20  Statistical Methods

Statistics is often introduced through methods:

But this order, while practical, can be misleading.

It makes methods look like recipes.
And statistics is not a cookbook.

A method is not just a sequence of operations.
It is a disciplined response to a kind of problem.

🏡 Statistical methods are best understood as structured paths from question to judgement.

That path matters more than any isolated formula.

20.1 πŸ”° Problem before technique

A method has meaning only relative to a problem.

Before asking:

  • what test?
  • what estimator?
  • what model?

one must ask:

  • what is the question?
  • what is the process?
  • what is being measured?
  • what level of uncertainty matters?
  • what kind of answer is actually needed?

This sounds obvious, but it is one of the most neglected truths in practice.

A powerful technique applied to the wrong question is not sophistication.
It is misalignment.

20.2 ☯ The statistical path

A statistical inquiry often moves through stages like these:

  1. Identification
    What phenomenon is being studied?

  2. Measurement
    How will it be observed or recorded?

  3. Compilation
    How will observations be gathered and organized?

  4. Summarization
    What compressed views reveal structure?

  5. Analysis
    What relations, patterns, or deviations are present?

  6. Modeling
    What formal structure could generate such data?

  7. Inference
    What can be said beyond the observed sample?

  8. Interpretation
    What does the result mean in the world?

  9. Decision
    What should be done, believed, revised, or tested next?

This is not always linear.
Often the process loops back on itself.

A failed model changes the summary.
A poor measurement changes the entire question.
A surprising residual plot reopens the structure of the process.

⚜️ Statistics is not a straight corridor.
It is a feedback system.

20.3 βš™οΈ Scientific method and statistics

Statistics lives naturally inside the broader scientific method.

Science asks:

  • what is happening?
  • why?
  • under what conditions?
  • what explanation survives confrontation with evidence?

Statistics provides the disciplined language through which evidence is organized.

In that sense, it is close to the Popperian spirit, though not reducible to it.

A good model is not β€œproven true.”
It is tested for adequacy, challenged by data, compared to alternatives, and judged by how well it survives criticism.

This makes statistics one of the great instruments of intellectual restraint.

It does not let explanation drift free from evidence.
It forces ideas to pass through contact with variation.

20.4 πŸ’‘ Statistical thinking as process-thinking

The deepest statistical habit is not calculation.

It is process-thinking.

Data is not an isolated object.
It is the trace of something that happened.

So statistical reasoning asks:

  • what produced these values?
  • what changed through time?
  • what constraints shaped the observations?
  • what feedbacks are hidden inside the variation?
  • what would repetition look like?

This is why good statistical work is always more than numerical.

It is causal in curiosity, structural in attention, and modest in conclusion.

20.5 πŸͺ„ Interconnection

Statistical ideas are not separate islands.

Measurement affects variation.
Variation shapes uncertainty.
Uncertainty constrains inference.
Inference depends on models.
Models depend on assumptions.
Assumptions depend on how the process is understood.

Everything touches everything.

This is one of the reasons statistics can feel difficult at first.
Its parts only fully make sense together.

But this is also what makes it powerful.

Statistics is not a bag of tools.
It is an interconnected way of seeing processes through data.

20.6 ⚠️ The violinist lesson

Sometimes the failure of a method is not lack of precision, but wrong alignment.

Imagine an orchestra that has accelerated slightly.

A violinist who follows the original rhythm with perfect exactness is now making a mistake.

Why?

Because correctness in isolation is no longer correctness in context.

This is one of the best metaphors for bad statistical practice.

A method may be internally exact and externally wrong.

A model can fit a local target and miss the process.
A measurement can be precise and irrelevant.
A summary can be mathematically correct and conceptually unfaithful.

⚜️ Statistical method is not the worship of exactness.

It is the discipline of relevant exactness.

That is much harder, and much wiser.

20.7 πŸ”° Methods are choices under loss

Every method chooses what to preserve and what to sacrifice.

A mean preserves center and loses shape.
A variance preserves spread and loses direction.
A model preserves selected structure and discards the rest.
A hypothesis test preserves a decision rule while compressing richer uncertainty into a statistic.

So no method is neutral.

Every method is an act of reduction guided by purpose.

This is why statistical education should never ask only:

how do we compute this?

It should also ask:

what does this method see?
what does it ignore?
what kind of mistake is it willing to make?

20.8 ☯ Description, explanation, inference

Statistical methods operate across at least three modes:

20.8.1 Description

What does the data look like?

20.8.2 Explanation

What structure might account for that shape?

20.8.3 Inference

What can be said beyond the observed data?

These are related, but not identical.

A histogram describes.
A regression model may explain.
A confidence interval supports inference.

Confusion begins when one mode is mistaken for another.

A description is not automatically an explanation.
An explanation is not automatically a justified inference.
A formal inference is not automatically meaningful in the real process.

Good method keeps these layers visible.

20.9 βš™οΈ The method must fit the question

Different questions need different methods.

  • If the question is about center, use methods that stabilize center.
  • If it is about spread, emphasize variation.
  • If it is about shape, distributions matter.
  • If it is about relationship, modeling matters.
  • If it is about decision under evidence, testing or interval methods may matter.
  • If it is about process change, dynamic methods matter.

This sounds simple, but it is the heart of applied statistics.

There is no universal best method.
There is only a better or worse fit between method and problem.

20.10 πŸ’‘ What makes a method good?

A good statistical method is not merely one that is mathematically elegant.

It should also be:

  • appropriate to the process
  • transparent in its assumptions
  • sensitive to relevant uncertainty
  • robust enough for the scale of the problem
  • interpretable in context

Sometimes the simplest method is the best because it preserves meaning.
Sometimes a more sophisticated method is necessary because the structure demands it.

The answer is not given in advance.
It emerges from the interplay of question, data, and world.

20.11 🏡 Final thought

Statistical methods are not rituals for producing respectable-looking numbers.

They are disciplined routes by which questions encounter evidence.

At their best, they do not merely calculate.
They orient thought.

They teach us to move carefully from observation to structure, from structure to uncertainty, and from uncertainty to judgement.

And that is why statistics, properly understood, is not only technical.

It is a method of intellectual responsibility.