Mira had three explanations for the same mistake.
The first was short.
“I was careless.”
The second was more sophisticated.
“I probably lost concentration because the question came after a difficult section, and then I rushed because I felt behind, which made me anxious, which caused the sign error.”
The third was less dramatic.
“I consistently distribute the negative sign incorrectly when the bracket contains two terms.”
Her tutor asked which explanation made the clearest prediction.
If the third explanation were correct, Mira should make the same error again in a calm, untimed question containing the same structural feature.
They tested it.
She did.
The simplest useful explanation is not the shortest sentence. It is the model that explains the evidence without carrying machinery the evidence does not need.
The 60-Second Route
Model parsimony is the preference, other things being sufficiently equal, for an explanation or model that accounts for the relevant evidence with fewer unnecessary assumptions, exceptions or moving parts.
The phrase “other things being equal” matters enormously.
A simpler model is not automatically better if it explains less, predicts worse, ignores important conditions or erases real complexity. A complicated model is not automatically better merely because it sounds sophisticated. High-performance reasoning asks how much explanatory and predictive work each part of a model is doing.
The learner should prefer a model that is as simple as the evidence permits, but no simpler than the problem permits.
Parsimony Is Not “Always Choose the Simplest Answer”
Occam’s razor is frequently paraphrased too aggressively.
Students hear “the simplest explanation is usually correct” and turn a useful heuristic into a law.
That is not safe.
A simple explanation that fails the evidence is merely wrong.
A simple model that works only by hiding important exceptions is not parsimonious in the useful sense; it is incomplete.
A complex system may genuinely require a complex model.
A 2025 review in Proceedings of the National Academy of Sciences revisited the principle of parsimony in modern scientific modelling and argued precisely this point: simplicity remains useful, but increasingly complex scientific models can be essential, and parsimony should be treated as one desirable property among others such as predictive accuracy, interpretability and research usefulness.
The educational rule is therefore conditional:
Prefer fewer assumptions only after the model still fits what matters.
What Counts as Unnecessary Complexity?
Complexity can enter an explanation in many ways.
- too many independent causes;
- too many exceptions;
- too many free assumptions;
- too many special rules for individual cases;
- too many parameters whose values are chosen only after the answer is known;
- too much detail that does not alter prediction;
- too many steps when one structural relationship would explain the same result.
Complexity is not synonymous with length.
A long explanation can describe one coherent mechanism carefully.
A short explanation can hide five unsupported assumptions inside one sentence.
The Assumption Budget
One practical way to teach parsimony is to give every explanation an assumption budget.
For each claim, ask:
- What must be true for this explanation to work?
- Which of those conditions are directly supported?
- Which are merely being assumed?
- Can one assumption be removed without reducing fit?
- Does another model explain the same evidence with fewer unsupported commitments?
This turns “simple versus complicated” into a more disciplined comparison.
Fit Comes Before Elegance
A model earns the right to be compared on simplicity only after it fits the important evidence reasonably well.
Suppose two explanations compete.
Model A uses one mechanism but cannot explain three central observations.
Model B uses two mechanisms and explains all three.
Choosing A merely because it is simpler is not parsimony.
It is underfitting.
The model has been simplified past the point where it can represent the phenomenon adequately.
Overfitting: When a Model Explains the Past Too Perfectly
The opposite problem is overfitting.
A model becomes so flexible that it can explain every observed detail after the fact.
That sounds impressive until a new case arrives.
Because the model captured accidental features rather than stable structure, prediction collapses.
Students overfit too.
- They memorise every surface feature of a worked example rather than the governing relationship.
- They invent a special explanation for each practice error instead of identifying a recurring mechanism.
- They learn essay phrases tied to one topic rather than rhetorical functions that transfer.
- They build Science explanations around the exact apparatus rather than the causal structure.
Parsimony is one defence against this accumulation of case-specific patches.
Underfitting: When Simplicity Erases the Problem
Students also underfit.
“I was careless” is a classic underfit model.
It can explain almost any mistake because it explains almost nothing.
“I need more practice” is another.
Practice of what?
Recognition?
Retrieval?
Method selection?
Execution?
Transfer?
Pacing?
A parsimonious explanation is compact but discriminating.
It removes unnecessary detail without deleting the mechanism that determines the next action.
Parsimony and Knowledge Compression Are Related but Different
Knowledge Compression concerns turning many details into compact usable structures.
Model parsimony concerns choosing among competing explanatory structures.
A learner may compress ten examples into one rule.
Parsimony then asks whether that rule is sufficient, whether it requires exceptions, and whether a different rule explains the same examples with fewer unsupported assumptions.
Compression is about representation economy.
Parsimony is about explanatory economy.
Parsimony and Evidence Weighting
Simplicity should never outweigh decisive evidence.
This is where Evidence Weighting enters.
Suppose the simple model explains eight weak clues but conflicts with one direct measurement.
A more complex model explains the direct measurement and the broader pattern.
The learner should not count clues.
They should weight them.
Parsimony operates inside evidence discipline.
Parsimony and Prediction Error
A simple model becomes valuable when it predicts.
If a compact explanation makes clear predictions and those predictions repeatedly survive, confidence can rise.
If prediction errors accumulate, simplicity no longer rescues the model.
See Prediction Error.
A parsimonious model should compress past evidence and generate useful expectations about new cases.
Parsimony and Boundary Conditions
One way a model becomes falsely simple is by hiding its boundaries.
“All questions with this keyword use this method” is simple.
It is also wrong if the keyword appears across several structural categories.
A better model might be slightly more complex:
Use this method when the keyword appears together with these two structural conditions; otherwise inspect the relationship first.
The additional condition increases complexity but reduces exceptions and negative transfer.
This is why Boundary Conditions are part of good parsimony.
Parsimony in Mathematics: Prefer the Structure That Does the Work
Mathematics rewards parsimonious representation.
A complicated arithmetic route may solve one instance.
An equation may reveal the general relationship.
A graph may expose the same relationship more directly than several symbolic manipulations.
A factorised expression may reveal zeros and multiplicity that an expanded form hides.
The most parsimonious mathematical form depends on the job.
There is no universally simplest representation.
There is a representation that exposes the relevant structure with less unnecessary work.
Mathematical Elegance Is Not the Same as Pedagogical Parsimony
An expert may compress a solution into three elegant lines.
A novice may need eight lines to keep the reasoning observable.
The three-line solution is locally shorter.
The eight-line solution may be more parsimonious for learning because each line performs a necessary diagnostic function.
Parsimony must therefore be relative to purpose.
Do not remove steps that are doing important cognitive work merely because an expert no longer needs them.
Parsimony in Mathematical Modelling
A model of a real situation should include enough variables to explain what matters and no more than necessary for the question being answered.
Too few variables and the model misses important structure.
Too many and the model becomes difficult to estimate, interpret or use.
Students can ask:
- Which variable changes the prediction materially?
- Which variable is merely decorative?
- Which assumption simplifies without distorting the purpose?
- Which omitted factor would change the conclusion?
This connects with Sensitivity Analysis.
Parsimony in Science: Models Are Tools, Not Miniature Reality
Scientific models simplify.
That is not a defect by itself.
A model leaves things out so the relationship of interest becomes tractable.
The educational mistake is to think a more detailed model is automatically more scientific.
More detail is useful only if it improves explanation, prediction or the question being investigated.
A good Science student learns to ask:
- What phenomenon is this model trying to explain?
- Which variables are essential?
- What has been deliberately ignored?
- Where does the simplification stop working?
- Would adding another mechanism change a prediction?
This is parsimony with explicit limits.
The Scientific Exception Count
A model can look simple until exceptions accumulate.
Suppose one explanation requires a new special rule for every unexpected result.
At some point, the collection of exceptions is itself complexity.
Another model may initially look more complex but explain all those cases through one additional mechanism.
The second model can be more parsimonious overall.
Students should therefore count hidden exception costs, not just visible variables.
Parsimony in English Reading: Do Not Invent What the Text Does Not Need
Reading comprehension often tempts students to construct elaborate psychological stories around limited evidence.
A character pauses before answering.
The learner decides the character is jealous, insecure, hiding a family secret and afraid of future rejection.
The text supports only hesitation.
Parsimonious reading keeps interpretation proportional.
Choose the smallest inference that explains the available evidence without inventing unsupported machinery.
If later evidence supports a richer interpretation, expand the model then.
Parsimony in Literary Interpretation Does Not Mean One Correct Meaning
Some texts legitimately support multiple interpretations.
Parsimony does not eliminate ambiguity by decree.
It asks which interpretation accounts for the textual evidence with fewer unsupported assumptions and fewer contradictions.
Two interpretations may remain viable.
Then confidence should remain proportional.
Parsimony is a comparison tool, not a machine that manufactures certainty.
Parsimony in Argument Writing: One Claim Should Do More Work
Weak arguments often accumulate points.
Point one.
Point two.
Point three.
Point four.
The essay becomes a list because no central model organises the evidence.
A parsimonious argument looks for a smaller number of claims with greater explanatory reach.
Can one underlying mechanism explain several observations?
Can two repetitive paragraphs be unified under one stronger distinction?
Can an exception be handled by refining the thesis rather than adding a completely separate argument?
Parsimony gives writing architecture.
Parsimony in Narrative Writing: Remove What Does Not Change the Story
Narratives also contain models.
Every scene, detail and character should ideally contribute to causality, tension, meaning or atmosphere.
A scene that can disappear without changing anything important may be ornamental load.
The parsimonious writer asks:
If I remove this, what explanatory, emotional or causal work disappears?
If the answer is “almost none,” the story may become stronger by subtraction.
Parsimony in Study Diagnosis: Stop Explaining Everything with Motivation
Families often use large global explanations.
“He is lazy.”
“She lacks discipline.”
“He just needs confidence.”
These models feel simple because they use one cause.
But they are often not parsimonious because they generate too many unexplained exceptions.
The learner works hard in Science but avoids Mathematics.
The learner is confident in class but freezes on mixed questions.
A more specific mechanism may explain the evidence better:
Method selection becomes slow when several algebraic strategies compete, so homework feels disproportionately effortful.
That explanation is slightly longer but much more useful.
The Diagnostic Parsimony Rule
A good learning diagnosis should explain several observations and make a testable prediction.
If the diagnosis is “retrieval access is weak after delay,” then:
- warm performance should be stronger than cold performance;
- minimal cues should restore performance quickly;
- mixed delayed retrieval should expose the weakness.
If those predictions fail, update the model.
A diagnosis that can explain every possible result is not diagnostic.
Parsimony and Information Gain
When several models remain plausible, do not gather information randomly.
Use Information Gain.
Ask which question produces different predictions under the competing models.
A simple model and a complex model may explain the current data equally well.
Design a test where they disagree.
The result tells you whether the additional complexity is earning its place.
Parsimony and Counterfactual Testing
Counterfactual Testing is another useful tool.
Remove one assumption from the explanation.
Does the prediction change?
If not, the assumption may be redundant.
Change one condition.
Does the model still predict coherently?
Counterfactuals expose which parts of the model are doing causal work.
Parsimony and Sensitivity Analysis
Some model components barely affect the conclusion.
Others dominate it.
Sensitivity Analysis helps identify which variables deserve representation.
If a parameter can vary widely without changing the decision, precise modelling of that parameter may add complexity with little value.
If a small change reverses the conclusion, the parameter cannot be simplified casually.
Parsimony and Analogical Mapping
Analogical Mapping benefits from parsimonious source models.
If the source case is stored with dozens of irrelevant surface details, transfer becomes difficult.
If it is represented through a small set of important relationships, mapping to a new case becomes easier.
Parsimony therefore helps identify what should travel.
Parsimony and Negative Transfer
Oversimplified rules can create Negative Transfer.
“All rate questions use the same setup.”
“All inference questions require the same sentence frame.”
“All strong essays need three arguments.”
These models are simple because they erase boundaries.
Useful parsimony preserves the distinctions that change the decision.
The Model Comparison Table
When students compare explanations, use a five-column mental table.
- Fit: how much of the important evidence does the model explain?
- Assumptions: how many unsupported commitments does it require?
- Exceptions: how many special patches are needed?
- Predictions: what does it say should happen next?
- Usability: can the learner apply and communicate it?
Parsimony is the relationship among these columns, not one number.
The One-Mechanism Test
When several errors appear, ask whether one mechanism could explain them together.
Suppose a student loses marks in:
- percentage change;
- ratio comparison;
- speed questions;
- graph gradients.
Four topics are visible.
Perhaps one deeper weakness explains all four: the learner does not reliably identify what quantity is being measured relative to what base.
If that single mechanism predicts the error pattern across topics, it is a strong parsimonious diagnosis.
The Exception Test
Now look for an exception.
If the learner handles graph gradient perfectly, the “relative quantity” model may need refinement.
Perhaps the weakness is specifically verbal identification of the reference quantity rather than the mathematical relation itself.
Exceptions are not annoyances to be ignored.
They tell us whether the simple model has been simplified too far.
The Predictive Test
A diagnosis should predict an unseen case.
If the model says the learner’s difficulty is selecting a base quantity in verbal problems, create a fresh problem with an unfamiliar context but the same structural decision.
If the error reappears, the model gains support.
If performance is strong, the explanation needs revision.
This is how parsimony avoids becoming storytelling.
The Intervention Test
A useful model should also suggest an intervention.
If “carelessness” is the diagnosis, what exactly should the student practise?
The answer is unclear.
If “negative-bracket sign distribution is not automated and fails under compression” is the diagnosis, the intervention becomes specific:
- contrast correct and erroneous expansions;
- make the sign transition visible;
- build a detector;
- practise to automaticity;
- retest under mixed conditions.
A model that guides repair has operational value.
When Complexity Is Worth Paying For
There are good reasons to use a more complex model.
- The simple model misses important evidence.
- The simple model repeatedly produces prediction error.
- The extra mechanism explains several previous exceptions at once.
- The decision is sensitive to a variable the simple model ignores.
- The stakes justify more precise modelling.
- The complex model is still usable enough for the purpose.
The key is that complexity must pay rent.
Every added component should improve fit, prediction, interpretation or decision quality enough to justify its cost.
When Simplicity Is Worth Paying For
Simpler models can have important advantages.
- They are easier to retrieve.
- They are easier to communicate.
- They make causal structure more visible.
- They can generalise better when unnecessary case-specific detail has been removed.
- They reduce decision latency.
- They are easier to falsify because predictions are clearer.
But these advantages matter only while the model remains adequate.
Parsimony and High-Stakes Examinations
Examinations reward usable models.
A student may know five equivalent methods, ten exceptions and twelve explanatory details.
Under time pressure, the strongest internal representation is often a compact model that preserves the decision-changing structure.
This is not about writing less merely to be fast.
It is about carrying a mental model small enough to retrieve and rich enough to govern the task.
The Exam Model Card
For a difficult recurring concept, a student can create a compact model card.
- Core relation: what is the central mechanism?
- Trigger: what cue suggests it?
- Boundary: when does it fail?
- Prediction: what should happen if it applies?
- Check: what cheap observation would reveal an error?
This is parsimonious because every item has a function.
Remove one and the model becomes less operational.
Parsimony and Graceful Degradation
The first Batch 15 article on Graceful Degradation asks what should remain when resources worsen.
Parsimonious models are valuable in degraded conditions because they preserve high explanatory value with lower cognitive cost.
But this creates a boundary:
Do not degrade into an oversimplified rule that no longer fits.
Compress optional detail.
Preserve decision-changing distinctions.
Parsimony and Error Detectability
The second Batch 15 article on Error Detectability asks how mistakes can become visible.
A parsimonious model often improves detectability because the learner knows what should happen if the model is true.
When prediction and outcome diverge, the contradiction is easier to locate.
An overcomplicated model can hide error because almost any outcome can be rationalised through one of its many moving parts.
Parsimony and Self-Explanation
Self-explanation can reveal whether complexity is necessary.
Ask the learner to explain why each step or assumption is present.
If they cannot say what explanatory work a component performs, it may be redundant.
If removing it breaks a prediction or boundary, it is earning its place.
This makes Self-Explanation a tool for model pruning.
The Parent Version: Replace Labels with Testable Models
Parents naturally use shorthand.
“Careless.”
“Unmotivated.”
“Weak foundation.”
Before treating the label as an explanation, ask whether it predicts anything specific.
If the child has a “weak foundation,” which prerequisite should fail on a clean diagnostic probe?
If the child is “careless,” what error pattern should recur across tasks?
If the child is “unmotivated,” why does effort vary sharply by subject or task type?
Good explanations reduce blame by becoming more precise.
The Tutor Version: Find the Smallest Model That Predicts the Error Pattern
A tutor can build candidate models from observed work.
Suppose a learner fails four questions.
Do not assume four separate weaknesses.
Ask whether one mechanism explains several failures.
Then deliberately seek a case where that mechanism predicts success and another where it predicts failure.
If the prediction holds, the model gains support.
If it fails, add only the complexity demanded by the new evidence.
This is parsimonious diagnostic tutoring.
The Four-Model Exercise
To teach parsimony directly, give students one dataset, passage or error pattern and four explanations.
- One is too simple and misses evidence.
- One fits well with few assumptions.
- One fits but uses unnecessary mechanisms.
- One can explain anything because it is too flexible.
Ask students to compare:
- fit;
- assumptions;
- exceptions;
- predictions;
- usability.
The exercise makes parsimony a reasoning skill rather than a slogan.
The Remove-One-Part Test
Take a model and remove one assumption.
What breaks?
If nothing important changes, the assumption may be unnecessary.
If a key observation becomes unexplained, restore it.
This exercise is especially useful in essays, Science explanations and complex problem-solving methods because students often accumulate steps without knowing which are structurally necessary.
The Add-One-Part Test
Now add one plausible mechanism or condition.
Does prediction improve?
Does an exception disappear?
Does the model become more interpretable?
Or does the new part merely make the explanation sound sophisticated?
Complexity that produces no measurable explanatory gain should be questioned.
The Transfer Test
A parsimonious model should often transfer better because it represents stable structure rather than one case’s decorative details.
Test this.
After teaching the model, change the context, numbers, wording or representation.
If the model still guides the correct decision, it has captured something portable.
If it collapses immediately, the simplification may have removed the very distinction that allowed transfer.
The Adversarial Test
Try to break the model.
Find the strongest counterexample.
Construct a boundary case.
Change one assumption.
Look for evidence the model does not naturally explain.
A simple model that survives strong tests earns more confidence than a simple model protected from difficult cases.
The Communication Test
A useful model should be communicable at the level appropriate to the learner.
If an explanation requires ten minutes to reconstruct every time, it may be too cumbersome for examination performance.
If it is so compressed that another student cannot tell why it works, it may be too thin for teaching.
Parsimony balances compactness and explanatory sufficiency.
The Retrieval Test
Models live in memory too.
A model with twelve loosely connected rules is harder to retrieve than one organised around two relationships and three boundaries.
But compressing too aggressively can create cue overload if several methods collapse into one vague label.
Useful parsimony preserves discriminating cues.
The model should be small enough to retrieve and sharp enough to select.
The Parsimony Ladder
- Case description: list what happened.
- Candidate mechanisms: identify plausible explanations.
- Evidence fit: eliminate explanations that miss decisive evidence.
- Assumption audit: expose unsupported commitments.
- Exception audit: count special patches.
- Prediction test: ask what each model predicts in a fresh case.
- Complexity comparison: prefer the model that achieves sufficient fit with less unnecessary machinery.
- Boundary statement: state where the model stops applying.
- Operational compression: represent the model in a form the learner can retrieve and use.
- Revision: add complexity only when new evidence earns it.
Field Manual: Use Parsimony to Diagnose a Study Problem
Suppose a student is taking too long to finish Mathematics homework.
Possible explanations include weak content knowledge, poor arithmetic fluency, distraction, perfectionism, slow method selection, excessive checking, weak reading comprehension or simple overload from too much work.
Do not pick the most fashionable explanation.
Collect discriminating evidence.
Time separate phases.
- reading;
- method selection;
- calculation;
- checking.
If reading and calculation are fast but method selection is consistently slow, several broad explanations become unnecessary.
Now test a clean hypothesis: similar methods are competing and discrimination is weak.
Give contrast pairs.
If decision time improves while calculation remains unchanged, the parsimonious model earns support.
This is a better learning loop than concluding “needs confidence” because the homework looked stressful.
Field Manual: Use Parsimony to Improve an Essay
Take a draft with five main claims.
Ask whether two claims are symptoms of one deeper mechanism.
Ask whether two examples prove the same point and whether one can be removed.
Ask whether an exception requires an entirely new paragraph or merely a refinement of the thesis.
Ask whether a sophisticated term clarifies a distinction or simply renames it.
The aim is not a shorter essay by default.
The aim is an essay in which each paragraph earns its place by advancing the explanatory architecture.
Field Manual: Use Parsimony to Learn Science Models
For each model, make four boxes.
- What it explains.
- What assumptions it makes.
- Where it works well.
- Where it breaks or needs extension.
Then compare models across the same phenomenon.
Which model predicts more with fewer special rules?
Which model is easier to use but less precise?
Which model becomes necessary only when a new scale or condition matters?
This teaches students that scientific progress often involves changing the model because the question or evidence changed—not because the earlier model was useless.
Field Manual: Use Parsimony in Revision Planning
A revision plan can also overfit.
The student creates separate routines for every topic, every worksheet, every teacher comment and every small weakness.
Soon the plan itself becomes unmanageable.
Look for reusable mechanisms.
- Several topics fail because retrieval is weak after delay.
- Several error types fail because negative signs are compressed too early.
- Several comprehension questions fail because evidence and inference are not separated.
- Several Science answers fail because mechanism is replaced by keywords.
Train the mechanism across cases.
A parsimonious revision plan repairs fewer deeper causes instead of maintaining dozens of disconnected patches.
Common Failure: Simplicity by Vocabulary Removal
Students sometimes make explanations “simple” by removing technical language.
That can help if jargon is hiding weak understanding.
It can hurt if the technical term carries a distinction ordinary language cannot express precisely.
Parsimony removes unnecessary complexity, not necessary vocabulary.
Common Failure: Simplicity by Removing Exceptions
Another false simplification says, “Ignore the exceptions for now,” and never returns to them.
This can create fast beginner performance and brittle later knowledge.
Exceptions should be introduced when they become decision-relevant.
The learner needs enough simplicity to build a model and enough boundary detail to avoid negative transfer.
Common Failure: Complexity as Status
Students can associate complicated language with intelligence.
They produce explanations with many clauses, abstract nouns and imported terminology.
The reader works harder while the idea becomes less precise.
A high-performance explanation should be as complex as the idea requires, not as complex as the writer can make it.
Common Failure: One-Cause Stories
Parsimony can tempt people into monocausal explanations.
“One thing caused everything.”
Sometimes that is true.
Often human learning is multi-causal.
A student’s performance may depend on knowledge, retrieval access, task interpretation, attention, time pressure and confidence simultaneously.
Parsimony does not forbid several causes.
It forbids adding causes that do no explanatory work.
Common Failure: Story After the Result
After an outcome is known, it is easy to create a detailed explanation that fits it.
The test is whether the model could have predicted something beforehand.
Ask what the explanation would have expected on a fresh case.
If no possible result could weaken the explanation, the model is too flexible.
Common Failure: Parsimony Without a Reference Class
A model can look beautifully simple inside one case and still be wrong about how common the case is.
Suppose a student expects to complete a six-week revision project because their internal plan is elegant.
How long did comparable projects actually take?
How often did earlier schedules survive school events, fatigue and competing deadlines?
That leads to the final Batch 15 article: Reference Class Reasoning.
Mira Replaces “Careless” with a Better Model
Mira returned to her error history.
The word “careless” appeared beside several questions.
They separated them.
Two were arithmetic slips.
Five involved negative brackets.
Three happened after method switching.
The global label dissolved.
Two mechanisms remained.
Negative-sign distribution was not sufficiently automated.
Switching between representations caused her to lose one intermediate condition.
Those models made predictions.
Those predictions could be tested.
Those mechanisms could be trained.
The explanation became slightly more complex than “careless” and dramatically more useful.
The Model Parsimony Test
- What evidence must the model explain?
- Does the model actually fit the decisive evidence?
- How many unsupported assumptions does it require?
- How many special exceptions must be added?
- Can one mechanism explain several observations?
- Does removing a component change prediction or explanatory fit?
- Does adding a component earn its complexity?
- Can the model make a prediction about a fresh case?
- Does the model survive counterexamples and boundary tests?
- Is it simple because it is well organised or because it ignores important structure?
- Does it underfit?
- Does it overfit?
- Can it transfer across changed surface details?
- Can the learner retrieve and use it under realistic conditions?
- Does the model make important errors easier to detect?
- Does it guide a specific intervention?
- Does stronger evidence override preference for simplicity?
- Are boundaries stored with the model?
- Is complexity being added only when evidence demands it?
- Can the learner explain why each major component earns its place?
Research Notes and Evidence Boundary
Parsimony is a long-established principle in scientific reasoning and model selection. A 2025 PNAS review, Is Ockham’s razor losing its edge? New perspectives on the principle of model parsimony, re-examines the role of simplicity in modern science. The authors argue that parsimony remains useful as a proxy for properties such as interpretability, predictive performance, research usefulness and resource efficiency, but that complex models are sometimes essential and simplicity is not universally beneficial.
A useful educational statement of the Law of Parsimony is also provided in the New South Wales Department of Education resource The Foundations of Scientific Thinking, which describes the preference for explanations with fewer assumptions and fewer exceptions when other considerations are sufficiently equal.
The broader educational framework in this article is an eduKatePunggol synthesis. “Model parsimony” is not being presented as a universal scoring formula for student explanations. Its purpose is to train a comparison habit: fit the evidence first, then ask whether each additional assumption, exception or mechanism earns its cognitive and explanatory cost.
Series Note
“High performance learning” is used descriptively throughout this eduKatePunggol series. The series does not claim affiliation with or reproduce any third-party branded educational framework using similar terminology.
Next: Compare This Case With the Right Family Before Predicting
Parsimony helps the learner choose among explanations inside a case.
The final Batch 15 article changes perspective.
Before predicting what will happen in this case, ask what usually happens in a genuinely comparable family of cases.
That is Reference Class Reasoning.
