Occam's Razor Without the Bumper Stickers: Simplicity, Models, and What Counts as an Explanation

Popular science loves a slogan: “the simplest explanation is usually true.” That bumper sticker is doing more work than it should. Occam’s razor (often traced to William of Ockham’s preference not to multiply entities beyond necessity) is a rule of thumb about how we build and compare explanations — not a metaphysical guarantee that the universe is tidy, and not a license to reject every detailed model because a one-line story feels cleaner. This is a philosophy-of-science walk through what “simplicity” can mean when you are trying to understand the world carefully. No politics, no invented study percentages, and no claim that philosophers settled the matter once and for all. ### What the razor is — and is not At its best, Occam’s razor is a preference among competing explanations that already fit the same observations: do not add extra moving parts that do no explanatory work. If two models predict the same data equally well, the one that introduces fewer unsupported assumptions is usually the better place to start — because unsupported assumptions are places you can be wrong without noticing. What the razor is not: 1. **A proof that nature is simple.** Nature can be messy. The razor is about our theories, not a decree that complexity is forbidden. 2. **A reason to ignore better-fitting evidence.** A “simple” story that fails a clean observation loses to a more detailed story that survives it. Fit and risk of error both matter. 3. **A warrant for vague slogans.** “Energy wants to…” or “evolution wants to…” can sound simple while smuggling agency the data never justified. Short is not the same as clear. ### Two kinds of “simple” that get confused Philosophers and scientists often mix at least two notions: - **Ontological thrift:** fewer kinds of things or forces. Preferring one mechanism over two independent coincidences is thrift of this sort. - **Descriptive or computational thrift:** fewer free parameters, shorter equations, easier predictions. A model can be ontologically rich (many particles) yet descriptively tight (one elegant framework), or the reverse. When someone says “simpler,” ask which ledger they mean. A classroom that only counts “how many words are on the slide” will reward slogans. A lab that counts “how many knobs we must tune without independent constraint” is closer to the useful scientific habit. ### Models are tools, not photographs A scientific model is usually a deliberate simplification: a map that keeps the relations you need and drops the rest. Maps are not the territory — and that is a feature when the map lets you navigate. Newtonian gravity is “wrong” as a final description of spacetime, yet it remains an excellent model for countless engineering and orbital tasks within its domain. Calling it “just a model” is not an insult; it is a reminder to state the domain. Good explanatory practice therefore includes: - **Scope.** Where does the model apply, and where do we already know it fails? - **Risk.** Which assumptions, if false, would quietly ruin the conclusion? - **Comparators.** What alternative would we have expected to see if this mechanism were absent? Those questions are more useful than arguing whether a model is “real” in some absolute sense. ### Explanation versus storytelling An explanation, in the sense scientists often care about, does more than narrate. It connects phenomena to mechanisms or patterns that also constrain new cases. “The leaves turned red because autumn arrived” is a calendar story. “Chlorophyll broke down while carotenoids remained visible, and in some species anthocyanins were synthesized under light and sugar conditions” is closer to mechanism — and it suggests what to check on a shaded branch or a different species. Bumper-sticker Occamism sometimes prefers the calendar story because it is shorter. Mechanism-first Occamism prefers the account that earns its parts: each piece should either improve fit, unify cases, or make a risky prediction you could in principle test. ### Where beginners get burned 1. **Equating unfamiliar with complex.** A vector equation can look “complicated” and still be the thrifty description; a paragraph of metaphors can look “simple” and hide three free parameters. 2. **Treating consensus rhetoric as thrift.** “Everyone knows” is not an entity count. Demand the mechanism or the measurement. 3. **Using the razor to shut down curiosity.** “That’s needlessly complicated” is sometimes a power move, not a comparison of models that both fit the data. Ask for the competitor that fits equally well with fewer risks. 4. **Forgetting measurement.** Two theories can look equally simple on paper until you ask what apparatus would distinguish them. Explanations that never touch a possible observation drift toward aesthetics. ### A portable checklist (no fake percentages) When you meet a neat explanation — in a textbook, a thread, or a keynote — try this: 1. What observations does it actually constrain? 2. What could I drop without changing any prediction? 3. What would surprise this account if it happened tomorrow? 4. Is “simple” here about fewer entities, fewer knobs, or just fewer words? If you cannot answer (3), you may have a story rather than an explanation. If you cannot answer (2), you may be carrying spare parts. ### Discussion questions 1. Pick a model you trust in your field (ideal gas, ray optics, Hardy–Weinberg, Keplerian orbits, a circuit equivalent). Which deliberate falsehoods does it keep, and why are those falsehoods useful rather than dishonest? 2. Give an example where the rhetorically “simpler” public story and the working technical model diverge — and say which ledger of simplicity each side was optimizing. 3. When has adding a new entity or parameter genuinely earned its keep in something you study (by unifying cases or surviving a test), and when was it decorative? Clear thinking beats bumper stickers. Corrections to the framing are welcome; claims that need clinical, legal, or engineering authority belong with qualified professionals and primary data — not with a philosophy tour.