Named for William of Ockham (Summa Logicae, c. 1323), though the principle of parsimony is older than he is.
Prefers the explanation that needs the fewest added assumptions when competing explanations fit the evidence equally well.
Three explanations for a sales dip, and how many unshown assumptions each needs
Each claim an explanation needs, and the evidence does not yet show, scores 1 under it and costs a point (weight −1); an explanation that fails a fact is out. The price rise needs the fewest, so it comes out on top. Then the store data arrives showing sales fell as much where prices did not rise: set the price rise to Fail on that row and the mark passes to the next simplest. Illustrative.
Fits the facts: sales fell 12 percent from May on
MustThe May price riseA rival's new podsShift to whole beanFits the store data, once it arrives
MustThe May price riseA rival's new podsShift to whole beanAssumes shoppers balk at a 10 percent rise
Weight−1The May price riseA rival's new podsShift to whole beanAssumes the rival's pods reached the brand's shelves in May
Weight−1The May price riseA rival's new podsShift to whole beanAssumes shoppers who tried them kept buying them
Weight−1The May price riseA rival's new podsShift to whole beanAssumes coffee drinkers began moving to whole bean this spring
Weight−1The May price riseA rival's new podsShift to whole beanAssumes the move reached the brand's pods before other brands
Weight−1The May price riseA rival's new podsShift to whole beanAssumes it showed in sales within weeks
Weight−1The May price riseA rival's new podsShift to whole beanWeighted total
The May price rise−1OutA rival's new pods−2OutShift to whole bean−3OutSimplest explanation that fitsNo explanation fits every fact: look for a new one.The May price riseA rival's new podsShift to whole bean
Fewest assumptions first
- −1The May price riseA rival's new podsShift to whole beanOut: fails a must (Fits the facts: sales fell 12 percent from May on).Out: fails a must (Fits the store data, once it arrives).
- −2The May price riseA rival's new podsShift to whole beanOut: fails a must (Fits the facts: sales fell 12 percent from May on).Out: fails a must (Fits the store data, once it arrives).
- −3The May price riseA rival's new podsShift to whole beanOut: fails a must (Fits the facts: sales fell 12 percent from May on).Out: fails a must (Fits the store data, once it arrives).
An example
A maker of premium coffee pods saw sales fall 12 percent in the quarter after a 10 percent price rise in May. The leadership team had three stories. The price rise was the obvious one; the head of sales blamed a rival's new pods, launched the same month; a board member argued that coffee drinkers were moving to whole bean coffee. The planning lead wrote out what each story needed that the evidence did not already show. The price rise needed one claim, that shoppers balk at a 10 percent increase. The rival's pods needed two, that they reached the same shelves in May and that shoppers who tried them stayed. The whole bean shift needed three. The team adopted the price story.
Six weeks later, store-level data arrived. Sales had fallen just as far at retailers that had absorbed the increase and kept the old shelf price. The price story could not survive that, so the team moved to the rival's pods, which fit everything. A field check found them beside the brand in four of five stores.
Why it works
Almost any explanation can be made to fit the facts if it is allowed enough extra assumptions. Each assumption is a separate chance to be wrong, so an account that needs three claims, each fairly likely, is less likely overall than one that needs a single claim. Preferring the leaner account is a bet on that arithmetic, and it also aims the next step: the simplest story is the cheapest to test, and its one assumption is where to look first.
The razor only chooses among explanations that fit the evidence equally well, and its choice is provisional. When a new fact contradicts the simple account, the account goes and the next simplest takes its place. Patching the favorite with a fresh assumption to keep it alive is the habit the rule exists to stop.
The thinking behind it
William of Ockham William of Ockham, an English Franciscan and scholastic logician, wrote in his Summa Logicae (about 1323) and in his lectures on the Sentences of Peter Lombard that plurality must never be posited without necessity, and that it is futile to do with more what can be done with fewer. The rule is older than he is: Aristotle, Ptolemy, Robert Grosseteste, and Duns Scotus all stated a version of it. The familiar wording about entities is John Punch's, from 1639, and the name Occam's razor first appears in Libert Froidmont's Philosophia Christiana de Anima (1649).
How to run it
- List the candidate explanations. Write down every account anyone in the room believes, including the dull and the unwelcome ones.
- Write the evidence each must fit. List the established facts with dates and sources, so every account is tested against the same set.
- Drop any that contradicts a fact. An explanation that cannot fit an established fact is out before any counting begins.
- List each survivor's assumptions. Name every claim it needs that the evidence does not already show, one claim per line.
- Adopt the leanest, provisionally. Act on the account with the fewest assumptions, and record that it is a working choice.
- Test its weakest assumption. Find the cheapest check of the claim it rests on, and run it.
- Move on when the facts demand it. When new evidence contradicts the account, switch to the next leanest instead of adding an assumption to save it.
Where it goes wrong
Counting gets gamed. One explanation hides three claims inside a single phrase, such as market conditions, while a rival's are listed one by one; spot it when an assumption contains the word and, and split compound assumptions before counting.
The second is reading simple as familiar. The story the team already believes feels simpler because it needs no new thinking, even when it rests on more unshown claims. Count the assumptions on paper, one claim per line.
The third is using the razor where the fit is unequal. A complex account that explains a fact the simple one cannot is the better account, however many assumptions it carries. Check fit first, and use the count only among accounts that fit equally well.
When to use something else
- Hanlon's Razor. When the competing explanations are bad intent and ordinary error, and the simpler reading is usually the mistake.
- Bayesian Updating. When base rates and evidence of known strength exist, so the explanations can be weighed by probability.
- Abductive Reasoning. When no good candidate explanation exists yet and one has to be built from surprising facts.
Where it sits in SOLVE eX
Go to the source
- Summa Logicae, c. 1323