Name each variable's role, know its level of measurement, and operationalize it well — three decisions that quietly determine which analyses are even possible.
The same construct can play different roles in different studies; the role is assigned by your theory and question, and it determines where the variable sits in the analysis.
| Independent variable (IV) | The presumed cause or predictor. Manipulated in experiments; measured in correlational work. Example: training method; perceived organizational support. |
|---|---|
| Dependent variable (DV) | The outcome the study explains or predicts. Example: 90-day sales performance; turnover intention. |
| Mediator (M) | The mechanism through which the IV works: X → M → Y. Example: POS → affective commitment → turnover intention. A mediation claim is a causal-chain claim — it needs strong design, not just a significant indirect effect. |
| Moderator (W) | A variable that changes the strength or direction of the X–Y relationship (an interaction). Example: the autonomy–engagement link is stronger for experienced staff. Moderators answer “when / for whom”. |
| Control variable | A rival explanation held constant statistically or by design (tenure, firm size). Choose controls from theory; do not stuff the model — controlling for a mediator or a consequence of Y can bias the focal estimate. |
Each variable's level of measurement constrains the mathematics that make sense on it — and therefore the statistics available.
| Nominal | Categories without order: gender, department, brand chosen. Meaningful summaries: counts, mode, proportions. Association: chi-square family. |
|---|---|
| Ordinal | Ordered categories with unequal or unknown spacing: seniority bands, single Likert items, satisfaction rankings. Summaries: median, percentiles. Association: rank correlations. |
| Interval | Equal spacing, no true zero: temperature in °C; by common (debated) convention, multi-item Likert scale scores. Means and SDs become meaningful; most parametric tests open up. |
| Ratio | Equal spacing with a true zero: salary, tenure, sales units, response time. All arithmetic is meaningful, including ratios (“twice as much”). |
Operationalization is the explicit recipe that turns an abstract construct into recorded data: conceptual definition → dimensions → indicators/items → scoring rule. “Employee engagement” might be defined via vigor, dedication and absorption, measured with the 9-item UWES on a 7-point frequency scale, scored as the item mean. Two studies with the same construct name but different operationalizations can legitimately reach different conclusions — which is why the recipe must be reported in full.
Prefer validated instruments over home-made items whenever they exist: they arrive with evidence of reliability and validity, published norms, and comparability to prior studies. Adapt wording only when necessary, report every change, and re-check reliability in your own sample.
| Multi-item scales | The default for psychological constructs (attitudes, perceptions, traits). Multiple items sample the construct's breadth, average out idiosyncratic wording effects, and allow reliability and factor-structure checks. |
|---|---|
| Single items | Acceptable for concrete, unambiguous facts (age, tenure, frequency of a behaviour) and defensible for globally understood constructs (overall job satisfaction) when survey length is at a premium — but no internal-consistency check is possible. |
Reliability asks whether the measure is consistent; validity asks whether it measures the intended construct. A measure can be reliable yet invalid (consistently measuring the wrong thing); it cannot be valid without being reliable. These Atlas methods are the standard toolkit:
Each link below opens the matching method page (opens the Statistical Methods Atlas).