What is an Independent Variable?

The independent variable (IV) is the factor in a study that is changed or chosen by the researcher in order to observe its effect: the presumed cause in the causal question. Its counterpart, the dependent variable, is the measured outcome, the presumed effect. In a true experiment the IV is actively manipulated (which onboarding flow a user sees, which headline a visitor gets) and assignment to its levels is randomised, which is what licenses causal reading. Many studies instead use selected IVs the researcher cannot assign (tenure, plan tier, device): the analysis machinery looks identical, but the causal licence lapses, and the relationship reverts to the careful vocabulary of association.

Levels, Conditions, and Operationalisation

An IV takes levels: the specific values compared. "Onboarding flow" with levels {current, redesigned}; "price point" with levels {£9, £12, £15}; the levels define the experiment's actual question, which is narrower than the concept's name suggests. That gap is the operationalisation problem: "simplified navigation" as a concept must become one concrete alternative design as a level, and the experiment tests the design, not the concept. Two disciplines follow. Vary one thing per level (a redesigned flow that also changes the copy and the colours is three IVs wearing one label, and a winning result cannot say which mattered), and check the manipulation landed (did participants actually experience the intended difference?), the humble manipulation check that saves misread experiments.

Independent Variables in Product Research

Every A/B test has exactly this structure: variant is the IV, conversion the DV, random assignment the licence. Comparative prototype studies make design version the IV, with task success and SUS as DVs; in a Ballpark comparison, randomising which prototype each participant sees (and, within-subjects, which comes first) is the assignment discipline in miniature. Survey experiments manipulate wording or framing as the IV to measure question effects directly. And segmentation analyses ("do power users rate this higher?") use selected IVs, valuable, and permanently one confounder away from a causal misreading.

The Craft Rules

1. Name the IV and its levels in one sentence.
"We vary X between {A, B} to see its effect on Y." If the sentence won't write cleanly, the design isn't clean yet.

2. Make levels differ by the tested thing only.
Bundled changes test bundles. When the bundle is the realistic unit (a whole redesign), accept that the experiment answers the bundle question, and say so.

3. Choose levels that span the decision.
Two prices you'd never charge test nothing; levels should be the genuine candidates, plus a control that anchors the comparison.

4. Keep everything else constant or randomised.
Whatever isn't the IV is either held fixed (a control variable) or left to randomisation to balance; the leftovers that vary systematically are confounders, the IV's illegitimate rivals.

The Takeaway

The independent variable is the experiment's deliberate difference: define it in levels, operationalise it without bundling, randomise its assignment where causality is the claim, and confirm it actually varied. Get the lever right and the dial's movements mean something; get it muddled and the whole apparatus measures a question nobody quite asked.