What are Variables?

A variable is any characteristic that can take different values across the units being studied: task time, plan tier, satisfaction rating, country, whether a feature was used. Studies are relationships among variables, and clarity about each variable's role and measurement type is what separates a design from a data pile. The two taxonomies below cover both, and nearly every methodological rule in this glossary is a rule about one of them.

The Roles

Independent and dependent. The lever and the dial: the manipulated (or selected) presumed cause, and the measured outcome, each with its own entry.

Control variables. Factors deliberately held constant (same tasks, same device class, same time window) so they cannot masquerade as effects. What can't be held constant is left to randomisation to balance, or measured and adjusted for statistically.

Confounding variables. The saboteurs: factors related to both the IV and the DV that manufacture or mask associations, the engine of every correlation-causation disaster. Experiments neutralise them by randomisation; observational work hunts them one by one and never finishes.

Moderators. Variables that change an effect's strength or direction: the redesign that helps novices and hinders experts has expertise as a moderator. Moderation is the honest version of subgroup analysis, hypothesised in advance rather than mined after.

Mediators. Variables that carry an effect: onboarding improves retention because it increases early activation, activation mediating the path. Mediators are the mechanism story, the "why" that qualitative work proposes and statistical mediation analysis formally tests.

The Measurement Types

The classic ladder: nominal (categories without order: country, browser), ordinal (ordered without equal spacing: satisfaction bands, the eternal Likert debate), interval (equal spacing, arbitrary zero: temperature in Celsius), and ratio (true zero: task time, counts, revenue). The practical cut is coarser: categorical versus continuous, because that distinction picks the analysis: proportions and chi-square machinery for categories, means and t-tests for continuous measures, and specialised handling for the ordinal middle. Declaring a variable's type before analysis is a small act of hygiene that prevents a large class of nonsense, like averaging category codes because they happened to be stored as numbers.

The Craft Rules

1. Cast every variable in its role, in writing.
A one-line design statement (IV, DV, controls, suspected confounders, hypothesised moderators) is the cheapest design review a study can get, and the fastest way to notice the confounder nobody planned for.

2. Operationalise before you field.
Every abstract variable needs its concrete measure and rubric, the construct validity and reliability work done up front.

3. Match analysis to type.
The measurement scale is a contract with the statistics; honour it, and flag the ordinal grey zones instead of averaging through them silently.

4. Pre-name the moderators you'll test.
Every additional slice spends the false-positive budget; hypothesised moderators are analysis, discovered ones are next study's hypotheses.

The Takeaway

Variables are the parts list of research: name each one's role (lever, dial, held-constant, saboteur, amplifier, carrier), declare its measurement type, and write the casting down before data arrives. Most methodological failures are miscast variables discovered too late; most rigour is just this taxonomy, applied early.