Retrospective Moral Judgment

Abstract

Retrospective moral judgment is the after-the-fact evaluation of a past action, and of the agent who performed it, as right or wrong and as blameworthy or praiseworthy. Because it is made once the outcome is known, it draws on information the agent lacked, and cognitive psychology has charted how this hindsight inflates judged foreseeability and biases blame toward bad results. This article treats the process as a problem in the psychology of judgment: how observers read a norm violation, weigh the agent's causation, intention, and control, and combine them into a graded verdict. The dominant view casts moral judgment as fast and intuitive, rationalized afterward, tuned to the perceived mind behind the act, and revised as new evidence arrives. Interactive demonstrations isolate outcome bias, the blame calculus, and the effect of psychological distance.

Keywords: retrospective moral judgment, blame, outcome bias, hindsight bias, moral cognition

What Retrospective Moral Judgment Is

Retrospective moral judgment is the assessment of a completed action, and of its agent, against a moral standard: the verdict that what was done was right or wrong, and that the person who did it deserves praise or blame. The defining feature is its backward orientation. Unlike a prospective moral choice, made under uncertainty about what will happen, a retrospective judgment is delivered once the consequences are known, and that asymmetry of information is the source of both its social usefulness and its characteristic errors. In the Medical Subject Headings vocabulary the term is filed under ethical analysis, where it carries the philosopher's sense of applying present moral standards to past conduct. This article treats the adjacent and larger psychological question: how the human mind actually forms moral verdicts about the past, and why those verdicts depart in lawful ways from the information the agent originally had.

Two things are being judged at once, and the literature is careful to separate them. The first is the act — whether the behaviour violated a moral norm, and how severely. The second is the agent — how much blame or credit the person deserves for it, which depends not only on what happened but on what the person caused, intended, foresaw, and could have prevented (Malle et al., 2014). A single event can draw a strong verdict on the act and a weak one on the agent, as when a serious harm results from a genuine accident, or a mild transgression is judged harshly because it was deliberate and malicious (Cushman, 2008). Keeping the two apart is what lets the psychology of retrospective judgment be more than a restatement of the moral facts.

The judgment also serves a forward-looking function that explains much of its structure. Evaluating a past act is, in practice, a way of learning what an agent is like and predicting how they will behave next, so retrospective moral judgment is tightly bound to the inference of character and trustworthiness (Everett et al., 2016). This social-regulatory purpose — deciding whom to cooperate with, whom to punish, and whom to avoid — is a recurring theme in the sections that follow, because it accounts for why the mind privileges intention over accident and why blame is more finely graded than praise.

Table 1 maps the phenomena and models the rest of this article develops, from the biases that distort the verdict to the mechanisms that compute it.

Outcome Knowledge and the Hindsight Problem

The foundational cognitive result behind retrospective moral judgment concerns memory and probability rather than morality. Fischhoff showed that once people learn how an uncertain situation turned out, they overestimate how predictable that outcome was all along — and misremember their own prior expectations as having been closer to the truth than they were. Reporting an outcome as having happened, he found, increases its judged prior probability, a distortion he named the creeping determinism of hindsight (Fischhoff, 1975). Because the outcome now feels as though it should have been foreseen, whoever failed to foresee it looks negligent, and this is the mechanism that couples an innocent fact about memory to a moral verdict about a person.

Its most direct moral consequence is outcome bias. Baron and Hershey presented identical decisions paired with either good or bad outcomes and asked participants to rate the quality of the decision and of the decision maker. The judgments tracked the outcome: a physician who chose a reasonable operation was rated a worse decision maker when the patient happened to die than when the same patient happened to live, even though the information available at the moment of choice was the same (Baron & Hershey, 1988). Normatively, a decision should be evaluated by what was knowable when it was made; outcome bias is the failure to hold the outcome constant, letting luck contaminate the assessment of both the choice and the chooser.

Outcome information does more than shade a rating up or down — it reorganizes the whole attribution of responsibility. Alicke's culpable control model describes how a salient bad outcome, and the negative affect it provokes, biases every upstream judgment that feeds into blame: the observer inflates the agent's causal contribution, downplays situational constraints, and reads intention into behaviour, so that the desire to blame someone for a bad result recruits the very evidence that would justify blaming them (Alicke, 2000). Retrospective judgment, on this account, is not a dispassionate reconstruction that happens to be biased at the margin; the affective response to the outcome is doing much of the causal work, with the reasoning arriving to ratify a conclusion already reached.

Computing Blame from a Past Act

Beyond its biases, retrospective moral judgment has a positive structure, and the most developed description of it is Malle, Guglielmo, and Monroe's theory of blame. Blame, on this model, is not an immediate affective flash but the output of a sequence of social-cognitive judgments arranged as a path. The observer first detects that an event violated a norm; then assesses whether the agent caused it; then whether the agent brought it about intentionally. If the act was intentional, blame turns on the agent's reasons and their justification; if it was unintentional, blame turns on whether the agent should and could have prevented it — its foreseeability and the obligation to have avoided it (Malle et al., 2014). Each branch draws on different information, which is why the same harm can be blamed heavily or lightly depending on where it enters the path.

The intentional and unintentional branches are separable in the mind, not merely on paper. Cushman had participants judge wrongness and punishment for cases that independently varied what an agent believed, intended, and caused, and found that verdicts of wrongness weighted the agent's mental states heavily, while judgments of deserved punishment gave more weight to the actual causal outcome. Moral evaluation, in other words, runs partly on an analysis of intention and partly on an analysis of causation, and these two analyses can pull a judgment in different directions (Cushman, 2008). The dissociation explains a familiar asymmetry: an attempted harm that fails is judged very wrong yet lightly punished, while an accidental harm that succeeds is judged less wrong yet more heavily penalized.

Control is the hinge on which the unintentional branch turns, and its centrality to blame has a long attributional lineage. Weiner's theory of perceived responsibility holds that an observer first judges whether an outcome was controllable by the agent, and that this controllability judgment governs the responsibility ascribed and the social emotions that follow — anger toward an agent held responsible, sympathy toward one seen as not (Weiner, 1993). Martin and Cushman sharpened the point experimentally, showing that people withhold blame in proportion to how little control an agent had over an outcome: harms that the agent could not have foreseen or prevented are forgiven, because blaming them would neither reflect the agent's character nor teach a useful lesson (Martin & Cushman, 2016). This forgiveness of the uncontrollable is the normative complement to outcome bias — the same machinery that overblames when hindsight makes an outcome look foreseeable correctly exonerates when the outcome is transparently beyond control. Retrospective blame, at its best, is a function-tracking device: it targets the agent to the degree that targeting them carries information about future behaviour.

Intuition, Reason, and the Mind Behind the Act

How the blame calculus is executed — deliberately or automatically — is the subject of the field's central theoretical dispute, and the modern answer is that moral verdicts are largely intuitive. Haidt's social intuitionist model holds that a moral judgment typically appears in consciousness as a quick, affect-laden intuition about right and wrong, with explicit reasoning arriving afterward to construct a justification for a conclusion already reached (Haidt, 2001). On this view the reasons people give for a retrospective verdict are frequently post hoc, which is one reason outcome-driven affect can steer a judgment while the person believes they are reasoning from principle. Greene and colleagues gave the picture a neural anchor, showing with functional imaging that moral dilemmas engaging strong emotional responses recruit distinct brain systems from those engaged by more impersonal moral problems, evidence for competing intuitive and controlled contributions to a single judgment (Greene et al., 2001).

Yet intuition is not immune to reason, and controlled work shows the two can be dissociated by content. Cushman, Young, and Hauser tested three principles that distinguish permissible from impermissible harms — the intention, action, and contact principles — and found that people applied some of them without being able to articulate why, while others were accompanied by sufficient justification, evidence that intuition and conscious reasoning both operate and do so over different rules (Cushman et al., 2006). Crockett situated this within a broader computational framework, mapping moral judgment onto the distinction between model-free and model-based valuation and giving the intuition-versus-reasoning contrast a formal learning-theoretic reading (Crockett, 2013).

What the intuitive system is most attuned to is the mind it perceives behind the act. Gray, Young, and Waytz argue that the perception of mind is the essence of moral judgment: an agent seen as having intentions and the capacity to know better is judged as a moral agent responsible for harm, and the more mind an observer attributes, the more blame a transgression attracts (Gray et al., 2012). The dependence runs in both directions. Although the blame path treats intentionality as an input that precedes the verdict, Knobe showed that the moral character of an outcome reaches back and colours the intentionality judgment itself: people say an agent who foresaw but did not care about a harmful side effect brought it about intentionally, yet deny that an identical but beneficial side effect was intentional — the asymmetry now known as the side-effect effect (Knobe, 2003). The content of the norms whose violation triggers the judgment is itself plural and partly cultural: moral foundations theory catalogues several distinct concerns — care, fairness, loyalty, authority, sanctity — on which an act can be evaluated, so that whether a past deed is condemned depends on which foundation the observer brings to it (Graham et al., 2013). Psychological distance modulates which level of construal is applied: Eyal, Liberman, and Trope found that acts judged from a temporally or socially distant standpoint are evaluated more by abstract moral principle and less by concrete mitigating circumstance, so the same vice is condemned more harshly, and the same virtue praised more warmly, when it is viewed from afar (Eyal et al., 2008).

Retrospective Moral Judgment in Motion

The three demonstrations below make the core mechanisms manipulable. The first isolates outcome bias, holding a decision's quality fixed while its outcome varies and showing the human verdict drift with the luck of the result. The second turns the theory of blame into a controllable path, letting the reader set causation, intention, and preventability and watch the graded verdict assemble. The third makes psychological distance concrete, showing why the same past act is judged more extremely from afar than up close.

Outcome bias: judging a decision by how it turned out

The decision is identical in both outcomes, so the normative blame stays 30. The human verdict is 52, a shift of +22 toward the bad outcome. That gap is outcome bias: luck contaminating the judgment of a choice fixed before the result was known.

The outcome-bias demonstration reproduces the logic of Baron and Hershey. The decision quality — the probability of a good result that was knowable at the moment of choice — is set by the reader and does not change. Toggling the realized outcome between good and bad leaves the normative evaluation untouched, because the choice was the same, yet the modelled human verdict swings, rating the identical decision and decider more harshly after a bad result. The gap between the two bars is outcome bias made visible (Baron & Hershey, 1988).

The blame calculus: from a past act to a graded verdict

The graded verdict is 72.9 of 100. On the intentional branch, strong justification of the reasons is what pulls blame down.

The blame-calculus demonstration puts the theory of blame under the reader's hand. Setting the norm violation's severity and the agent's causal contribution, then choosing whether the act was intentional, routes the judgment down one of two branches: an intentional act is mitigated by strong justification, an unintentional one is aggravated by high preventability. The resulting blame score shows why intention and accident are weighted so differently and how a graded verdict emerges from a few underlying inputs (Malle et al., 2014).

Psychological distance and moral extremity

From close up the transgression is judged at 55 of 100. Up close, concrete mitigating circumstances temper the verdict.

The construal-distance demonstration makes Eyal, Liberman, and Trope's finding tangible. A slider sets the psychological distance from which a past act is viewed; as distance grows, the weight on abstract moral principle rises and the weight on concrete mitigating circumstance falls, so the judged severity of a vice increases and the judged merit of a virtue increases with it. It shows why a transgression viewed from a temporal or social remove is condemned more absolutely than the same transgression seen close up (Eyal et al., 2008).

Worked Example

Take the blame calculus the second demonstration builds and give it numbers. Model a graded blame verdict on a 0-100 scale as the product of three inputs and a branch-specific weight: B = 100 · s · c · m, where s is the norm violation's severity (0 to 1), c is the agent's causal contribution (0 to 1), and m is a branch-specific moral-responsibility factor. For an intentional act, m = w_i · (1 − j), where the intentional-branch weight w_i = 1.0 and j is the justification of the agent's reasons (0 to 1); for an unintentional act, m = w_u · p, where the unintentional-branch weight w_u = 0.6 and p is the outcome's preventability (0 to 1).

Consider a serious harm, s = 0.9, that the agent clearly caused, c = 0.9. If the act was intentional and largely unjustified, j = 0.1, then m = 1.0 · (1 − 0.1) = 0.90 and B = 100 · 0.9 · 0.9 · 0.90 = 72.9 — a heavy verdict. Now hold the harm and causation fixed but make it an accident the agent could largely have prevented, p = 0.5: m = 0.6 · 0.5 = 0.30 and B = 100 · 0.9 · 0.9 · 0.30 = 24.3. The identical harm, identically caused, draws roughly a third of the blame once it is unintended, quantifying the intention effect that Cushman isolated experimentally (Cushman, 2008).

The control principle falls out of the same formula. Drive preventability toward zero for the accidental case — an outcome the agent could not have foreseen or averted, p = 0.05 — and blame collapses to B = 100 · 0.9 · 0.9 · (0.6 · 0.05) = 2.4, the near-total forgiveness of the uncontrollable that Martin and Cushman documented (Martin & Cushman, 2016). Outcome bias, in these terms, is a distortion of a single input: hindsight inflates the perceived preventability p of a harm whose bad outcome is now known, pushing the unintentional branch toward a verdict the pre-outcome facts did not warrant (Fischhoff, 1975). The arithmetic does not settle which weights the mind actually uses, but it makes precise how intention, causation, and control combine — and where luck sneaks in.

Discussion

The psychology of retrospective moral judgment resolves an apparent paradox: people believe they are judging a past act on its merits, yet their verdicts are demonstrably swayed by information the agent never had. The resolution is that moral judgment is not primarily a reconstruction of the past but a fast, intuitive, affect-driven appraisal aimed at the future — deciding what an agent's behaviour reveals about their character and how much they should be trusted or deterred (Haidt, 2001). Read this way, outcome bias and hindsight are not incidental flaws but the cost of a system that recruits the most salient available signal, the outcome, to a judgment that in the ancestral case had to be made quickly and defended socially (Alicke, 2000).

The same functional logic explains the structure that survives scrutiny. Blame is graded rather than binary, weighted toward intention over accident, and withheld for the uncontrollable, because those are exactly the features that make a verdict diagnostic of the agent rather than of their luck (Malle et al., 2014). When the machinery works, it forgives the unforeseeable and condemns the deliberate; when it fails, it is because a vivid outcome has hijacked the inputs — inflating perceived causation, intention, and foreseeability — so that the agent is blamed for the world's contingency rather than their own conduct (Martin & Cushman, 2016).

For cognitive psychology the enduring interest of retrospective moral judgment is that it is a judgment under uncertainty wearing a moral face. The biases that afflict it — hindsight, outcome bias, the affective contamination of causal attribution — are the same biases that afflict probabilistic judgment generally, which is why the field's foundational result came from the study of judgment under uncertainty rather than from ethics (Fischhoff, 1975). What the moral case adds is the perception of mind and intention, and the social stakes of getting the verdict right, which together give retrospective moral judgment its distinctive combination of automaticity, sensitivity to mental states, and resistance to the very outcome information it cannot help using (Gray et al., 2012).

Current Directions

One active line treats the retrospective verdict as a signal read by others rather than a private appraisal. Everett, Pizarro, and Crockett showed that agents who reject harming one person to save several — the characteristically deontological choice — are judged more trustworthy and are preferred as social partners, evidence that observers use a person's past moral judgments to infer character and to decide whom to cooperate with (Everett et al., 2016). This reframes retrospective moral judgment as part of a partner-selection system, and predicts that the features a judgment weights should be those that best forecast future cooperation.

A second direction examines how verdicts change as evidence accumulates. Monroe and Malle found that people systematically revise their blame judgments when new information arrives about an agent's mental states or the situation, and that this moral judgment updating follows the structure of the theory of blame rather than drifting arbitrarily — blame is revised, not merely anchored (Monroe & Malle, 2019). Alongside this, Guglielmo and Malle documented an asymmetry between the negative and positive cases: blame is more finely differentiated across levels of intentionality and cause than praise is, consistent with blame being the more computationally elaborated and socially consequential of the two (Guglielmo & Malle, 2019).

A third strand asks whether the judgments studied in the laboratory predict real moral behaviour. Bostyn, Sevenhant, and Roets compared hypothetical trolley-style judgments with choices in a consequential analogue and found that what people say they would do in a described dilemma does not straightforwardly match what they do when a real, if mild, consequence is at stake — a caution that the retrospective verdicts elicited by vignettes may diverge from the judgments people act on (Bostyn et al., 2018). Across these strands the trajectory is toward treating retrospective moral judgment as a dynamic, socially embedded inference rather than a static reading of a fixed moral fact.

Common Misconceptions

People judge a past decision purely by the information available when it was made.

They do not. Outcome bias is the well-documented tendency to rate an identical decision as worse, and the decider as more culpable, when it happens to turn out badly, even though the outcome was unknowable at the moment of choice (Baron & Hershey, 1988).

Hindsight bias is just a memory error with no moral consequences.

The memory error is the engine of a moral one. Because outcome knowledge inflates how foreseeable an outcome seems, it makes whoever failed to foresee it look negligent, converting a distortion of probability judgment into an inflation of blame (Fischhoff, 1975).

Moral verdicts are reached by conscious reasoning from principles.

Reasoning usually arrives second. The social intuitionist model holds that a moral judgment typically surfaces as a fast affective intuition, with explicit justification constructed afterward, which is why outcome-driven feeling can steer a verdict a person believes they reasoned into (Haidt, 2001).

Blame is an all-or-nothing reaction to a bad act.

Blame is graded and computed. It emerges from a path of distinct judgments — causation, intentionality, reasons, and, for accidents, preventability — so the same harm can draw very different verdicts depending on the agent's mental states and control (Malle et al., 2014).

Glossary

Attribution theory of responsibility.

Weiner's account on which an observer's judgment of whether an outcome was controllable by the agent governs the responsibility ascribed and the social emotions, such as anger and sympathy, that follow.

Blame.

A graded negative moral verdict directed at an agent for a norm violation, computed from judgments of causation, intentionality, reasons, and preventability rather than issued all at once.

Causal contribution.

The degree to which an agent's action, rather than other factors, produced an outcome; one of the inputs to blame, and a judgment that vivid bad outcomes tend to inflate.

Construal-level theory.

The account on which greater psychological distance shifts thought toward abstract, high-level features; applied to morality, it predicts more principle-based and more extreme judgments of distant acts.

Culpable control model.

Alicke's model in which the affective reaction to a bad outcome biases the upstream judgments of causation and intention that feed into blame, so the wish to blame recruits its own justification.

Dual-process moral judgment.

The view that moral verdicts arise from competing contributions of a fast, emotional system and a slower, controlled one, supported by imaging evidence that different dilemmas recruit different neural systems.

Foreseeability.

The extent to which an agent could have anticipated an outcome; central to blaming unintended harms, and the judgment most distorted by hindsight.

Hindsight bias.

The tendency, once an outcome is known, to overestimate how predictable it was and to misremember one's own prior expectations as closer to it; the cognitive root of outcome-driven blame.

Intentionality.

Whether an agent brought an outcome about on purpose; the pivot of the blame path, routing judgment to reasons for intentional acts and to preventability for unintentional ones.

Mind perception.

The attribution of intention and the capacity to know better to an agent; on Gray and colleagues' account, the perception that underlies whether and how much an act is morally judged.

Moral foundations theory.

Graham and colleagues' catalogue of distinct moral concerns — care, fairness, loyalty, authority, sanctity — that determines which norm an observer treats a past act as violating.

Moral intuition.

A fast, affect-laden appraisal of right or wrong that appears in consciousness without deliberate reasoning and that reasoning is typically marshalled to justify.

Moral judgment updating.

The systematic revision of a blame verdict as new evidence about an agent's mental states or situation arrives, following the structure of the blame path rather than drifting arbitrarily.

Outcome bias.

The judging of a decision and its maker by how the decision turned out rather than by what was knowable when it was made; the direct moral consequence of hindsight.

Presentism.

In the ethical sense that MeSH files under the descriptor, the application of current moral standards to judge past actions, institutions, or persons.

Preventability.

Whether an agent could and should have averted an unintended outcome; the input on the unintentional branch of the blame path, and the one hindsight tends to overstate.

Retrospective moral judgment.

The after-the-fact evaluation of a completed action, and of its agent, as right or wrong and as blameworthy or praiseworthy.

Side-effect effect.

The Knobe-effect asymmetry whereby a foreseen harmful side effect is judged intentional while an identical beneficial one is not, showing that moral valence shapes attributions of intentionality.

Social intuitionist model.

Haidt's account on which moral judgment is driven by quick intuition, with explicit reasoning serving mainly to justify a conclusion already reached.

Trustworthiness inference.

The use of an agent's moral judgments and past acts to predict their character and future cooperation, the forward-looking function that shapes retrospective evaluation.

Key Researchers

Molly J. Crockett. Neuroscientist at Princeton University whose synthesis of moral-judgment models and experiments on how moral verdicts signal trustworthiness connect retrospective evaluation to its social function of predicting whom to trust. ORCID - Google Scholar - Faculty page - Wikipedia

Fiery Cushman. Psychologist at Harvard University who dissected moral judgment into separable causal and intentional analyses, tested the harm principles behind intuitive verdicts, and showed that people forgive what an agent could not control. ORCID - Google Scholar - Faculty page - Wikipedia

Baruch Fischhoff. Decision scientist at Carnegie Mellon University whose demonstration of hindsight bias is the foundational cognitive result behind retrospective moral judgment, explaining why a known outcome is misremembered as having been foreseeable. ORCID - Google Scholar - Faculty page - Wikipedia

Kurt Gray. Social psychologist at The Ohio State University whose mind-perception account of morality explains why a past agent's presumed intentions and capacity to know better drive how harshly the deed is later judged. Google Scholar - Faculty page

Joshua D. Greene. Psychologist and neuroscientist at Harvard University whose functional imaging work grounded the dual-process picture of moral judgment, showing distinct emotional and controlled contributions to evaluating a transgression. Google Scholar - Faculty page - Wikipedia

Jonathan Haidt. Social psychologist at New York University whose social intuitionist model reframed moral judgment as fast intuition rationalized after the fact, and whose moral foundations theory supplies the changing normative content by which a past act is re-judged. ORCID - Google Scholar - Faculty page - Wikipedia

Bertram F. Malle. Psychologist at Brown University whose theory of blame formalizes the information-processing path from a norm violation to a graded verdict, and whose experiments show people systematically update those verdicts as evidence arrives. ORCID - Google Scholar - Faculty page

Frequently Asked Questions

What is retrospective moral judgment?
It is the evaluation of a past action, and of the agent who performed it, as morally right or wrong and as blameworthy or praiseworthy. Because the judgment is made after the fact, it draws on information the agent lacked at the time, chiefly the outcome, which is what gives it its distinctive biases (Malle et al., 2014).

How does knowing the outcome change the judgment?
It shifts the verdict in the outcome's direction. Baron and Hershey found that an identical decision is rated worse, and its maker judged more culpable, when it happens to turn out badly, a distortion called outcome bias because it lets the result contaminate the assessment of a choice that was fixed before the result was known (Baron & Hershey, 1988).

Why does hindsight matter for blame?
Once an outcome is known, people overestimate how foreseeable it was and misremember their own prior expectations as closer to it. That inflation of perceived foreseeability makes whoever failed to foresee the outcome look negligent, turning a memory bias into a moral one (Fischhoff, 1975).

Is blame a single reaction or a computed judgment?
It is computed. The theory of blame describes a path in which an observer detects a norm violation, then evaluates causation, intentionality, the agent's reasons, and, for accidents, preventability, combining these into a graded verdict rather than an all-or-nothing reaction (Malle et al., 2014).

Why is an intentional harm blamed more than an accidental one?
Because intention and causation are analyzed separately, and wrongness weights mental states heavily. Cushman showed that judgments of wrongness track what an agent intended and believed, so a deliberate harm draws far more blame than an identical harm caused by accident (Cushman, 2008).

Do people forgive harms an agent could not control?
Yes. Martin and Cushman found that blame falls as an agent's control over an outcome falls, so harms that could not be foreseen or prevented are largely forgiven, because blaming them would reveal nothing about the agent's character (Martin & Cushman, 2016).

Are moral judgments mostly emotional or mostly reasoned?
The dominant view is that they are mostly intuitive. Haidt's social intuitionist model holds that a moral verdict appears as a quick affective intuition, with reasoning arriving afterward to justify it, and imaging work shows emotionally engaging dilemmas recruit distinct neural systems (Haidt, 2001).

Does how people judge dilemmas predict how they actually behave?
Not straightforwardly. Bostyn, Sevenhant, and Roets compared hypothetical trolley-style judgments with behaviour in a consequential analogue and found the two diverge, a caution that verdicts elicited by vignettes may not match the judgments people act on (Bostyn et al., 2018).

References