An examined document · read the margins

Occam's Razor

The tendency to over-prefer the simplest explanation or plan—the one positing the fewest causes or unobserved effects—even when probability information shows a more complex account is at least as likely or more likely to be true.

Filed · occam-s-razorAnnotated in the gutter

Occam's Razor is a normative philosophical principle ("do not multiply entities beyond necessity"), but psychologists have shown that as a cognitive default it can bias judgment: people assign simpler causal explanations an inflated prior probability and demand disproportionate evidence before switching to a more complex one (Lombrozo, 2007). A related "narrow latent scope bias" shows people favor explanations that predict fewer unobserved effects, violating the laws of probability—though whether this is a robust fallacy or largely an artifact of forced-choice tasks is actively disputed (Stephan, 2023; Khemlani et al., 2024). In medicine the same instinct ("when you hear hoofbeats, think horses") collides with Hickam's dictum, and documented cases show single-diagnosis reasoning can miss coexisting independent diseases.

The dominant account is that simplicity functions as a heuristic cue to prior probability: fewer-cause explanations are assigned higher priors, so they need disproportionate evidence to be displaced (Lombrozo, 2007). For latent scope, the 'inferred evidence' account holds that when evidence is unobserved but potentially diagnostic, people guess it is absent (often using base rates that are not actually relevant), ruling out the broader explanation (Johnson et al., 2016). A 2024 account reframes the simplicity preference as a special case of a general preference for completing goals efficiently—people judge explanations the way they judge methods/processes (Sehl et al., 2024). A Bayesian model-selection account argues the preference is largely adaptive: integrating over latent causes penalizes overly flexible complex hypotheses fit to noise (Piasini et al., 2025).

Disagreement centers on whether the latent-scope effect is a genuine reasoning fallacy or a rational/pragmatic response: Stephan (2023) argues much of it reflects sensible inferences about feature diagnosability and forced-choice artifacts, while Khemlani et al. (2024) argue it persists across formats. Separately, mechanism accounts conflict on whether simplicity preference is heuristic over-reliance (Lombrozo), efficiency-goal transfer (Sehl et al.), or near-optimal Bayesian inference (Piasini et al.). Causal mechanism information can even reverse the bias toward complexity (Zemla et al., 2023).

The correction

Force the probability comparison explicitly rather than choosing by feel: list the competing explanations, ask 'what would I expect to observe under each?', and treat unobserved evidence as genuinely unknown (50/50) rather than absent. Allow 'both/multiple' as an answer (the latent-scope bias largely disappears when 'equally likely' is an option), and ask whether causal mechanism detail would actually justify the more complex account. In diagnosis, deliberately invoke Hickam's dictum for elderly/comorbid patients.

Stephan (2023, N=2200) showed the bias is strong under forced choice but is eliminated or reduced when participants can give the normatively correct 'equally likely' answer; Lombrozo (2007) showed the preference reverses when probability information unambiguously favors the complex account; Zemla et al. (2023) showed adding causal mechanism information increases preference for complex explanations.

Further annotations

Stephan (2023) — bias may be an artifact of task format. The narrow latent scope bias was far less robust than claimed: it was strong only under forced choice, largely disappeared when participants could give the correct answer, and was driven by a minority of participants plus sensible pragmatic inferences rather than uniform fallacy.

Simon Stephan, 2023 · not reported as a single standardized value; bias 'eliminated' (Exp 1) or 'reduced' (Exps 2–3) under scale format

Khemlani, Johnson, Oppenheimer & Sussman (2024) — robustness rebuttal. Narrow-scope responses dominated (55% in Exp 1, 62% in Exp 2) and 127 of 187 participants showed the bias (binomial test p<.001); the authors conclude the bias is robust and replicable across many task types, not just forced choice.

Sangeet Khemlani, Samuel G.B. Johnson, Daniel M. Oppenheimer, Abigail B. Sussman, 2024 · 127/187 participants biased; 55% and 62% narrow-scope responses

Piasini, Liu, Chaudhari, Balasubramanian & Gold (2025) — Occam's razor in perceptual decisions. Participants showed simplicity preferences consistent with Bayesian model selection, with significant sensitivity to all four complexity features; accuracy declined when individual sensitivity departed from theoretically optimal values, suggesting an adaptive (rather than purely fallacious) simplicity bias.

Eugenio Piasini, Shuze Liu, Pratik Chaudhari, Vijay Balasubramanian, Joshua I. Gold, 2025 · population sensitivities: dimensionality 4.66±0.96, robustness 2.21±0.12, boundary 1.12±0.10, volume 0.23±0.12

case · Occam's razor versus Hickam's dictum: two very rare tumours in one single patient (2019)

A 71-year-old man with a month of anorexia, fatigue, fever and weight loss—classic single-syndrome constitutional symptoms—was found on investigation to have two independent rare tumors (a right-atrial cardiac hemangioma and a neoplastic appendiceal mucocele), each requiring separate surgery. The authors use the case to argue that anchoring on one unifying diagnosis (Occam) can miss coexisting independent disease (Hickam).

case · Hickam's Dictum: An Analysis of Multiple Diagnoses (2024)

A structured study (review of 83 published case reports, 220 New England Journal of Medicine cases, and a survey of 265 physicians) documenting that real patients frequently have multiple coexisting diagnoses in predictable patterns (incidentalomas, preexisting conditions, causally related diseases, coincidental independent diseases), and that the simplest single-cause framing systematically underweights this—an empirical caution against reflexive Occam reasoning, especially in older/comorbid patients.

Catch it in the act

You're choosing among explanations or plans and the simplest one just 'feels right'—one cause, one culprit, one clean story—before you've actually checked the probabilities. Red flags: you treat unobserved/missing evidence as if it's absent; you'd need overwhelming proof to accept a two-cause account; you say 'horses not zebras' about a patient who is old or has many systems involved; or you pick the tidy narrative because it's manageable, not because it's more likely.

Cross-references in the margin

Narrow latent scope biaschildA specific, experimentally studied form of over-preferring simplicity: favoring explanations that predict fewer unobserved effects.
Hickam's dictumoppositeClinical counter-maxim ('a patient can have as many diseases as he pleases') that warns against forcing one diagnosis.
Anchoring / premature closuremechanistically-linkedOnce a simple explanation is adopted it becomes an anchor; clinicians and reasoners stop searching, amplifying the simplicity trap.
Conjunction fallacyeasily-confusedThe conjunction fallacy is over-preferring a more specific/complex story; it is in tension with—not the same as—the simplicity preference, and which dominates depends on framing and mechanism cues.
Less-is-better / minimal-cause heuristicsiblingRelated preference for parsimonious, manageable accounts; may share an 'efficiency of goal completion' mechanism (Sehl et al., 2024).

Open the same case on your own draft.