The replication crisis has been framed primarily as a methodological scandal—a story of p-hacking, underpowered studies, and questionable research practices finally exposed. This framing, while not wrong, is incomplete. It treats symptoms while obscuring a more fundamental diagnosis: psychology's theoretical infrastructure may be inadequate to the phenomena it purports to explain.

Consider what a failed replication actually tells us. The standard interpretation assumes a well-specified theoretical prediction that either holds or fails across contexts. But psychological theories rarely achieve this level of specification. When a priming study fails to replicate, we cannot determine whether the original finding was spurious, whether contextual moderators differ, or whether our theoretical understanding of priming itself is too vague to generate precise predictions. The crisis, in this light, reveals not merely sloppy methods but theoretical underspecification masquerading as empirical claims.

This essay argues that the replication crisis presents an opportunity for metatheoretical reflection that the field has largely avoided. By examining how theoretical assumptions shape replication logic, what effect sizes actually signify, and what genuine theoretical reconstruction would require, we can understand why methodological reforms alone—preregistration, increased power, open data—cannot resolve problems rooted in theoretical architecture. The crisis is not a temporary embarrassment to be managed but a diagnostic window into psychology's epistemological foundations.

Philosophy of science has long recognized that observation is theory-laden—what we see depends on the conceptual frameworks we bring to looking. This insight, central to the work of Kuhn, Hanson, and Feyerabend, applies with particular force to replication in psychology. What counts as a 'same' study, a 'same' manipulation, or a 'same' population cannot be determined without theoretical commitments that are themselves under dispute.

Consider the extensive debates surrounding failed replications of ego depletion effects. Critics of the original findings argue that null results demonstrate the phenomenon is not robust. Defenders counter that replications used different populations, different dependent measures, or different motivational contexts. Both positions are internally coherent—and the disagreement cannot be resolved by methodological standards alone. The question of what constitutes a theoretically relevant difference between original and replication requires a theory of ego depletion that specifies boundary conditions, moderators, and mechanisms with precision that the literature has never provided.

This is not an isolated case. Social priming, stereotype threat, facial feedback, and numerous other research programs face identical interpretive ambiguities when replications fail. The problem is structural: psychological theories typically specify that some manipulation X produces some outcome Y, without articulating the causal pathway, contextual dependencies, or measurement invariance assumptions that would allow us to determine whether a given replication is a genuine test. We lack what philosophers call auxiliary hypotheses—the connecting assumptions that link abstract theoretical claims to concrete experimental procedures.

The methodological reform movement implicitly assumes that if we simply do studies correctly—with adequate power, preregistration, and transparency—truth will emerge. But this assumption presupposes that our theoretical claims are sufficiently well-articulated to generate precise, testable predictions. When theories remain at the level of verbal propositions connecting vaguely defined constructs, no amount of methodological rigor can compensate. We can perfectly execute a test of an imprecise prediction and learn very little.

The theory-ladenness problem suggests that replication debates often function as proxy wars for deeper theoretical disagreements that the field has not confronted directly. Rather than asking whether a finding replicates, we might more productively ask: what would our theory have to specify for replication success or failure to be interpretable? This shift in framing moves attention from the symptom to the disease.

Takeaway

A failed replication is not a straightforward empirical verdict but an interpretive puzzle whose resolution depends on theoretical commitments that psychology's major theories rarely make explicit.

Much discussion of the replication crisis focuses on effect sizes—original studies claimed large effects that replications reveal to be small or nonexistent. The standard interpretation frames this as a calibration problem: researchers overestimated how strongly their manipulations influenced outcomes. But a deeper question lurks beneath: what does a psychological effect size actually represent, and what should we expect it to be?

In mature sciences, effect sizes connect to theoretical parameters with interpretable meanings. The gravitational constant has a specific value because of the underlying physics; deviations from expected values would indicate either measurement error or new phenomena requiring theoretical revision. Psychology rarely achieves this integration. When we report that a manipulation produces d = 0.3, we typically cannot say whether this represents a strong influence constrained by measurement noise, a weak influence operating through the hypothesized mechanism, or an artifact of the particular operationalization chosen. Effect sizes float free of theoretical anchoring.

Consider what small effect sizes might signify if taken seriously as theoretical information. One possibility is that psychological manipulations genuinely produce modest influences on behavior—that human psychology involves many weak causes whose effects combine in complex ways. If true, this would have profound implications for intervention design, for the relationship between laboratory and field effects, and for what kind of science psychology can be. Another possibility is that our measurements are too noisy, our manipulations too imprecise, or our constructs too poorly specified to capture the genuine causal structure underlying behavior. These possibilities require different responses.

The field has largely defaulted to treating small effects as embarrassments to be overcome through better methodology—larger samples, more precise measures, cleaner manipulations. This response assumes that the effects we seek are there to be found if only we look carefully enough. But it may be that small and variable effects are telling us something true about psychological causation: that context-sensitivity, individual differences, and developmental history matter so much that population-level regularities are inherently weak. Taking this possibility seriously would require theoretical frameworks that embrace complexity rather than seeking simple main effects.

The methodological reform agenda focuses on accurately estimating effect sizes through improved practices. This is valuable but insufficient. We also need theoretical work that tells us what effect sizes should mean—what underlying causal structure they reflect, how they should vary across contexts, and what magnitude would constitute strong versus weak evidence for different theoretical claims. Without such frameworks, ever-more-precise estimation of theoretically unanchored parameters produces knowledge that is accurate but shallow.

Takeaway

Small effect sizes may reflect not methodological failures but genuine features of psychological causation that our theories have not been designed to accommodate or explain.

If the replication crisis stems partly from theoretical inadequacy, then resolution requires theoretical reconstruction, not merely methodological reform. But what would such reconstruction involve? Drawing on philosophy of science and examples from other disciplines, we can identify several features that more adequate psychological theories would possess—features largely absent from current theoretical discourse.

First, adequate theories would specify mechanisms rather than merely associations. Knowing that X correlates with or causes Y provides limited insight; understanding how X produces Y enables prediction across contexts, identification of moderators, and intervention design. Psychological theories often skip this step, moving directly from manipulation to outcome without articulating the intervening process. When mechanisms are specified, they frequently remain at the level of verbal description rather than formal models that can generate quantitative predictions.

Second, adequate theories would integrate across levels of analysis. Psychology studies behavior, cognition, affect, neural activity, social interaction, and developmental trajectories—but theories typically address single levels in isolation. A priming theory that does not connect to neural mechanisms of memory, individual difference factors, and social contextual variables will necessarily be incomplete. The replication crisis has revealed how much our findings depend on unspecified contextual factors; integrative theories would make these dependencies explicit and predictable.

Third, adequate theories would be formally specified in ways that allow derivation of precise predictions. The most successful sciences express theories mathematically, enabling deduction of consequences that can be tested quantitatively. Psychology's verbal theories permit only vague directional predictions—X should increase Y—that are difficult to falsify and impossible to precisely replicate. Computational modeling offers a path toward formalization, but remains peripheral rather than central to mainstream psychological theorizing.

What prevents such reconstruction? Incentive structures reward novelty over depth, empirical production over theoretical development, and quick publishable studies over long-term programmatic work. But the barriers are not merely institutional. Genuine theoretical progress may require acknowledging that much of what we thought we knew lacks the foundations to count as knowledge. This is a harder acknowledgment than admitting that our methods need improvement—it questions the epistemic status of our accumulated findings rather than merely their precision. The replication crisis invites this more radical reflection, but the field has largely declined the invitation.

Takeaway

Methodological reforms address symptoms while leaving the underlying disease—theoretical underspecification—untreated; genuine progress requires building theories that specify mechanisms, integrate levels, and generate precise predictions.

The replication crisis has prompted substantial methodological reflection and reform—more transparency, larger samples, preregistration, and collaborative replication efforts. These developments represent genuine progress in research practice. But they leave untouched the deeper theoretical problems that failed replications expose.

Psychology faces a choice it has largely deferred. We can continue treating the crisis as a methodological embarrassment to be managed through better practices, preserving our theoretical frameworks intact. Or we can accept the more uncomfortable diagnosis: that our theories have been insufficiently specified to generate the kind of knowledge we assumed we were producing. The crisis reveals not just that we were wrong about particular findings, but that our theoretical apparatus was inadequate to tell us clearly what we had found.

Embracing this diagnosis opens possibilities for reconstruction—for building psychological theory that specifies mechanisms, integrates levels, and connects to formal frameworks capable of generating precise predictions. Such reconstruction is difficult, long-term work without guaranteed success. But it is the work that the crisis, properly understood, demands.