Coercion Changes the Sense of Agency in the Human Brain

Abstract

People may deny responsibility for negative consequences of their actions by claiming that they were “only obeying orders.” The “Nuremberg defense” offers one extreme example, though it is often dismissed as merely an attempt to avoid responsibility. Milgram’s classic laboratory studies reported widespread obedience to an instruction to harm, suggesting that social coercion may alter mechanisms of voluntary agency, and hence abolish the normal experience of being in control of one’s own actions. However, Milgram’s and other studies relied on dissembling and on explicit measures of agency, which are known to be biased by social norms. Here, we combined coercive instructions to administer harm to a co-participant, with implicit measures of sense of agency, based on perceived compression of time intervals between voluntary actions and their outcomes, and with electrophysiological recordings. In two experiments, an experimenter ordered a volunteer to make a key-press action that caused either financial penalty or demonstrably painful electric shock to their co-participant, thereby increasing their own financial gain. Coercion increased the perceived interval between action and outcome, relative to a situation where participants freely chose to inflict the same harms. Interestingly, coercion also reduced the neural processing of the outcomes of one’s own action. Thus, people who obey orders may subjectively experience their actions as closer to passive movements than fully voluntary actions. Our results highlight the complex relation between the brain mechanisms that generate the subjective experience of voluntary actions and social constructs, such as responsibility.

Full text

Article
Coercion Changes the Sense of Agency in the Human
Brain
Highlights
dResponsibility for action is a key feature of human societies
dIt depends on association between actions and outcomes in
the brain
dClaims of reduced responsibility are sometimes based on
‘‘only obeying orders’’
dTwo experiments suggest coercion can reduce implicit
measures of sense of agency
Authors
Emilie A. Caspar, Julia F. Christensen,
Axel Cleeremans, Patrick Haggard
Correspondence
p.haggard@ucl.ac.uk
In Brief
Acting under coercion modifies the
subjective experience of being the author
of an action, reducing the perceived
temporal association between actions
and outcomes. Caspar et al. show that the
neural processing of action outcomes
under coercion more closely resembles
situations of passive movement than
actions carried out intentionally.
Caspar et al., 2016, Current Biology 26, 1–8
March 7, 2016 ª2016 The Authors
http://dx.doi.org/10.1016/j.cub.2015.12.067

Current Biology
Article
Coercion Changes the Sense of Agency
in the Human Brain
Emilie A. Caspar,
1,2
Julia F. Christensen,
2
Axel Cleeremans,
1
and Patrick Haggard
2,
*
1
Consciousness, Cognition, and Computation Group (CO3), Center for Research in Cognition and Neurosciences (CRCN), ULB Neuroscience
Institute (UNI), Universite
´libre de Bruxelles (ULB), Avenue F.D. Roosevelt 50, CP191, 1050 Brussels, Belgium
2
Institute of Cognitive Neuroscience, University College London (UCL), Queen Square 17, London WC1N 3AR, UK
*Correspondence: p.haggard@ucl.ac.uk
http://dx.doi.org/10.1016/j.cub.2015.12.067
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
SUMMARY
People may deny responsibility for negative conse-
quences of their actions by claiming that they were
‘‘only obeying orders.’’ The ‘‘Nuremberg defense’’ of-
fers one extreme example, though it is often dis-
missed as merely an attempt to avoid responsibility.
Milgram’s classic laboratory studies reported wide-
spread obedience to an instruction to harm, suggest-
ing that social coercion may alter mechanisms of
voluntary agency, and hence abolish the normal
experience of being in control of one’s own actions.
However, Milgram’s and other studies relied on
dissembling and on explicit measures of agency,
which are known to be biased by social norms.
Here, we combined coercive instructions to admin-
ister harm to a co-participant, with implicit measures
of sense of agency, based on perceived compres-
sion of time intervals between voluntary actions
and their outcomes, and with electrophysiological
recordings. In two experiments, an experimenter
ordered a volunteer to make a key-press action that
caused either financial penalty or demonstrably
painful electric shock to their co-participant, thereby
increasing their own financial gain. Coercion
increased the perceived interval between action
and outcome, relative to a situation where partici-
pants freely chose to inflict the same harms. Interest-
ingly, coercion also reduced the neural processing of
the outcomes of one’s own action. Thus, people who
obey orders may subjectively experience their ac-
tions as closer to passive movements than fully
voluntary actions. Our results highlight the complex
relation between the brain mechanisms that
generate the subjective experience of voluntary ac-
tions and social constructs, such as responsibility.
INTRODUCTION
In Milgram’s classic experiments on obedience [1, 2], an exper-
imenter ordered volunteer participants to inflict allegedly painful
shocks to a third party. These studies focused on participants’
readiness to conform to authority and obey coercive instructions
to perform harmful actions. Interestingly, participants’ subjective
experience in such situations has not been systematically inves-
tigated, even though the legal defense of ‘‘only obeying orders’’
implies a loss of voluntary agency with coercion.
Sense of agency refers to the subjective experience of control-
ling one’s actions, and, through them, external events. Explicit
reports of perceived agency are modulated by numerous biases
[3], notably social desirability and cognitive dissonance effects
[4]. For example, individuals coerced into harmful actions might
report reduced sense of agency for secondary gain, such as
avoiding blame or punishment. Implicit measures may provide
more direct access to the cognitive mechanisms underlying
sense of agency, since these measures are less affected by
task demands and social factors such as desirability. Here we
used the perceived compression of time between a voluntary ac-
tion and its outcome [5] as an appropriate implicit marker of
sense of agency, and we investigated how coercion influenced
this measure. Action-outcome intervals are perceived as shorter
for intentional actions than for unintended actions such as pas-
sive movements [5, 6]. Therefore, if coercion indeed reduces
the core experience of agency, interval estimates should be
longer in the coercive than in the free-choice condition.
In a first experiment, participants were tested in pairs. They
took turns being ‘‘agent’’ and ‘‘victim,’’ ensuring reciprocity. In
a first group of participants, the agent could freely choose on
each trial to increase her own remuneration by taking money
from the ‘‘victim’’ (financial harm). In a second, smaller group,
the agent could freely choose to administer an electric shock
to the ‘‘victim’’ (physical pain), again increasing her own remu-
neration. This free-choice condition was compared to a coercive
condition, in which the experimenter stood next to the agent and
ordered her before each trial whether to take money or not, or
whether to shock the ‘‘victim’’ or not (see Figures 1 and 2).
RESULTS AND DISCUSSION
Experiment 1: Results
No participants withdrew from the experiment, and none re-
ported any distress either after testing or at follow-up. In the
financial harm group, agents freely chose to take money from
the ‘‘victim’’ in 33.97/60 trials (95% confidence interval [CI] =
29.07–38.88, min 0, max 60). In the physical pain group, agents
freely chose to give painful electric shocks to the ‘‘victim’’ in
31.37/60 trials (95% CI = 24.96–37.78, min 6, max 60). In
Current Biology 26, 1–8, March 7, 2016 ª2016 The Authors 1
Please cite this article in press as: Caspar et al., Coercion Changes the Sense of Agency in the Human Brain, Current Biology (2016), http://dx.doi.org/
10.1016/j.cub.2015.12.067

addition, our free-choice condition captured key features of
interpersonal choice, such as social reciprocity. In particular,
experiencing pain as a ‘‘victim’’ guided subsequent free choices
whether to inflict pain on one’s co-participant. Regression anal-
ysis showed that, within the subgroup of participants who were
first ‘‘victims’’ and then agents, participants who initially received
high numbers of shocks as ‘‘victim’’ subsequently gave more
shocks (t(9) = 4.776, p = 0.001, R
2
= 0.860). Such vindictive
behavior is consistent with previous reports in economic
games [7].
We analyzed agents’ interval estimates using ANOVA, with
condition (free choice, coercive) and outcome (harm, no harm)
as within-subject factors, and group (financial harm, physical
pain) as a between-subjects factor. The main effect of condition
was significant (F(1,50) = 22.740, p < 0.001, h
2partial
= 0.313),
with coercion leading to longer interval estimates than free
choice (437 ms, 95% CI = 399–475, and 370 ms, 95%
Figure 1. Experimental Setup
Schematic representation of the coercive condi-
tion (top) and the free-choice condition (bottom). In
this condition, the experimenter looked elsewhere.
In the coercive condition, the experimenter or-
dered the agent at each trial either to take money
from her co-participant (financial harm group) or to
deliver a shock (physical pain group). The experi-
menter stood next to the agent and looked at her
throughout the whole condition.
CI = 338–402, respectively; see Figure 4).
There was a main effect of group, with
lower interval estimates for the financial
harm group than for the physical pain
group (F(1,50) = 6.042, p = 0.017,
h
2partial
= 0.108) but no evidence for
any interaction between group and con-
dition (F(1,50) = 0.073, NS). The main ef-
fect of outcome was not significant (p >
0.3). The interaction condition 3out-
come was not significant (p = 0.099; for
full ANOVA table and further results, see
the Supplemental Experimental Proce-
dures). Interestingly, there was a signifi-
cant interaction between outcome and
group (F(1,50) = 6.201, p < 0.02, h
2partial
=
0.110). In the financial harm group, inter-
val estimates decreased on trials when
participants actually delivered harm
(352 ms, 95% CI = 316–388) compared
to when they did not (375 ms, 95% CI =
336–414; t(34) = 2.699, p = 0.01, Co-
hen’s d = 0.456). However, in the phys-
ical pain group, this difference was not
significant (p > 0.2). No other interactions
with group were significant (all ps > 0.3).
Importantly, there was no three-way
interaction with choice condition—so
we found no evidence that sense of
agency varied specifically as a function
of freely chosen outcomes. The effect of coercion was thus
not related to whether or not agents delivered harm on any spe-
cific trial or to the content of any individual instruction, but was
rather a contextual effect of receiving coercive instructions. This
result also rules out explanations based on the attentional or
arousing effects of harming others leading to altered time
perception.
We then performed planned comparisons with our control
conditions. Data for one participant in the active control condi-
tion was not available because of technical problems during
testing. Planned comparisons showed that the free-choice con-
dition produced shorter interval estimates than the active control
condition (t(56) = 2.809, p = 0.007, Cohen’s d = 0.372).
Conversely, the coercive condition did not differ significantly
from the passive condition (p > 0.6). These results suggest that
voluntary actions made under coercion are experienced in
some ways as if they were passive.
2Current Biology 26, 1–8, March 7, 2016 ª2016 The Authors
Please cite this article in press as: Caspar et al., Coercion Changes the Sense of Agency in the Human Brain, Current Biology (2016), http://dx.doi.org/
10.1016/j.cub.2015.12.067

Because participants’ free choices varied, we investigated
whether any difference between free-choice and coercive condi-
tions could simply reflect differences between these conditions
in the number of harmful actions. We therefore added the differ-
ence between the number of harmful actions freely chosen by
each participant and the number ordered by the coercive exper-
imenter as a covariate. The covariate was not significant (p > 0.7),
and the overall pattern of conclusions remained unchanged.
Thus, our results reflect a specific effect of coercive instruction
on the subjective experience of agency, occurring at the moment
of voluntary action, rather than any difference between the con-
tents of coercive instruction and free choices. In summary, when
the agent was coerced, they experienced less agency than when
they freely chose between the same options. This difference be-
tween coercion and choice did not interact with whether harm
was actually inflicted or not.
Pre-session questionnaire responses allowed us to investi-
gate whether sense of agency under coercion could be related
to personality or trait empathy [8, 9]. We therefore explored
whether personality and empathy measures were related to the
‘‘harm effect,’’ i.e., the main effect difference between interval
estimates associated with harmful actions and interval estimates
associated with non-harmful actions. Questionnaire scores of
trait empathy were positively and significantly related to the
extent to which a harmful outcome event reduced participants’
individual agency estimates (Table S1). More empathetic individ-
uals showed a more dramatic reduction in sense of agency when
their actions had harmful outcomes, compared to less harmful
outcomes. Correlations with personality factors were generally
weaker.
Additionally, we investigated whether the coercion effect, cor-
responding to the main effect difference between interval esti-
mates associated with the coercive condition and interval esti-
mates associated with the free-choice condition, correlated
Figure 2. Schematic Representation of the
Apparatus during the Experiment
The agent saw trial-by-trial feedback on the com-
puter screen, whereas the ‘‘victim’’ did not. The
agent pressed ‘‘F’’ on a keyboard to inflict harm
and earn money or ‘‘H’’ not to inflict harm/earn
money. Both the agent and the ‘‘victim’’ gave in-
dependent written estimates of action-tone in-
tervals on an answer sheet. Electrodes connected
to the stimulator were placed on the ‘‘victim’s’’ left
hand, which was clearly visible to the agent.
with personality and trait empathy (Table
S2). Correlations were generally weaker
than for the harm effect.
Experiment 1: Discussion
Why do people so readily comply with
coercive instructions? This question re-
mains central to historical [10] and
psychological [11] investigations. The
experience of agency under coercion
has been surprisingly neglected in previ-
ous discussions, despite its obvious
relation to personal responsibility. Here, we observed that being
ordered to perform an action reduces the subjective experience
of agency over the outcome in comparison with being free to
choose between outcomes, as shown by reduced estimates of
the temporal interval between action and outcome. Crucially,
the effect of coercion was not related to whether harm actually
occurred on any specific trial or to the content of any individual
instruction (financial loss versus painful shock), but was rather
a contextual effect of receiving coercive instructions. Sense of
agency was previously shown to increase with the size of the
‘‘response space’’ of action choices [12]. A similar cognitive
mechanism may explain why coercion both reduces the basic
experience of agency and simultaneously increases compliance
with instructions. ‘‘Only obeying orders’’ may not merely be a
retrospective narrative of behavior, aimed at secondary gain
such as blame avoidance, but may rather reflect a genuine differ-
ence in subjective experience of agency. Coercive instructions
appear to induce a passive mode of processing in the brain
compared to free choice between alternatives.
Experiment 2
In a second experiment, we investigated whether coercion
changes brain activity, by focusing on electroencephalogram
(EEG) potentials evoked by action outcomes. Filevich and col-
leagues [13] showed greater event-related potential (ERP) ampli-
tudes for outcomes when participants freely chose an action
compared with being instructed. We therefore predicted that co-
ercive instructions should reduce outcome-evoked ERP ampli-
tudes relative to free choices.
Experiment 2: Results
Behavioral Results
No participants withdrew from the experiment, and none re-
ported any distress either after testing or at follow-up. Agents
Current Biology 26, 1–8, March 7, 2016 ª2016 The Authors 3
Please cite this article in press as: Caspar et al., Coercion Changes the Sense of Agency in the Human Brain, Current Biology (2016), http://dx.doi.org/
10.1016/j.cub.2015.12.067

freely chose to administer painful electric shocks to the ‘‘victim’’
in 18.75/60 trials (95% CI = 11.46–26.04, min 0, max 52). Regres-
sion analysis again showed modest support for vindictive
behavior, as in experiment 1, with a trend for participants who
served first as ‘‘victims’’ tending to choose to administer shocks
in proportion to the number they had previously received (t(9) =
1.681, p = 0.1, R
2
= 0.561).
We also assessed how responsible participants felt during
each condition by asking them to rate, in a post-session ques-
tionnaire, their responsibility as a percentage score in each con-
dition. As expected, participants reported a higher degree of re-
sponsibility in the active condition (56.15%, SD = 39.70) than in
the passive condition (17.92%, SD = 24.12; t(19) = 3.792, p =
0.001, Cohen’s d = 0.847). Interestingly, they also reported
feeling more responsible in the free-choice condition (86.85%,
SD = 16.31) than in the coercive condition (34.80%, SD =
22.53; t(19) = 9.832, p < 0.001, Cohen’s d = 2.19), but also
than the active control condition (56.15%, SD = 39.70; t(19) =
3.562, p = 0.002, Cohen’s d = 0.796). In addition, the degree of
responsibility was higher in the coercive condition (34.80%,
SD = 22.53) than in the passive control condition (17.92%,
SD = 24.12; t(19) = 2.699, p = 0.014, Cohen’s d = 0.603; see
Figure 3).
Agents’ interval estimates were analyzed using repeated-
measures ANOVA, with condition (free choice, coercive) and
outcome (harm, no harm) as within-subject factors. The main ef-
fect of condition was significant (F(1,16) = 15.123, p = 0.001,
h
2partial
= 0.486), with free choice producing shorter interval judg-
ments than coercion (366 ms, 95% CI = 288–444, and 424.5 ms,
95% CI = 351–498, respectively; see Figure 4). The main effect of
outcome was not significant (p > 0.8), nor was the interaction
condition 3outcome (p > 0.5, for a full ANOVA table and further
results, see the Supplemental Experimental Procedures).
Planned comparisons with our control conditions showed that
the passive and the coercive conditions did not differ (p > 0.9)
and that the free-choice condition did not differ from the active
condition (p > 0.09).
Event-Related Potentials
Standard ERP recording and processing methods were used
(see the Supplemental Experimental Procedures).
We applied repeated-measures ANOVA to the auditory N1
amplitude [14], with the same ANOVA design used for interval es-
timates. The main effect of condition was significant (F(1,16) =
8.009, p = 0.012, h
2partial
= 0.334), with free choice producing
more negative N1 amplitudes than coercion (10.70 mv, 95%
CI = 13.81 to 7.60, and 8.15 mv, 95% CI = 10.83 to
5.47, respectively). The main effect of outcome was not signif-
icant (p > 0.1), nor was the interaction condition 3outcome (p >
0.4; for a full ANOVA table and further results, see the Supple-
mental Experimental Procedures; see Figure 5).
Interestingly, comparison between our active and passive
control conditions showed a similar, but smaller, effect. Specif-
ically, N1 amplitude was reduced in the passive condition
(9.99 mv, SD = 4.39) relative to the active condition
(11.11 mv, SD = 3.76; t(19) = 1.814, p = 0.086, Cohen’s d =
0.405; see Figure 6). Thus, the psychological effect of coercion
on sensory processing was similar to the physiological effect
of subtracting voluntary motor commands from bodily move-
ment [15]. Further, a direct comparison between the free-coer-
cive difference, and the active-passive difference, expressed
as the interaction term of a 2 32 ANOVA, showed that the mod-
ulation of sensory processing due to coercion was significantly
stronger than the modulation by the voluntary motor command
(F(1,17) = 4.878, p = 0.041, h
2partial
= 0.223). Direct planned com-
parisons between experimental and control conditions showed
that the free-choice experimental condition did not differ from
EXPERIMENT 1 EXPERIMENT 2
Estimated Interval (ms)
100
200
300
400
500
600
700
more agency less agency
200 500 800
Action-tone interval (ms)
AB
Financial harm (Group 1)
Estimated interval (ms)
100
200
300
400
500
600
700
Action-tone interval (ms)
200 500 800
Free-Choice
Coercive
Physical pain (Group 2)
100
200
300
400
500
600
700
more agency less agency
200 500 800
***
Action-tone interval (ms)
Free-Choice
Coercive ***
Free-Choice
Coercive ***
Figure 3. Interval Estimation Results
(A) Effects of coercion on interval estimates in experiment 1 in each group. The data for different action-tone intervals are shown to demonstrate interval esti-
mation performance, but this factor was not central to our predictions. Error bars show SEs. Coercion consistently prolonged interval estimates. *** indicates a
p value <0.001.
(B) Effects of coercion on interval estimates in experiment 2. *** indicates a p value <0.001.
4Current Biology 26, 1–8, March 7, 2016 ª2016 The Authors
Please cite this article in press as: Caspar et al., Coercion Changes the Sense of Agency in the Human Brain, Current Biology (2016), http://dx.doi.org/
10.1016/j.cub.2015.12.067

the active control condition (p > 0.6). The auditory N1 was more
pronounced in the passive condition (9.99 mv, SD = 4.39) than
in the coercive condition (7.93 mv, SD = 4.85; t(19) = 3.377, p =
0.003, Cohen’s d = 0.755).
Experiment 2: Discussion
The behavioral results replicated those of experiment 1.
Receiving coercive orders again reduced the sense of agency
over potentially harmful actions, according to our implicit mea-
sure based on intentional binding. Again, coercion produced an
experience that more closely resembled passive movement
than freely chosen voluntary action. This effect was again linked
to the context of coercive instruction, rather than the actual harm
resulting from any particular action. Explicit judgments of agency
provided an important cross-validation of our implicit measures.
Coercive orders thus influenced both explicit judgments of
agency and the low-level subjective feeling of agency on which
such judgments may be based [16]. The defense of ‘‘only obeying
orders’’ is often treated with suspicion in law because of the clear
secondary gain associated with denying responsibility. However,
our result suggests that primary feelings and neurophysiological
processing of agency are indeed reduced by coercion.
Analysis of the auditory N1 amplitude showed that coercion
reduced processing of action outcomes compared to conditions
in which participants freely chose what action to perform. This
finding was again independent of whether the tone was accom-
panied by a painful electric shock to the co-participant or not.
Importantly, the auditory tones were physically identical and
equally predictable in all conditions. We suggest that coercive
contexts produce an anticipatory reduction of sensory process-
ing for action outcomes. This involves both downregulation of
perceptual gains and temporal distancing. Indeed, the passive
condition also displayed a reduction of the auditory N1 ampli-
tude in comparison with the active condition, suggesting that
the brain may treat consequences of one’s actions under coer-
cion as if they were passively triggered.
Interestingly, we found no evidence for sensorimotor attenua-
tion of outcome processing [14, 17] when comparing either
active versus passive movements or free versus coerced ac-
tions. However, the outcomes in our study occurred later than
the short post-action window within which sensorimotor attenu-
ation operates [18].
General Discussion
Issues of personal responsibility, moral action, and social influ-
ence are central to many accounts about human nature. Previ-
ous behavioral experiments have studied these issues using
laboratory experiments [1]. However, the design and interpreta-
tion of those studies have been criticized. For example, Mil-
gram’s participants did not actually deliver pain, and pain re-
sponses of the third party were faked by an actor. It remains
unclear whether Milgram’s participants really believed they
were delivering severe pain or whether they had some intuition
that they were part of a simulation. In our design, participants
acted reciprocally as agent and ‘‘victim,’’ both delivering and
receiving harm. They knew from direct sensory experience
how their actions would affect their co-participant. This experi-
ence demonstrably influenced their free choices. Moreover,
the combination of money and pain in our experiments ensured
that our participants were motivated by greed and fear, factors
that may pervasively influence many human choices [19].
Legal, historical [10], and psychological [11] thought have all
considered how obeying orders influences personal responsibil-
ity. Social constructs, including power [20] and authority [1], are
often invoked. Using an implicit marker of sense of agency based
on time perception, we showed that coercive instructions
caused participants to experience less agency over the harmful
outcomes of their actions. The results generalized over implicit
and explicit measures of agency, and also over financial harm
and physical pain, and they were also found on subsets of trials
where no harm was actually delivered. Our results suggest that
‘‘only obeying orders’’ may not merely be a retrospective narra-
tive of behavior, adopted for secondary gain such as mitigation,
but may rather reflect a genuine difference in subjective experi-
ence of agency at the point of action itself. A previous study
[12] reported that sense of agency decreased as the size of the
‘‘response space’’ [21] or alternative actions decreased. We sug-
gest that this cognitive mechanism may also underlie the effects
of coercion on sense of agency reported here.
Milgram reported that ‘‘ordinary’’ people frequently comply
with coercive instructions [1]. Interestingly, our effects of coer-
cion on sense of agency were quite general across individuals
and were not strongly associated with particular personality
traits or with empathy. This was not merely due to insensitivity
of agency measures, since the effects of harm versus non-
harm on sense of agency were higher in those with more
empathic traits, as might be predicted. Rather, our result clearly
suggests one reason why so many people can be coerced. Spe-
cifically, coercion may reduce the linkage that normally binds the
experience of actions to their outcomes. Indeed, emotional
distancing from distasteful outcomes of one’s own necessary
actions forms a specific part of training and professional culture
in medicine [22] and in the military [23]. Training effects might
0
20
40
60
80
100
Conditions
Active Passive Free-choice Coercive
***
not responsible fully responsible
Judgment of responsibility (in %)
***
**
Figure 4. Judgments of Responsibility in Each Condition in Experi-
ment 2
*** indicates a p value %0.001. ** indicates a p value between 0.001 and 0.01.
Error bars show SEs.
Current Biology 26, 1–8, March 7, 2016 ª2016 The Authors 5
Please cite this article in press as: Caspar et al., Coercion Changes the Sense of Agency in the Human Brain, Current Biology (2016), http://dx.doi.org/
10.1016/j.cub.2015.12.067

also work in the opposite direction: learning the true valence of
one’s actions’ outcomes might potentially make the sense of
agency more resilient to the undermining effects of coercion.
We showed that acting under coercion deeply modifies the
sense of being responsible for outcomes of one’s actions. It
also attenuates the neural processing of outcomes. Both results
can be interpreted as a cognitive operation of ‘‘distancing,’’ or
reducing the linkage between one’s own decision-making, ac-
tion, and outcome. Our results may have profound implications
for social and legal responsibility. Laws are culturally evolved
rules for managing impact of individuals’ behaviors on others.
Laws must therefore engage with the psychological and neuro-
cognitive mechanisms that drive individual actions. Our finding
of reduced experience of agency under coercion does not legit-
imate Nuremberg-type defenses: society could still expect
agents to try to resist evil [10, 24]. However, our results do sug-
gest that people may indeed experience reduced agency at the
point of being coerced to perform abhorrent actions. Clearly, so-
ciety needs protection from harm, irrespective of whether the
perpetrators experienced agency at the time of the act, or not.
For example, the law argues that informed, rational agents
should know they remain responsible for their actions, even if
Figure 5. Neural Response to Outcome
Tones: Experimental Conditions
Graphical representation of the auditory N1
amplitude in the free-choice (light blue) and the
coercive (dark blue) conditions when (A) an elec-
trical shock was delivered at the same time as the
tone or (B) no electrical shock occurred. Topo-
graphical representations display the activity along
the whole scalp.
they have minimal experience of agency
at the time of action. Further, the law
could shift its focus away from those
who obey orders toward those who give
them, to prevent them from abusing a po-
sition that allows them to coerce others.
Our research on the experience of agency
highlights the fundamental link between
law and cognitive neuroscience. The law
has to engage with the human capacity
to control action if it is to fulfil its function
of allowing individuals to live together in
societies.
EXPERIMENTAL PROCEDURES
Experiment 1
The principles of the 2013 Declaration of Helsinki
were followed. The study was approved by Univer-
sity College London Research Ethics Committee
(0847/006). All participants provided written
informed consent prior to the experiment. No
participant withdrew, and no participant reported
distress at debriefing or at later follow-up.
Participants
Sixty right-handed student female participants
were recruited in pairs and were paid £15–£25 for
their participation. Only female participants were
tested in order to control for potential effects of gender, both within the partic-
ipant pairs, and also between participants and the (female) experimenters.
Data from a number of standard questionnaires, including Big Five personality
and trait empathy, were available from collection prior to participation. The
protocol for matching participants in pairs stipulated that participants could
not be relatives, friends, or from the same course or faculty. Thus, there was
no particular relation between co-participants prior to the experiment. Data
exclusion criteria were decided in advance of the experiment: failure to pro-
duce temporal intervals covarying monotonically with actual action-tone inter-
val and any general failure to follow instructions. To identify participants for
whom the action-tone intervals did not gradually increase with action-tone in-
tervals, we performed linear trend analyses with contrast coefficients 1, 0, 1
for the three delays of the action-tone intervals (see the next section). Two par-
ticipants were excluded due to non-significant linear trend. After this proce-
dure, 39 participants remained in group 1 (mean age = 22.92, SD = 3.82)
and 19 in group 2 (mean age = 22.79, SD = 3.63).
Materials and Procedure
On arrival at the experimental laboratory, participants read an information
sheet about the experimental procedure and the aim of the experiment. The
two co-participants signed their individual consent forms simultaneously,
ensuring that they were both aware of the other’s consent. Roles were as-
signed randomly so that one of the participants was told they were the agent
and the other was the ‘‘victim.’’ These roles were reversed for the second half
of the experiment, making the procedure fully reciprocal. Participants sat at a
table, face to face. An external, silent SODIAL Flexible Foldable USB keyboard
6Current Biology 26, 1–8, March 7, 2016 ª2016 The Authors
Please cite this article in press as: Caspar et al., Coercion Changes the Sense of Agency in the Human Brain, Current Biology (2016), http://dx.doi.org/
10.1016/j.cub.2015.12.067

was placed between them, oriented toward the agent but visible to both. The
experimental task ran on a computer, which was located on the agent’s right
side with the screen visible only to the agent (see Figure 2). The agent was in-
structed to press a key on the keyboard at a time she chose after the start of
the trial, using the right index finger. This caused a tone to occur. The delay be-
tween key press and tone was set to vary randomly between 200, 500, and
800 ms. The participants’ task was to estimate the delay between the key
press and the tone. They were informed that the delay would vary randomly
on a trial-by-trial basis, between 0 and 1,000 ms (they were reminded that
1,000 ms makes 1 s). Participants were also told to make use of all possible
numbers between 1 and 1,000, as appropriate, avoiding restricting their
answer space (e.g., not only use numbers between 1 and 100), and avoiding
rounding (cf. [25]). Each participant received a paper sheet with 60 empty
boxes for their time estimates in each condition of the task. Participants’ an-
swers were hidden from view of the other participant by a barrier, so as to avoid
participants being biased by the other participant’s answers.
There were two experimental and two control conditions. In the active con-
trol condition, the agent pressed the space key whenever she wanted. In the
passive control condition, the experimenter pressed the agent’s index finger
down on the space bar, making sure to be unpredictable in her movements
so as to minimize motor preparation in the agent. During these conditions,
nothing was displayed on the computer screen. The predictions focused on
two experimental conditions: free-choice and coercion. Participants were
informed that they would both start with a specific amount of money (i.e.,
£20 for group 1 and £15 for group 2; this difference was due to the fact that
participants in group 1 could lose money, but we needed to make sure partic-
ipants would leave the experiment with the mandated minimum payment of
£7.50/hr). In the free-choice condition, agents were instructed that they could
freely choose to increase their remuneration for the experiment by delivering,
or not delivering, a financial harm (group 1) or a physical pain (group 2) to the
other participant, using the appropriate keys on the keyboard. They were told
that they were totally free to choose how to act. The computer screen dis-
played which key press would be associated with which action (for instance,
the ‘‘F’’ key for taking money/delivering a shock and the ‘‘H’’ key for refraining
from taking money/delivering a shock). In group 1, the agents earned 5p each
time they chose to inflict financial pain to the ‘‘victim,’’ who then lost 5p. In
group 2, the agents earned 5p each time they decided to deliver a painful elec-
tric shock to the ‘‘victim.’’ They earned no money if they decided not to deliver
a shock. During this condition, the experimenters did not look at the partici-
pants, but focused their attention on task irrelevant objects. In the coercive
condition, the experimenters stood up next to the agent and ordered her, on
each trial, to take money or not (group 1) or to administer a shock or not (group
2) to the ‘‘victim.’’ The tone played after the key press was the same for both
keys. In group 1, the ‘‘victim’’ did not know on each trial whether the agent
Time (in ms)
-0.1 0 0.1 0.2 0.3 0.4 0.5
Amplitude
(
in mv
)
-10
-5
0
5
10
Active condition
Passive condition
Amplitude (mv)
0
5
10
15
Conditions
Free-choice Coercive Active Passive
N1 Amplitude
p = .086
** **
Active condition
Passive condition
auditory N1
B
AC
tone
Amplitude (in v)
Figure 6. Neural Response to Outcome
Tones: Control Conditions
(A) Auditory N1 amplitude in active (light green) and
the passive (dark green) conditions.
(B) Mean amplitude of the auditory N1 in all con-
ditions. ** indicates a significant difference (two-
tailed, p %0.01). Error bars show SEs.
(C) Topographical maps in active and passive
conditions.
had chosen to inflict financial harm or not and did
not know which of the agent’s response keys was
mapped to financial harm—this information was
available only on the agent’s feedback display
screen. Two new response keys (‘‘J’’ and ‘‘L’’)
were used when roles were reversed in group 1,
so that former ‘‘victims’’ could not now simply
repeat harm done to them by repeating the agent’s
previous key presses. Thus, group 1 was prevented
from imitative behavior. In group 2, participants
inevitably experienced the action selected by the
agent on each trial, in the form of physical pain. Therefore, imitative behav ior
would become unavoidable: we thus decided to use the same key-press map-
pings throughout.
The gains and losses were displayed on the screen visible to the agent. In
group 1, the agent saw two moneybags, 1 with her own money and 1 with
the ‘‘victim’s’’ money. Each time the agent inflicted financial harm, a coin
was shown moving from the ‘‘victim’s’’ bag to her bag, and the total amount
of money increase was displayed. When the agent did not take money, no an-
imations were displayed. In group 2, the agent only saw her own moneybag on
the screen since the ‘‘victim’’ did not lose money, but instead received a painful
shock to their left hand. The shock caused a twitch of the ‘‘victim’s’’ hand that
was readily visible to the agent.
Two experimenters participated in group 1, each testing half the sample.
One experimenter instructed agents to take money 50/60 times. The other
experimenter instructed agents to take money 30/60 times. This variation
allowed some control over possible effects of experimenter’s nastiness on
participants’ behavior and experiences under coercion. In group 2, both ex-
perimenters were simultaneously present, but one gave the coercive instruc-
tions. Both experimenters instructed agents to deliver shocks 30/60 times.
We used a partially randomized order of conditions. Participants performed
the active or the passive control condition first, then the free-choice condition,
then the coercive condition, and then the remaining control condition, either
active or passive. We chose not to randomize free-choice and coercive condi-
tions in order not to bias participants in the free-choice condition from previous
experience in the coercive condition (e.g., attempting to match the coercive
experimenter’s instructions in their ‘‘free’’ choices). Participants went through
the same four conditions twice, once as agent and once as ‘‘victim,’’ that is,
eight conditions in total. There were 60 trials per condition (20 trials at each ac-
tion-tone delay, in randomized order), giving a total of 480 trials. Participants
performed 240 actions as agents and observed 240 actions as ‘‘victims.’’
The order of the conditions was the same within each pair.
Details of the painful stimulation and other measures are given inthe Supple-
mental Experimental Procedures.
At the end of the experiment, participants were paid separately based on
their earned financial gain during the experiment. For one dyad in group 1,
the experimenter judged that the relation between the agent and the ‘‘victim’’
had become conflictual and hostile. The experimenter made an on-the-spot
decision to pay both of these participants the same amount (£20), to reduce
the possibility of subsequent distress or conflict.
Experiment 2
The principles of the 2013 Declaration of Helsinki were followed. The study was
approved by the ethical committee of the Universite
´libre de Bruxelles
(018/2015). All participants provided written informed consent prior to the
Current Biology 26, 1–8, March 7, 2016 ª2016 The Authors 7
Please cite this article in press as: Caspar et al., Coercion Changes the Sense of Agency in the Human Brain, Current Biology (2016), http://dx.doi.org/
10.1016/j.cub.2015.12.067

experiment. No participant withdrew, and no participant reported distress at
debriefing or at later follow-up.
Participants
Twenty-two right-handed student female participants were recruited in pairs
and paid V25–V31 for their participation. The same protocol for matching par-
ticipants and the same data exclusion criteria than in experiment 1 were used.
Two participants were excluded because no relation was found between
perceived and actual action-shock intervals. After this procedure, 20 partici-
pants remained (mean age = 23.15, SD = 3.183).
Materials and Procedure
We used the same method as for the physical pain group in experiment 1, with
modifications for EEG recording. Participants were instructed to wait a mini-
mum of 2 s in a relaxed position before pressing a key, so as to obtain a consis-
tent and noise-free baseline. Participants were further instructed not to move
for up to 2 s after the tone. Participants first performed 30 trials in the active and
the passive conditions, then 60 trials in the two experimental conditions, and
then again 30 trials in the active and the passive conditions. These combina-
tions of conditions were counterbalanced across participants. In order to
have the same number of choices between the control and the experimental
conditions, participants could choose between pressing ‘‘F’’ or ‘‘H’’ in the
active condition. In the passive condition, the agent was asked to position
two fingers (the index finger and the middle finger) on the two keys and the
experimenter pressed down on one of the agent’s fingers at an unpredictable
moment in time. In the post-session questionnaire, we additionally asked par-
ticipants to rate (from 0, ‘‘not responsible at all,’’ to 100, ‘‘entirely responsible’’)
how much they felt responsible in each condition. We also asked participants
to rate how frequently they would have disobeyed if they could have (from 3,
‘‘almost never,’’ to +3, ‘‘almost all the time’’). In this experiment, the mean
stimulation level selected by this procedure was 18.3 mA (SD = 6.7, pulse
duration = 200 ms).
Source Data
The behavioral data reported in this paper have been published in Mendeley
Data and are available at http://dx.doi.org/10.17632/322y43x9b7.1.
SUPPLEMENTAL INFORMATION
Supplemental Information includes Supplemental Experimental Procedures
and two tables and can be found with this article online at http://dx.doi.org/
10.1016/j.cub.2015.12.067.
AUTHOR CONTRIBUTIONS
E.A.C. developed the study concept. E.A.C. and P.H. created the study
design, and J.F.C. and A.C. provided critical comments. E.A.C. and J.F.C.
ran experiment 1. E.A.C. ran experiment 2. E.A.C. performe d the data analysis
and interpretation under the supervision of J.F.C., P.H., and A.C. E.A.C.
drafted the manuscript, and J.F.C., P.H., and A.C. provided critical revisions.
All authors approved the final version of the manuscript for submission.
ACKNOWLEDGMENTS
E.A.C. was supported by the FRS-F.N.R.S (Belgium). P.H. and J.F.C. were
supported by AHRC grant L015145/1 to P.H. and ERC Advanced Grant HUM-
VOL. P.H. was additionally supported by an ESRC Professorial Research
Fellowship. A.C. is a Research Director with the F.R.S.-FNRS (Belgium). This
work was partly supported by BELSPO IAP grant P7/33 and by ERC Advanced
Grant RADICAL to A.C.
Received: November 19, 2015
Revised: December 11, 2015
Accepted: December 23, 2015
Published: February 18, 2016
REFERENCES
1. Milgram, S. (1963). Behavioral study of obedience. J. Abnorm. Psychol.
67, 371–378.
2. Milgram, S. (1974). Obedience to Authority: An Experimental View (Harper
and Row).
3. Wegner, D.M. (2002). The Illusion of Conscious Will (The MIT press).
4. Bandura, A. (2006). Toward a psychology of human agency . Perspect.
Psychol. Sci. 1, 164–180.
5. Haggard, P., Clark, S., and Kalogeras, J. (2002). Voluntary action and
conscious awareness. Nat. Neurosci. 5, 382–385.
6. Engbert, K., Wohlschla
¨ger, A., and Haggard, P. (2008). Who is causing
what? The sense of agency is relational and efferent-triggered.
Cognition 107, 693–704.
7. Singer, T., Seymour, B., O’Doherty, J.P., Stephan, K.E., Dolan, R.J., and
Frith, C.D. (2006). Empathic neural responses are modulated by the
perceived fairness of others. Nature 439, 466–469.
8. Koban, L., Corradi-Dell’Acqua, C., and Vuilleumier, P. (2013). Integration
of error agency and representation of others’ pain in the anterior insula.
J. Cogn. Neurosci. 25, 258–272.
9. Lepron, E., Causse, M., and Farrer, C. (2015). Responsibility and the sense
of agency enhance empathy for pain. Proc. Biol. Sci. 282, 20142288.
10. Browning, C.R. (1998). Ordinary Men: Reserve Police Battalion 101 and
the Final Solution in Poland (Harper Perennial).
11. Haslam, S.A., and Reicher, S. (2007). Beyond the banality of evil: three dy-
namics of an interactionist social psychology of tyranny. Pers. Soc.
Psychol. Bull. 33, 615–622.
12. Barlas, Z., and Obhi, S.S. (2013). Freedom, choice, and the sense of
agency. Front. Hum. Neurosci. 7, 514.
13. Filevich, E., Ku
¨hn, S., and Haggard, P. (2013). There is no free won’t: ante-
cedent brain activity predicts decisions to inhibit. PLoS ONE 8, e53053.
14. Timm, J., SanMiguel, I., Keil, J., Schro
¨ger, E., and Scho
¨nwiesner, M.
(2014). Motor intention determines sensory attenuation of brain responses
to self-initiated sounds. J. Cogn. Neurosci. 26, 1481–1489.
15. Wittgenstein, L. (1953). Philosophical Investigations (Blackwell).
16. Synofzik, M., Vosgerau, G., and Newen, A. (2008). Beyond the comparator
model: a multifactorial two-step account of agency. Conscious. Cogn. 17,
219–239.
17. Blakemore, S.J., Wolpert, D., and Frith, C. (2000). Why can’t you tickle
yourself? Neuroreport 11, R11–R16.
18. Williams, S.R., Shenasa, J., and Chapman, C.E. (1998). Time course and
magnitude of movement-related gating of tactile detection in humans. I.
Importance of stimulus location. J. Neurophysiol. 79, 947–963.
19. Coombs, C.H. (1973). A reparameterization of the prisoner’s dilemma
game. Behav. Sci. 18, 424–428.
20. Foucault, M. (1977). Discipline and Punish: The Birth of the Prison (Allen
Lane).
21. Nathaniel-James, D.A., and Frith, C.D. (2002). The role of the dorsolateral
prefrontal cortex: evidence from the effects of contextual constraint in a
sentence completion task. Neuroimage 16, 1094–1102.
22. Tattersall, A.J., Bennett, P., and Pugh, S. (1999). Stress and coping in hos-
pital doctors. Stress Health 15, 109–113.
23. Johnsen, B.H., Laberg, J.C., and Eid, J. (1998). Coping strategies and
mental health problems in a military unit. Mil. Med. 163, 599–602.
24. Arendt, H. (1963). Eichmann in Jerusalem: A Report on the Banality of Evil
(Penguin).
25. Caspar, E.A., Cleeremans, A., and Haggard, P. (2015). The relationship be-
tween human agency and embodiment. Conscious. Cogn. 33, 226–236.
8Current Biology 26, 1–8, March 7, 2016 ª2016 The Authors
Please cite this article in press as: Caspar et al., Coercion Changes the Sense of Agency in the Human Brain, Current Biology (2016), http://dx.doi.org/
10.1016/j.cub.2015.12.067