The principle of parsimony says that when multiple explanations fit the same evidence, you should prefer the simplest one. More precisely, it tells you not to multiply assumptions beyond what’s necessary to explain what you observe. This idea runs through nearly every scientific discipline, from physics to medicine to statistics, though it plays out differently in each one.
You’ve probably heard it called Occam’s Razor. The principle traces back to the medieval philosopher William of Ockham (1285–1347), who stated it in Latin: Pluralitas non est ponenda sine necessitate, or “Plurality should not be posited without necessity.” A more common English phrasing is “Entities are not to be multiplied beyond necessity.” The core idea is straightforward: don’t invent extra causes, forces, or mechanisms when fewer ones already explain what you see.
Parsimony as a Rule of Scientific Reasoning
Parsimony isn’t just a philosophical preference. It’s baked into the foundations of modern science. Isaac Newton included a version of it as the very first of his three “Rules of Reasoning in Philosophy” at the start of his Principia Mathematica in 1687: “We are to admit no more causes of natural things than such as are both true and sufficient to explain their appearances.” He added that “Nature is pleased with simplicity, and affects not the pomp of superfluous causes.” Galileo held a similar view.
The principle doesn’t claim simpler explanations are always correct. It’s a guideline for choosing between competing hypotheses when the evidence doesn’t clearly favor one over another. If two theories predict the same observations equally well, parsimony says you should go with the one that makes fewer assumptions, because each additional assumption is another place where you could be wrong.
Building Evolutionary Trees
One of the most concrete applications of parsimony is in evolutionary biology, where researchers use it to figure out how species are related. When building a phylogenetic tree (a branching diagram of evolutionary relationships), scientists compare different possible arrangements of species and count how many evolutionary changes each arrangement requires.
The tree that requires the fewest changes wins. For example, imagine two possible trees explaining how vertebrates are related. One requires six evolutionary changes. The other requires seven, including a bony skeleton evolving independently in two separate lineages. Both trees technically fit the data, but parsimony favors the first because it doesn’t require the same complex trait to evolve twice from scratch. The method, called maximum parsimony, has no built-in model of how evolution works at the molecular level. It simply minimizes the total number of changes needed to explain the observed differences between species.
Medical Diagnosis: When Parsimony Helps and Hurts
In medicine, diagnostic parsimony means trying to explain all of a patient’s symptoms with a single disease rather than invoking multiple unrelated conditions. Medical students are taught this principle early, and it works well in many situations. Some conditions are mutually exclusive by nature: a jaw can’t be both abnormally small and abnormally large, and Addison’s disease (where the adrenal glands produce too little cortisol) can’t coexist with Cushing’s syndrome (where they produce too much).
But parsimony has real limits in the clinic. Many diseases cluster together. High blood pressure and diabetes frequently occur in the same patient. An older, frail person may have three or four conditions at once, none of which fully explains the others. The counterargument to parsimony in medicine has a name: Hickam’s dictum, which simply states that a patient can have as many diseases as they please.
Leaning too hard on parsimony can lead to what’s called premature diagnostic closure, where a doctor settles on a single diagnosis and stops looking. A study examining hypothetical COVID-19 scenarios found exactly this pattern: when health care workers were presented with a patient who already had a diagnosis of mononucleosis, they estimated the odds of that patient also having COVID-19 as 73% lower than they would have otherwise. The presence of one explanation made clinicians less likely to consider a second, potentially serious one.
Preventing Overfitting in Statistics
In statistics and machine learning, parsimony solves a very practical problem: overfitting. When you build a predictive model, you can always improve its fit to your existing data by adding more variables. But a model with too many variables starts capturing random noise rather than real patterns. It performs beautifully on the data you built it with, then falls apart when it encounters new data.
A parsimonious model uses only the variables that genuinely help predict the outcome and drops the rest. Researchers balance this tradeoff using tools like the Bayesian Information Criterion (BIC), which penalizes models for complexity. The goal is a sweet spot: enough variables to capture real relationships, but not so many that the model becomes fragile. Techniques like LASSO regression enforce parsimony directly by shrinking less useful variables toward zero, effectively removing them from the model. The result is a simpler model that tends to perform better in the real world, and one that clinicians or decision-makers can actually interpret and use.
Animal Behavior and Morgan’s Canon
Psychology has its own version of parsimony called Morgan’s Canon, first published in 1894 by the British psychologist C. Lloyd Morgan. The rule states: “In no case may we interpret an action as the outcome of the exercise of a higher psychical faculty, if it can be interpreted as the outcome of one which stands lower in the psychological scale.”
In plain terms, if a dog learns to open a gate latch, you shouldn’t assume it’s reasoning through the problem like a human unless you have strong evidence for that. The simpler explanation, that the dog learned through trial and error, should be your starting point. Morgan’s Canon was developed during a period when many writers attributed rich human-like thought to animals based on anecdotes alone, and it pushed researchers toward more conservative, evidence-based interpretations.
Interestingly, Morgan himself was clear that his canon wasn’t purely about simplicity. He acknowledged that attributing human-like reasoning to animals can sometimes seem like the simpler explanation. His point was more specific: when you don’t yet have evidence that an animal possesses a higher cognitive ability, you should default to a lower-level process whose existence in that animal is already established. He even cautioned that “the simplicity of an explanation is no necessary criterion of its truth.”
What Parsimony Does Not Mean
The most common misunderstanding of parsimony is that it claims simple explanations are always true. It doesn’t. Nature is often complicated, diseases cluster, and simple theories sometimes turn out to be wrong. Parsimony is a starting point, not a conclusion. It tells you which explanation to prefer when you lack evidence to distinguish between alternatives, essentially placing the burden of proof on the more complex hypothesis.
It also doesn’t mean you should strip away complexity that the evidence actually supports. If your data clearly requires three variables to explain a pattern, a two-variable model isn’t more parsimonious; it’s just wrong. Parsimony only applies when the simpler and more complex explanations both fit the available evidence equally well. The moment new evidence favors the more complex explanation, parsimony steps aside.