on Duns Scotus by the Irish Franciscan John Punch. The name Occam's razor is later still, and the word razor refers to shaving away unnecessary assumptions. The principle itself is older than Ockham, with versions in Aristotle and the medieval philosophers before him.

The modern study of parsimony has two branches. In philosophy of science, Sober's Ockham's Razors: A User's Manual, published in 2015, takes the principle apart into its separate uses and asks what justifies each; its answer is that parsimony has no one justification, and that where it is justified it is because of a specific connection, such as the one between model simplicity and out-of-sample prediction formalized in Akaike's information criterion. In psychology, Lombrozo's 2007 experiments in Cognitive Psychology asked people to choose between a one-cause and a two-cause explanation for a set of symptoms, with base rates supplied, and found that participants leaned on simplicity beyond what the probabilities supported, a demonstration of the razor working as a bias rather than a rule.

The razor is in use whenever two accounts compete, and it is misused whenever simplicity substitutes for a look at the evidence.

The counter is to put the razor back in its place: after the evidence, as a tie-breaker.

• Check the tie first. Before preferring the simpler explanation, list the facts and ask whether each explanation accounts for all of them. If one leaves a fact unexplained, the razor does not apply yet.
• Count assumptions, not words. A short story can hide many assumptions and a long one can rest on few. The razor counts things that must be true, not sentences.
• Ask what each explanation predicts. Two accounts that fit the same past facts often predict different future ones; a test that separates them is worth more than any amount of parsimony.
• Allow for more than one cause. Where the base rate of each ordinary cause is high, two ordinary causes together can be more likely than one rare cause that explains everything.
• Treat the conclusion as provisional. The razor picks the best current bet, and a bet is revised when the next fact arrives. A simple explanation held after the facts have moved is the bias.

Several biases share the razor's taste for the clean story, and the differences are worth keeping.

• It is not the conjunction fallacy, in which a detailed story is judged more probable than one of its own parts. That bias runs the other way, toward the elaborate; the razor properly applied is its corrective, and the razor misapplied is its mirror.
• It is not belief bias, the acceptance of an argument because its conclusion is believable. The razor concerns how many assumptions an explanation makes, not how congenial its conclusion is.
• It is not the rhyme-as-reason effect or the broader preference for fluent statements. A simple explanation and a fluent one are different things; the razor is about structure, not phrasing.
• It is not Hickam's dictum, which is its counterweight in medicine rather than a bias: the reminder that a person can have several conditions at once, and that the simplest single diagnosis is not always the right one.