All Models Are Wrong (But Some Useful)

Scientific Reasoning

Medium
This framework reminds you that every model is a simplification, which means every model is incomplete, but some are still useful if their limits are understood. It matters because people often swing between naive trust in models and total cynicism about them, when the disciplined position is conditional usefulness.
Reasoning type
Meta-model evaluation
Certainty level
Purpose- and domain-dependent
Cognitive load
Medium
Formality
Medium

Core Idea

Definition

All Models Are Wrong (But Some Useful) is the principle that representations of reality are always incomplete and should be judged by fit-for-purpose usefulness rather than by whether they capture reality perfectly.

In Plain English

A model does not need to be literally true in every detail to help you think well, but it does need to be used within its limits.

Framework Structure

Components

Model
Simplifying Assumptions
Useful Domain
Failure Modes

Flow

Identify the model -> Clarify what it simplifies -> Define where it helps -> Watch for where it breaks

How to Apply

  • 1.Ask what the model leaves out or compresses
  • 2.Define the domain where the model is still useful
  • 3.Use it as a tool for thinking, not as a perfect mirror of reality
  • 4.Look for situations where the omitted details start to matter
  • 5.Switch or combine models when one no longer fits the problem well

When to Use

  • Evaluating frameworks, theories, and dashboards
  • Comparing multiple ways to represent a problem
  • Avoiding overconfidence in neat abstractions
  • Explaining why a useful model still has edge cases
  • Any context where representation and reality can be confused

When NOT to Use

  • When it is used lazily to excuse bad models
  • When a model's failure modes are obvious enough that it should simply be rejected
  • When the phrase becomes a slogan instead of a real diagnostic question
  • When skeptical distance turns into decision paralysis

Example

Problem

A team uses a simple funnel model to understand user conversion.

Application

  • 1.Recognize the model is useful for highlighting drop-off points
  • 2.Notice that it hides repeat behavior, social influence, and cross-device complexity
  • 3.Use it where it clarifies the main flow while avoiding total reliance on it for everything
  • 4.Add richer models when those omitted factors start driving decisions

Conclusion

The team gets the benefit of the model without mistaking it for full reality.

Takeaway

Model usefulness depends on fit, limits, and humility, not on perfection.

Common Mistakes

  • Trusting a model outside the domain it was built for
  • Dismissing a useful model because it is not perfect
  • Treating every model as equally flawed and equally useful
  • Forgetting that model choice should depend on purpose
  • Using abstraction to avoid observing reality directly

How to Practice

what does it hide

For any model you use, list at least three important things it leaves out.

fit for purpose check

Ask what decision or understanding task the model helps with and where it stops helping.

model switch drill

Try describing the same problem through a second model and compare what each one reveals or distorts.

Related Cognitive Biases

overconfidence

People often trust neat models more than the complexity of reality deserves.

false dichotomy

People may think a model is either totally true or totally useless instead of conditionally helpful.

abstraction bias

Elegant representations can overshadow messy but relevant facts.

Related Frameworks

Related Skills

evaluating reliability
fact inference separation
systems thinking
uncertainty tolerance

Variants & Extensions

Fit-for-purpose modeling
Abstraction-limit checks
Representation humility
Multi-model reasoning

Typical Failure Modes

  • Slogan-only use
  • Bad-model excuse
  • All-models-equivalent thinking

Further Reading

  • The Model Thinker by Scott E. Page
  • Thinking in Systems by Donella H. Meadows
  • The Signal and the Noise by Nate Silver