A method is a structured way of doing, finding, testing, building, measuring or deciding something.
Interviewing, random sampling, long division, controlled experimentation, debugging, peer review, close reading, simulation and cost-benefit analysis are all methods. What distinguishes them is not just the sequence of steps, but their purpose, evidence requirements, assumptions, repeatability, scope and failure modes.
A useful classification frame
- Purpose: discover, measure, explain, predict, compare, create, verify or decide?
- Procedure: fixed, adaptive, iterative or exploratory?
- Evidence: what observations or outputs count?
- Repeatability: can another competent user reproduce the procedure?
- Scope: where does the method apply?
- Resources: what tools, time, skills and data are required?
- Assumptions: what must be true for the method to work?
- Validity: does the method measure or test what it claims?
- Limits: where does it break down?
- Revision: how is the method updated when evidence changes?
This article extends How to Categorise Anything by treating methods as reusable procedures whose value depends on fit to purpose.
Method is not the same as task
A task is the work to be completed. A method is the approach used to complete it. “Estimate demand” is a task; time-series analysis, expert elicitation and market survey are different methods for attempting it.
Method is not the same as tool
A spreadsheet, microscope or language model is a tool. The method determines how that tool is used, what evidence is accepted and how outputs are checked.
Method is not the same as theory
Theory offers an explanatory framework. Method governs how we investigate, apply or test claims associated with that framework.
Classify first by reader or operator purpose
Discovery methods search for patterns or possibilities. Measurement methods assign values. Explanatory methods investigate mechanisms. Predictive methods estimate unseen outcomes. Design methods create candidate solutions. Verification methods check whether claims, products or processes meet requirements. Decision methods compare alternatives.
Procedural methods follow defined steps
Laboratory protocols, accounting workflows and standard calculations rely on explicit sequences. Their strength is consistency; their weakness can be brittleness when context departs from expected conditions.
Iterative methods learn through cycles
Debugging, design, editing and scientific investigation often alternate between action, observation and revision. The next step depends on what the previous cycle reveals.
Adaptive methods change route without changing purpose
A teacher may change explanation strategy when a student remains confused. A diagnostic method may choose the next test based on the previous result. Adaptation should be governed enough that the reasoning remains reconstructable.
Exploratory methods widen the search space
Brainstorming, open coding, field observation and broad literature scanning can reveal possibilities before a narrower hypothesis or taxonomy is chosen.
Confirmatory methods test pre-specified expectations
They work best when hypotheses, criteria or thresholds are declared before observing the decisive evidence.
Qualitative and quantitative methods answer different questions
Qualitative methods can preserve meaning, process and context. Quantitative methods can estimate magnitude, frequency and uncertainty. Neither category is inherently more rigorous; rigor depends on fit, execution and evidence quality.
Observational and experimental methods differ in intervention
Observational methods examine naturally occurring variation. Experimental methods deliberately alter conditions to test causal effects under controlled or structured designs.
Direct and indirect methods differ in inference distance
Direct measurement observes the target closely. Indirect methods infer the target from proxies, models or traces and therefore require stronger validation of the link.
Manual, automated and hybrid methods should remain distinct
A human-only review, fully automated classifier and human-plus-model workflow may share a goal but differ materially in speed, reproducibility, explainability and failure mode.
Repeatability is one quality dimension
A method is repeatable when the same operator can apply it consistently under similar conditions. Reproducibility goes further by asking whether independent operators or environments can obtain compatible results.
Validity is purpose-specific
A method can be reliable yet invalid for the intended question. A ruler reliably measures length but cannot validate a causal claim.
Sensitivity and specificity matter for detection methods
Some methods detect most true cases but create many false alarms. Others are highly selective but miss weak cases. The right trade-off depends on consequence.
Resource intensity belongs in the classification
Two methods can produce similar evidence but require very different time, expertise, equipment, money or data. Operational choice therefore depends on both quality and cost.
Method scope should be explicit
A method validated for one population, material, language or operating range should not be silently generalised beyond that domain.
Assumptions travel with methods
Sampling methods assume something about populations. Statistical methods assume something about data-generating processes. Search methods assume something about where relevant material can be found. These assumptions should remain visible.
Failure modes deserve their own field
A method can fail through bias, poor calibration, missing data, invalid assumptions, operator error, contamination, overfitting, weak controls or simple mismatch to the question.
Verification methods need acceptance criteria
Testing is meaningful only when the pass, hold or fail conditions are known. A vague “looks fine” check is not equivalent to a defined verification method.
High-stakes methods need independent review
Where safety, rights, finances or important decisions are affected, method choice and execution deserve stronger evidence and separation between producer and reviewer.
Methods evolve
Improved instruments, evidence or theory can make an older method obsolete, narrower or newly useful. Versioning prevents historical results from being interpreted under a method that did not yet exist.
A practical method record
- method ID and name;
- purpose;
- procedure type;
- inputs and prerequisites;
- tools and resources;
- assumptions;
- evidence produced;
- repeatability and reproducibility;
- validity evidence;
- scope and boundary conditions;
- failure modes;
- review requirements;
- owner;
- version and effective date.
The deeper idea
Methods are not interchangeable recipes. They are contracts between a question, a procedure and the kind of evidence the procedure can legitimately produce.
To categorise a method well is to know what job it performs, what assumptions it depends on, what evidence it can produce, where it fails and how reliably another competent operator can use it.
Final answer
Categorise methods by purpose, procedure, evidence, repeatability, scope, resources, assumptions, validity, limitations and revision. Keep method separate from task, tool and theory, and judge quality by fitness for the actual question rather than by prestige or complexity.