Content uploaded by Ivar R. Hannikainen
Author content
All content in this area was uploaded by Ivar R. Hannikainen on Apr 15, 2024
Content may be subject to copyright.
Understanding rule enforcement using drift diffusion models
Neele Engelmann (engelmann@mpib-berlin.mpg.de)
Max Planck Institute for Human Development, Center for Humans and Machines
Berlin, Germany
Ivar R. Hannikainen (ivar@ugr.es)
Department of Philosophy I
Granada, Spain
Carlos Gonz´
alez-Garc´
ıa (cgonzalez@ugr.es)
Department of Experimental Psychology
Granada, Spain
Mar´
ıa Ruz (mruz@ugr.es)
Department of Experimental Psychology
Granada, Spain
Abstract
Since their inception, drift diffusion models have been applied
across a wide range of disciplines within psychology to un-
cover the mental processes that underlie perception, attention,
and cognitive control. Our studies contribute to ongoing ef-
forts to extend these models to abstract, social reasoning pro-
cesses like moral or legal judgment. We presented participants
with a set of social rules, while manipulating whether various
behaviors violated the rule’s letter and/or its purpose–––two
independent standards by which to decide what constitutes a
transgression. In this framework, cases that violate or com-
ply with both a rule’s text and its purpose can be seen as
congruent or ‘easy’ cases, and cases that elicit opposing ver-
dicts as incongruent or ‘hard’ cases–––in a manner analogous
to widely-studied conflict tasks in cognitive psychology. We
recorded 34,573 decisions made by 364 participants under soft
time pressure, and investigated whether hierarchical drift diffu-
sion modeling could explain various behavioral patterns in our
data. This approach yielded three key insights: (1) judgments
of conviction were faster than judgments of acquittal owing to
an overall bias (zparameter) toward conviction; (2) incongru-
ent cases produced longer reaction times than congruent cases
(an interference effect), due to differences in the rate of evi-
dence accumulation (vparameter) across case-types; and (3)
increases in the ratio of congruent-to-incongruent cases am-
plified the interference effect on reaction times, by fostering
greater response caution—revealed by a larger threshold (or a
parameter). Thus, our studies document dissociable effects of
the drift diffusion components on rule-based decision-making,
and illustrate how the cognitive processes that subserve ab-
stract and social decision-making tasks, such as the enforce-
ment of communal and legal rules, may be illuminated through
the drift diffusion framework.
Keywords: cognitive control, conflict task, statutory inter-
pretation, legal reasoning, drift diffusion modeling
The tension between norm adherence and moral virtue is a
recurring theme in moral philosophy. Though abiding by the
maxim “do not lie” arguably serves us well in most contexts
(Sunstein, 2005), there may be circumstances under which ly-
ing is permissible or even obligatory (Engelmann, 2023). A
broader literature has now documented the tendency for peo-
ple to perceive certain rule violations as morally acceptable
(Awad et al., 2022; Kwon, Zhi-Xuan, Tenenbaum, & Levine,
2023; Kwon, Tenenbaum, & Levine, 2022), and certain in-
stances of seeming compliance i.e., ‘loopholes’, as neverthe-
less undermining the rule’s deeper spirit (Bridgers, Taliaferro,
Parece, Schulz, & Ullman, 2023). This body of evidence can
be fruitfully understood by considering the instrumental di-
mension of rules—that, as well as having a specific formula-
tion or text (e.g., ”No shoes in the house”), rules are generally
intended to serve a legislative goal or purpose (i.e., of keep-
ing the floors clean). In the legal domain, where the ques-
tion whether a person’s conduct was in violation of the law
or not can have grave consequences, this division between
a rule’s letter and its spirit has given rise to rival theories
of interpretation—often referred to as textualism (Schauer,
1991) and purposivism (Fuller, 1957) respectively—that re-
main hotly contested today.
A number of empirical reports have turned attention to the
way in which people apply written rules, providing evidence
that people prioritize a rule’s text (Struchiner, Hannikainen,
& de Almeida, 2020), but also attend to the rule’s purpose,
especially when the lawmaker’s goal is seen as morally good
(Flanagan, Almeida, Struchiner, & Hannikainen, 2023). This
pattern reflects the fact that people simultaneously consider
the rule’s literal formulation, but also their broader moral at-
titude toward the case at hand, and consequently waver when
tasked with applying rules to ’hard’ cases, in which a behav-
ior complies with the rule’s purpose despite violating its text
or vice versa (Almeida, Struchiner, & Hannikainen, 2023). In
these circumstances, the opportunity to reflect appears to bol-
ster adherence to the rule’s text (Flanagan et al., 2023), per-
haps in recognition that text provides a focal point (Schelling,
1960) that facilitates coordinated interpretation (Hannikainen
et al., 2022).
Overall, these studies provide convergent evidence that
people are divided between applying textualist and purpo-
sivist standards, and that they subjectively recognize this ten-
sion. These findings motivated the objectives of our present
research: to examine the extent to which cognitive conflict
arises in rule enforcement by exploiting an analogy with ba-
sic interference tasks in cognitive psychology and leveraging
the toolkit of cognitive modeling to help characterize the psy-
chological processes that subserve rule-based reasoning.
Interference tasks, such as the Stroop and Flanker tests,
examine people’s ability to classify target stimuli along a
task-relevant dimension, while simultaneously manipulating
a task-irrelevant dimension between the congruent and incon-
gruent conditions. Repeatedly, researchers have documented
longer reaction times and reduced accuracy in incongruent tri-
als relative to congruent trials—a phenomenon known as the
interference effect—and efforts to apply drift diffusion mod-
eling (Ratcliff, Smith, Brown, & McKoon, 2016; Ratcliff,
1978; Ratcliff & Smith, 2004; Ratcliff & McKoon, 2008; My-
ers, Interian, & Moustafa, 2022; Voss, Rothermund, & Voss,
2004) have yielded a deeper understanding of the cognitive
processes that underlie this effect. Taking binary responses
and response times as input, drift diffusion models estimate
four main parameters which jointly characterize the decision-
making process. The drift rate, v, captures the speed at which
evidence toward a decision is accumulated; that is, drift rates
are typically faster in easy tasks (e.g., congruent trials) than
in difficult tasks (e.g., incongruent trials). The threshold, a,
represents the degree of response caution, with higher values
indicating that decision-makers adopt a more stringent stan-
dard of evidence before making a decision (and lower values
indicating that less evidence suffices). Non-decision time, t,
captures the amount of time employed in parsing the relevant
stimuli before initiating the decision process, plus executing
the required motor response (e.g., pressing a button in an ex-
periment). Finally, the bias parameter, z, captures whether
people are initially biased toward either response option, be-
fore having accumulated any evidence toward a decision.
In short, the goal of this work is to investigate whether ten-
sion between the text and purpose of written rules generates
interference effects similar to those found on basic conflict
tasks. We predict that interference will manifest in longer
reaction times when applying rules to hard or ’incongruent’
cases (i.e., in which someone’s behavior violates the rule’s
text yet complies with its purpose, or vice versa) than in
easy or ’congruent’ cases (i.e., in which a behavior violates
both the rule’s text and its purpose, or complies with both).
Derivatively, our studies aim to understand the cognitive pro-
cesses that underlie people’s application of rules through the
lens of drift diffusion modeling.
Experiment 1:
Violating and complying with rules
Experiments 1a and 1b were designed as a first exploration
of judgment and reaction time data using the drift diffusion
framework. The experiments are identical except for the test
question (”did this person violate the rule?” vs. ”did this
person comply with the rule?”). Both experiments were pre-
registered (Exp. 1a: https://aspredicted.org/4QM SMC,
Exp. 1b: https://aspredicted.org/4CS W73). We imple-
mented the tasks in jsPsych (https://www.jspsych.org/
7.3/), and hosted them on cognition.run. Data were an-
alyzed using R and RStudio, as well as the hddm package
in python (for hierarchical drift diffusion modeling). Mate-
rials, data, and code for all experiments presented in this pa-
per are available at https://osf.io/wsejx/?view only=
c27142c28d0d4f21bcbd4902f12592e8. Online demos of
all experiments are available at https://anonymous.4open
.science/w/2024 cogsci rules-26A0/.
Experiment 1a: Violating rules
Design, Material and Procedure We used a 2 (text viola-
tion: yes vs. no) ×2 (purpose violation: yes vs. no) ×3
(items) ×8 (scenario) design, with all manipulations admin-
istered within subjects. Thus, each subject saw 12 trials in
each of the 8 scenario blocks, for a total of 96 trials. The
order of scenario blocks and the order of trials within blocks
was randomized. Each block began with the presentation of
a rule and its purpose, e.g.: “To keep her house clean, Mary
announced: No one may wear shoes in the house.” On the fol-
lowing pages, participants were presented with example be-
haviors (one per page) that either violated the rule’s text e.g.,
“A guest wears his new sneakers and the floor stays clean”),
its purpose (e.g., “A barefoot guest has a bleeding toe and
stains the carpets”), both (e.g., “A guest wears muddy sneak-
ers and dirties the carpets”), or neither (e.g., “A guest walks
around barefoot and keeps the carpet clean”). On each page,
we asked “Did this person violate the rule?”, and participants
were instructed to answer yes or no as fast as possible by
pressing either the e key or the i key on their keyboard. As-
signment of response keys was counterbalanced across par-
ticipants. After a time limit of 8 seconds without a response,
a trial was recorded as invalid. A fixation cross was presented
between trials with a random duration between 250 and 2,000
ms. Before the main experiment, participants completed a
practice block (without time pressure or incongruent trials) to
familiarize themselves with the layout and task. Participants
had the opportunity to take a break after each block. The ex-
periment ended with the assessment of demographic variables
and a debriefing.
Participants 119 participants completed the survey on pro-
lific.com (Mage = 35.9, SDage = 14 years, 55% women, 44%
men, 1% non-binary or no answer). Inclusion criteria (for all
three experiments) were: being a native English speaker, not
having participated in previous studies using similar materi-
als, and an approval rate of at least 90% on previous tasks on
the platform. Participants received a compensation of £1.50
for an estimated ten minutes of their time.
Results and Discussion See Figure 1a for an overview. As
expected, participants overwhelmingly indicated that the rule
was violated when the target behavior conflicted with both its
text and its purpose (95% yes, 95% CI: 94-96%), and that
the rule was not violated (4% yes, 95% CI: 3-5%) in absence
of either text or purpose violation. Replicating previous re-
sults (Struchiner et al., 2020; Hannikainen et al., 2022), vi-
olation of text alone more often led people to judge that the
rule was violated (65% yes, 95% CI: 63-67%) than violation
of purpose alone (26% yes, 95% CI: 23-29%). The data were
best described by a mixed logistic model with fixed effects of
text (χ2
d f =1= 6700.3, p<.001) and purpose (χ2
d f =1= 1646.7,
p<.001), in addition to random intercepts for participant and
scenario.
Turning to response times, participants reacted faster to
congruent cases of violation (median = 2,307 ms) or com-
pliance (median = 2,285 ms) than to conflict cases (text viola-
tion: median = 2,999 ms, purpose violation: median = 2,658
ms). Accordingly, the data were best described by a linear
mixed model containing fixed effects for text (χ2
d f =1= 96.39,
p<.001), purpose (χ2
d f =1= 40.97, p<.001),and their two-way
interaction (χ2
d f =1= 740.75, p<.001) in addition to random
intercepts for participant and scenario.1
Visual inspection of Figure 1a revealed that ‘yes’ responses
tended to be faster than ‘no’ responses (full violation: ∆RT =
1,063 ms, text-only violation: ∆RT = 473 ms, purpose-only
violation: ∆RT = 124 ms), except in the compliance condi-
tion (∆RT = -196 ms). As an exploratory analysis, we thus
tested whether adding a three-way interaction between text,
purpose, and response (yes vs. no) improved model fit, and
found that it did (χ2
d f =4= 158.56, p<.001).
Next, we ran a series of drift diffusion models in the hddm
Python library (Wiecki, Sofer, & Frank, 2013), with 10,000
samples, a burn-in rate of 10% and a thinning factor of 3.
We forced a constant boundary separation across case-types,
under the plausible assumption that participants would not
adjust their degree of response caution flexibly on each trial,
given that trial type was randomized and thus unpredictable.
On the basis of the Deviance Information Criterion, and vi-
sual evidence of convergence in the trace plots (see Online
Materials), we compared models that allowed the drift rate,
non-decision time, and/or bias to vary across case-types.
The best-fitting model revealed a positive bias (z= .53,
95% HDI [.53, .54]), i.e., a bias toward the ’Yes’ response,
and a boundary separation of a= 3.16 (95% HDI [3.07,
3.25]). Non-decision times differed modestly across condi-
tions, between 1,014 and 1,171 ms, as shown in Figure 2C.
Meanwhile, drift rates were faster in congruent cases than
incongruent cases (see Figure 2D)—both in the comparison
between full violation (v= 1.02, 95% HDI [0.95, 1.10]) and
literal violation (v= 0.24, 95% HDI [0.17, 0.30]), and be-
tween full compliance (v= -1.18, 95% HDI [-1.25, -1.11])
and literal compliance (v= -0.51, 95% HDI [-0.58, -0.45]).
Relaxing the assumption of constant boundary separation did
not qualitatively affect this pattern of results (see Online Ap-
pendix 1).
Experiment 1b: Complying with rules
In Experiment 1b, we sought to generalize the behavioral
results of Experiment 1a to judgments of compliance, and
1Deviating from our preregistration, we report a linear model of
reaction times due to lack of convergence in an ordinal model.
to assess the degree of convergence between the drift diffu-
sion parameters across both experiments. Additionally, Ex-
periment 1b allowed us to disambiguate whether the faster
’yes’ responses in Experiment 1a reflect general acquiescence
(a preference for affirmative responses), or instead an “ac-
cusatory stance” (preferring to convict than acquit, indepen-
dently of how judgments are elicited).
Design, Material and Procedure This experiment was
identical to Experiment 1a in every respect, except for the
test question. We now asked: ‘Did this person comply with
the rule?´.
Participants 124 participants completed the survey on pro-
lific.com (Mage = 36, SDage = 12.3 years, 49% women, 48%
men, 3% non-binary or no answer). Participants received a
compensation of £1.50 for an estimated ten minutes of their
time.
Results and Discussion See Figure 1A for an overview.
With the inverse dependent measure, participants almost al-
ways indicated that agents in congruent cases complied with
the rule when neither text nor purpose had been violated (93%
yes, 95% CI: 92-94%), and that they did not comply with the
rule when violating both text and purpose (6% yes, 95% CI:
5-7%). Mirroring the pattern of results in Experiment 1a, vi-
olations of purpose alone were more often judged in compli-
ance with the rule (73% yes, 95% CI: 70-76%) than violations
of text (28% yes, 95% CI: 26-30%). The data were best de-
scribed by a mixed logistic model with fixed effects for text
(χ2
d f =1= 45.16, p<.001), purpose (χ2
d f =1= 26.75, p<.001),
and their two-way interaction (χ2
d f =2= 776.23, p<.001), in
addition to random intercepts for scenario and participant.
As in Experiment 1a, participants were faster to decide
congruent cases of rule violation (median = 2,335 ms) and
compliance (median = 2,296 ms) than incongruent cases (pur-
pose violation: median = 2,710 ms, text violation: median =
2,940 ms). Across all conditions, “no” responses were now
faster than “yes” responses, though the size of this median
difference varied across conditions (violation: ∆RT = 346 ms,
compliance: ∆RT = 158 ms, text violation: ∆RT = 534 ms, pur-
pose violation: ∆RT = 201 ms). The data were best described
by a linear mixed model with fixed effects for text (χ2
d f =1=
45.16, p<.001), purpose (χ2
d f =1= 26.75, p<.001), response
(yes vs. no) (χ2
d f =1= 75.53, p<.001), as well as the text ×
purpose interaction (χ2
d f =1= 700.7, p<.001), and the three-
way text ×purpose ×response interaction (χ2
d f =3= 41.31,
p<.001), in addition to random intercepts for participant and
scenario. Thus, the faster violation judgments observed in
Experiment 1a were not due to acquiescence, but rather to a
greater ease in indicating that a target behavior is against the
rule (rather than in compliance).
Again, we fit a series of hierarchical drift diffusion models,
with a constant boundary separation across case-types, and
applied the same criteria for model specification and selection
as in Experiment 1a. The best-fitting model revealed a bias
Figure 1: Reaction Time Distributions in Experiments 1 and 2. The figure displays grouped histograms of ‘Yes’ (blue) and
‘No’(red) responses, separately for each experiment and case type. Overlaid vertical lines represent (A) the median, 1st and 3rd
quartiles, and (B) the median in the congruence dominant (solid) and incongruence dominant (dashed) conditions.
toward the ‘no’ response (z= .46, 95% HDI [.45, .47]) with
a boundary separation of a= 3.23 (95% HDI [3.11, 3.36]).
Non-decision times again differed modestly across conditions
(see Figure 2C), ranging between 1019 and 1120 ms. Drift
rates were substantially slower for literal violation (v= -0.36,
95% HDI [-0.44, -0.29]), than full violation cases (v= -1.03,
95% HDI [-1.12, -0.95]), and for literal compliance (v= 0.55,
95% HDI [0.46, 0.62] than full compliance cases (v= 1.15,
95% HDI [1.05, 1.22]). Relaxing the assumption of constant
boundary separation did not qualitatively affect this pattern of
results (see Online Appendix 2).
Thus, Experiments 1a and 1b revealed an interference ef-
fect on reaction times. In both experiments, drift diffusion
modeling uncovered that this effect was primarily attributable
to changes in the average drift rate across case-types. Non-
response times, by contrast, independently contributed to
longer reaction times only in cases of literal violation. In light
of previous validation studies on the drift diffusion model,
this pattern of results can be interpreted as revealing cognitive
conflict in the application of rules to hard cases. Addition-
ally, both experiments uncovered shorter reaction times for
judgments of conviction than of acquittal and, accordingly,
the drift diffusion model retrieved a bias toward (i.e., greater
initial proximity to) the decision boundary representing con-
viction in both experiments.
Alternatively, it is also possible that people exhibit in-
creased response caution when faced with such scenarios, that
is, they want to be more sure to get things ”right” in more
complicated cases. This would be reflected in differences in
boundary separation, i.e., the aparameter.
Experiment 2:
Varying the proportion of incongruent trials
In Experiment 1, case-types were presented in a random order
and with an equal probability, precluding the possibility that
participants would adapt their response strategy to character-
istics of the task. Accordingly, we assumed that the boundary
separation parameter—a relative stable aspect of participants’
response strategy—would remain constant across trials.
The purpose of Experiment 2 is to further validate the
application of the drift diffusion framework to rule-based
decision-making by leveraging a manipulation of the propor-
tion of congruency trials—known, in the cognitive control lit-
erature, to influence the magnitude of the interference effect.
Typically, a low proportion of incongruent trials increases the
interference effect (producing longer reaction times and more
errors on incongruent trials), while a high proportion of in-
congruent trials reduces them (Bugg & Crump, 2012). Drift
diffusion modeling has shown, in this context, that block-
wide manipulations of the congruency proportion allow par-
Figure 2: Posterior Distributions of the HDDM Parameter Estimates in Experiments 1 and 2. (A) Bias and (B) boundary
separation in each experiment, with boundary separation displayed separately for congruent dominant (△) and incongruent
dominant (▽) blocks. (C) Non-decision times and (D) absolute drift rates are displayed separately for congruent (light shade)
and incongruent (dark shade) cases of violation (brown) and compliance (green). Note: bias in Experiment 1b has been
reversed to facilitate visual comparison.
ticipants to adapt their degree of response caution to changes
in the likelihood of encountering congruent vs. incongru-
ent trials across experimental blocks. Our goal was to test
whether and how boundary separation is affected by this ma-
nipulation, and to study the behavior of the remaining param-
eters under these conditions. The experiment is preregistered
at https://aspredicted.org/blind.php?x=QV6 XDN.
Design, Material and Procedure The experiment was
identical to the previous studies in all aspects except we in-
troduced a manipulation of the proportion of congruent (i.e.,
violation and compliance) vs. incongruent (i.e., literal viola-
tion and literal compliance) trials per rule. Participants now
saw two blocks consisting of four rules each: a block with
a high proportion of congruent trials (10 out of 12 trials for
each rule; 5 violation, 5 compliance, 1 literal violation, and 1
literal compliance) and another block with a low proportion
of congruent trials (2 out of 12 trials for each rule). The or-
der of the congruency proportion blocks was randomized be-
tween subjects. To ensure that no rule appeared twice for any
given participant, the rules that made up the first and second
blocks were randomly drawn from the set of eight rules with-
out replacement. Participants were given no indication of the
manipulation of congruency proportion, and all instructions
remained identical to the previous studies. The test question
was “Did this person violate the rule?”, as in Experiment 1a.
Participants 121 participants completed the survey on pro-
lific.com (Mage = 35.2, SDage = 13.3 years, 51% women, 48%
men, 1% non-binary or no answer) and received £1.50 as
compensation for an estimated ten minutes of their time.
Results and Discussion Replicating the pattern of results
observed in previous experiments (see Figure 1b), literal vio-
lations were seen as rule violations less often (65% yes, 95%
CI: 63-67%) than full violations (95% yes, 95% CI: 94-96%),
while cases of literal compliance were seen as violating the
rule more often (18% yes, 95% CI: 15-21%) than cases of full
compliance (4% yes, 95% CI: 3-5%). The data were best de-
scribed by a model that contained fixed effects of text (χ2
d f =1
= 6857.6, p<.001), purpose (χ2
d f =1= 1183.2, p<.001), pro-
portion of congruency (χ2
d f =1= 9.0, p= .003), and the text
×purpose ×proportion of congruency interaction (χ2
d f =3=
12.11, p= .007), in addition to random intercepts for partici-
pant and scenario. 2
For our preregistered analysis of reaction times, we col-
lapse the types of trials into the class of congruent (full vio-
lation and compliance) and incongruent (text-only violation
and purpose-only violation) trials. Responses to congruent
trials were faster than responses to incongruent trials in both
congruency proportion blocks, but the median difference was
larger when incongruent trials were rare (∆RT = 822 ms) than
when they were frequent (∆RT = 395 ms). Accordingly, the
data were best described by a linear model containing fixed
effects of congruency (χ2
d f =1= 476.37, p<.001), congruency
proportion (χ2
d f =1= 10.57, p= .001), and the congruency ×
proportion interaction (χ2
d f =1= 63.06, p<.001), in addition
to random intercepts for participant and scenario.
In our drift diffusion models, we now freed the boundary
separation parameter, allowing congruency proportion to im-
pact individuals’ response caution (as well as non-decision
times and drift rates). Boundary separation was larger in
blocks with a majority of congruent trials (a= 3.56 (95%
HDI [3.11, 3.36]) than in blocks with a majority of incongru-
ent trials (a= 3.14 (95% HDI [3.00, 3.29]). The proportion
manipulation also influenced non-decision times, which were
generally longer in the incongruent dominant block (between
1,058 and 1,146 ms) than in the congruent dominant block
(between 792 and 922 ms), except in literal violation cases
where the difference was negligible (from 1,111 to 1,114 ms;
see Figure 2d).
The proportion manipulation revealed differential effects
on evidence accumulation during congruent versus incongru-
ent cases. Specifically, drift rates during congruent trials were
faster in blocks where they were frequent (violation: v= 1.24,
95% HDI [1.16, 1.33]; compliance: v= -1.40, 95% HDI [-
1.48, -1.30]) than in blocks where they were infrequent (vi-
olation: v= 0.99, 95% HDI [0.86, 1.11]; compliance: v=
-1.16, 95% HDI [-1.30, -1.04]). By contrast, the proportion
of congruency had no effect on drift rates for either literal vi-
olation (high congruency v = 0.23, 95% HDI [0.16, 0.31]; vs.
low congruency v = 0.27, 95% HDI [0.18, 0.37]) or literal
compliance (high congruency v = -0.75, 95% HDI [-0.83, -
0.67]; low congruency v = -0.72, 95% HDI [-0.84, -0.61])
cases. Lastly, the best-fitting model revealed a prosecution
bias (z= .53, 95% HDI [.52, .54]) as in Experiment 1.
Thus, we reproduced the widely-documented proportion
congruency effect (Bugg & Crump, 2012) on reaction times in
the domain of rule violation judgments: namely, the pattern of
interference was amplified during blocks in which most cases
were congruent. A drift diffusion model revealed that these
high congruency blocks fostered a greater emphasis on ac-
curacy than did low congruency blocks (in which most cases
were incongruent), while also accelerating evidence accumu-
2The interaction effect involving the congruency proportion ma-
nipulation reflected a weak tendency toward elevated rule violation
judgments in congruent-dominant blocks for all case-types (ORs
ranging from 1.06 to 1.38) except full compliance (where OR =
0.52). However, given the magnitude of these effects, we take them
to be theoretically irrelevant.
lation selectively on congruent trials. Meanwhile, the manip-
ulation of congruency proportion did not affect drift rates on
incongruent trials.
General Discussion
Multiple recent studies provide evidence that moral cogni-
tion can come into conflict with people’s literal understand-
ing of legal and communal rules (Struchiner et al., 2020;
Turri, 2019; Hannikainen et al., 2022). Meanwhile, develop-
ments in drift diffusion modeling have afforded a clearer un-
derstanding of the cognitive processes that facilitate decision-
making on basic perceptual tasks evoking response competi-
tion. In this work, we brought together these parallel lines
of research, and contributed to ongoing efforts to apply drift
diffusion modeling to higher-order reasoning tasks (Cohen &
Ahn, 2016; Yu, Siegel, Clithero, & Crockett, 2021; Siegel,
van der Plas, Heise, Clithero, & Crockett, 2022; Engelmann
& Hannikainen, 2024).
Behaviorally, we replicated typical effects of interference
and congruency proportion observed in basic conflict tasks
(e.g., Stroop and Flanker tasks) (Bugg & Crump, 2012).
Cases in which the application of purposivist and textualist
standards would yield opposing verdicts produced longer re-
action times, and this tendency was amplified when the pro-
portion of incongruent trials was low.
Drift diffusion models indicated that the interference effect
in both Experiments 1a and 1b was attributable to a slower
rate of evidence accumulation (or drift rate), and this reduc-
tion persisted when experimentally elevating the proportion
of incongruent trials in Experiment 2. Thus, greater expo-
sure to incongruent cases did not appear to facilitate evidence
accumulation on incongruent trials—which further supports
the interpretation of slower drift rates as a marker of response
competition. Additionally, the bias parameter indicated that
the starting point of evidence accumulation was closer to the
conviction boundary than the acquittal boundary—a tendency
that dovetails with various asymmetries in the attribution of
third-party blame versus praise (Guglielmo & Malle, 2019),
and explains the tendency for conviction to occur faster than
acquittal in all three experiments.
In sum, our present studies exploited a fruitful, though im-
perfect, analogy between letter vs. spirit conflicts in statu-
tory interpretation and lower-level forms of response compe-
tition. Still, whether domain-general mechanisms of conflict
monitoring and control guide decision-making across these
disparate tasks cannot be gleaned from the present evidence
alone. Additionally, understanding the trajectory of evidence
accumulation over time in a time-varying framework (Ulrich,
Schr¨
oter, Leuthold, & Birngruber, 2015) may provide novel
insights into the onset of competing textualist and purposivist
responses (see also (Flanagan et al., 2023)). Nevertheless, our
studies raise the prospect of better understanding how peo-
ple reason about everyday transgressions of written rules and
laws by applying insights from the the drift diffusion frame-
work.
References
Almeida, G., Struchiner, N., & Hannikainen, I. R. (2023).
Rule is a dual character concept. Cognition,230, 105259.
Awad, E., Levine, S., Loreggia, A., Mattei, N., Rahwan, I.,
Rossi, F., . .. Kleiman-Weiner, M. (2022). When is it ac-
ceptable to break the rules? Knowledge representation of
moral judgement based on empirical data. Retrieved from
https://arxiv.org/abs/2201.07763
Bridgers, S. E. C., Taliaferro, M., Parece, K., Schulz, L., &
Ullman, T. (2023). Loopholes: A window into value align-
ment and the communication of meaning. Retrieved from
https://osf.io/preprints/psyarxiv/cnxzv
Bugg, J. M., & Crump, M. J. (2012). In support of a dis-
tinction between voluntary and stimulus-driven control: A
review of the literature on proportion congruent effects.
Frontiers in Psychology,3, 367.
Cohen, D. J., & Ahn, M. (2016). A subjective utilitarian
theory of moral judgment. Journal of Experimental Psy-
chology: General,145(10), 1359.
Engelmann, N. (2023). Murderer at the door! To lie or to
mislead? In A. Wiegmann (Ed.), Lying, fake news, and
bullshit. Bloomsbury. (In press)
Engelmann, N., & Hannikainen, I. R. (2024). Does moral
valence influence the construal of alternative possibilities?
Possibility Studies & Society. (in press)
Flanagan, B., Almeida, G., Struchiner, N., & Hannikainen,
I. R. (2023). Moral appraisals guide intuitive legal deter-
minations. Law and Human Behavior,47(2), 367.
Fuller, L. L. (1957). Positivism and fidelity to law – A reply
to Professor Hart. Harvard Law Review,71, 630.
Guglielmo, S., & Malle, B. F. (2019). Asymmetric morality:
Blame is more differentiated and more extreme than praise.
PloS one,14(3), e0213544.
Hannikainen, I. R., Tobia, K. P., de Almeida, G. d. F.,
Struchiner, N., Kneer, M., Bystranowski, P., .. . others
(2022). Coordination and expertise foster legal textualism.
Proceedings of the National Academy of Sciences,119(44),
e2206531119.
Kwon, J., Tenenbaum, J., & Levine, S. (2022). Flexibility
in moral cognition: When is it okay to break the rules? In
J. Culbertson, H. Rabagliati, V. Ramenzoni, & A. Perfors
(Eds.), Proceedings of the annual meeting of the cognitive
science society (Vol. 44, pp. 905–911).
Kwon, J., Zhi-Xuan, T., Tenenbaum, J., & Levine, S. (2023).
When it’s not out of line to get out of line: The role
of universalization and outcome-based reasoning in rule-
breaking judgments. Retrieved from https://osf.io/
preprints/psyarxiv/n8bjr
Myers, C. E., Interian, A., & Moustafa, A. A. (2022). A
practical introduction to using the drift diffusion model
of decision-making in cognitive psychology, neuroscience,
and health sciences. Frontiers in Psychology,13, 1039172.
Ratcliff, R. (1978). A theory of memory retrieval. Psycho-
logical Review,85(2), 59.
Ratcliff, R., & McKoon, G. (2008). The diffusion deci-
sion model: Theory and data for two-choice decision tasks.
Neural Computation,20(4), 873–922.
Ratcliff, R., & Smith, P. L. (2004). A comparison of sequen-
tial sampling models for two-choice reaction time. Psycho-
logical Review,111(2), 333.
Ratcliff, R., Smith, P. L., Brown, S. D., & McKoon, G.
(2016). Diffusion decision model: Current issues and his-
tory. Trends in Cognitive Sciences,20(4), 260–281.
Schauer, F. (1991). Playing by the rules: A philosophical
examination of rule-based decision-making in law and in
life. Clarendon Press.
Schelling, T. C. (1960). The strategy of conflict. Harvard
University Press.
Siegel, J. Z., van der Plas, E., Heise, F., Clithero, J. A., &
Crockett, M. (2022). A computational account of how
individuals resolve the dilemma of dirty money. Scientific
Reports,12(1), 18638.
Struchiner, N., Hannikainen, I. R., & de Almeida, G. d. F.
(2020). An experimental guide to vehicles in the park.
Judgment and Decision Making,15(3), 312–329.
Sunstein, C. R. (2005). Moral heuristics. Behavioral and
Brain Sciences,28(4), 531–541.
Turri, J. (2019). Excuse validation: A cross-cultural study.
Cognitive Science,43(8), e12748.
Ulrich, R., Schr¨
oter, H., Leuthold, H., & Birngruber, T.
(2015). Automatic and controlled stimulus processing in
conflict tasks: Superimposed diffusion processes and delta
functions. Cognitive Psychology,78, 148–174.
Voss, A., Rothermund, K., & Voss, J. (2004). Interpreting the
parameters of the diffusion model: An empirical validation.
Memory & Cognition,32, 1206–1220.
Wiecki, T. V., Sofer, I., & Frank, M. J. (2013). Hddm: Hier-
archical bayesian estimation of the drift-diffusion model in
python. Frontiers in Neuroinformatics, 14.
Yu, H., Siegel, J. Z., Clithero, J. A., & Crockett, M. J. (2021).
How peer influence shapes value computation in moral
decision-making. Cognition,211, 104641.
