Categories › Methods & inference
Category
Methods & inference
Cross-cutting tools for deciding what a result actually says, where it travels, and whether it applies to a particular person or situation.
Terms
Heterogeneity of treatment effectsVariation in the size or direction of an intervention effect across people, subgroups, baseline risks, or contexts.
Reference-class problemThe problem of choosing which comparison population is actually relevant to the case being evaluated.
Average treatment effect versus individual responseThe distinction between an estimated population-average effect and the response of a particular individual.
N-of-1 trialA repeated within-person experimental design used to compare interventions for one individual when outcomes can be measured across periods.
External validity / transportabilityWhether an empirical result remains informative when moved from the original sample and setting to another population or context.
Ecological fallacyInferring an individual-level relationship directly from a group-level association.
Atomistic fallacyInferring a group-level relationship directly from relationships measured among individuals.
Selection effects / self-selectionDifferences created partly because people nonrandomly enter, remain in, or leave environments, institutions, treatments, or samples.
Collider biasA spurious association created by conditioning on a variable that is influenced by two other variables.
Regression to the meanThe tendency for unusually extreme measurements to be followed by measurements closer to the average, even without intervention.
Reverse causationA causal ambiguity in which the proposed outcome may partly cause, precede, or signal the proposed exposure.
Measurement invarianceThe requirement that a measure represent the same latent construct in different groups, times, or contexts before comparisons are interpreted straightforwardly.
Construct validity versus predictive validityThe distinction between measuring the intended theoretical construct and successfully forecasting an outcome.
Mediator versus moderatorA mediator represents a pathway through which an effect operates; a moderator represents conditions under which the effect changes.
Trait–state distinctionThe distinction between relatively stable dispositions and temporary states induced by immediate circumstances.
Dose–response nonlinearityA relationship in which changing exposure by the same amount does not have the same consequence across the entire exposure range.
Base-rate dependenceThe dependence of the practical meaning of a signal or test result on the prevalence of the underlying condition in the relevant population.
Within-person versus between-person inferenceThe distinction between how variables change together inside one person over time and how different people compare with one another.
Prediction versus explanationA model may forecast an outcome accurately without identifying the mechanism producing it.
Causal inferenceThe family of designs and assumptions used to distinguish causal effects from association, selection, confounding, and reverse causation.
Research trail
This category page is a wiki-authored organizing layer. Follow individual term pages for supplied sources, provenance notes, related concepts, and reading trails.