Across decades of meta-analyses, the same pattern keeps appearing: the typical effect in psychology is small. Richard, Bond and Stokes-Zoota synthesised roughly a century of social psychological research and found an average correlation of about .21. Individual differences research clusters around similar values. The large, clean effects of introductory textbooks turn out to be rare on closer inspection, and when replicated with adequate samples they often shrink.

The conventional response treats this as an indictment. Small effects are read as symptoms of noisy measurement, underpowered designs, or a discipline that has not yet found its proper variables. On this view, psychology is still waiting for its Newtonian moment, after which explained variance will climb as it once did in physics. Much of the reform movement of the replication era implicitly shares this hope: fix the methods, and the real effects will finally emerge at full strength.

The question deserves a more fundamental framing. Is the smallness of psychological effects a fact about our methods, or a fact about minds? The answer matters, because each interpretation implies a different picture of what psychological causation is and what psychological theory should be trying to do. If small effects are substantive rather than artefactual, then the discipline's explanatory ambitions may need to be recalibrated, not simply realised.

Begin with arithmetic rather than epistemology. Ahadi and Diener observed in 1989 that when an outcome has multiple independent determinants, each one faces a ceiling on how much variance it can explain. Suppose a behaviour is jointly produced by ten roughly equal, independent causes. Then no single cause can account for much more than ten percent of the variance, which corresponds to a correlation of about .32. This smallness is a mathematical consequence of multicausality, not evidence of weak measurement.

Meehl's notion of the crud factor points the same way. In what he called soft psychology, nearly everything correlates with nearly everything else to some degree. The causal web is dense, and its strands are interwoven. Physics could isolate its systems through shielding, vacuums and controlled conditions. A psychological system cannot be shielded from its developmental history, its current mood, its cultural scaffolding or the interpretive frame it brings to the laboratory.

The methodological explanation still deserves its due. Measurement error attenuates correlations, and Spearman's correction for unreliability raises them. Yet disattenuation rarely moves effects into the range that textbook narratives imply. Correcting a .20 for typical reliabilities might yield something near .28, not .70. Even after measurement has been given its full share of the blame, a residual smallness remains, and that residue calls for a substantive account.

Comparison with neighbouring sciences supports this reading. Rosenthal noted that the aspirin trial for preventing heart attacks, stopped early because its benefit seemed so clear, corresponded to a correlation of roughly .03. Meyer and colleagues later showed that many psychological effects are comparable in magnitude to accepted medical ones. Epidemiology, ecology and economics share the same pattern. Small effects are the characteristic signature of open systems with many interacting determinants.

The expectation of large effects therefore rests on a hidden assumption: a closed-system ontology inherited from classical mechanics. Here Kuhn's insight is instructive. A paradigm supplies more than theories. It supplies exemplars that define what a good result looks like. Psychology adopted the exemplars of physics and then judged its own findings by an aesthetic that may simply not fit its subject matter.

Takeaway

A small effect may be an accurate measurement of a genuine cause operating in a crowded causal field. The disappointment it provokes may say more about our inherited image of science than about the phenomenon itself.

An average effect conceals the distribution it summarises. A mean treatment effect of d = 0.2 could arise because everyone shifts slightly. It could arise because a minority shifts substantially while most remain unaffected. Or it could arise because some people improve markedly while others deteriorate. Group-level statistics alone cannot tell these structures apart, yet they imply profoundly different accounts of the underlying causation.

Molenaar's ergodicity argument sharpens the problem. Inferences from variation between individuals to processes within individuals are valid only when those processes are ergodic, meaning homogeneous across persons and stable over time. Psychological processes rarely meet either condition. Fisher, Medaglia and Jeronimus examined intensive longitudinal data and found that individual-level variability often greatly exceeded what group-level estimates suggested. Population structure and personal structure can diverge.

This revives an old tension. Cronbach's 1957 address on the two disciplines of scientific psychology pitted the experimentalist's treatment effects against the correlationist's individual differences. By 1975 he warned of a hall of mirrors, with interactions qualified by further interactions. Lewin's formula, behaviour as a function of person and environment, already implied as much. When person characteristics moderate a causal relation, the main effect becomes an average taken over many different causal structures.

Mischel and Shoda's if-then behavioural signatures illustrate the point. People show stable but idiosyncratic patterns of responding across situations. One person becomes assertive under criticism, while another withdraws. Each pattern may be strong and reliable within the individual. Aggregated across a sample, however, the situational effect can appear small or even vanish, because heterogeneity in direction cancels at the level of the mean.

This interpretation requires discipline of its own. Heterogeneity is a hypothesis, not a convenient excuse. Variance-ratio analyses of some clinical trials suggest that response differences are sometimes more modest than the rhetoric of precision approaches implies. Separating true heterogeneity from noise demands repeated measurement, N-of-1 designs and idiographic modelling. Done well, such work could show where small averages truly conceal large individual effects and where they do not.

Takeaway

An average effect is a property of a population, not necessarily of any person within it. The small number on the page may be the shadow of many large, divergent causal stories.

If smallness is substantive, the aim of explanation must shift. Variance accounted for is not a property of a causal relation. It is a joint property of the cause, the population and the design. Lewontin made this point decisively against analysis-of-variance reasoning in behavioural genetics. A cause can be real, mechanistically important and even necessary while still explaining little variance, simply because it varies little in the sample studied.

Small effects also expose a weakness in how theories are tested. Meehl argued that soft psychology progresses slowly because weak theories are corroborated merely by rejecting null hypotheses. With a large enough sample, the crud factor guarantees that some directional difference will appear. Small effects combined with vague theory produce an illusion of confirmation. The remedy is riskier prediction: specifying functional forms, moderators, scope conditions and the circumstances in which an effect should disappear.

There is a constructive side. Abelson's paradox showed that batting skill explains about one third of one percent of the variance in a single at-bat, yet it decisively shapes outcomes across a season. Götz, Gosling and Rentfrow argue accordingly that small effects are the indispensable foundation of a cumulative psychological science. Modest causes compound across time, across repeated occasions and across populations into consequences of real magnitude.

That tolerance carries a risk. Small effects are also where artefacts live. Analytic flexibility, publication bias and confounding can all produce small spurious associations that are hard to distinguish from small genuine ones. Funder and Ozer's call to treat effect sizes realistically therefore comes with an obligation: tiny effects require large samples, preregistration and replication. Accepting smallness must never become accepting fragility. Small and real must be kept apart from small and illusory.

The deepest implication may be a shift in disciplinary self-understanding. If psychology resembles meteorology or evolutionary biology more than mechanics, its explanations will often take the form of configurations rather than laws. Network models, dynamic systems approaches and Bruner's narrative modes of understanding all treat the mind as an ensemble of interacting influences. This is less a lowering of ambition than a different ambition: to model the architecture of many causes rather than to search for the master variable.

Takeaway

The test of a theory is not how much variance it explains but whether it accurately describes its causal structure, and whether it states in advance where its effects should be large and where they should vanish.

The smallness of psychological effects is overdetermined. Measurement error, underpowered designs and publication bias contribute, but they do not exhaust the explanation. Behind them lie two substantive facts: psychological outcomes arise from crowded causal fields, and causal structure varies across persons.

Taken seriously, these facts reposition psychology among the sciences of complex open systems. Its proper exemplars may be epidemiology and ecology rather than mechanics. Its proper ambitions may be the mapping of causal architecture, the specification of heterogeneity and the modelling of cumulative influence, rather than the prediction of single acts from single variables.

This does not license complacency about weak findings. It demands more rigour, because small effects are both scientifically meaningful and easy to fake. What emerges is a quieter but more honest conception of psychological knowledge: modest numbers describing a genuinely intricate mind, in which the task of theory is to explain why the numbers take the shape they do.