Hypotheses, Theories, and Models

Scientific language is often distorted in everyday speech.

“That’s just a theory.”

“I have a hypothesis.”

“The model says…”

These phrases can refer to very different things.

A hypothesis is not simply an immature theory.

A theory is not merely a guess.

A model is not always a simplified toy.

Science uses these structures for different jobs.

Understanding the difference helps us understand how scientific knowledge is built.

What Is a Hypothesis?

A hypothesis is a proposed claim that can be assessed through evidence.

Examples:

• a particular gene affects a trait,
• a planet exists around a star,
• a drug reduces blood pressure,
• dark matter consists of a certain particle,
• a geological layer was formed by a volcanic eruption.

A hypothesis is specific enough to generate expectations.

Its scientific value depends on whether evidence can bear on it.

Hypotheses Can Be Broad or Narrow

Some hypotheses are highly specific.

“This compound inhibits enzyme X.”

Others are broad.

“Life may exist beneath the icy surface of Europa.”

A broad hypothesis often needs to be broken into testable subclaims.

Scientific progress usually requires translating vague possibilities into measurable consequences.

Null Hypotheses

Statistical testing often uses a null hypothesis.

The null may state:

• no difference,
• no association,
• no effect.

Researchers then ask how surprising the observed data would be if the null were true.

Rejecting a null does not automatically prove one preferred alternative.

It only tells us the data are difficult to reconcile with the specific null under the assumptions of the test.

What Is a Theory?

A scientific theory is a structured explanatory framework connecting many observations, principles, and mechanisms.

Examples include:

• evolution by natural selection,
• general relativity,
• quantum theory,
• germ theory,
• plate tectonics.

Theories organize large bodies of knowledge.

They explain why many separate facts fit together.

A mature theory usually contains many hypotheses, models, laws, and auxiliary assumptions.

“Just a Theory” Is a Category Error

In ordinary language, theory often means speculation.

In science, a well-supported theory can be among the strongest forms of knowledge available.

Evolution is not less established because it is a theory.

General relativity is not a guess.

The word refers to explanatory structure, not evidential weakness.

Theories Do Not Become Laws

A common school misconception says:

hypothesis → theory → law

as if scientific ideas graduate through stages.

That is wrong.

Laws and theories serve different roles.

A law often describes a regular pattern.

A theory may explain the pattern.

Newton’s law of gravitation did not become general relativity.

General relativity explains gravity through a deeper framework and reproduces Newtonian behavior approximately in suitable regimes.

What Is a Model?

A model is a representation of a system.

Models can be:

• physical,
• mathematical,
• computational,
• conceptual,
• statistical,
• diagrammatic.

A climate model.

A model atom.

A neural-network model.

A population model.

A cosmological model.

Models simplify or structure reality for a purpose.

Models Omit

A model that included every detail of reality would be useless.

A subway map omits building shapes, tree locations, and most roads.

Those omissions are not mistakes.

They make the map useful for navigation.

Scientific models work similarly.

The central question is:

Which details matter for the problem?

Models Can Be False but Useful

The ideal gas model assumes point particles with simplified interactions.

No real gas matches it perfectly.

Yet it predicts many phenomena well under suitable conditions.

Newtonian mechanics is not fundamentally exact at relativistic speeds.

It remains useful for bridges, machines, and ordinary planetary calculations.

A model can contain literally false assumptions while capturing the right effective behavior.

Idealization

Science often uses idealizations.

Examples:

• frictionless surfaces,
• perfectly rigid bodies,
• point masses,
• infinite populations,
• isolated systems,
• perfectly rational agents.

No real system satisfies these exactly.

Idealization removes complications so that structure becomes visible.

The danger comes when we forget which assumptions were idealized away.

Approximation

An approximation is not the same as an error.

Approximations deliberately ignore small effects.

A physicist may neglect air resistance because it changes the result by less than the required precision.

An economist may aggregate individual behaviors.

A biologist may model a population continuously rather than as discrete organisms.

Good approximation is controlled simplification.

Scale Determines the Model

Different models can describe the same system at different scales.

A gas can be modeled as:

• molecules,
• a continuous fluid,
• thermodynamic variables.

A brain can be modeled as:

• ion channels,
• neurons,
• networks,
• cognitive functions.

A galaxy can be modeled as:

• individual stars,
• a smooth gravitational potential,
• a dark-matter halo plus baryonic components.

No one model must dominate every scale.

Model Families

Scientists rarely rely on one model alone.

They compare families of models.

Different parameter values.

Different assumptions.

Different mechanisms.

Model comparison helps answer:

Which structure is actually needed to explain the data?

This is especially important when flexible models can fit almost anything.

Overfitting

A model can fit known data too well.

If it contains too many adjustable parameters, it may reproduce random noise.

This is overfitting.

An overfit model performs impressively on past data but poorly on new data.

Prediction on unseen cases is therefore a powerful test.

Complexity must be justified by improved generalization.

Underfitting

A model can also be too simple.

If it ignores important structure, it fails even on basic patterns.

This is underfitting.

Scientific modeling therefore involves a balance:

simple enough to generalize, rich enough to capture the phenomenon.

This tradeoff appears in machine learning as well as classical science.

Parameters

Models often contain parameters.

A parameter is a quantity whose value helps determine model behavior.

Examples:

• Hubble constant,
• reaction rate,
• growth rate,
• diffusion coefficient,
• neural-network weights.

Some parameters are measured independently.

Others are estimated by fitting data.

A model with many free parameters can become difficult to test strongly.

Initial Conditions

A model may also require initial conditions.

The same laws can produce very different outcomes from different starting states.

Weather prediction is a classic example.

The equations may be known well, yet small uncertainties in initial conditions grow through chaos.

Model, parameter, and initial condition are distinct sources of prediction.

Assumptions

Every model has assumptions.

Some are explicit.

Others are hidden.

A population model may assume random mating.

A cosmological model may assume large-scale homogeneity.

A statistical model may assume independent errors.

A result is only as strong as the assumptions required to obtain it.

Good science exposes assumptions to scrutiny.

Mechanistic Models

A mechanistic model represents causal processes.

For example:

enzyme binds substrate → reaction occurs → product forms.

Mechanistic models aim to explain how a result is produced.

They can support intervention.

If the mechanism is correct, changing one component should alter the outcome in predictable ways.

Phenomenological Models

A phenomenological model describes patterns without necessarily identifying deep mechanism.

A fitted equation may accurately describe a relationship.

A scaling law may summarize behavior.

These models can be scientifically valuable even when the underlying cause is unknown.

Description can precede explanation.

Statistical Models

Statistical models represent probability distributions and relationships in data.

Regression.

Bayesian models.

Time-series models.

Survival models.

They help distinguish signal from noise and quantify uncertainty.

Statistical models are not merely mathematical afterthoughts.

They often define the inferential link between observations and conclusions.

Computational Models

Some systems are too complex for closed-form equations.

Scientists simulate them.

Climate.

Galaxy formation.

Protein folding.

Epidemics.

Fluid turbulence.

A computational model lets rules unfold over time.

But simulation output is still model-dependent.

A computer does not turn assumptions into truth.

Toy Models

A toy model is deliberately simplified to reveal one conceptual mechanism.

Physicists often learn from models that are obviously unrealistic in detail.

A two-particle system may expose a principle hidden in a thousand-particle system.

Toy models are valuable because understanding often requires subtraction.

The Standard Model

The phrase Standard Model can confuse beginners because here “model” names a very deep physical theory.

The Standard Model of particle physics is not a casual simplified diagram.

It is a sophisticated quantum field theory describing known elementary particles and three fundamental interactions.

Scientific terminology evolves historically.

Words do not always obey one rigid taxonomy.

The Standard Cosmological Model

ΛCDM is often called the standard cosmological model.

It combines:

• general relativity,
• cold dark matter,
• a cosmological constant,
• specified initial conditions,
• measured parameters.

It is extremely successful.

But it still contains unresolved ingredients, especially the fundamental nature of dark matter and dark energy.

A successful model need not be a final theory.

How Hypotheses, Models, and Theories Interact

A theory provides broad explanatory structure.

A model represents a specific system under specified assumptions.

A hypothesis makes a testable claim.

For example:

Theory

General relativity.

Model

A cosmological spacetime with matter, radiation, and dark energy.

Hypothesis

The dark-energy equation of state is exactly consistent with a cosmological constant.

These layers interact but are not interchangeable.

One Theory, Many Models

General relativity admits many spacetime models.

Expanding universes.

Black holes.

Gravitational waves.

Different matter distributions.

The theory provides equations.

Models specify particular solutions or approximations.

This is a general pattern across science.

One Model, Multiple Theoretical Interpretations

Sometimes the same mathematical model can fit within different deeper theories.

A statistical relationship may be compatible with several mechanisms.

This creates underdetermination.

Good evidence should therefore test what distinguishes the deeper explanations, not only whether one model can fit the data.

Models and Reality

Is a successful model literally true?

Sometimes only approximately.

A map can represent a city accurately without becoming the city.

A model may capture one structure and omit another.

Scientific realism does not require treating every component of every useful model as a literal object.

We need to ask which parts are representational convenience and which correspond to genuine features of nature.

The Lifecycle of a Model

Models change.

They are:

• proposed,
• calibrated,
• tested,
• compared,
• revised,
• sometimes abandoned.

A model may survive because it predicts well.

It may be replaced because another explains more with fewer assumptions.

Scientific progress often consists of improving representations rather than accumulating isolated facts.

From Models to Experiment

Once we have a hypothesis, theory, or model, we need a way to confront it with reality.

One of the most powerful tools is controlled intervention.

Change one factor.

Hold others fixed.

Observe what happens.

This is the logic of experiment.

So the next question is:

What makes an experiment scientifically powerful?