Occam's razor
Occam's razor is the rule that entities should not be multiplied beyond necessity: when two explanations account for the same evidence equally well, prefer the one that requires fewer assumptions. It is a tie-breaker, and both halves of that description carry weight. It applies only when the explanations really are level on the evidence, and it selects for fewer assumed things rather than for a shorter or friendlier answer. Almost everything that goes wrong with the razor in practice comes from dropping one of those two conditions.
What the principle actually says
The principle says something narrower than its reputation suggests. It ranks explanations by what they ask you to take on: the number of independent entities, mechanisms, coincidences and special conditions each one has to assume before it can produce the observations you already have. The explanation carrying the smaller load is preferred. Preferred is not the same as true, and preference here is provisional, a place to stand while better evidence is collected rather than a verdict on the world.
The name comes from William of Ockham, an English Franciscan friar and logician of the 14th century, who used the principle repeatedly in his arguments. The Latin formula usually quoted alongside it, entia non sunt multiplicanda praeter necessitatem, does not appear in his surviving writings and was put into that exact form by later authors. Versions of the idea are older still, and Ockham's contribution was to wield it consistently rather than to invent it. It is called a razor because a razor shaves: the metaphor is about cutting away assumptions that are doing no work, not about being sharp or decisive.
Ockham or Occam, and how to say it
Both spellings refer to the same person and the same principle. Ockham is the spelling of the Surrey village he came from and the form used in most scholarly writing; Occam is the older Latinized rendering, and it is the one that survived in the phrase Occam's razor. There is no difference in meaning, so ockham razor versus occam razor is a question about orthography and not about ideas. It is said OCK-um, with the stress on the first syllable and the second reduced almost to nothing.
The misreading: the simplest answer is not automatically right
The popular version of the razor, that the simplest explanation is usually the correct one, is a different claim and a false one. It fails in 3 distinct ways, and each failure is worth being able to name.
- Simplicity is not the criterion. The razor counts assumed entities, not words, steps or how easy the explanation is to state. An explanation can be short to say and expensive to believe.
- It only applies to a tie. The condition is that both explanations account for the evidence equally well. When one explains more, it wins on evidence and the razor never gets picked up.
- It selects, it does not verify. A preference is not a demonstration. The world is under no obligation to be economical, and complicated explanations are true all the time.
A worked case shows the shape of it. An office printer jams on Monday mornings and on no other day. Explanation A: paper left in the tray absorbs moisture over the weekend and feeds unevenly. Explanation B: a colleague deliberately misaligns the tray each Friday, has done so for months and has told nobody. Both fit the pattern exactly. B assumes an additional actor, a sustained intention, a repeated action and continued secrecy, which is 4 assumed entities against 1 for A, so A is preferred. Now change the evidence: the jam also happens on the first morning after a public holiday, when nobody has been in the building. A survives untouched and B needs a new assumption, so the gap widens. Change it again: a jam occurs on a Wednesday afternoon in dry conditions, 10 minutes after the tray was refilled. The explanations are no longer tied at all, and the razor has nothing to do, because the evidence has started discriminating between them by itself. That is the normal outcome. The razor is for the stretch of an inquiry when the evidence is silent, and it hands the floor back the moment the evidence speaks.
Is Occam's razor a theory
No. It is a heuristic, a rule of method, and it belongs to the same family as other working principles rather than to the family of theories about the world. A theory makes claims that observations can contradict. The razor makes no claim about what exists, so no observation can contradict it, and calling it Occam's razor theory misdescribes what kind of thing it is. It is also not a law and not a proof, and it cannot function as a step in an argument. Nobody has ever established a conclusion by pointing at the razor.
The justification for it is practical rather than empirical. An explanation with fewer independent assumptions has fewer places to be wrong, is easier to test, and can be corrected more cheaply when it fails. Working scientists use parsimony in exactly this way when several models predict the same observations, including in cosmology, where models of the early universe that reproduce the same measurements are routinely compared on how many free parameters each one needs. The model with fewer adjustable parts is preferred, and it is preferred until a measurement separates them, at which point the preference is abandoned without any embarrassment.
Counting assumptions is harder than it looks
Counting assumptions is harder than it looks because parsimony has to be measured before it can be applied, and the count depends on how the explanations are described. 3 problems recur.
| Problem | What goes wrong | What to do instead |
|---|---|---|
| Simple for whom | An explanation looks simple because it is familiar, not because it assumes less | List the entities each explanation requires and compare the lists, not the impressions |
| Simple in what | One explanation is simpler in mechanism while the other is simpler in mathematics | State which kind of parsimony is being invoked before invoking it |
| Hidden assumptions | The apparently plain explanation smuggles in a large unstated premise | Write out the premise and add it to the count |
There is also a standing counterweight, Hickam's dictum, which is the reminder that reality is under no obligation to arrange a single cause for everything that appears at once. Several unrelated things going wrong in the same week is a coincidence, and coincidences are common. Insisting on one explanation for a cluster of observations is a misapplication of the razor, because uniting them is itself an assumption and it belongs in the count. Combining the razor with a causal claim compounds the risk: a single tidy cause is more attractive than four separate ones, and attractiveness is not evidence, which is where correlation and causation get quietly confused.
How to use the razor on a claim you meet tomorrow
Use the razor in 4 steps, and stop at the first step that fails.
- Write both explanations out in enough detail that each one actually produces the observations, rather than gesturing at them.
- Check that they are tied. Does each account for every piece of evidence, including the awkward pieces? If not, the evidence decides and the razor is not needed.
- Count the entities. How many separate things, events, intentions or coincidences does each explanation require you to assume, that you did not already accept?
- Prefer the lighter one provisionally and name the observation that would overturn the preference.
The question to carry away is the second step in plain form: are these two explanations really equally supported, or have I called them equal because I prefer the conclusion of the lighter one? The razor is a genuinely powerful instrument and it is most often used badly by people who have already decided, because it can be pointed at whichever explanation they like least and made to look like an argument. It is not an argument. It is a way of allocating attention while the evidence is being gathered, and what counts as evidence is settled elsewhere and by different standards. When someone answers a detailed case by saying that the simpler explanation is more likely, they have named a preference and produced no support for it, and the conclusion does not follow from the premise: that is a non sequitur wearing a respectable name. Epistemology is the field that examines why a preference of this kind is reasonable to hold at all, and it does not conclude that the world is simple. It concludes that our explanations should not be more complicated than the evidence requires, which is a statement about us.