Occam's Razor: The Simplest Explanation is Usually Right

Around 1320, the Franciscan friar William of Ockham articulated a principle that would outlast empires: 'Entities should not be multiplied beyond necessity.' The original Latin — entia non sunt multiplicanda praeter necessitatem — was a methodological rule for philosophical arguments, but its application extends far beyond medieval theology. Ockham was not arguing that the world is simple. He was arguing that explanations should not be more complex than the evidence demands. If two theories account for the same observations equally well, prefer the one with fewer assumptions — not because complexity is wrong, but because every unnecessary assumption is an unforced error in your reasoning. The distinction is important. Occam's Razor is not a claim about reality. It's a claim about the optimal strategy for navigating uncertainty. You have limited time, limited information, and limited cognitive bandwidth. When faced with competing explanations, you need a tiebreaker. Ockham's contribution was recognising that parsimony — all else being equal — is that tiebreaker. The principle earned its name as a 'razor' because it shaves away unnecessary assumptions, leaving only what the evidence requires. Seven centuries later, it remains the single most widely applied heuristic in scientific reasoning.

There is a common misunderstanding that Occam's Razor is about elegance — that scientists prefer simple theories because they're more beautiful. The real justification is probabilistic, and it's rigorous. Every additional assumption in an explanation is another point of potential failure. If explanation A requires three assumptions and explanation B requires seven, explanation B has more than twice as many junctures where reality can diverge from the model. Each assumption must be independently true for the overall explanation to hold, so the joint probability of B being correct drops with each added element. Bayesian probability formalises this intuition. In Bayesian reasoning, every hypothesis begins with a prior probability before any evidence is examined. Simpler hypotheses receive higher priors not because the universe prefers elegance, but because there are combinatorially fewer ways for a simple model to be configured — and therefore fewer ways for it to be wrong. As evidence accumulates, Bayes' theorem updates these priors, but the simpler hypothesis starts with a structural advantage. This is why Occam's Razor survives in an age of data science and machine learning. Regularisation techniques in statistical modelling — which penalise model complexity to prevent overfitting — are formalised versions of the same principle Ockham articulated with pen and parchment.

The history of physics offers the clearest demonstration of how Occam's Razor operates — and where its limits appear. For over two centuries, Newtonian mechanics was the simpler, more parsimonious explanation for the motion of objects. It required fewer assumptions than any competing framework, and it predicted observations with extraordinary accuracy. Then anomalies emerged. Mercury's orbit precessed slightly more than Newton's equations predicted. The speed of light appeared constant regardless of the observer's motion. These were small deviations, but they were real, and no amount of tinkering with Newton's model could account for them without adding ad hoc assumptions. Einstein's general relativity was, by any measure, more complex than Newtonian mechanics. It required curved spacetime, a new understanding of gravity, and mathematics that took years for even specialists to absorb. But it explained everything Newton explained plus the anomalies that Newton could not — without requiring the additional patches and exceptions that would have made a modified Newtonian model even more complex. This is the razor operating correctly. The goal is not to pick the theory with the fewest equations. It is to pick the theory that accounts for all the evidence with the fewest unexplained exceptions. When the simple model requires patches to handle anomalies, the more complex model that handles them natively may actually be the more parsimonious one.

Occam's Razor is a working tool in any field that involves troubleshooting. When your car won't start, a mechanic checks the battery before inspecting the engine computer. When a website goes down, the engineer verifies the server is running before reviewing application code. When a patient presents with a cough and fever, the physician tests for a common respiratory infection before ordering scans for rare autoimmune conditions. Medical training encodes this explicitly. Dr. Theodore Woodward at the University of Maryland coined the aphorism: 'When you hear hoofbeats, think horses, not zebras.' The common, simple explanation — dead battery, crashed server, viral infection — is statistically more likely than the exotic alternative. Starting with the probable diagnosis saves time, money, and the patient's anxiety. Software engineers apply the same logic daily. When a feature breaks after a deployment, the first question is 'What changed?' — not 'Is there a race condition in the distributed cache layer?' The change that immediately preceded the failure is the most parsimonious explanation for it. If that doesn't pan out, you move to the next simplest hypothesis. The pattern is universal: rank your hypotheses by complexity, investigate them in order, and stop when you find one that accounts for the evidence. You won't always be right on the first try, but you'll be right on the first try more often than any other approach.

Occam's Razor has a critical caveat that is frequently ignored by people who invoke it: the competing explanations must account for the same evidence equally well. If the simpler explanation leaves anomalies unexplained, it isn't simpler — it's incomplete. And an incomplete explanation, no matter how parsimonious it appears, is not what the razor recommends. Darwin's theory of evolution by natural selection is vastly more complex than 'all species were created in their present form.' But creation accounts for almost none of the evidence that evolution explains: the fossil record's progression, DNA homology across species, geographical distribution of related organisms, observed instances of speciation. The simpler claim isn't competing on equal terms — it's ignoring most of the data. The same principle applies in business diagnostics. 'Our sales are down because the economy is weak' is a simpler explanation than 'Our sales are down because our onboarding flow has a 60% drop-off rate at step three, our pricing doesn't match the value perception of our core segment, and our main competitor launched a superior product in Q2.' But if the second explanation accounts for specific, measurable patterns that the first one ignores, the razor actually favours the more complex story. Simplicity is a tiebreaker between equally explanatory hypotheses, not a licence to ignore inconvenient evidence.

Before accepting any explanation — for why a deal fell through, why a product failed, why a team is underperforming — list the assumptions each competing explanation requires. Count them. The explanation with fewer assumptions deserves investigation first, not because it's guaranteed to be correct, but because it's the most efficient use of your limited diagnostic effort. When your startup's growth stalls, check whether the product experience has degraded before theorising about macroeconomic headwinds. When a team member's performance drops, ask whether their workload or role clarity changed before speculating about motivation or loyalty. When a marketing campaign underperforms, verify that the tracking is working correctly before redesigning the creative strategy. The razor is especially valuable in high-stakes environments where the temptation toward complex narratives is strongest. Conspiracy theories, elaborate corporate strategies, and multi-causal explanations all have a seductive quality — they make the explainer feel sophisticated. But sophistication that doesn't improve accuracy is just noise dressed up as signal. The practical discipline is straightforward: generate your hypotheses, rank them by the number of assumptions each requires, and test them in that order. When you find an explanation that accounts for the evidence without leftover anomalies, you've likely found your answer. If it doesn't hold, move to the next hypothesis on the list. This is not glamorous thinking. It is effective thinking.