No Luck for Moral Luck

Abstract

Moral philosophers and psychologists often assume that people judge morally lucky and morally unlucky agents differently, an assumption that stands at the heart of the puzzle of moral luck. We examine whether the asymmetry is found for reflective intuitions regarding wrongness, blame, permissibility and punishment judgments, whether people's concrete, case-based judgments align with their explicit, abstract principles regarding moral luck, and what psychological mechanisms might drive the effect. Our experiments produce three findings: First, in within-subjects experiments favorable to reflective deliberation, wrongness, blame, and permissibility judgments across different moral luck conditions are the same for the vast majority of people. The philosophical puzzle of moral luck, and the challenge to the very possibility of systematic ethics it is frequently taken to engender, thus simply does not arise. Second, punishment judgments are significantly more outcome-dependent than wrongness, blame, and permissibility judgments. While this is evidence in favor of current dual-process theories of moral judgment, the latter need to be qualified since punishment does not pattern with blame. Third, in between-subjects experiments, outcome has an effect on all four types of moral judgments. This effect is mediated by negligence ascriptions and can ultimately be explained as due to differing probability ascriptions across cases.

Full text

Contents lists available at ScienceDirect
Cognition
journal homepage: www.elsevier.com/locate/cognit
Original Articles
No luck for moral luck
Markus Kneer
a,⁎,1
, Edouard Machery
b,1
a
University of Zurich, Rämistrasse 66, 8001 Zurich, Switzerland
b
University of Pittsburgh, 4200 Fifth Avenue, 15260 Pittsburgh, USA
ARTICLE INFO
Keywords:
Moral luck
Moral judgment
Outcome effect
Dual process theory of moral judgment
Hindsight bias
ABSTRACT
Moral philosophers and psychologists often assume that people judge morally lucky and morally unlucky agents
differently, an assumption that stands at the heart of the Puzzle of Moral Luck. We examine whether the
asymmetry is found for reflective intuitions regarding wrongness, blame, permissibility, and punishment judg-
ments, whether people’s concrete, case-based judgments align with their explicit, abstract principles regarding
moral luck, and what psychological mechanisms might drive the effect. Our experiments produce three findings:
First, in within-subjects experiments favorable to reflective deliberation, the vast majority of people judge a
lucky and an unlucky agent as equally blameworthy, and their actions as equally wrong and permissible. The
philosophical Puzzle of Moral Luck, and the challenge to the very possibility of systematic ethics it is frequently
taken to engender, thus simply do not arise. Second, punishment judgments are significantly more outcome-
dependent than wrongness, blame, and permissibility judgments. While this constitutes evidence in favor of
current Dual Process Theories of moral judgment, the latter need to be qualified: punishment and blame
judgments do not seem to be driven by the same process, as is commonly argued in the literature. Third, in
between-subjects experiments, outcome has an effect on all four types of moral judgments. This effect is
mediated by negligence ascriptions and can ultimately be explained as due to differing probability ascriptions
across cases.
1. Introduction
1.1. Moral luck
Sally had a few too many at the bar yet decides to drive home. On
the way, a child stumbles in front of her car, is run over, and dies. John,
having had just as much to drink as Sally, also drives home yet arrives
at his house without incident. According to philosophers, people are
inclined to judge Sally as more blameworthy than John, or her action as
morally worse than John’s. We will call the distinct assessment of the
two actions or of the two agents the Difference Intuition. The Difference
Intuition appears to be in conflict with the widely held Control Principle,
according to which agents are only morally responsible for features of
their actions that are under their control. The Puzzle of Moral Luck arises
due to the fact that both the Difference Intuition and the Control
Principle are plausible, yet appear in conflict with one another (Nagel,
1979; Williams, 1981; for a review, see Zimmerman, 2006; we limit
ourselves to resultant moral luck, and do not address any of the other
kinds discussed by Nagel, 1979).
What could explain the Difference Intuition? Much evidence
supports the hypothesis that outcome has an effect on moral judgment
(Alicke & Davis, 1989; Alicke, Davis, & Pezzo, 1994; Baron & Hershey,
1988; Lowe & Medway, 1976; McKillip & Posavac, 1975; Walster, 1966;
for a meta-analysis, see Robbennolt, 2000). According to the Simple
View (not necessarily endorsed by the authors just listed), the Difference
Intuition is just the product of an unmediated effect of outcome on
moral judgment (Fig. 1).
More complex models propose that the effect of outcome on moral
judgments is mediated by the attribution of certain mental states.
According to the Epistemic Single Pathway Model, the difference in moral
judgment results from a difference in epistemic states ascribed to the
agent (Fig. 2; our typology extends the one proposed by Nichols,
Timmons, & Lopez, 2014).
To take an example, Royzman and Kumar (2004) suggest that the
Difference Intuition is due to an “epistemically corrupted evaluation” of
the agent and her situation. Falling prey to the “I know, you know” bias
(on which see Royzman, Cassidy, & Baron, 2003), people unconsciously
and inappropriately project their superior knowledge about the out-
come onto the agent: They are more inclined to ascribe knowledge
about the harmful outcome hor awareness of a substantial probability
https://doi.org/10.1016/j.cognition.2018.09.003
Received 20 August 2017; Received in revised form 27 August 2018; Accepted 3 September 2018
⁎
Corresponding author.
1
Both authors contributed equally.
E-mail addresses: markus.kneer@uzh.ch (M. Kneer), machery@pitt.edu (E. Machery).
Cognition 182 (2019) xxx–xxx
0010-0277/ © 2018 Published by Elsevier B.V.
T

of h’s occurrence to the agent in the unlucky condition than in the lucky
condition. The more severe moral evaluation of the unlucky agent is
thus not due to a direct impact of outcome on moral judgment, but
instead due to a higher propensity to attribute the inculpating mental
state of foresight (knowledge that hwill occur) or recklessness
(awareness of a substantial probability of h).
A third model, the Probabilistic Single Pathway Model, explains the
difference in moral judgment across cases in terms of the hindsight bias
(on the hindsight bias, see Fischhoff, 1975; for a review see Hawkins &
Hastie, 1990; for a meta-analysis see Christensen-Szalanski & Willham,
1991). According to this third model, the impact of outcome on moral
judgment is due not to an outcome-driven asymmetry in epistemic state
ascriptions, but due to an outcome-driven asymmetry in the assessment
of the likelihood of the harmful consequence. In other words, outcome
knowledge triggers a bias not with respect to the assessment of the
agent’s mental states, but—in line with Fischhoff’s original findings of
“creeping determinism”—a bias with respect to the perceived prob-
ability of events. The effect of outcome on moral judgment would
consequently not be mediated by foresight or recklessness attributions,
but rather by negligence attributions, i.e. judgments about whether an
agent should have been aware of a certain risk (a risk being the prob-
ability of a harmful event), see Fig. 3.
In retrospect, the probability of an accident is perceived as higher in
the unlucky case, the unlucky agent is judged as more negligent than
the lucky one, and the unlucky agent—who is perceived as more neg-
ligent—is judged more harshly. Given that it is commonplace to ex-
pound the Puzzle of Resultant Moral Luck with reference to acts of
negligence, it is surprising that the Probabilistic Single Pathway Model
has never been tested.
The analysis in terms of an epistemic or probabilistic bias is not the
only way to account for the Difference Intuition. Instead, one might
deem outcome a valid source of information for the ascription of
epistemic states to the agent (for suggestions along these lines, see
Heider, 1958; Richards, 1986; Thomson, 1993; Rosebury, 1995; for
discussion see Nichols et al., 2014 and Young, Nichols, & Saxe, 2010).
John’s belief that no accident will occur, if well justified, can be taken
to constitute knowledge. Sally’s belief, by contrast, simply cannot
amount to knowledge: Knowledge is a factive mental state and Sally’s
belief that no accident will occur turns out false.
The probabilistic model might also be given a rationalist spin: The
fact that an accident did indeed occur in Sally’s case might be taken as
evidence that the probability of such an accident was high, whereas the
fact that it did not occur might constitute good grounds to infer that the
probability of an accident was low. Consequently, one might be dis-
posed to hold that Sally should have been aware of a substantial risk of
an accident (and hence acted negligently). It is, however, not the case
that John should have been aware of a substantial risk of an accident,
for the simple reason that the risk is not deemed substantial in the first
place.
Considerations of this sort engender a debate as to the appropriate
type of probability for moral assessment: ex ante probability, according
to which what is probable or not is assessed at the time of action, or ex
post probability, where the probability of the event in question takes
outcome information into consideration. Note that both in criminal law
and ethics (except for hard-nosed consequentialist views) the definition
of negligence draws on an ex ante concept of probability. Although it
might thus be in some sense rational to ascribe different probabilities of
an accident across the two cases, this does not yet make it rational to
judge one agent more negligent than the other.
In contrast to the Single Pathway accounts just introduced, the Dual
Pathway Model proposes that moral judgment is sensitive to both mental
states (first pathway) and causal factors and outcomes (second
pathway). Cushman (2008, 2013) proposes a view of this sort, though it
is complicated by a further, orthogonal, distinction according to which
blame and punishment judgments are sensitive to causal factors and
mental states, whereas wrongness and responsibility judgments are
sensitive mainly to mental states. Cushman calls his influential account
the Dual Process Model of moral judgment (Fig. 4) since different types
of judgment are processed with respect to different factors. It should not
be confused with Dual Pathway Models of moral luck.
According to the Epistemic Dual Pathway Model (discussed, though
not endorsed, by Nichols et al. (2014)), the impact of outcome is par-
tially mediated by epistemic state ascriptions: Outcome directly influ-
ences moral judgment, but also affects the degree to which knowledge
and belief are attributed, influencing moral judgment indirectly
(Fig. 5). Just as for Single Pathway Models, an analogue drawing on
probability ascriptions rather than epistemic state attributions can be
envisioned (Fig. 6). Both types of Dual Pathway Models again allow for
a rationalist or a bias-driven interpretation.
2
Fig. 1. The Simple View.
Fig. 2. The Epistemic Single Pathway Model.
Fig. 3. The Probabilistic Single Pathway Model.
Fig. 4. Cushman’s Dual Process Model of moral judgment.
Fig. 5. The Epistemic Dual Pathway Model.
2
Yet other models exist. Young et al. (2010), for instance, propose an account
according to which the asymmetry in blame ratings across the lucky and un-
lucky cases is driven by epistemic factors which (though dependent on truth-
conferring features of outcome) are independent of the severity of outcome. The
unlucky agent, they argue, frequently has a false outcome-related belief (e.g.
that the bad consequence will not occur) whereas the lucky agent does not hold
a false belief, and the more severe moral evaluation of unlucky agents is
principally due to their false outcome-related beliefs. We will set this model
aside for two reasons: First, we think that the model only applies to the subclass
of moral luck cases in which a relevant outcome-related target belief can be
clearly singled out. However, moral luck is believed to arise principally in si-
tuations of negligence, situations where the agent should have had, but did not
actually entertain a belief about a substantial risk regarding a harmful outcome
(and hence a fortiori did not have a true or false outcome-related belief).
Second, in two experiments (Studies 3 and 4a) where we explicitly tested a wide
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
2

1.2. Empirical data
Several empirical studies address moral luck directly (Cushman,
Dreber, Wang, & Costa, 2009; Gino, Moore, & Bazerman, 2009; Lench,
Domsky, Smallman, & Darbor, 2015; Young et al., 2010) or employ
moral luck scenarios to explore related topics in moral psychology
(Cushman, 2008; Gino, Shu, & Bazerman, 2010; Schwitzgebel &
Cushman, 2012). Across a broad range of scenarios, the unlucky agent
is robustly judged more harshly than the lucky agent (for a recent re-
view of the empirical literature on moral luck, see Martin & Cushman,
2015). However, the literature suffers from a variety of shortcomings.
First, most experiments use a between-subjects design, in which parti-
cipants are assigned to either the good luck or the bad luck condition.
While studies of this sort testify to an effect of outcome on moral
judgment, they do not explore whether reflective judgments also
manifest the Difference Intuition. Designs allowing participants to
compare the two conditions are more likely to elicit such reflective
judgments (e.g., Hsee, 1996; Baron, 2000; Baron & Ritov, 2004). In fact,
in the few within-subjects studies (Spranca, Minsk, & Baron, 1991;
Schwitzgebel & Cushman, 2012) about two thirds of the subjects judge
the lucky and the unlucky agent identically in moral respects.
Second, most studies examine judgments about concrete moral luck
situations, though less is known about people’s abstract views about
moral luck. Some evidence suggests that most people do not, in fact,
endorse the Difference Intuition when explicitly asked about it (90%
according to Nichols, 2009; 80% according to Schwitzgebel &
Cushman, 2012; 85% and 63% in Studies 2 and 3 of Lench et al., 2015).
The apparent conflict between explicitly endorsed abstract principles
and judgments about concrete cases requires further exploration.
Third, there is a staggering variety in types of judgments that are
invoked across the different experiments. Participants are asked to rate
whether the agents’ behavior is “wrong,” “fair,” “unethical,” “blame-
worthy,” “bad,” “should be a crime,” or “deserves punishment.” This
variety complicates the comparative assessment of results across stu-
dies, in particular since the different types of judgment are frequently
conflated. Schwitzgebel and Cushman (2012), for instance, ask about
blame in their study invoking concrete cases, yet explore punishment
judgments when they ask participants about their explicitly endorsed
abstract principles. Lench et al.’s (2015) questionnaire about partici-
pants’ explicit commitment to the abstract principle of moral luck in-
vokes at least four different types of moral judgment, which leaves it
obscure what the dependent variable actually is. The Puzzle of Re-
sultant Moral Luck and the Difference Intuition are often formulated in
terms of blame and wrongness, and not just any blame-related judg-
ment will do (for discussion, see Enoch & Marmor, 2007; Levy, 2005;
but see, e.g., Kumar, 2018).
Fourth, the existing empirical literature is inconclusive as to which
of the psychological models of moral luck, if any, is correct. Some
studies suggest that outcome has a strong effect on the relevant judg-
ments (Gino et al., 2009); others suggest it does not (Lench et al., 2015),
although design and variables differed across experiments. In support of
the Dual Pathway Model, Cushman (2008) reports that outcome
strongly affects blame and punishment, while it affects wrongness and
permissibility judgments only marginally. The findings of Gino et al.
(2009), however, suggest that judgments of wrongness or “ethicality”
are just as affected by outcome as blame and punishment. Young et al.
(2010) report that moral evaluation is partially mediated by the per-
ceived justification of the agent’s belief, though outcome also seems to
have a partial direct effect on blame. Entertaining counterfactuals—e.g.
thinking about a negative outcome in a neutral outcome scenario—is
sometimes found to affect moral judgment (Lench et al., 2015).
Finally, there has been little work exploring whether intuitions re-
garding moral luck correlate with psychological, ideological and de-
mographic factors. An exception is Gino et al. (2009), who report that
participants primed to engage in slow, analytic cognition manifest the
Difference Intuition to a lesser degree than those primed to engage in
intuitive cognition.
In the following, we will report a systematic series of studies that
explore moral luck intuitions for two distinct scenarios and address
these shortcomings. For each scenario, we ran a between-subjects ex-
periment, a within-subjects experiment, and a contrastive design ex-
periment (i.e., an experiment in which participants are presented with a
single question comparing the morally lucky and unlucky agent instead
of two questions across and within participants; see Section 3). We also
collected data on people’s abstract views about moral luck in order to
compare them with their concrete, case-based judgments. Following
Cushman (2008, 2013), each experiment focused on four distinct de-
pendent variables: wrongness, blame, permissibility, and punishment.
Furthermore, several studies explicitly targeted the question which, if
any, of the psychological models surveyed above best explains the effect
of outcome on moral judgment. We also employed a variety of ques-
tionnaires to investigate whether moral luck intuitions correlate with
particular psychological, ideological, or demographic characteristics,
namely the Rational-Experiential Inventory (Epstein, Pacini, Denes-Raj,
& Heier, 1996; Pacini & Epstein, 1999), the Belief in a Just World Scale
(Rubin & Peplau, 1975), the 12-item Social and Economic Con-
servativism Scale (Everett, 2013), and the 20-item Moral Foundations
Questionnaire (Graham, Haidt, & Nosek, 2009; Graham et al., 2011).
The results are reported in Sections 3.2.2 and 7.3 of the Appendix.
The article will proceeds as follows: Studies 1a, 1b, and 2 explore
the robustness of the effect of outcome on moral judgment by em-
ploying a between-subjects design, a within-subjects design, and a
contrastive design respectively. Study 3 investigates whether the effect
of outcome on moral judgment is mediated by epistemic states and
probability ascriptions. Studies 4a, 4b, 4c, and 4d, report replications of
the findings. Study 5 focuses on participants’ abstract views about
moral luck. The general discussion examines the significance of these
findings for moral psychology and moral philosophy and points out
future avenues of research on the topic.
2. Study 1a (between-subjects design) and Study 1b (within-
subjects design)
2.1. Methods and materials
For Study 1a, we used a mixed-factorial design (between-subjects
factor: Moral Luck—lucky vs. unlucky; within-subjects factor:
Judgment Type—wrongness vs. permissibility vs. blame vs. punish-
ment). Having passed an attention check, participants were randomly
presented with one of the following two vignettes:
Bad Luck
Anna is at home, giving her 2-year-old son a bath. She fills the bath,
while her son stands near the tub. The phone rings in the next room.
Anna tells her son to stand near the tub while she answers the
phone. Anna believes her son will stand near the tub for a few
Fig. 6. The Probabilistic Dual Pathway Model.
(footnote continued)
range of epistemic state ascriptions (knowledge, belief, and awareness of a
substantial risk), we found no evidence for a significant difference across the
lucky and unlucky cases.
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
3

minutes and wait for her to return. Anna leaves the room for 5 min.
When Anna returns, her son is in the tub, dead, face down in the
water.
Good Luck
Beth is at home, giving her 2-year-old son a bath. She fills the bath,
while her son stands near the tub. The phone rings in the next room.
Beth tells her son to stand near the tub while she answers the phone.
Beth believes her son will stand near the tub for a few minutes and
wait for her to return. Beth leaves the room for 5 min. When Beth
returns, her son is still standing near the tub, where she left him. The
boy then enjoys his bath.
Participants were asked to answer four moral judgment questions.
The wrongness question read: “How wrong was Anna [Beth] to leave
her son alone in the above scenario?”. Responses were recorded on a 7-
point Likert scale anchored at 1 with “not wrong at all” and at 7 with
“extremely wrong.” The three other questions (with adapted labels of
the scale) read:
(1) To what extent was Anna [Beth] blameworthy for leaving her son
alone in the above scenario? (not at all blameworthy/extremely
blameworthy)
(2) To what extent was it permissible for Anna [Beth] to leave her son
alone in the above scenario? (entirely permissible/entirely im-
permissible)
(3) How much punishment does Anna [Beth] deserve for leaving her
son alone in the above scenario? (no punishment at all/very severe
punishment)
These four questions were presented in randomized order on sepa-
rate screens. They were followed by a comprehension check and a de-
mographic questionnaire.
Study 1b was identical in all respects, except that participants were
presented with both vignettes (bad luck always first), and then had to
judge both Anna and Beth’s actions in terms of all four measures
(wrongness, blame, permissibility, and punishment), presented in ran-
domized order. The wrongness question, for instance, read “How wrong
were Anna and Beth to leave their sons alone in each of the above
scenarios?”. Participants had to rate Anna’s action, and thereafter
Beth’s action, on separate Likert scales ranging from 1 (“not wrong at
all”) to 7 (“extremely wrong”).
2.2. Participants
For Study 1a, 241 participants were recruited online via Amazon
Mechanical Turk. The IP address location was restricted to the USA.
Participants who failed the attention check or the comprehension
question were excluded, leaving a sample of 196 participants (male:
40.8%; mean age: 30.1, range: 18–73).
For Study 1b, 120 participants were recruited online via Amazon
Mechanical Turk. The IP address location was again restricted to the
USA. Participants who failed the attention check or the comprehension
question were excluded, leaving a sample of 95 participants (male:
48.4%; mean age: 37.2, range: 20–69). Analyses with the complete data
sets are reported in Sections 1 and 2 of the Appendix (excluding in-
attentive subjects did not make an important difference).
2.3. Results for Study 1a
A mixed-design ANOVA determined that, aggregating across the
four dependent variables, participants judged the action of the morally
unlucky agent to be worse than the action of the morally lucky agent (F
(1,194) = 2831.62, p< .001, η
2
= .94, Fig. 7). Furthermore, it re-
vealed that, aggregating across the two moral luck conditions, the dif-
ference in judgment type was significant (F(3,582) = 162.77,
p< .001, η
2
= .46). Bonferroni-corrected post hoc tests showed that
the only significant differences were (1) between answers to the pun-
ishment question and answers to the three other questions and (2)
between answers to the permissibility question and the wrongness
question (uncorrected ps < .001). We did not find any evidence that
participants responded differently to the other pairs of questions (blame
and wrongness: uncorrected p= .50; wrongness and permissibility:
uncorrected p= .01; blame and permissibility: uncorrected p= .008).
Importantly, the two main effects were qualified by an interaction (F
(3,582) = 9.89, p< .001, η
2
= .04). To analyze this interaction, we
compared the difference between the morally lucky and the morally
unlucky conditions for each of the four questions. We then computed
the effect size (Cohen’s d) for each effect (Morris and DeShon, 2002) as
well as the confidence intervals for each of these four effect sizes
(Wuensch, 2012). The results are summarized in Table 1.
2.4. Results for Study 1b
To examine the influence of moral luck on moral judgments, we
analyzed participants’ answers by means of a two-way (Moral Luck:
lucky vs. unlucky; Judgment Type: wrongness vs. permissibility vs.
blame vs. punishment) repeated-measures ANOVA. We found that ag-
gregating across the four judgment types, participants’ mean responses
for the lucky condition differed significantly from the unlucky condition
(F(1,94) = 25.00, p< .001, η
2
= .21; Fig. 7). We also found that, ag-
gregating across the two moral luck conditions, participants’ mean
answers to the wrongness, permissibility, blame, and punishment
questions differed significantly (F(3,282) = 105.55, p< .001,
η
2
= .34). Bonferroni-corrected post hoc tests revealed that the only
significant differences were between answers to the punishment ques-
tion and answers to the three other questions (uncorrected ps < .001;
all the other ps > .5). That is, participants answered the punishment
question differently from the wrongness, permissibility, and blame
questions. We did not find any evidence indicating that participants
responded differently to the wrongness, permissibility, and blame
questions. Importantly, the two main effects we observed were quali-
fied by a two-way interaction (F(3,282) = 16.18, p< .001, η
2
= .15).
To analyze this interaction, we compared the difference between the
morally lucky and the morally unlucky conditions for each of the four
questions. We then computed the effect size for each effect as well as
the confidence intervals for each of these four effect sizes, using the
2016 ESCI software. The results are summarized in Table 1.
To investigate whether the observed aggregate influence of moral
luck on wrongness, blame, and punishment judgments is widespread
among participants or is rather due to a small minority, we compared
the proportion of participants manifesting and failing to manifest the
Difference Intuition. Participants who judged the lucky and the unlucky
agents identically were counted as manifesting “no Difference
Intuition” (No DI for short in Fig. 8). The vast majority of participants
did not share the Difference Intuition for wrongness (89%), blame
(87%), and permissibility (96%), though the proportion was sub-
stantially smaller for punishment (64%). All proportions differed sig-
nificantly from chance (binomial test p< .001, two-tailed). Punish-
ment judgments differed significantly from wrongness, blame, and
permissibility judgments (binomial tests, test proportion = .64,
p< .001). Permissibility judgments differed significantly not only
from punishment judgments, but also from wrongness judgments (bi-
nomial test, test proportion = .96, p= .010) and blame judgments
(binomial test, test proportion = .96, p< .001). Wrongness judgments
did not differ significantly from blame judgments (binomial test, test
proportion = .89, p= .702, all two-tailed).
2.5. Discussion
Taken together, Studies 1a and 1b revealed four main findings. First,
at the aggregate level an effect of outcome on moral judgment was
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
4

found in both between- and within-subjects designs, although the effect
sizes in the within-subjects design were smaller than in the between-
subjects design. Second, for a very large majority of people the outcome
of the agent’s actions did not influence permissibility, blame, and
wrongness judgments when they read about both a morally lucky and
unlucky situation. This result is in line with previous findings by
Spranca, Minsk, and Barron (1991) and Schwitzgebel and Cushman
(2012) reported in the introduction. It suggests that outcome does not
influence most people’s reflective (i.e. within-subjects design) judg-
ments about concrete cases of moral luck. Third, the influence of moral
luck varied across types of moral judgment: We found a medium to
large effect of outcome on punishment judgments, but only a small to
medium effect of outcome on wrongness and blame judgments. Ad-
mittedly, the confidence intervals for the effect sizes overlap in Study
1a, which prevents us from concluding that the effect size of punish-
ment is significantly larger than the effect sizes for wrongness and
blame. But the lower bound of the confidence intervals for the effect
size of punishment is larger than the higher bound of the effect size of
wrongness in Study 1b. We will come back to this question in Study 4.
We failed to find any significant effect of outcome on permissibility
judgments. Fourth, in light of the existing literature, the way the four
different types of moral judgment related to one another is unexpected.
Cushman (2008, 2013) reports that punishment patterns with blame,
but we observed that blame patterns with wrongness and differs from
punishment.
Across the first two studies, we have varied the factor of explicit
comparability. Study 1a used a between-subjects design (low compar-
ability), Study 1b a within-subjects design (medium comparability). In
Study 2 we explore participants’ responses to a question that explicitly
contrasts the two agents (high comparability), i.e. a question that asks
whether Anna and Beth should be judged identically. We will refer to
this type of design as a “contrastive design.”
Fig. 7. Mean wrongness, blame, permissibility, and punishment judgments for the between-subjects and within-subjects designs; error bars denote 95% confidence
intervals.
Table 1
Effect of outcome on wrongness, blame, permissibility, and punishment judgments in a between-subjects and a within-subjects design; 95% confidence intervals are
given for the effect sizes.
Between-subjects Design Within-subjects Design
Variable t(194) pCohen's d95% CI t(94) pCohen's d95% CI
Wrongness 3.08 .002 .44 [.16;.72] 2.80 .006 .16 [.004;.27]
Blame 2.09 .04 .39 [.17;.58] 3.28 .001 .24 [.09;.38]
Permissibility 1.85 .07 .26 [−.02;.55] 1.55 .12 .06 [−.02;.14]
Punishment 5.52 < .001 .79 [.50;1.08] 5.84 < .001 .47 [.30;.64]
Fig. 8. Proportions of participants who judged the lucky and unlucky agents
identically (No DI) with respect to wrongness, blame, permissibility, and pun-
ishment; errors bars denote 95% confidence intervals; Wilson method, see
Brown, Cai, & DasGupta, 2001.
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
5

3. Study 2: Contrastive design
Study 2 employed a contrastive design to further increase partici-
pants’ reflective comparison of the two types of cases.
3.1. Methods and materials
Following an attention check, participants were presented with both
vignettes of Study 1b (the order was fixed: bad luck first). The for-
mulation of the wrongness, blame, permissibility, and punishment
questions differed from Study 1b. In Study 2, participants were pre-
sented with statements explicitly comparing Anna and Beth with re-
spect to each of the four moral measures:
(1) It was just as wrong for Anna to leave her son as it was for Beth.
(2) Anna is just as blameworthy as Beth for leaving her son.
(3) It was just as impermissible for Anna to leave her son as it was for
Beth.
(4) Anna deserves just as much punishment as Beth for leaving her son.
Each statement (presented in randomized order) was followed by a
7-point Likert scale anchored at 1 with “completely agree” and at 7 with
“completely disagree.”
3.2. Participants
120 participants were recruited online via Amazon Mechanical
Turk. We restricted the IP address location to the USA. Participants who
failed the attention check or the comprehension question were re-
moved, leaving a sample of 89 participants (male: 48.3%; mean age:
36.9, SD = 12.1; range: 18–65). Analyses with the complete data sets
are reported in the Appendix.
3.3. Results
A one-way repeated-measures ANOVA determined that participants’
mean answers to the wrongness, blame, permissibility, and punishment
questions differed significantly (F(3,264) = 39.51, p< .001, η
2
= .24;
Fig. 9). Post hoc tests using the Bonferroni correction revealed that the
only significant comparisons were those between answers to the pun-
ishment question and answers to the three other questions (uncorrected
ps < .001; wrongness and permissibility, uncorrected p= .38; wrong-
ness and blame, uncorrected p= .05; blame and permissibility,
uncorrected p= .34). Participants disagreed more with the claim that
the morally lucky and morally unlucky agents deserved the same
punishment than they disagreed with the claims that their actions were
equally wrong, equally impermissible, and equally blameworthy, and
we did not find any evidence that they responded differently to these
three latter claims. The means for all four measures were significantly
below the neutral midpoint 4 and significantly above the endpoint 1
(complete agreement), see Section 3.1 of the Appendix.
Following Lench et al. (2015), we calculated the percentage of
participants who agreed with the claim that the two agents should be
judged identically. We also aggregated the number of participants who
responded with “completely agree” (Likert scale endpoint 1). The re-
sults, presented in Table 2, were consistent with the findings from Study
1b: For wrongness, blame, and permissibility, the large majority of
participants (over 80%) agreed that the two agents should be judged
identically (Likert scale < 4). Over half of the participants chose the
endpoint of the Likert scale. As regards punishment, by contrast, only
about half of the participants agreed that the two agents deserve the
same punishment, and less than a third completely agreed with an as-
sessment of this sort.
3.4. Discussion
Study 2 replicates and extends the findings of Study 1b. People ra-
ther uniformly hold that the two agents should be judged the same in
terms of wrongness, blame, and permissibility. Mean agreement levels
differ significantly from the scale midpoint; over 80% of the responses
are on the “agree” spectrum of the Likert scale and the majority
“completely agrees” with an identical assessment in terms of wrongness
(62%), blame (60%), and permissibility (57%). As in Studies 1a and 1b,
outcome had a much larger impact on punishment judgments than on
wrongness, blame, and permissibility judgments, and only a minority of
participants (31%) completely agreed that the actions of the morally
lucky and unlucky agents deserve equal punishment. Consistent with
Studies 1a and 1b, blame judgments patterned with wrongness judg-
ments (and in this experiment with permissibility, too), but differed
from punishment judgments.
According to Studies 1b and 2, which employed a within-subjects
and contrastive design respectively, most people do not share the
Difference Intuition for wrongness, blame, and permissibility. This
raises the question of whether there is a Puzzle of Resultant Moral Luck
in the first place, a question we address in the general discussion. While
the majority of previous moral luck experiments have consistently
found that outcome influences moral judgment, this is presumably due
to the fact that they employ a between-subjects design. Our between-
subjects results from Study 1a replicate these between-subjects findings,
which constitute an important phenomenon in its own right: In or-
dinary life, moral and legal judgment resembles the between-subjects
design, as we rarely compare actual situations to counterfactual ones
with alternative outcomes. In Study 3 we turn to the question of what
explains the effect of outcome on moral judgment in non-comparative
contexts.
4. Study 3: Mediators of the impact of outcome on moral judgment
What explains the different moral assessment of the lucky and
Fig. 9. Mean disagreement with the claim that the two agents should be judged
identically with respect to wrongness, blame, permissibility, and punishment;
error bars denote 95% confidence intervals.
Table 2
Proportions of participants who agreed (Likert scale < 4) or completely agreed
(endpoint 1) that the two agents should be judged identically with respect to
wrongness, blame, permissibility, and punishment.
Measure Wrongness Blame Permissibility Punishment
Likert scale < 4 87% 87% 82% 52%
Endpoint 1 62% 60% 57% 32%
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
6

unlucky agents in non-comparative contexts? In the introduction we
have surveyed several accounts. According to the Simple View, the
difference in moral judgment arises due to a brute effect of outcome on
moral judgment. Alternatively, the effect of outcome on moral judg-
ment might be mediated either by ascriptions of epistemic states or else
perceptions of the probability of a harmful outcome across cases. The
mediation could be complete (Single Pathway Model) or partial (Dual
Pathway Model), in which case parts of the outcome effect on judgment
would be direct, and parts of it mediated by epistemic state ascription
or probability assessment.
4.1. Methods and materials
In a 2 (Moral Luck—lucky vs. unlucky) × 2 (Judgment
Type—wrongness vs. punishment) between-subjects design study em-
ploying the two vignettes from Study 1a, participants were randomly
assigned to one of the four conditions. They responded either to the
wrongness question or the punishment question on a 7-point Likert
scale. Thereafter, participants were presented with six questions re-
garding the agent’s epistemic states (belief and knowledge), negligence,
and the probability of an accident (the order was randomized). The
questions (except for the last one) asked to what extent participants
agreed or disagreed with the following statements on a 7-point Likert
scale anchored at 1 with “completely disagree” and at 7 with “com-
pletely agree”:
(1) Anna knew an accident would happen.
(2) Anna knew there was a high probability of an accident.
(3) Anna believed an accident would happen.
(4) Anna believed that there was a high probability of an accident.
(5) Anna should have believed that there was a high probability of an
accident.
(6) On a scale of 0–100%, how high do you think was the probability of
an accident? N.B. 0% means the accident was completely improb-
able, 100% means it was certain to occur.
Before the main task, all participants completed an attention check.
Following the main task, they were presented with a comprehension
check.
4.2. Participants
650 participants were recruited online via Amazon Mechanical
Turk. We restricted their IP address locations to the USA. Participants
failing the attention or comprehension checks were excluded, leaving a
sample of 566 participants (male: 51%, mean age: 37.3, SD = 12.4,
range: 18–84).
4.3. Results
A 2 (Outcome: lucky vs. unlucky) × 2 (Judgment Type: wrongness
vs. punishment) ANOVA revealed a significant main effect for judgment
type (F(1,562) = 134.23, p< .001, η
2
= .193), a significant main ef-
fect for outcome (F(1,562) = 59.20, p< .001, η
2
= .095), and a sig-
nificant interaction (F(1,562) = 4.50, p= .034, η
2
= .008). Wrongness
judgments in the unlucky condition (M= 6.00, SD = 1.52) significantly
exceeded those in the lucky condition (M= 5.17, SD = 1.66), t
(280) = 4.341, p< .001, d= .52. Punishment judgments in the un-
lucky condition (M= 4.60, SD = 1.97) also significantly exceeded
those in the lucky condition (M= 3.15, SD = 1.83), t(282) = 6.407,
p< .001, d= .76 (Fig. 10).
Turning to the ascription of epistemic states and probability judg-
ments (Figs. 10 and 11), knew p ascriptions did not differ significantly
across the unlucky (M= 2.22, SD = 1.65) and lucky conditions
(M= 2.25, SD = 1.68), t(564) = −.216, p= .83, d= −.03; knew
probably p ascriptions did not differ significantly across the unlucky
(M= 3.38, SD = 2.00) and lucky conditions (M= 3.14, SD = 2.05), t
(564) = 1.423, p= .16, d= .09; believed p ascriptions did not differ
significantly across the unlucky (M= 2.06, SD = 1.50) and lucky con-
ditions (M= 2.00, SD = 1.47), t(564) = .482, p= .63, d= .04; believed
probably p ascriptions did not differ significantly across the unlucky
(M= 2.61, SD = 1.80) and lucky conditions (M= 2.50; SD = 1.81), t
(564) = .701, p= .48, d= .06. By contrast, should have believed that
probably p ascriptions (i.e., judgments of negligence) differed sig-
nificantly across the unlucky (M= 5.72, SD = 1.70) and lucky condi-
tions (M= 5.18, SD = 1.99), t(564) = 3.506, p< .001, d= .30, and
probability judgments differed significantly across the unlucky
(M= 58.84, SD = 28.41) and lucky conditions (M= 47.40,
SD = 27.74), t(564) = 4.839, p< .001, d= .41.
Several multiple mediation analyses were conducted using Preacher
and Hayes’s (2008) macro for SPSS with 20,000 bootstrap samples,
testing for all potential mediators simultaneously. Negligence attribu-
tion proved a significant mediator between outcome and wrongness
ascription. None of the other proposed mediators were significant, al-
though the a path of probability reached significance (Fig. 12). A
follow-up analysis revealed probability to be a strong mediator between
outcome and negligence (Fig. 13).
For punishment, another mediation analysis with 20,000 bootstrap
samples was conducted. Both negligence attribution and probability
judgments proved significant mediators between outcome and punish-
ment (Fig. 14). Probability again proved a strong (close-to-complete)
mediator between outcome and negligence (Fig. 15).
4.4. Discussion
Section 1 introduced several competing explanations of the influ-
ence of outcome on moral judgments. One type of model assumes a
single pathway of the impact of outcome on moral judgment, which can
either be direct (the Simple View), mediated by epistemic state as-
criptions (the Epistemic Single Pathway Model) or by probability as-
sessments (the Probabilistic Single Pathway Model). According to the
Dual Pathway Model, some of the impact of outcome on judgment is
direct, and some of it is mediated (either by epistemic state ascriptions
or probability assessments). Study 3 shows that the impact of outcome
on moral judgment is not epistemically mediated, which undermines
the corresponding Single and Dual Pathway Models. Participants as-
cribed the same degree of outcome knowledge and risk awareness to the
unlucky and lucky agent. Hence, the effect of outcome on moral judg-
ment does not result from different attributions of the mens rea of
foreknowledge (knowledge that a harmful outcome will occur) or
recklessness (awareness of the risk of a harmful outcome) across con-
ditions. (For the definitions of the mens reas of foreknowledge and
recklessness, see the US Model Penal Code (1985), 2.02 sections b and
c.)
The data makes a strong case in favor of the probabilistic models.
On average, when people are presented with a single situation (either
the lucky or the unlucky situation), they agree more with the statement
that unlucky Anna should have been aware of a high probability of an
accident than with the statement that lucky Beth should have been
aware of such a risk. That is, people judge the unlucky agent as more
negligent than the lucky agent (see the US Model Penal Code, 2.02 (d)
for the standard definition of negligence). Negligence attributions
proved a partial mediator of the impact of outcome on wrongness and
punishment judgments. The impact of outcome on negligence, in turn,
is fully mediated by the diverging assessments of the probability of an
accident. Overall, the findings are consistent with a Probabilistic Dual
Pathway Model, where some of the effect of outcome on wrongness and
punishment is mediated by negligence, and some might be direct
(“might” because we cannot rule out that mediators distinct from the
ones here tested were at play).
A limitation of the studies reported so far is that we used a single
vignette to study moral luck, and one may worry that our results
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
7

depend on the specific formulations of this vignette. Study 4 addresses
this question and replicates our findings.
5. Replications
The goal of the following series of studies was to replicate the
previous results with a different scenario and an alternative within-
subjects design inspired by Experiment 1 of Spranca et al. (1991).
5.1. Study 4a: Between-Subjects replication
5.1.1. Methods and materials
In a mixed-factorial design study (between-subjects factor: Moral
Luck—lucky vs. unlucky; within-subjects factor: Judgment
Type—wrongness vs. permissibility vs. blameworthiness vs.
punishment), participants were randomly presented with one of the
following two vignettes (variations in square brackets):
Bad [Good] Luck
John [Frank] is an artisan toymaker. For his new puppet collection,
he changes the paint supplier. He picks a company from abroad that
has previously had problems with lead in their paint. Lead is poi-
sonous. John [Frank] believes that the paint no longer has the lead
problem, and does not have it tested. Unfortunately, the paint does
contain lead. John’s puppets sell out and three children who play
with them die from lead poisoning. [Fortunately, the paint does not
contain any lead. Frank's puppets sell out and none of the children
playing with them die from lead poisoning.]
Participants were asked to answer four moral questions about the
agents (John or Frank) and their four moral judgment questions. The
wrongness question read: “How wrong was John [Frank] in using the
new paint without testing it for lead?” Responses were recorded on a 7-
point Likert scale anchored at 1 with “not wrong at all” and at 7 with
“extremely wrong.” The three other questions, whose scale labels were
adapted, read:
(1) To what extent is John [Frank] blameworthy for using the new
paint without testing it for lead? (not at all blameworthy/extremely
blameworthy)
(2) To what extent was it permissible for John [Frank] to use the new
paint without testing it for lead? (entirely permissible/entirely
impermissible)
(3) How much punishment does John [Frank] deserve for using the
new paint without testing it for lead? (no punishment at all/very
severe punishment)
The order of the four questions was randomized. The main task was
preceded by an attention check and followed by a comprehension
question. Having completed the task, participants were presented with
questions about the agent’s epistemic states and the likelihood of a bad
outcome. On a 7-point Likert scale anchored at 1 with “completely
disagree” and at 7 with “completely agree” participants had to report
their agreement with statements 1–5, and to report their answer to
question 6 on a scale of 0% (“no probability at all”) to 100% (“certain”).
The items were presented in randomized order:
Fig. 10. Mean ascription of wrongness, punishment, epistemic states, and negligence; error bars denote 95% confidence intervals.
Fig. 11. Mean ascription of probability of the occurrence of a harmful outcome;
error bars denote 95% confidence intervals.
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
8

(1) John [Frank] knew the paint still contained lead.
(2) John [Frank] knew there was a high probability that the paint
would still contain lead.
(3) John [Frank] believed the paint still contained lead.
(4) John [Frank] believed that there was a high probability that the
paint would still contain lead.
(5) John [Frank] should have believed that there was a high prob-
ability that the paint would still contain lead.
(6) On a scale of 0–100%, how high do you think was the probability of
the paint still containing lead? N.B. 0% means no probability at all,
100% means it was certain that the paint would contain lead.
At the end of the study, participants were also asked to complete a
biographical questionnaire.
5.1.2. Participants
332 participants were recruited online via Amazon Mechanical
Turk. We restricted their IP address location to the USA. Participants
who failed the attention check or the comprehension check were re-
moved, leaving a sample of 254 participants (male: 46.9%; mean age:
37.2, range: 19–81).
5.1.3. Results
A mixed design ANOVA determined that, aggregating across the
four judgment types, participants considered the action of the unlucky
agent as worse than the action of the lucky agent (F(1,252) = 24.46,
p< .001, η
2
= .09), and that, aggregating across the two moral luck
conditions, they answered the four moral questions differently (F
(3,756) = 83.11, p< .001, η
2
= .25; Fig. 16). Post hoc tests using the
Bonferroni correction revealed that the significant differences were (1)
between answers to the punishment question and answers to the three
other questions, and (2) between answers to the permissibility question
and answers to the three other questions. We did not find any evidence
that the participants responded differently to the other pairs of ques-
tions.
Importantly, the two main effects were qualified by an interaction
(F(3,756) = 23.38, p< .001, η
2
= .09). To analyze this interaction, we
compared the difference between the lucky and the unlucky conditions
for each of the four questions. We then computed the effect size
(Cohen’s d) for each effect as well as the confidence intervals for each of
these four effect sizes (Wuensch 2012). The results are summarized in
Table 3.
Figs. 16 and 17 present the results for epistemic state ascriptions
and probability assessments; Table 4 reports effect sizes and confidence
intervals for each of the six measures.
A series of multiple mediation analyses was conducted to investigate
whether the relation between outcome (lucky vs. unlucky) and moral
judgment (wrongness, blame, and punishment) was mediated by epis-
temic state ascriptions or assessments of probabilities. Since for per-
missibility there was no significant outcome effect, we did not run a
Fig. 12. Mediation of the relationship between outcome (lucky vs. unlucky) and wrongness judgments by probability, negligence, knowledge, and belief ascriptions.
Fig. 13. Mediation of the relationship between outcome (lucky vs. unlucky) and negligence ascriptions (“should have believed there was a high probability of an
accident”) by probability judgments.
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
9

mediation analysis. The simultaneously tested potential mediators were
that Frank/John (i) knew that p, (ii) knew that probably p, (iii) believed
that p, (iv) believed that probably p, (v) should have believed that probably
p, and (vi) the perceived probability of p, where pstands for “the paint
contained lead.” The mediation analyses were conducted using
Preacher and Hayes’s (2008) SPSS macro with 20,000 bootstrap sam-
ples.
For all three DVs, negligence (“should have believed that probably
p”) and probability were significant mediators, whereas the other pro-
posed mediators proved insignificant. For both wrongness and blame
the unmediated, significant relation between outcome and moral
judgment (the c path) was rendered insignificant once the mediators
were taken into account (the c’ path; Figs. 18 and 19).
Importantly, the relation between outcome and negligence was it-
self mediated by probability. The significant unmediated impact of
outcome on negligence (the c path) was rendered insignificant once
probability was taken into account as a mediator (the c’ path; Fig. 20).
Punishment was once again an outlier. Whereas for wrongness and
blame the c’ path was insignificant, for punishment it remained sig-
nificant even after the mediators had been taken into account (Fig. 21).
The mediation analyses replicate the findings from Study 3:
Whereas we could find no evidence that the relation between outcome
and moral judgment is mediated by epistemic state ascriptions, negli-
gence and probability were found to be significant mediators. The effect
of outcome on wrongness and blame judgments fit the Probabilistic
Single Pathway Model: Once negligence has been controlled for as a
mediator, the impact of outcome on wrongness and blame is no longer
significant. Punishment judgments, by contrast, fit the Probabilistic
Dual Pathway Model: The effect of outcome on punishment is partly
mediated by negligence and probability ascriptions, and partly direct
(at least with respect to the mediators tested, though others might be
possible).
5.2. Study 4b: Within-subjects replication
5.2.1. Methods and materials
To replicate Study 1b, the Toymaker scenario was used in a within-
subjects design identical in all respects to the design of Study 1b.
Participants were presented with both the Bad Luck and Good Luck
vignettes (in that order) and had to assess the action of John and Frank
with respect to all four moral judgment types (wrongness, blame, per-
missibility, and punishment). To take an example, participants were
asked “How wrong were John and Frank in using the new paint without
testing it for lead?” and had to answer separately for John and Frank on
a 7-point scale anchored at 1 with “not wrong at all” and at 7 with
“extremely wrong.” The order of the four questions was again rando-
mized. The main task was preceded by an attention check and followed
by a comprehension question.
Having completed the main task, participants were asked explicitly
about their abstract views regarding moral luck. They read the
Fig. 14. Mediation of the relationship between outcome (lucky vs. unlucky) and punishment judgments by probability, negligence, knowledge, and belief ascriptions.
Fig. 15. Mediation of the relationship between outcome (lucky vs. unlucky) and negligence ascriptions (“should have believed there was a high probability of an
accident”) by probability judgments.
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
10

following prompt and had to respond to four questions concerning
wrongness, blame, permissibility, and punishment on separate screens
in randomized order:
Suppose two people do the exact same thing, with the exact same
frame of mind. Then, due entirely to matters of chance beyond their
control, one of them produces a very bad outcome, but the other
does not. To what extent do you agree or disagree with the following
statements:
(1) The action of the one person is exactly as wrong as that of the other
person.
(2) Both people are equally blameworthy.
(3) The action of the one person is as permissible or impermissible as
that of the other person.
(4) Both people deserve an equal amount of punishment.
Responses were collected on a 7-point Likert scale anchored at 1
with “completely disagree” and at 7 with “completely agree”. Following
the main task, participants completed the 20-item Moral Foundations
Questionnaire (Graham et al., 2009, 2011).
5.2.2. Participants
396 participants were recruited online via Amazon Mechanical
Turk. We restricted their IP address location to the USA. Participants
who failed an attention check or the comprehension question were
removed, leaving a sample of 320 participants (male: 48.4%; mean age:
45.6, range: 20–75).
5.2.3. Results
To examine the influence of moral luck on moral judgments, we
analyzed participants’ answers by means of a two-way (Moral Luck:
lucky vs. unlucky; Judgment Type: wrongness vs. permissibility vs.
blameworthiness vs. punishment) repeated-measures ANOVA (Fig. 21).
We found that aggregating across the four questions, participants’ an-
swers for the unlucky agent (John) differed significantly from those
concerning the lucky agent (Frank), F(1,319) = 78.0077, p< .001,
η
2
= .20. We also found that, aggregating across the two levels of Moral
Luck, participants’ mean answers to the wrongness, permissibility,
blameworthiness, and punishment questions differed significantly
Fig. 16. Mean wrongness, blame, permissibility, and punishment judgments as well as mean epistemic state and negligence ascriptions; error bars denote 95%
confidence intervals.
Table 3
Effect of outcome on wrongness, blame, permissibility, and punishment judg-
ments.
Variable t(252) pCohen's d 95% CI
Wrongness 2.80 .006 .36 [.16;.63]
Blame 2.67 .008 .33 [.13;.63]
Permissibility 1.53 .13 .06 [−.08;.48]
Punishment 8.50 < .001 1.07 [.84;1.39]
Fig. 17. Mean probability judgments; error bars denote 95% confidence in-
tervals.
Table 4
Effect of outcome on ascriptions of epistemic states, negligence, and prob-
ability.
Variable t(252) pCohen's d 95% CI
Knew 1.67 .10 .21 [−.08;.48]
Knew probably 1.15 .25 .15 [−.17;.52]
Believed .64 .53 .08 [−.20;.36]
Believed probably 1.70 .09 .05 [−.25;.39]
Should have believed 2.11 .04 .36 [.01;.60]
Probability 6.13 < .001 .77 [.51;1.02]
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
11

(F(3,957) = 66.14, p< .001, η
2
= .17). Post hoc tests using the Bon-
ferroni correction revealed that the significant differences were (1)
between answers to the punishment question and answers to the three
other questions, and (2) between answers to the wrongness question
and answers to the three other questions (Fig. 22). We did not find any
evidence that participants responded differently to the last pair of
questions.
Importantly, the two main effects we observed were qualified by a
two-way interaction (F(3,957) = 60.61, p< .001, η
2
= .16). To ana-
lyze this interaction, we compared the difference between the morally
lucky and the morally unlucky conditions for each of the four questions.
Probably because of rounding issues, we were unable to compute 95%
confidence intervals for these effect sizes using 2016 ESCI. The results
are summarized in Table 5.
To investigate whether the observed aggregate influence of moral
luck on wrongness, blame, permissibility, and punishment judgments is
widespread among participants or due to a small minority, we com-
pared the proportion of participants who manifested a Difference
Intuition with those who did not. Participants who judged the lucky and
the unlucky agents identically were counted as No DI (Fig. 23). As in
Study 1b, the vast majority of participants did not share the Difference
Intuition for wrongness (89%), blame (83%), and permissibility (90%);
all distributions differed significantly from chance (binomial test,
p< .001, two-tailed). Consistent with the findings from Study 1b, the
Fig. 18. Mediation of the relationship between outcome (lucky vs. unlucky) and wrongness judgments by probability, negligence, knowledge, and belief ascriptions.
Fig. 19. Mediation of the relationship between outcome (lucky vs. unlucky) and blame judgments by probability, negligence, knowledge, and belief ascriptions.
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
12

ratio was considerably smaller for punishment (55%), which did not
differ significantly from chance (binomial test, p= .103, two-tailed),
and which differed significantly from wrongness, blame, and permis-
sibility (binomial tests, test proportion = .55, p< .001, two-tailed).
Blame also differed significantly from wrongness (binomial test, test
proportion = .83, p= .004, two-tailed) and permissibility (binomial
test, test proportion = .83, p< .001, two-tailed).
Next, we turned to people’s abstract views about moral luck. A re-
peated measures ANOVA with a Greenhouse-Geisser correction re-
vealed a significant difference across types of moral judgments (F
(2.543, 752.80) = 29.58, p< .001, η
2
= .091; Fig. 24). The order of
display was not significant (F(23, 296) = 1.26, p= .193, η
2
= .089). In
Bonferroni corrected comparisons, all pairs differed significantly from
one another except wrongness and blame judgments. The means for all
four measures differed significantly from the midpoint 4 of the Likert
scale, as well as from the endpoint 7 (see Appendix section 5.1).
To investigate the fit of judgments about moral luck in the concrete
and abstract conditions, we tested for correlations between, on the one
hand, the difference in the judgments about the unlucky and unlucky
agent and, on the other, responses in the abstract condition for all four
measures. For all but permissibility, abstract rejection of moral luck
correlates with a less pronounced difference in the assessment of the
unlucky agent in contrast to the lucky one (for N = 320, wrongness:
r = −.38, p< .001; blame: r = −.28, p< .001; permissibility:
r = −.10, p= .090; punishment: r = −.54, p< .001, all two-tailed).
Fig. 20. Mediation of the relationship between outcome (lucky vs. unlucky) and negligence ascriptions (“should have believed there was a high probability of an
accident”) by probability judgments.
Fig. 21. Mediation of the relationship between outcome (lucky vs. unlucky) and punishment by probability, negligence, knowledge, and belief ascriptions.
Fig. 22. Mean wrongness, blame, permissibility, and punishment judgments in
the within-subjects design for the Toymaker scenario; error bars denote 95%
confidence intervals.
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
13

The results suggest that in concrete cases, people judge largely in har-
mony with their abstract views about moral luck.
Following Lench et al. (2015), we aggregated the number of parti-
cipants who agreed that the two agents should be judged identically if
their actions differed only with respect to outcome-related factors be-
yond their control (answers > 4) and those who completely agreed
with this claim (endpoint 7). The results are summarized in Table 6.
To further explore the fit between concrete and abstract judgments
about moral luck, we aggregated the number of participants who
judged the two agents identically in the two concrete cases and who,
when asked explicitly about their abstract principles, agreed that the
two agents should be judged the same, thus uniformly disavowing the
Difference Intuition. We also aggregated the number of participants
who judged the two agents differently in the two concrete cases, and
who, when asked explicitly about their abstract principles, disagreed
with the claim that the two agents should be judged the same, thus
uniformly endorsing the Difference Intuition. The sum of the two pro-
portions constitutes the overall ratio of participants whose judgments
about concrete cases fit their abstract principles, i.e. who judged uni-
formly (Fig. 25).
For the vast majority of participants judgments about wrongness,
blame, and permissibility were in harmony with their abstract princi-
ples, whereas just about half of punishment judgments were uniform.
Perhaps even more importantly, for wrongness, blame, and permissi-
bility, over three quarters of the participants uniformly disavow the
Difference Intuition.
5.3. Study 4c: Alternative within-subjects design
Given the potentially controversial nature of our findings—an ab-
sence of a Difference Intuition regarding wrongness, blame, and per-
missibility for the vast majority of people—we ran a second replication
with a different design. Following Experiment 1 of Spranca, Minsk, and
Barron (1991), 140 participants were given a single vignette (a varia-
tion of the Toymaker scenario) with the instruction to imagine two
distinct endings: one in which some children die of lead poisoning,
another in which no harm occurs.
3
Subsequently they were asked to
judge wrongness, blame, responsibility, and punishment for the first
and the second ending. This design has the advantage that a single
agent is used, which makes it perfectly transparent to the reader that
the general situation of the protagonist and their mental states are held
fixed. As in Study 1b and 4b, participants were also asked about their
abstract views regarding moral luck. Methods, materials, and detailed
results are reported in the Appendix (Section 6).
The findings were consistent with the results of Studies 1b and 4b.
First, at the aggregate level, an effect of outcome on moral judgment
was found for all four measures. Second, for the vast majority of people,
the outcome of the agent’s actions did not influence permissibility,
blame, and wrongness judgments when they read about both a morally
lucky and unlucky situation or when they report their abstract views on
the matter. Third, the influence of moral luck varies across types of
moral judgment: We found a large effect of outcome on punishment
judgments, but only medium effects of outcome on wrongness, blame,
and permissibility judgments. Fourth, wrongness, blame, and permis-
sibility form one cluster of moral judgments, while punishment alone
forms a distinct cluster. Finally, abstract views were by and large cor-
related with judgments about concrete cases.
5.4. Study 4d: Contrastive design
Study 4d replicated Study 2. 386 participants recruited on Amazon
Mechanical Turk were presented with both the Good Luck and Bad Luck
versions of the Toymaker vignette. Thereafter they had to state to what
Table 5
Effect of outcome on wrongness, blame, permissibility, and punishment judg-
ments.
Variable t(319) p Cohen's d
Wrongness 4.20 <.001 .19
Blame 3.99 < .001 .30
Permissibility 5.86 < .001 .12
Punishment 11.27 < .001 .64
Fig. 23. Proportions of participants who judged the two agents identically (No
DI) for wrongness, blame, permissibility, and punishment in the Toymaker
scenario; errors bars represent 95% confidence intervals; Wilson method, see
Brown et al., 2001.
Fig. 24. Abstract denial of moral luck for wrongness, blame, permissibility, and
punishment; error bars denote 95% confidence intervals.
Table 6
Proportions of participants who agreed (above midpoint 4) and completely
agreed (endpoint 7) with the abstract denial of moral luck for wrongness,
blame, permissibility, and punishment.
Measure Wrongness Blame Permissibility Punishment
Answers > 4 81% 83% 76% 64%
Endpoint 7 54% 51% 54% 36%
3
We are grateful to Jonathan Baron for suggesting this experimental design.
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
14

extent they agreed with comparative claims about Frank and John’s
behavior, focusing on wrongness, blame, permissibility, and punish-
ment. The statements took the exact same form as in Study 2. For
wrongness, for instance, participants had to assess the claim “It was just
as wrong for John to use the new paint without testing it for lead as it
was for Frank” on a 7-point Likert scale anchored at 1 with “completely
agree” and at 7 with “completely disagree.” As in Study 2, the four
statements were presented in randomized order. The main task was
preceded by an attention check and followed by a comprehension
question. Methods, materials, and detailed results can be found in the
Appendix (Section 7).
All results from Study 2 were replicated: First, for wrongness,
blame, and permissibility, over 80% of the participants agreed (Likert-
scale < 4) and over 60% completely agreed (endpoint 1) that the two
agents should be judged in the same fashion. The ratios for punishment
were significantly lower: 62% agreed and 41% completely agreed with
an identical assessment of the two agents. Second, for wrongness,
blame, and permissibility mean disagreement with moral luck was
pronounced both in the case-based task and the abstract question; for
punishment mean disagreement was present though again significantly
lower. Third, for all four measures, abstract endorsement of moral luck
correlates with a more pronounced difference in the assessment of the
two agents in the case-based task.
5.5. Discussion
5.5.1. Evidence for the Difference Intuition across experimental designs
The results for the between-subjects Study 4a replicated those of
Study 1a perfectly. The effect size for the influence of outcome on
punishment judgment was again large (d= 1.07), whereas it was about
a third as pronounced yet also significant for wrongness and blame
judgments (d= .36 and d = .33). Furthermore, the lower bound of the
95% confidence interval for the effect size for punishment judgments
was larger than the higher bounds of the 95% confidence intervals for
the effect sizes for the three other judgments. As in Study 1a, outcome
had no statistically significant effect on permissibility judgments.
The basic pattern of the within-subjects Study 1b was replicated by
Study 4b: The influence of outcome was again large for punishment
judgment (d= .61) and moderate for wrongness (d= .30) and blame
(d= .35) judgments. In contrast to the other studies, however, we
found a small effect for permissibility judgments (d= .15). Since the
confidence intervals for the effect of outcome on permissibility judg-
ments in our previous studies barely included 0, we suspect that there is
a real, although small effect of outcome on permissibility judgments.
Study 4c, which used an alternative design invoking two distinct end-
ings rather than two different vignettes, was also consistent with the
findings of Study 1b and 4b.
The results of the contrastive design Study 4d were consistent with
those of Study 2. There was solid agreement with the claims that the
two agents were equally blameworthy, their actions equally wrong, and
equally impermissible, although for all three measures the average
differed significantly from the endpoint 1. Most participants “com-
pletely agreed” with the claim that the actions of the morally lucky and
unlucky agents are equally permissible, wrong, and blameworthy.
Punishment judgments differed from the three other types of judgment.
Consistent with the findings of Study 2, there was only moderate
agreement with the statement that the two agents deserve the same
degree of punishment. Only a minority of participants “completely
agreed” with the claim that the morally luck and unlucky agents de-
serve equal punishment.
Taken together, Studies 4b, 4c, and 4d confirm that for the vast
majority of people outcome has no influence on permissibility,
wrongness, and blame judgment when they are presented with both the
Good Luck and Bad Luck scenarios. By contrast, even in such com-
parative circumstances, outcome still influences judgments about pun-
ishment. Study 4a confirms that in a non-comparative task, where
people judge only one of the two types of agents, outcome influences
wrongness, blame, and punishment judgments (the latter more than the
former two).
5.5.2. Mediation analyses
The second part of Study 4a, which focused on potential mediators,
replicated the findings of Study 3. For all four types of moral judgment,
the epistemic mediators proved once more insignificant. As in Study 3,
we found strong support in favor of the Probabilistic Model: For
wrongness, blame, and punishment, probability and negligence as-
criptions mediated the impact of outcome on moral judgment, and
negligence was near-completely mediated by probability. Importantly,
the impact of outcome on wrongness and blame were reduced to in-
significance once the mediators had been taken into account, which
suggests that for wrongness and blame a Single Probabilistic Pathway
Model might be most appropriate. Punishment, by contrast, again best
fits a Dual Probabilistic Pathway Model. While the effect of outcome on
moral judgment was significantly mediated by probability, a strong
direct effect of outcome on moral judgment remained (note again that it
might disappear if other potential mediators were tested for).
Fig. 25. Proportions of uniform judgments across outcomes for wrongness, blame, permissibility, and punishment.
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
15

6. Study 5: Abstract views on moral luck
In Studies 4b, 4c and 4d, we examined not only people’s judgments
regarding moral luck in concrete cases, but also their abstract views on
the subject. Since in all three studies, the questions about participants
abstract views about moral luck were preceded by the concrete, case-
based judgment tasks, responses to the latter might have influenced re-
sponses to the former. We ran another study with 121 participants re-
cruited on Amazon Mechanical Turk, who were presented with the ab-
stract questions (Section 5.2.1), this time not preceded by judgments
tasks invoking concrete cases. The results are by and large the same as in
Studies 4b, 4c and 4d (see Section 8 of the Appendix for further detail):
About three quarters of participants agree that the two agents should be
judged identically in terms of wrongness, blame, and permissibility,
whereas for punishment it is just over half of the participants.
Together with Studies 4b, 4c and 4d, Study 5 suggests that there is
no abstract/concrete difference for moral luck as there is in other areas
of moral psychology (Nichols & Knobe, 2008). On average, people agree
with the abstract claim that the morally lucky and unlucky individuals
should be treated similarly. Punishment differs from the three other
moral judgment types: wrongness, blame, and permissibility.
7. General discussion
7.1. Moral luck and moral competence
Our studies suggest that when confronted with both a Good Luck
and a Bad Luck scenario, a very large majority of people judge the two
agents and their actions identically in terms of wrongness, blame, and
permissibility. That is, for most people outcome does not influence
these moral judgments in such circumstances. This is true both when
people are asked to make a moral judgment about a concrete situation
and when they are asked to consider abstract views about moral luck.
Our mediation analyses cast light on the mechanisms behind the
influence of outcome on moral judgment when people consider only
one situation. According to one view, people ascribe different epistemic
states to the two agents: The difference in the moral assessment of the
morally lucky and unlucky agents would thus be due to different as-
criptions of foresight (knowledge that the bad outcome will occur) or
recklessness (awareness of a substantial probability of the bad outcome
occurring). Our results suggest that such models are incorrect since
people ascribe foresight and recklessness to similar degrees across the
lucky and unlucky situations. Rather, our mediation analyses suggest
that, in non-comparative situations, outcome affects blame and
wrongness judgments because people view the unlucky agent as more
negligent than the lucky agent, and they view the former as more
negligent because they consider the probability of the bad outcome’s
occurrence in the unlucky situation as higher than in the lucky situa-
tion. Once these mediators are taken into consideration, the effect of
outcome on blame and wrongness judgments turns insignificant.
The mediation analysis results of Study 4a suggest a Probabilistic
Single Pathway Model for blame and wrongness (though for wrongness,
the results of Study 3 are more in tune with a Probabilistic Dual
Pathway Model). Both Studies 3 and 4a suggest a Probabilistic Dual
Pathway Model for punishment. These findings can be interpreted in
two distinct ways: On a rationalist approach, outcome information can
be taken to constitute evidence regarding an event’s likelihood: It is
only reasonable, one might argue, for a judging subject to assess the
probability of a contextually salient event higher in a situation in which
it has just occurred than in a situation in which it just failed to occur.
This justified difference in perceived probability gives rise to an equally
innocuous perceived difference in negligence ascriptions across lucky
and unlucky agents. But if the unlucky agent was indeed more negli-
gent, it is appropriate to consider her more blameworthy than the
lucky, and less negligent, agent.
On a bias approach, by contrast, the differing perceptions of accident
probability might be seen as a performance error. A performance error
is a judgment that is not reflective of people’s domain-specific compe-
tence, but rather results from the characteristics of the processes in-
volved in making the judgment. For instance, the judgment that a
center-embedded sentence like “The rat the cat the dog bit chased es-
caped” is not acceptable is a performance error: it does not reflect
people’s grammatical competence, but rather results from the proces-
sing limitations of working memory. Other types of sentences such as
garden-path sentences elicit performance errors, too. A garden-path
sentence is grammatical but sounds unacceptable because its beginning
is similar to a misleading salient syntactic construction and thus elicits
an interpretation that ends up being incorrect. “The horse raced past
the barn fell” and “The old man the boat” are classic examples of
garden-path sentences. On reflection, however, competent speakers can
see that such sentences are acceptable because they overcome the si-
milarity between their beginning and the misleading syntactic con-
structions. Certain types of moral judgments in non-comparative con-
texts, we propose, give rise to a similar phenomenon. On reflection, we
hypothesize, the folk conceptions of probability and negligence in
moral contexts pattern with their legal analogues (the basic legal fra-
mework regarding mens rea and culpability presumably derives from
folk morality and psychology, see Moore (2011)). According to the folk
(and legal) view, what matters for moral evaluation is the ex ante
probability of an accident (the probability at the context of action) and
a concept of negligence defined in terms of the latter, not ex post
probability (the probability at the context of assessment). When at-
tempting to assess the ex ante probability of an accident in non com-
parative situations, however, people fall prey to the hindsight bias (a
performance error), as their judgments are inappropriately influenced
by outcome information. They exaggerate the accident probability in
the unlucky case, deem the unlucky agent as more negligent than the
lucky one, and consequently judge her more harshly. In short, the in-
fluence of outcome on blame and wrongness judgments results from a
domain-general, irrational bias regarding probability assessments, and
is not expressive of people’s reflective moral views. The distinct as-
sessment of the morally lucky and unlucky agents in non-comparative
contexts, this is to say, is not a manifestation of moral competence; it is
a performance error.
Even if the effect of outcome on blame and wrongness judgments in
non-comparative contexts is a performance error, it is a phenomenon of
great significance for everyday life. Moral and legal judgments typically
take place in such non-comparative contexts, and one would thus ex-
pect that outcome frequently influences judgment in everyday cir-
cumstances (cf. also Kneer & Bourgeois-Gironde, 2017 for the effect of
outcome on mens rea attribution amongst professional judges). The
phenomenon should thus be of concern to legal professionals, but also
to all of us when we engage in moral judgment.
7.2. Punishment vs. blame and wrongness
Punishment judgments systematically diverged from wrongness and
blame judgments in all our studies: (i) Outcome has a systematically
larger effect on aggregate punishment judgments than on aggregate
wrongness and blame judgments in both the between-subjects and
within-subjects designs. (ii) In contrast to wrongness and blame judg-
ments, a sizeable proportion of people judge that the morally lucky and
unlucky agents should be punished differently in comparative contexts
and when reporting their abstract principles. (iii) The mechanisms be-
hind punishment judgments differ from the mechanisms behind
wrongness and blame judgments, as revealed by the mediation ana-
lyses. The influence of outcome on wrongness and blame judgments is
principally due to the different ascriptions of negligence, which in turn
are driven by distinct perceptions of outcome probability across con-
ditions (Study 4a, but see Study 3). By contrast, the influence of out-
come on punishment judgments is only partly due to the varying as-
criptions of negligence. With the mediators tested, outcome continues
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
16

to have a direct effect on punishment judgments even after negligence
and probability ascriptions have been taken into account.
The systematic divergence of punishment judgments from wrong-
ness and blame judgments suggests a Dual Process Model of moral
judgment. Blame and wrongness judgments result from a first process,
while punishment judgment results from a second process. The key
distinction between these two processes regards a partial direct effect of
outcome on moral judgment. Whether an outcome is bad or good was
shown to have a considerable direct effect on punishment judgments
(Studies 3 and 4a). In Study 4a, however, no direct effect of outcome on
wrongness and blame judgments could be detected; the entire effect of
outcome on wrongness and blame was mediated by negligence (though
there was some direct effect on wrongness in Study 3).
Our findings are partly consistent with current influential Dual
Process Models of moral judgment (Cushman, 2008; see Fig. 4 above),
according to which wrongness judgments are primarily influenced by
mental state ascriptions (belief and desire), while punishment judgments
are influenced by mental state ascriptions and the agent’s causal con-
tribution to the outcome. The respective effect sizes of outcome for
punishment and wrongness judgments are consistent with this model:
Since the mental states of the agents were held fixed across the lucky and
unlucky conditions and perceived to be the same for both agents (Studies
3 and 4a), their influence on moral judgment in the tested cases must be
limited. Outcome, by contrast, varies across conditions. Consistent with
Cushman’s theory, outcome had a more pronounced effect on punish-
ment judgments (deemed sensitive to both mental and causal factors)
than on wrongness (deemed sensitive principally to mental states). In
addition, while in Study 4a the impact of outcome on wrongness was
mediated in its entirety by probability and negligence ascriptions (in
conformity with the Probabilistic Single Pathway Model), we detected a
considerable direct effect of outcome on punishment judgments (which
thus fits the Probabilistic Dual Pathway Model).
In one respect, however, our findings are inconsistent with current
Dual Process Models of moral judgments. These models predict that
blame judgments pattern with punishment judgments: Blame judg-
ments are supposed to be influenced by outcome the way punishment
judgments are. The findings reported in this article as well as further
findings on the action/omission effect (Kneer et al., in preparation a)
suggest that this need not be the case; rather, blame often patterns with
wrongness judgments. Cushman (p.c.) draws our attention to the fact
that in his experiments, the questions were stated using the expression
“blame” whereas our questions were phrased in terms of “blame-
worthy.” As of yet unpublished findings of different researchers seem to
converge on the fact that judgments invoking “blame” are more out-
come-sensitive than judgments cast in terms of “blameworthy.”
7.3. Philosophical implications
The Control Principle and the Difference Intuition are inconsistent. A
puzzle arises if, as philosophers have commonly assumed, both the
Control Principle and the Difference Intuition are widely held. The phi-
losophical literature on the topic is vast since the Puzzle of Moral Luck is
deemed one of the most foundational challenges to systematic ethics. As
Bernard Williams observed, widespread support for the Difference
Intuition “cannot leave the concept of morality where it was” (1993, 54).
The Difference Intuition, as it features in the philosophical litera-
ture, requires the comparative assessment of two agents, who bring
about distinct outcomes in otherwise identical situations. The between-
subjects experiments frequently found in the empirical literature on the
topic are thus not, strictly speaking, pertinent to the debate regarding
the Puzzle of Resultant Moral Luck. We have used a variety of designs
(within-subjects, contrastive, and Baron’s design with alternative end-
ings in Study 4c) that are actually suited to test the Difference Intuition.
In contrast to what philosophers standardly assume, very few people
share the Difference Intuition. In the within-subjects studies for the
Bathroom and the Toymaker scenarios, the vast majority of participants
judged the two agents identically for wrongness (89% in the Bathroom
scenario/89% in the Toymaker scenario), blame (87%/83%), and per-
missibility (96%/90%). The same pattern of results was found in the
contrastive design experiments and when participants were explicitly
consulted on their abstract views about moral luck for all three mea-
sures. It is unclear whether the few participants apparently in agree-
ment with the Difference Intuition genuinely endorse it or instead
constitute statistical noise. In the latter case, there simply is no Puzzle
of Resultant Moral Luck; in the former, its significance would be dras-
tically reduced. Moral luck would no longer constitute a fundamental
challenge to the very possibility of systematic ethics, but merely an
oddity, or perhaps a bias, of the moral judgment of a small minority.
Whereas no Difference Intuition could be detected for wrongness,
blame, and permissibility judgments in comparative contexts, a sub-
stantive minority of people hold that the unlucky agent deserves a
harsher punishment. One might thus conclude that the Puzzle of Moral
Luck exists after all, though its scope is restricted to the punishment
judgments of a substantive minority. However, the problem of moral
luck is often cast in terms of blame, that is, a property the ascription of
which should exclusively depend on moral responsibility, rather than
‘blame-related’ notions such as deserved punishment, which may also
depend on considerations regarding, for instance, deterrence, preven-
tion, or what is socially desirable (Enoch & Marmor, 2007; but see
Kumar, 2018 for a punishment-centered approach to the Puzzle of
Moral Luck). Thus, the fact that unlucky reckless drivers are punished
more harshly than lucky ones in most Western legal systems, and that a
substantive minority deems this appropriate, need not raise a puzzle.
7.4. Limitations
Our studies suffer from two main limitations. First, the nature of
permissibility judgments is not entirely clear. The effect of outcome on
permissibility judgments was only significant in some studies although
the confidence intervals of the effect sizes barely included zero when
this effect failed to reach significance. We lean towards the view that
outcome does influence permissibility judgments in non-comparative
contexts, although its effect size is smaller than the effect size for blame
and wrongness judgments. If this were correct, a question arises as to
whether permissibility judgments result from the same process as
wrongness and blame judgments or from a distinct process. Future re-
search should investigate this matter.
Second, the studies reported in this article as well as past studies
have principally investigated the influence of moral luck on the moral
judgments of Westerners (an important exception is Barrett et al., 2016,
who explored the impact of intent and accidental factors on moral
judgment in small-scale societies). In light of the variation of psycho-
logical findings across cultures in general (Fessler & Machery, 2012;
Henrich, Heine, & Norenzayan, 2010) and moral psychology in parti-
cular (Abarbanell & Hauser, 2010; Hannikainen et al., in preparation),
the universality of our findings should be investigated cross-culturally
and cross-linguistically (Kneer et al., in preparation b).
8. Conclusion
Philosophers have argued that the Puzzle of Resultant Moral Luck
challenges the very possibility of systematic ethics. Our findings suggest
there is no such challenge since most people simply do not share the
Difference Intuition. A very large majority of people do not make different
wrongness, blame, and permissibility judgments when they are reflectively
considering the Good Luck and the Bad Luck scenarios side by side, and
this pattern of concrete judgments is consistent with their abstract prin-
ciples. It is only when people consider a single situation in isolation that
their judgment may be influenced by factors that the agent does not
control, such as the action’s outcome. While such circumstances are
common in everyday life, our mediation analyses suggest that the resulting
judgments are performance errors: They result from the hindsight bias,
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
17

that is, from a retroactive overestimation of the probability that an out-
come would occur if it does indeed occur, which leads people to view the
morally unlucky agent as more negligent than the lucky one.
Punishment judgments differ from the other types of moral judg-
ments including blame, an unexpected finding in light of standard dual-
process models, according to which punishment judgments pattern with
blame judgments. We found that both in comparative contexts and
when consulted about their abstract principles, people consider an
unlucky agent more deserving of punishment than a lucky agent. While
this finding might call for a refinement of current Dual Process Models
of moral judgment in moral psychology, it is not philosophically puz-
zling since on many normative views, deserved punishment does de-
pend on the consequences of actions.
Acknowledgments
We would like to thank Jonathan Baron, Fiery Cushman, Ivar
Hannikainen, Shaun Nichols, Mike Stuart, Juri Viehoff, the members of
the Pittsburgh Empirical Philosophy Lab (www.philosophicalexper-
iments.com), particularly Hannah Kim and Wes Buckwalter, the mem-
bers of the Moral Psychology Research Group, particularly Victor
Kumar, and the referees for Cognition for very helpful feedback.
Appendix A. Supplementary materials
Supplementary data associated with this article can be found, in the
online version, at https://doi.org/10.1016/j.cognition.2018.09.003.
References
Abarbanell, L., & Hauser, M. D. (2010). Mayan morality: An exploration of permissible
harms. Cognition, 115(2), 207–224.
Alicke, M., & Davis, T. (1989). The role of a posteriori victim information in judgments of
blame and sanction. Journal of Experimental Social Psychology, 25(4), 362–377.
Alicke, M., Davis, T. L., & Pezzo, M. V. (1994). A posteriori adjustment of a priori decision
criteria. Social Cognition, 12(4), 281–308.
American Law Institute (1985). Model penal code: Official draft and explanatory notes:
Complete text of model penal code as adopted at the 1962 Annual Meeting of The
American Law Institute at Washington, D.C., May 24, 1962. Philadelphia: American Law
Institute.
Baron, J. (2000). Thinking and deciding. Cambridge University Press.
Baron, J., & Hershey, J. C. (1988). Outcome bias in decision evaluation. Journal of
Personality and Social Psychology, 54(4), 569.
Baron, J., & Ritov, I. (2004). Omission bias, individual differences, and normality.
Organizational Behavior and Human Decision Processes, 94(2), 74–85.
Barrett, H. C., Bolyanatz, A., Crittenden, A. N., Fessler, D. M. T., Fitzpatrick, S., Gurven,
M., & Scelza, B. A. (2016). Small-scale societies exhibit fundamental variation in the
role of intentions in moral judgment. Proceedings of the National Academy of Sciences,
113(17), 4688–4693.
Brown, L. D., Cai, T. T., & DasGupta, A. (2001). Interval estimation for a binomial pro-
portion. Statistical Science, 101–117.
Christensen-Szalanski, J. J., & Willham, C. F. (1991). The hindsight bias: A meta-analysis.
Organizational Behavior and Human Decision Processes, 48(1), 147–168.
Cushman, F. (2008). Crime and punishment: Distinguishing the roles of causal and in-
tentional analyses in moral judgment. Cognition, 108(2), 353–380.
Cushman, F. (2013). Action, outcome, and value a dual-system framework for morality.
Personality and Social Psychology Review, 17(3), 273–292.
Cushman, F., Dreber, A., Wang, Y., & Costa, J. (2009). Accidental outcomes guide pun-
ishment in a “trembling hand” game. PloS one, 4(8), e6699.
Enoch, D., & Marmor, A. (2007). The case against moral luck. Law and Philosophy, 26(4),
405–436.
Epstein, S., Pacini, R., Denes-Raj, V., & Heier, H. (1996). Individual differences in in-
tuitive–experiential and analytical–rational thinking styles. Journal of personality and
social psychology, 71(2), 390.
ESCI Exploratory Software for Confidence Intervals (2016), accompanies. Cumming, G. &
Calin-Jagerman, R. (2017), Introduction to the New Statistics: Routledge.
Everett, J. A. (2013). The 12 item social and economic conservatism scale (SECS). PloS
one, 8(12), e82131.
Fessler, D. M., & Machery, E. (2012). Culture and cognition. In E. Margolis, R. Samuels, &
S. P. Stich (Eds.). The Oxford handbook of philosophy of cognitive science. Oxford
University Press.
Fischhoff, B. (1975). Hindsight is not equal to foresight: The effect of outcome knowledge
on judgment under uncertainty. Journal of Experimental Psychology: Human Perception
and Performance, 1(3), 288.
Gino, F., Moore, D. A., & Bazerman, M. H. (2009). No harm, no foul: The outcome bias in
ethical judgments. Harvard Business School NOM Working Paper, 08–080.
Gino, F., Shu, L. L., & Bazerman, M. H. (2010). Nameless + harmless = blameless: When
seemingly irrelevant factors influence judgment of (un)ethical behavior.
Organizational Behavior and Human Decision Processes, 111(2), 93–101.
Graham, J., Haidt, J., & Nosek, B. A. (2009). Liberals and conservatives rely on different
sets of moral foundations. Journal of Personality and Social Psychology, 96(5), 1029.
Graham, J., Nosek, B. A., Haidt, J., Iyer, R., Koleva, S., & Ditto, P. H. (2011). Mapping the
moral domain. Journal of Personality and Social Psychology, 101(2), 366.
Hannikainen, I., Machery, E., Stich, S. P., Rose, D., Alai, M., Angelucci, A., … Zhu, J. (In
preparation). Sourcehood, alternate possibilities, and the moderating role of reflection.
Hawkins, S. A., & Hastie, R. (1990). Hindsight: Biased judgments of past events after the
outcomes are known. Psychological Bulletin, 107(3), 311.
Heider, F. (1958). The psychology of interpersonal relations. New York: Wiley.
Henrich, J., Heine, S. J., & Norenzayan, A. (2010). Most people are not WEIRD. Nature,
466(7302), 29.
Hsee, C. K. (1996). The evaluability hypothesis: An explanation for preference reversals
between joint and separate evaluations of alternatives. Organizational Behavior and
Human Decision Processes, 67(3), 247–257.
Kneer, M., & Bourgeois-Gironde, S. (2017). Mens rea ascription, expertise and outcome
effects: Professional judges surveyed. Cognition, 169, 139–146.
Kneer, M., Olivola, C. Y., Kim, H., Machery, E., Stich, S. P., Rose, D. … Zhu, J. (In pre-
paration a). The action/omission distinction across cultures.
Kneer, M., Machery, E., Stich, S. P., Rose, D., Alai, M., Angelucci, A. … Zhu, J. (In pre-
paration b). Moral luck across cultures.
Kumar, V. (2018). Empirical vindication of moral luck. Nous.
Lench, H. C., Domsky, D., Smallman, R., & Darbor, K. E. (2015). Beliefs in moral luck:
When and why blame hinges on luck. British Journal of Psychology, 106(2), 272–287.
Levy, N. (2005). The good, the bad, and the blameworthy. Journal of Ethics and Social
Philosophy, 1(2), 1–16.
Lowe, C. A., & Medway, F. J. (1976). Effects of valence, severity, arid relevance on re-
sponsibility and dispositional attribution. Journal of Personality, 44(3), 518–538.
Martin, J. W., & Cushman, F. (2015). The adaptive logic of moral luck. In J. Sytsma, & W.
Buckwalter (Eds.). The Blackwell companion to experimental philosophy. Wiley-
Blackwell.
McKillip, J., & Posavac, E. J. (1975). Judgments of responsibility for an accident. Journal
of Personality, 43(2), 248–265.
Moore, M. S. (2011). Intention as a marker of moral culpability and legal punishability. In
A. Duff, & S. P. Green (Eds.). Philosophical foundations of criminal law. Oxford: Oxford
University Press.
Morris, S. B., & DeShon, R. P. (2002). Combining effect size estimates in meta-analysis
with repeated measures and independent-groups designs. Psychological Methods, 7(1),
105.
Nagel, T. (1979). Mortal Questions (Vol. 89, Vol. 3). Cambridge University Press.
Nichols, S., & Knobe, J. (2008). Moral responsibility and determinism. In J. Knobe and S.
Nichols (Eds.), Experimental Philosophy (pp. 105-128).
Nichols, S., Timmons, M., & Lopez, T. (2014). Using experiments in ethics–ethical con-
servatism and the psychology of moral luck. In M. Christen, C. vanSchaik, J. Fischer,
M. Huppenbauer, & C. Tanner (Eds.). Empirically informed ethics: Morality between
facts and norms (pp. 159–176). Springer International Publishing.
Nichols, S. (2009). Ethics and the psychology of moral luck. In Presented at the Pacific
American Psychological Association, Vancouver, BC.
Pacini, R., & Epstein, S. (1999). The relation of rational and experiential information
processing styles to personality, basic beliefs, and the ratio-bias phenomenon. Journal
of Personality and Social Psychology, 76(6), 972.
Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and resampling strategies for assessing
and comparing indirect effects in multiple mediator models. Behavior Research
Methods, 40(3), 879–891.
Richards, N. (1986). Luck and desert. Mind, 65, 198–209.
Robbennolt, J. K. (2000). Outcome severity and judgments of “Responsibility”: A meta-
analytic review 1. Journal of Applied Social Psychology, 30(12), 2575–2609.
Rosebury, B. (1995). Moral responsibility and moral luck. Philosophical Review, 104,
499–524.
Royzman, E., Cassidy, K. W., & Baron, J. (2003). “I know, you know”: Epistemic ego-
centrism in children and adults. Review of General Psychology, 7(1), 38.
Royzman, E., & Kumar, R. (2004). Is consequential luck morally inconsequential? em-
pirical psychology and the reassessment of moral luck. Ratio, 17(3), 329–344.
Rubin, Z., & Peplau, L. A. (1975). Who believes in a just world? Journal of Social Issues,
31(3), 65–89.
Schwitzgebel, E., & Cushman, F. (2012). Expertise in moral reasoning? Order effects on
moral judgment in professional philosophers and non-philosophers. Mind & Language,
27(2), 135–153.
Spranca, M., Minsk, E., & Baron, J. (1991). Omission and commission in judgment and
choice. Journal of Experimental Social Psychology, 27, 76–105.
Walster, E. (1966). Assignment of responsibility for an accident. Journal of Personality and
Social Psychology, 3(1), 73.
Thomson, J. J. (1993). Morality and bad luck. In D. Statman (Ed.). Moral luck (pp. 195–
215). SUNY Press.
Williams, B. (1981). Moral luck: Philosophical papers 1973–1980. Cambridge: Cambridge
University Press.
Williams, B. (1993). Moral luck. In D. Statman (Ed.). Moral Luck. SUNY Press.
Wuensch, K. (2012). Using SPSS to obtain a confidence interval for Cohen’s d. http://core.
ecu.edu/psyc/wuenschk/SPSS/CI-d-SPSS.pdf.
Young, L., Nichols, S., & Saxe, R. (2010). Investigating the neural and cognitive basis of
moral luck: It’s not what you do but what you know. Review of Philosophy and
Psychology, 1(3), 333–349.
Zimmerman, M. J. (2006). Moral Luck: A Partial Map. Canadian Journal of Philosophy,
36(4), 585–608.
M. Kneer, E. Machery Cognition 182 (2019) xxx–xxx
18