1

Prefrontal connections express individual differences in honesty

Azade Seid-Fatemi1, Yosuke Morishima

2 3, Felix Heise

4, Carmen Tanner

5 6, Rajna Gibson

7,

Alexander F. Wagner5, Philippe N. Tobler

1

1 Department of Economics, University of Zurich

2 Division of Systems Neuroscience of Psychopathology, University Hospital of Psychiatry,

University of Bern

3 Japan Science and Technology Agency, PRESTO, Saitama, Japan

4 Department of Experimental Psychology, University of Oxford

5 Department of Banking and Finance, University of Zurich

6 Leadership Excellence Institute Zeppelin, Zeppelin University

7 Geneva Finance Research Institute, University of Geneva

Correspondence

Should be addressed to A.S. (azade.seid-fatemi@econ.uzh.ch) or P.N.T. (phil.tobler@econ.uzh.ch)

Laboratory for Social and Neural Systems Research

Department of Economics

University of Zurich

Blümlisalpstrasse 10

8006 Zürich, Switzerland

2

Abstract

Individual differences in honesty may reflect differences in the weighting of honesty-related and

economic values. These differences should be expressed in the decision network, particularly in

prefrontal cortex. However, the interactions between the core players of the decision network during

honesty-related decisions remain poorly understood. To investigate these processes we used functional

magnetic resonance imaging and measured brain activations while participants made decisions

concerning honesty. We found that in participants who valued honesty highly, the information flow

from the inferior frontal gyrus to the dorsolateral prefrontal cortex was increased when economic costs

were high compared to when they were low. Moreover, the interaction between these two brain

regions was specific for honesty-related as opposed to control decisions, suggesting a dedicated

mechanism for honesty-related decisions. We thereby provide a novel account of the neural circuitry

that underlies honest decisions in the face of economic temptations.

3

Decisions concerning honesty are one of the most common moral decisions individuals make, not only

in everyday situations but also in contexts that have great economic and political impact. For example,

CEOs of large companies can, and did, in the case of Enron, WorldCom and many others, resort to

dishonest practices, such as announcing higher earnings than the company actually achieved. Financial

fraud (e.g., the cases of Bernard Madoff and Jerome Kerviel) and banking scandals (e.g., the LIBOR

and FX market manipulations) have rocked the business world as well. Deceptive behavior can have

devastating consequences for employees, shareholders, and the economy at large, as evidenced by the

cases mentioned. Importantly, even before entering the realm of fraud, there is an area where moral

judgments can play a big role. For example, the choice whether to use judgment in financial reporting

with the intention of misleading stakeholders (i.e. manage earnings (Healy and Wahlen, 1999)) is

essentially a moral decision, given that the practice itself is legal, at least up to a point. Critically,

CEOs are tempted to be dishonest because they can improve their personal finances considerably

through bonuses tied to announced earnings or to share prices, which are, at least in the short run,

positively affected by the announcement of higher earnings. Conversely, and surprisingly from the

point of view of standard economics, some individuals are able to resist the financial temptation and

remain honest. Whistleblowers who reveal misconduct risk careers and friendship; journalists who

report the truth about human rights violations risk their lives.

As exemplified with decisions to manage earnings, decisions about honesty often require

weighing the benefits of an option against its costs. Thus, while most individuals value the mere act of

telling the truth, they often also care about the consequences of their actions. Importantly, individuals

differ in the extent to which they consider the economic consequences of honesty, which leads to

individual differences in the tendency to tell the truth. When the economic costs of telling the truth are

low, the benefits of being truthful will outweigh the costs for many individuals. However, when these

costs increase, some individuals will no longer be willing to incur them, while others will stand by

their moral principles and continue to disregard the economic costs of doing so. Thus, individual

differences in honesty should be more pronounced when the cost of telling the truth is higher (Gibson

et al., 2013).

4

At the neural level, there is substantial evidence from recent neuroimaging studies that the

prefrontal cortex is critically involved when individuals make decisions concerning honesty.

Particularly the dorsolateral prefrontal cortex (DLPFC), the dorsomedial prefrontal cortex (DMPFC),

the ventrolateral prefrontal cortex (VLPFC) and the inferior frontal gyrus (IFG) have been implicated

in honesty decisions (Abe, 2011; Abe and Greene, 2014; Baumgartner et al., 2009; Ganis et al., 2003;

Greene and Paxton, 2009; Langleben et al., 2005; Sip et al., 2008; Spence et al., 2001). However,

given that these regions have different functions and are anatomically connected, it is likely that

honesty decisions are implemented not by a single one of them, but through dynamic interactions

among them. Therefore, understanding the neural mechanisms underlying decisions involving honesty

makes it necessary not only to identify the activated brain regions in isolation, but also to illuminate

how these regions flexibly interact with each other as a function of situation and honesty-related

motives. These types of interactions can be captured through changes in strength or direction of the

functional coupling between brain regions (Friston et al., 1997). It is unknown whether connectivity

changes during honesty decisions are modulated by honesty-related motives and situational costs, even

though previous behavioural research clearly showed that these factors play a major role in honesty

decisions (Gibson et al., 2013; Gneezy, 2005). In the present study, we set out to close these gaps.

Our first aim was to investigate how neural connectivity patterns change as a function of the

economic cost of truthtelling and of the individual propensity to tell the truth. We used functional

magnetic resonance imaging (fMRI) to measure brain activations while participants made decisions

concerning truthfulness. In each trial, participants had to decide between an honest and a dishonest

option. Importantly, the dishonest option led to higher actual payoffs for the participant, such that

choosing the honest option incurred an economic cost compared to choosing the dishonest option.

Across trials we varied these costs, which in turn allowed for trading-off economic with honesty-

related incentives. We investigated connection parameters (strength and direction of coupling between

prefrontal regions) that help participants to more often protect their honesty-related values and resist

the lure of economic incentives when the costs of truthfulness are high.

Our second aim was to relate the functional interactions between prefrontal regions to a

central question in the study of morality, namely the extent to which moral and non-moral decision

5

making relies on similar or different neural mechanisms. While it has been argued that moral decision

making relies on common (“domain-general”) rather than dedicated (“domain-specific”) neural

mechanisms (Greene and Haidt, 2002; Shenhav and Greene, 2010; Tobler et al., 2008), there have also

been proposals that morality forms a domain of its own (Mikhail, 2007). If so, we hypothesized that

particularly individuals with strong honesty-related values should tap into this specific domain and

prefrontal connectivity should express it.

Results

Behavioral results

Honesty-related values predict decisions in different cost situations. While being scanned,

participants made decisions concerning honesty (truthtelling task). We aimed to create trade-offs

between moral and economic motives by adopting a previously established behavioral paradigm

(Gibson et al., 2013; Figure 1A), which has revealed a relation between moral motives and

individuals’ propensity to be honest. In particular, the paradigm creates a trade-off situation that CEOs

actually face in real life. CEO compensation is frequently tied to the earnings companies announce.

Participants assumed the role of a CEO and decided whether to announce true earnings, which would

reduce their compensation. Alternatively they could falsely (but legally) report higher earnings which

would result in a higher compensation. Thus, the moral motive to tell the truth competes with the

economic motive to achieve higher earnings. Our participants indeed traded off these two motives (SI

Results).

To assess moral domain-specificity versus domain generality, participants had to perform two

control tasks. In these tasks, too, the decisions the participants made as a CEO affected their

compensation (see Experimental procedures - Task instructions and payments). Importantly, in all

tasks the variable economic value difference between the two options was the same, but only the

truthtelling task involved a trade-off with honesty-related values. After scanning, participants

completed a questionnaire that assessed their honesty-related values regarding honest and dishonest

decisions with economic benefits (see Experimental Procedures - Measurement of honesty-related and

6

effort-related values). Specifically, our measure of honesty-related values emphasizes the idea that

individuals differ in the extent to which they regard honesty values as protected, that is, as priceless

and not subject to trade-offs.

We expected that strong honesty-related values should more often allow participants to resist

the temptation of not being truthful for economic benefits particularly when the economic costs for

telling the truth are high. Thus compared to weaker honesty-related values, strong honesty-related

values should lead to relatively more honest decisions in higher cost conditions. We assessed this

hypothesis in an interaction analysis using our honesty-related value questionnaire. In agreement with

the hypothesis, we observed a positive interaction between honesty-related values and economic costs

of truthfulness (β = 0.28 ± 0.13, t = 2.15, p < 0.05). In other words, when economic costs of

truthfulness were high, participants with stronger honesty-related values chose the truthful option

more often compared to participants with weaker honesty-related values (Figure 1B). Importantly, in

the effort and valuation task, we found no interaction effect such that participants with strong honesty-

related values chose the more costly option more often at higher costs (Figure S1 and SI Results -

Relation of honesty-related moral values to behavior in control tasks and to response times). Taken

together, these data suggest that strong honesty-related values reduced (but did not abolish)

participants’ willingness to trade off specifically honesty-related values against economic costs, in line

with the notion that the truthfulness task is specific for the domain of honesty-related decisions.

Neuroimaging results

DLPFC-IFG and DMPFC-IFG interactions increase with honesty-related values. Our behavioral

analyses indicated that compared to weaker values, stronger honesty-related values support honest

decisions as the cost of telling the truth increases. We therefore asked how stronger honesty-related

values affect neural interactions specifically when honest behavior is more costly. In particular, we

tested whether participants with stronger honesty-related values exhibit a specific prefrontal

connectivity pattern that promotes honest behavior. To do so, we first identified activation-based seed

regions where coding of the economic costs of telling the truth related to propensity for telling the

truth. The DLPFC and DMPFC coded the cost of telling the truth more strongly the more often

7

participants actually told the truth (both whole-brain FWE cluster-level corrected; Figure S2; SI

Experimental Procedures - Identification of the seed regions, and SI Results - Cost-related activations

in the DLPFC and the DMPFC for more details). We then investigated whether these seed regions

interact differently with other regions of the decision network as a function of cost and moral motives.

We therefore formed two first-level models with activity either from the DLPFC or the DMPFC seed

region as a physiological regressor, the cost-level as a psychological regressor (high vs. low cost), and

a psycho-physiological interaction (PPI) regressor corresponding to the product of the first two

regressors. The PPI regressor then served for a second-level correlation analysis with our participant-

specific measure of honesty-related values. We found that functional connectivity between the DLPFC

and the IFG as well as between the DMPFC and the IFG differed significantly as a function of cost-

level and honesty-related values (Figure 2A and B; DLPFC: -46, 28, 30; t(30) = 4.56; Figure 2C and D;

DMPFC: -44, 28, 28; t(30) = 5.12; p < 0.05, both whole-brain FWE cluster-level corrected; additionally,

whole-brain cluster-level corrected activation was found in the parietal cortex (-46, -32, 38;

t(30) = 4.29)). Specifically, both DLPFC and DMPFC showed stronger functional connectivity with the

IFG in high cost conditions compared to low cost conditions as honesty-related values increased

(Figure 2; see SI Results - Additional PPI analyses for control analyses). Thus, stronger honesty-

related values tightened the coupling between prefrontal decision regions particularly when the costs

for telling the truth were high.

DLPFC-IFG interactions are specific for honesty-based decisions. We next tested whether the

modulation of the DLPFC-IFG and DMPFC-IFG connectivity by honesty-related values is specific for

the truthtelling task by comparing the connectivity pattern with the other two tasks. The analysis was

based on the fact that in all tasks participants were confronted with the same economic cost-levels

when choosing between the honest, low effort, or low value option on the one hand and the dishonest,

high effort, or high value option on the other hand. Thus, the only additional component in the

truthtelling task was the moral nature of the decision made. We therefore included the two control

tasks in the PPI model with the cost-levels (high vs. low) of the effort and valuation task respectively

as psychological regressor and correlated the strength of connectivity with the individual differences in

8

honesty-related values for all tasks. We found that the cost-level dependent connectivity between

DLPFC and IFG related more strongly to honesty-related values in the truthtelling task as compared to

the other two tasks (Figure 3A and B; -46, 24, 16; t(60) = 4.52; p < 0.05, whole-brain FWE cluster-level

corrected). Confirming that the task-specific DLPFC-IFG connectivity identified here coincided with

the more general honesty value-dependent connectivity identified above (Figure 2), we found an

overlap between the two analyses (Figure 3C). Thus, DLPFC and IFG are particularly strongly

coupled in individuals with stronger honesty-related values when the economic costs are high in the

honesty task rather than in the control tasks. By contrast, the cost- and moral-value-related increase of

DMPFC connectivity with the IFG was not specific for the truthtelling task (p > 0.8). Thus, strong

honesty-related values appear to exert their honesty-promoting effects particularly via a connection

between the IFG and the DLPFC rather than the DMPFC.

Next, we contrasted the relation between the DLPFC-IFG coupling and individual differences

in honesty-related values in the truthtelling task against individual differences in effort-related, control

values. More specifically, we determined individual differences in subjects’ general tendency to trade-

off a non-honesty-related value (leisure time) against economic benefits in the effort task. This score

was assessed similarly as the honesty-related values score (Experimental Procedures). We aimed to

find out whether the DLPFC-IFG coupling relates more strongly to honesty-related values in the

truthtelling task than to effort-related values in the effort task. We therefore performed a second-level

correlation analysis in which we tested for a stronger correlation of cost-related connectivity strength

with the individual differences in honesty-related values in the truthtelling task than with individual

differences in effort-related values in the effort task. Albeit at a less stringent threshold, we found that

the cost-level dependent DLPFC-IFG coupling related more strongly to honesty-related values in the

truthtelling task than to effort-related values in the effort task (Figure S3; -30, 28, 14; t(29) = 3.67;

p < 0.001, uncorrected; resulting in overlap with the analysis of Figure 2 at -32, 26, 26). As these

results do not survive whole-brain correction, we refrain from drawing strong conclusions but they

tentatively suggest that primarily honesty-related values modulate prefrontal coupling during decisions

concerning honesty.

9

Connectivity from IFG to DLPFC is critical for honesty. Our PPI results show that in individuals

with stronger honesty-related values particularly the connection between the DLPFC and IFG is

crucial when honesty is costly. In order to better understand the nature of this coupling, in the final

step of our analysis we examined the directional effects of this connection by performing dynamic

causal modeling (DCM). As the cost-related DLPFC-IFG connection was crucial only for participants

with strong honesty-related values we performed the DCM analysis only in these participants. In

particular, we specified and estimated 3 model families that were grouped based on how the high cost

condition modulated the connection between the DLPFC and IFG (Figure 4A; see Experimental

Procedures, and SI Results - Additional DCM analysis for a refinement of the DCM analysis in which

we also consider different modulations of the DMPFC-IFG coupling). The first model family assumed

that high costs modulate the connection from the DLPFC to the IFG, the second assumed that they

modulate the connection from the IFG to the DLPFC and the third one that bidirectional connections

between these two regions are modulated by high costs. We used Bayesian model selection to identify

the most probable family of models. The most likely model family (family 2 in Figure 4A; exceedance

probability = 0.83) was the one with a unidirectional connection from the IFG to the DLPFC (Figure

4B). Finally, we used Bayesian model averaging to estimate the coupling parameters of the most likely

model family. We found that the coupling from the IFG to the DLPFC was positive (0.15 Hz) during

high cost conditions. This indicates that an increased rate of IFG activity under the high cost condition

results in an increase of DLPFC activity, suggesting that in high cost situations the activity in the

DLPFC is more responsive to signals in the IFG.

Discussion

In the present study we used two complementary methods, PPI and DCM, to reveal how task-related

connectivity relates to honesty costs and honesty-related values during truthtelling decisions. Our PPI

results show that in high, compared to low-cost situations, the DLPFC and DMPFC are more strongly

coupled with the IFG in individuals with stronger honesty-related values. Thus, individuals with

stronger honesty-related values may benefit from prefrontal interactions when the costs of honesty are

10

high. Importantly, the relation between DLPFC-IFG connectivity and individual differences in

honesty-related values was specific for honesty-related decisions, suggesting preferential involvement

of this link in moral as opposed to non-moral decisions. Furthermore, our DCM results specified the

direction of this interaction by showing that the coupling in participants with strong honesty-related

values was best explained by a model in which the DLPFC receives input from the IFG. Together, the

findings provide novel insights into how prefrontal connectivity underpins individual variability in

honesty-related values that promote honesty.

In line with previous behavioral research (Gibson et al., 2013) we find that both personal

factors (honesty-related values) and situational factors (cost-level) predict whether individuals behave

honestly. More specifically, when the economic costs of truthfulness were high, participants with

stronger honesty-related values chose the truthful option more often compared to participants with

weaker honesty-related values. This suggests that honesty-related values are instrumental in driving

truthful decisions particularly when truthfulness is more costly.

Our measure of honesty-related values assesses the extent to which individuals regard honesty

as protected and not subject to trade-offs. Thus, the present findings should be interpreted only with

respect to this type of honesty-related values. Indeed, different dimensions of honesty-related values

exist (Gibson et al., 2013) and supplementary analyses suggest that the prefrontal connectivity

modulations are specifically related to honesty-related values that are expressed by trade-off

reluctance as compared to honesty-related values that capture emotional aspects (see SI Results).

Therefore, future research should study the degree to which our conclusions extend to other concepts

of honesty-related values.

Given that the DLPFC, DMPFC and IFG have been consistently implicated in cognitive

control and response inhibition (Aron, 2007; Botvinick et al., 2001; Carter and van Veen, 2007; Miller

and Cohen, 2001), the increased connectivity between them suggests that an active control mechanism

might help bias honest individuals against dishonest behavior specifically at higher economic cost. We

note that this interpretation is different to that proposed by previous studies (Abe, 2011, 2009; Greene

and Paxton, 2009; Sip et al., 2008), which suggest that deceptive rather than honest behavior engages

the control network. The present study differs in several ways from the ones conducted previously.

11

First, in contrast to other studies (Abe et al., 2006; Ganis et al., 2003; Langleben et al., 2005; Spence et

al., 2001), our study involved spontaneous rather than instructed lying with a clear economic benefit

when choosing the dishonest option. Second, in a recent study that has investigated spontaneous lying

(Greene and Paxton, 2009), participants were given the opportunity to behave dishonestly by lying

about the accuracy of their predictions about the outcomes of computerized coin flips. Thus, in order

to lie participants had to override their previous prediction, which may require more cognitive control

than the lie decisions in our study. Third, in our study, participants were explicitly told that in real life

CEOs could report false earnings within accounting laws. Consequently, one could speculate that in

our experiment the default action may have been to choose the dishonest option and selecting the

honest option may therefore have required increased cognitive control. By extension, whether honesty

or dishonesty engages control regions would depend on which of the two represents the default action.

In line with this interpretation, recent reports have shown that active cognitive control can indeed

support honest behavior (Gino et al., 2011; Mead et al., 2009; Shalvi et al., 2012) and that disruption

of prefrontal control regions reduces honesty (Zhu et al., 2014).

It has been proposed that strong moral values can be derived from rules that prohibit certain

actions (Baron and Spranca, 1997). Such rules come in different forms (Ginges et al., 2007), in our

case for example the rule not to trade-off honesty for economic benefits. Indeed, the measure we use

for honesty-related values focuses on this idea of trade-off resistance. On a very basic level, these rules

might be stored as semantic knowledge, which is retrieved when individuals are faced with decisions

that are related to those rules. Interestingly, the IFG has been associated with semantic rule retrieval

and processing (Badre et al., 2005; Bunge, 2004; Souza et al., 2009). Our DCM analyses showed that

high truthtelling costs increased the connection from the IFG to the DLPFC. Thus, the IFG could

provide the DLPFC with input that represents semantic rules which in turn may bias participants

towards honesty irrespective of consequences. In agreement with the proposed role of the IFG,

activation in the IFG has been found to be stronger when individuals refuse to accept money for a

change in their previously reported moral views compared to when they accept (Berns et al., 2012)

(see also (Duc et al., 2013)). Our results reveal that the DLPFC-IFG connectivity in individuals with

12

strong honesty-related values is enhanced specifically when high economic costs are involved and thus

strong control mechanisms are needed to protect the honesty-related values.

Our results re-visit the question of whether moral decisions are represented by domain-

specific or domain-general neural mechanisms (Greene and Haidt, 2002; Mikhail, 2007). Our findings

suggest that individual differences in honesty-related values predict the strength of DLPFC-IFG

connectivity specifically in the honesty task, but not in the two control tasks. In other words, in

decisions concerning honesty particularly individuals with strong honesty-related values tapped into

the DLPFC-IFG connection. Thus, differences that arise from individually varying honesty-related

values specifically relate to connectivity patterns during honesty-related as opposed to non-honesty-

related decision making. Moreover, the cost-level dependent DLPFC-IFG coupling related more

strongly to honesty-related values in the truthtelling task than to effort-related values in the effort task.

This suggests that the cost-dependent coupling specifically relates to individual differences in honesty-

related values rather than non-honesty-related values. Thus, the way in which lateral prefrontal regions

interact with each other as a function of cost and honesty-related values can be preferentially related to

the moral aspects of decisions.

In contrast to the DLPFC-IFG connectivity the DMPFC-IFG connectivity did not show

specificity for moral values when comparing honesty-related decisions with non-moral control

decisions. It has been proposed that while the DLPFC is responsible for the regulative function of

control, the DMPFC plays a role in monitoring and specifying cognitive control (Botvinick et al.,

2004; Carter and van Veen, 2007; MacDonald et al., 2000; Shenhav et al., 2013). Thus, conflict

monitoring represented by honesty-related DMPFC-IFG connectivity appears to partly affect also

other domains of conflict, not only those involving moral aspects. In contrast, the implementation of

control as mediated by moral value-dependent information flow from the IFG to the DLPFC would

play a protective role specifically during honesty-related moral decisions. By extension, cognition

related to honesty-related motives has both domain-general (Greene and Haidt, 2002; Shenhav and

Greene, 2010; Tobler et al., 2008) and domain-specific (Mikhail, 2007) aspects, depending on the

specific mental function that is investigated.

13

Our control tasks did not contain a honesty-related component, and only the truthtelling task

involved moral decisions. It is therefore possible that our findings reflect more general characteristics

of moral decision making that are not limited to honesty. Therefore, future studies should investigate

whether our findings extend to other types of moral decisions or whether they are specific to honesty-

related decisions. Finally, one conceivable interpretation of our findings in the truthtelling task is that

participants may have perceived choosing the dishonest option as involving a reputation loss. This is

unlikely given that participants were told that choosing the dishonest option would be within legal

accounting limits. Moreover, if reputation concerns drove behavior in the truthtelling task, the effort

control task should have involved similar reputation concerns in that choosing the less demanding

option could be regarded as lazy. If anything, participants with strong moral values chose the low

effort option in the effort task more than participants with weak moral values (see Figure S1A),

suggesting that their behavior was not strongly driven by reputation concerns.

In conclusion, our results provide the first evidence that different prefrontal regions interact

during decisions concerning honesty. These interactions occur in a flexible manner, leading to

connectivity patterns that are situation-specific and depend on honesty-related motives. In particular,

individual differences in honesty-related values modulate DLPFC-IFG and DMPFC-IFG connectivity

in high cost situations. Moreover, we find that prefrontal connectivity patterns can be specific to

decisions concerning honesty, which is in line with domain-specificity. Our findings highlight the

relevance of studying prefrontal connectivity and provide an explanation why some people decide to

follow a moral principle whereas others do not. By extension, prefrontal coupling patterns could be

responsible for predisposing some CEOs to immoral behavior for high personal gains. Our findings

might also offer a framework to study the more extreme forms of immoral and deceptive decision

making in psychopathy and antisocial personality disorder.

Experimental Procedures

Participants

14

Thirty-two subjects (14 female; aged 22.0 ± 0.5 years; range: 18-28) took part in the experiment. None

of them had prior histories of neurological or psychiatric disorders and all had normal or corrected-to-

normal vision. The study was approved by the Research Ethics Committee of the Canton of Zurich,

and all subjects provided informed consent.

Experimental Design

Task instructions and payments. Participants performed a task that was adapted from a previously

established behavioral paradigm (Gibson et al., 2013). They were first asked to read the instructions

and answer control questions in order to ensure that they understood the tasks and the rules of the

experiment. Before entering the scanner the participants completed a series of practice trials. Inside the

scanner, participants performed the truthtelling task and two control tasks (the effort and the valuation

task) that were matched to the truthtelling task in terms of potential earnings and costs but did not

contain a moral component. Throughout the experiment, participants were placed in the situation of an

imaginary CEO who had to make various incentivized decisions. Participants received a fixed basic

payment of 25 Swiss Francs (CHF; at the time of the experiment, the exchange rate was about US $ 1

= CHF 0.91) and a variable additional payment that corresponded to the summed earnings from five

randomly selected trials per task and was determined with the actual decisions made in these trials.

The average variable payment was CHF 69.60.

In each trial of the truthtelling task (Figure 1A), participants had to announce the company’s

earnings per share for the previous quarter. The participants were informed that the variable

component of their payment as a CEO would be higher if they announced higher earnings. They were

also told that the market and the shareholders anticipated earnings of 35 cents per share, but that the

true earnings were 31 cents per share. Accordingly, participants could either announce 31 cents per

share at an economic cost to them (see below) or announce earnings of 35 cents per share while

knowingly remaining within legal accounting limits. Thus, participants had to choose whether to

honestly announce true earnings or to lie and falsely report higher earnings. Importantly, compared to

truthful reports, false reports led to higher actual payoffs for the participants, corresponding to a trade-

off between honesty-related and economic incentives. This setup parallels a real-life conflict that

15

CEOs face, as their variable compensation is often either tied to announced earnings directly or tied to

stock price performance, which in turn depends on the earnings they announce. Thus, despite knowing

that it is ethically wrong, the CEO has economic incentives to behave dishonestly and boost earnings

in order to increase his or her payout. Importantly, a CEO’s inclination to the truth should decrease as

the costs of telling the truth increase.

The bonus to be gained from giving truthful reports varied across trials between CHF 100,000

and 500,000. The bonus to be gained from giving false reports was fixed to CHF 500,000. These

amounts reflect the substantial monetary consequences that earnings management can have for CEOs

of public corporations. Participants were informed that for their experimental payoff the CEO bonus

would be converted into real money at a rate of CHF 100,000 = CHF 1.

In the two control tasks, too, the decisions the participants made as a CEO affected stock value

and therefore their compensation. In the effort task, participants chose how much work to invest as a

CEO. The participants were told that the more they worked the more the company would profit and

the more they would earn as a CEO. Specifically, in each trial participants decided how many simple

math problems to solve after the experiment outside of the scanner. They chose between solving only

one problem or five problems, which would take five times longer than solving one problem. Yet, the

five problems option typically led to a higher actual payment for the CEO (CHF 500,000) than the one

problem option (CHF 100,000 - 500,000), resulting in a trade-off between time and economic

incentives. In the valuation task, participants chose between two projects to invest in. The two projects

were similar in their properties (e.g. risk-free and realized after the same delay) but differed in their

profit and accordingly the CEO compensation. The high value project (labelled project “XIR” in the

task) typically led to a higher actual payment for the CEO (CHF 500,000) than the low value project

(labelled “ZEM”; CHF 100,000 - 500,000). Note that in all tasks and trials participants decided

between a variable (1-5 CHF) and a fixed payoff option (5 CHF).

Trial structure. In each trial of each task, participants first saw the variable, economically less

beneficial option followed by the constant, economically more beneficial (but, in the case of the

truthfulling task, false or, in the case of the effort task, more effortful option). The payoff of the first

options varied between 1 and 5 CHF whereas the second option always led to a fixed payoff of 5 CHF.

16

Therefore the cost of choosing the first option varied between 0 and 4 CHF. Both task type and

payoff-levels were randomly intermixed across trials during the experiment. Trials were separated

from each other by an intertrial interval (ITI) with a mean duration of 4 s, varying from 2 to 6 s

(Figure 1A). Each trial started with the presentation of a cue (1 s) shown at the top of the screen that

indicated which kind of task participants had to perform. The cue “Announce” referred to the

truthtelling task, “Calculate” to the effort task and “Perform” to the valuation task. Next, the first, i.e.

variable, option was shown for 3 s in the center of the screen. In the truthtelling task the true earning

(31 cents per share) was presented. In the effort task, the first option was one calculation and, in the

valuation task, the low value project. Below the option, the CEO compensation for that option was

indicated together with the corresponding participant payoff in parentheses. After an interstimulus

interval consisting of a blank screen (mean of 4 s, varying between 2 and 6 s), the second, i.e.

constant, option was presented, together with the first option. The second option was the false report

(35 cents per share) in the truthtelling task, the five calculations in the effort task, and the high value

project in the valuation task.

Participants were asked to make up their mind already during the presentation of the first

option. This was possible as only the first option varied across trials whereas the second option

remained constant. The decision was conveyed during the presentation of the second option. To

prevent motor preparation during presentation of the first option the two options were presented

randomly either on the left or the right side of the screen during the response phase. Participants had

2 s to indicate their choice by performing a button press. When subjects pressed a button, the color of

the written text on the screen changed from white to yellow to indicate that a response had been

recorded. To keep them engaged throughout the task, in each trial in which participants failed to

respond or responded too slowly, CHF 1 was deducted from their final monetary payment. These trials

were repeated later. On average the percentage of trials missed by participants was 1.5% ± 0.6%

(mean ± SEM; range 0 - 16%) with no significant difference between the tasks (F(1,32) = 0.35, p =

0.70). Over the course of the experiment, each participant completed 75 truthtelling trials, 75 effort

trials and 75 valuation trials. Each of the five payoff levels was presented 15 times per task. Stimulus

17

presentation and response recording were controlled using Cogent 2000 (Wellcome Department of

Imaging Neuroscience, London, UK) as implemented in Matlab.

Measurement of honesty-related and effort-related values. After scanning, we assessed participants’

honesty and effort-related values (Gibson et al., 2013; Hanselmann and Tanner, 2008) with two

questionnaires consisting of four items each. The questionnaires examined how much importance they

attributed to specific features of honesty/effort versus economic trade-offs (such as trade-off

reluctance, unwillingness to sacrifice a value, or incommensurability). That is, in line with our

research question on how agents differ in honest decision-making at high cost levels, these scales

capture the extent to which individuals consider the values in question as protected against trade-offs

or indeed “sacred” (Ginges et al., 2007; Hanselmann and Tanner, 2008). After recoding of some items,

an index of the degree of honesty/effort-related values was constructed, based on the means across the

items of each questionnaire. For ease of interpretation of the empirical results, we changed the scale to

range from 0 (for an individual with no honesty/effort-related values) to 6 (for an individual with

maximum honesty/effort-related values).

CEOs have an opportunity to modify information in the reports they provide to their shareholders.

Some view such modification as a violation of truthfulness, others regard it as acceptable protection of

personal interests. What do you think about the value of truthfulness in such a situation?

Truthfulness is something

1) … that one should not sacrifice, no matter what the (material or other) benefits.

strongly disagree 1 2 3 4 5 6 7 strongly agree

2) … for which I think it is right to make a cost-benefit analysis.

strongly disagree 1 2 3 4 5 6 7 strongly agree

3) … that cannot be measured in monetary terms.

strongly disagree 1 2 3 4 5 6 7 strongly agree

4) … about which I can be flexible if the situation demands it.

18

strongly disagree 1 2 3 4 5 6 7 strongly agree

Similarly, participants filled in a questionnaire assessing effort-related values, where subjects had to

indicate their views about spending more time to solve more math problems and earning more money.

The math problems are examples of decision situations in which individuals have to weigh between

time and money. Some people easily give up time in order to earn more money, however others don’t.

What do you think about the value of time in such a situation?

Time is something

1) … that one should not sacrifice, no matter what the (material or other) benefits.

strongly disagree 1 2 3 4 5 6 7 strongly agree

2) … for which I think it is right to make a cost-benefit analysis.

strongly disagree 1 2 3 4 5 6 7 strongly agree

3) … that cannot be measured in monetary terms.

strongly disagree 1 2 3 4 5 6 7 strongly agree

4) … about which I can be flexible if the situation demands it.

strongly disagree 1 2 3 4 5 6 7 strongly agree

Behavioral analysis

The effect of honesty-related values on choice behavior was analyzed using a series of logistic

regression analyses. For each of the analyses, the dependent variable corresponded to the decision in a

given trial. This was coded as a binary variable that took on the value of 1 if participants chose the

honest (low effort or low value) option, and the value of 0 if participants chose the dishonest (high

effort or high value) option. The independent variables (regressors) in the model included cost-level,

honesty-related values, and their interaction, while controlling for gender and age.

19

fMRI Data Acquisition

fMRI was performed on a Philips Achieva 3T whole-body scanner equipped with an eight-channel

head coil (Philips Medical Systems, Best, The Netherlands) at the Laboratory for Social and Neural

Systems Research, University of Zurich. We acquired gradient-echo T2*-weighted echoplanar images

(EPIs) with blood-oxygen-level-dependent (BOLD) contrast (slices/volume, 37; repetition time, 2 s).

Participants each completed five sessions of the experiment in the scanner, with short breaks between

each session. 315-360 volumes were collected per session (the variation was due to individual

differences in the number of repeated trials) along with 5 “dummy” volumes at the start of each

session to allow for magnetization to stabilize to a steady state. Scan onset times varied randomly

relative to stimulus onset times. Volumes were acquired at a 15° tilt to the anterior commissure-

posterior commissure line, rostral > caudal. Imaging parameters were the following: echo time, 30 ms;

field-of-view, 220 mm; in-plane resolution, 2.75 mm; slice thickness, 3 mm; interslice gap, 0.5 mm. A

T1-weighted structural image was also acquired for each participant. These high-resolution T1-

weighted structural scans were coregistered to their mean EPIs and averaged to permit anatomical

localization of the functional activations at the group level.

fMRI Data Analysis

fMRI data processing and statistical analyses were carried out using statistical parametric mapping

(SPM8; Wellcome Department of Imaging Neuroscience, London, UK). Data preprocessing consisted

of realignment, slice-time correction, coregistration, segmentation, spatial normalization using the

DARTEL toolbox and smoothing using a Gaussian kernel with a full width at half maximum of 10

mm.

To test for coupling differences due to variations in cost-level and honesty-related values we

conducted psycho-physiological interaction (PPI) analyses (Friston et al., 1997) using a generalized

PPI method (McLaren et al., 2012). Clusters in the DLPFC and DMPFC, in which activity for cost of

truthtelling correlated significantly (whole-brain FWE cluster-level corrected) with percent truthtelling

(SI Experimental Procedures for seed identification and SI Results), were used as seed regions. For

each subject, the average time series was extracted from these seed regions and this time series served

20

as a physiological regressor in a first-level general linear model for the PPI analysis. Moreover, we

entered cost-level as a psychological regressor (high cost, i.e. cost levels 3 and 4, vs. low cost, i.e. cost

levels 1 and 2), and a PPI regressor that corresponded to the product of the first two regressors. We

then used the PPI regressor to perform a second-level correlation analysis with individual honesty-

related values. Correction for multiple comparisons (familywise error, FWE; p < 0.05) was performed

in the whole brain at the cluster-level (cluster-inducing threshold: p < 0.005) unless otherwise stated.

To characterize prefrontal information flow in individuals with strong honesty-related values,

we examined the coupling direction between the IFG and the DLPFC by using dynamic causal

modelling (Friston et al., 2003) in the sixteen participants with honesty-related values above the

median (median = 3.88). We were specifically interested in these participants as the PPI analysis

revealed stronger cost-related connectivity with the IFG only in these subjects. For each participant,

we extracted activation time courses (principal eigenvariates) from a 4mm-radius sphere around the

peak activations identified in the preceding analyses. These regions were the DLPFC (-30, 56, 26),

DMPFC (-6, 16, 44) and the IFG (-44, 28, 28). Although we were primarily interested in the

information flow between the DLPFC and the IFG we also included the DMPFC in the model space as

our PPI analysis also identified this region to increase its connectivity with the IFG with respect to

cost-level and honesty-related values. The principal eigenvariates were adjusted for the F-contrast

capturing unspecific effects of interest.

At the outset, we ensured that all potential DCMs are anatomically plausible. This was indeed

the case, as direct connections between the three regions (DLPFC, DMPFC, and IFG) have been

demonstrated in humans and non-human primates (Barbas and Pandya, 1989; Beckmann et al., 2009;

Neubert et al., 2014; Petrides, 2005; Sallet et al., 2013; Yeterian et al., 2012). Further, we included

only models that contained a plausible information flow based on the combinatorial variations in

driving inputs and modulatory connections. Hence, we ensured that every region receives either a

driving input or input from another region and thus is not isolated from the information flow. For

example, we did not include a model (Figure S4) in which the driving input enters only the DLPFC

and the modulation of connectivity is from the IFG to the DLPFC but not in the inverse direction,

because modulation in the inverse direction should be present if the information entering only the

21

DLPFC is to exert any effect. Thus, the models in our DCM analysis were based on criteria of

functional and anatomical plausibility as well as internal consistency. As a result, we were left with 41

models. We categorized these 41 DCMs into three different families, with each family containing

between ten and seventeen models (depending on how many models ensured plausible information

flow). The categorization into families was based on connectivity patterns between the DLPFC and the

IFG (Figure 4A). The modulatory effect (modelled as a stick function) consisted of high cost

conditions and the families differed in the way by which this modulatory effect influenced the

connections between the DLPFC and the IFG (either from DLPFC to IFG, the IFG to the DLPFC or

bidirectional). Each family varied in terms of the modulatory connection between the DMPFC and the

IFG (either from DMPFC to IFG, from the IFG to the DMPFC or bidirectional) and the driving input.

Driving inputs represent extrinsic parameters that change the neuronal state of brain regions within a

model (here, visual presentation of the first option). We specified driving inputs either into the

DLPFC, the DMPFC, the IFG or any combination of these regions. We identified the most likely

family of models using Bayesian model selection (Penny et al., 2010; Stephan et al., 2009). This

allowed us to compare the model families based on their exceedance probabilities, a measure of

whether a model family is more likely to be correct than all other model families. Lastly, we used

Bayesian model averaging (Kasess et al., 2010) for the most likely family of models to compute the

magnitudes of coupling parameters.

References

Abe, N., 2011. How the brain shapes deception: an integrated review of the literature. Neuroscientist

17, 560–574.

Abe, N., 2009. The neurobiology of deception: evidence from neuroimaging and loss-of-function

studies. Curr. Opin. Neurol. 22, 594–600.

Abe, N., Greene, J.D., 2014. Response to anticipated reward in the nucleus accumbens predicts

behavior in an independent test of honesty. J. Neurosci. 34, 10564–10572.

22

Abe, N., Suzuki, M., Tsukiura, T., Mori, E., Yamaguchi, K., Itoh, M., Fujii, T., 2006. Dissociable

roles of prefrontal and anterior cingulate cortices in deception. Cereb. Cortex 16, 192–199.

Aron, A.R., 2007. The neural basis of inhibition in cognitive control. Neuroscientist 13, 214–228.

Badre, D., Poldrack, R.A., Paré-Blagoev, E.J., Insler, R.Z., Wagner, A.D., 2005. Dissociable

controlled retrieval and generalized selection mechanisms in ventrolateral prefrontal cortex.

Neuron 47, 907–918.

Barbas, H., Pandya, D.N., 1989. Architecture and intrinsic connections of the prefrontal cortex in the

rhesus monkey. J. Comp. Neurol. 286, 353–375.

Baron, J., Spranca, M., 1997. Protected values. Organ Behav Hum Dec 70, 1–16.

Baumgartner, T., Fischbacher, U., Feierabend, A., Lutz, K., Fehr, E., 2009. The neural circuitry of a

broken promise. Neuron 64, 756–770.

Beckmann, M., Johansen-Berg, H., Rushworth, M.F.S., 2009. Connectivity-based parcellation of

human cingulate cortex and its relation to functional specialization. J. Neurosci. 29, 1175–

1190.

Berns, G.S., Bell, E., Capra, C.M., Prietula, M.J., Moore, S., Anderson, B., Ginges, J., Atran, S., 2012.

The price of your soul: neural evidence for the non-utilitarian representation of sacred values.

Philos Trans R Soc Lond B Biol Sci 367, 754–762.

Botvinick, M.M., Braver, T.S., Barch, D.M., Carter, C.S., Cohen, J.D., 2001. Conflict monitoring and

cognitive control. Psychol Rev 108, 624–652.

Botvinick, M.M., Cohen, J.D., Carter, C.S., 2004. Conflict monitoring and anterior cingulate cortex:

an update. Trends Cogn. Sci. 8, 539–546.

Bunge, S.A., 2004. How we use rules to select actions: a review of evidence from cognitive

neuroscience. Cogn Affect Behav Neurosci 4, 564–579.

Carter, C.S., van Veen, V., 2007. Anterior cingulate cortex and conflict detection: an update of theory

and data. Cogn Affect Behav Neurosci 7, 367–379.

Duc, C., Hanselmann, M., Boesiger, P., Tanner, C., 2013. Sacred values: Trade-off type matters.

Journal of Neuroscience, Psychology, and Economics 6, 252–263.

23

Friston, K.J., Buechel, C., Fink, G.R., Morris, J., Rolls, E., Dolan, R.J., 1997. Psychophysiological and

modulatory interactions in neuroimaging. Neuroimage 6, 218–229.

Friston, K.J., Harrison, L., Penny, W., 2003. Dynamic causal modelling. NeuroImage 19, 1273–1302.

Ganis, G., Kosslyn, S.M., Stose, S., Thompson, W.L., Yurgelun-Todd, D.A., 2003. Neural correlates

of different types of deception: an fMRI investigation. Cereb. Cortex 13, 830–836.

Gibson, R., Tanner, C., Wagner, A., 2013. Preferences for truthfulness: heterogeneity among and

within individuals. Am. Econ. Rev. 103, 532–548.

Ginges, J., Atran, S., Medin, D., Shikaki, K., 2007. Sacred bounds on rational resolution of violent

political conflict. Proc. Natl. Acad. Sci. U.S.A. 104, 7357–7360.

Gino, F., Schweitzer, M.E., Mead, N.L., Ariely, D., 2011. Unable to resist temptation: How self-

control depletion promotes unethical behavior. Organ Behav Hum Dec 115, 191–203.

Gneezy, U., 2005. Deception: The Role of consequences. Am. Econ. Rev. 95, 384–394.

Greene, J.D., Paxton, J.M., 2009. Patterns of neural activity associated with honest and dishonest

moral decisions. Proc. Natl. Acad. Sci. U.S.A. 106, 12506–12511.

Greene, J., Haidt, J., 2002. How (and where) does moral judgment work? Trends Cogn. Sci. 6, 517–

523.

Hanselmann, M., Tanner, C., 2008. Taboos and conflicts in decision making: Sacred values, decision

difficulty and emotions. Judgm Decis Mak 3, 51–63.

Healy, P.M., Wahlen, J.M., 1999. A review of the earnings management literature and its implications

for standard setting. Accounting Horizons 13, 365–383.

Kasess, C.H., Stephan, K.E., Weissenbacher, A., Pezawas, L., Moser, E., Windischberger, C., 2010.

Multi-subject analyses with dynamic causal modeling. Neuroimage 49, 3065–3074.

Langleben, D.D., Loughead, J.W., Bilker, W.B., Ruparel, K., Childress, A.R., Busch, S.I., Gur, R.C.,

2005. Telling truth from lie in individual subjects with fast event-related fMRI. Hum Brain

Mapp 26, 262–272.

MacDonald, A.W., Cohen, J.D., Stenger, V.A., Carter, C.S., 2000. Dissociating the role of the

dorsolateral prefrontal and anterior cingulate cortex in cognitive control. Science 288, 1835–

1838.

24

McLaren, D.G., Ries, M.L., Xu, G., Johnson, S.C., 2012. A generalized form of context-dependent

psychophysiological interactions (gPPI): A comparison to standard approaches. Neuroimage

61, 1277–1286.

Mead, N.L., Baumeister, R.F., Gino, F., Schweitzer, M.E., Ariely, D., 2009. Too tired to tell the truth:

self-control resource depletion and dishonesty. J Exp Soc Psychol 45, 594–597.

Mikhail, J., 2007. Universal moral grammar: theory, evidence and the future. Trends Cogn. Sci.

(Regul. Ed.) 11, 143–152.

Miller, E.K., Cohen, J.D., 2001. An integrative theory of prefrontal cortex function. Annu. Rev.

Neurosci. 24, 167–202.

Neubert, F.-X., Mars, R.B., Thomas, A.G., Sallet, J., Rushworth, M.F.S., 2014. Comparison of human

ventral frontal cortex areas for cognitive control and language with areas in monkey frontal

cortex. Neuron 81, 700–713.

Penny, W.D., Stephan, K.E., Daunizeau, J., Rosa, M.J., Friston, K.J., Schofield, T.M., Leff, A.P.,

2010. Comparing families of dynamic causal models. PLoS Comput. Biol. 6, e1000709.

Petrides, M., 2005. Lateral prefrontal cortex: architectonic and functional organization. Philos Trans R

Soc Lond B Biol Sci 360, 781–795.

Sallet, J., Mars, R.B., Noonan, M.P., Neubert, F.-X., Jbabdi, S., O’Reilly, J.X., Filippini, N., Thomas,

A.G., Rushworth, M.F., 2013. The organization of dorsal frontal cortex in humans and

macaques. J. Neurosci. 33, 12255–12274.

Shalvi, S., Eldar, O., Bereby-Meyer, Y., 2012. Honesty requires time (and lack of justifications).

Psychol Sci 23, 1264–1270.

Shenhav, A., Botvinick, M.M., Cohen, J.D., 2013. The expected value of control: an integrative theory

of anterior cingulate cortex function. Neuron 79, 217–240.

Shenhav, A., Greene, J.D., 2010. Moral judgments recruit domain-general valuation mechanisms to

integrate representations of probability and magnitude. Neuron 67, 667–677.

Sip, K.E., Roepstorff, A., McGregor, W., Frith, C.D., 2008. Detecting deception: the scope and limits.

Trends Cogn. Sci. 12, 48–53.

25

Souza, M.J., Donohue, S.E., Bunge, S.A., 2009. Controlled retrieval and selection of action-relevant

knowledge mediated by partially overlapping regions in left ventrolateral prefrontal cortex.

Neuroimage 46, 299–307.

Spence, S.A., Farrow, T.F., Herford, A.E., Wilkinson, I.D., Zheng, Y., Woodruff, P.W., 2001.

Behavioural and functional anatomical correlates of deception in humans. Neuroreport 12,

2849–2853.

Stephan, K.E., Penny, W.D., Daunizeau, J., Moran, R.J., Friston, K.J., 2009. Bayesian model selection

for group studies. Neuroimage 46, 1004–1017.

Tobler, P.N., Kalis, A., Kalenscher, T., 2008. The role of moral utility in decision making: An

interdisciplinary framework. Cogn Affect Behav Neurosci 8, 390–401.

Yeterian, E.H., Pandya, D.N., Tomaiuolo, F., Petrides, M., 2012. The cortical connectivity of the

prefrontal cortex in the monkey brain. Cortex, Frontal lobes 48, 58–81.

Zhu, L., Jenkins, A.C., Set, E., Scabini, D., Knight, R.T., Chiu, P.H., King-Casas, B., Hsu, M., 2014.

Damage to dorsolateral prefrontal cortex affects tradeoffs between honesty and self-interest.

Nat Neurosci 17, 1319–1321.

Acknowledgments

We thank Todd Hare for helpful discussions. This work was supported with funding from the Swiss

National Science Foundation (PP00P1_128574 and PP00P1_150739). We would also like to

acknowledge the Neuroscience Center Zurich (ZNZ) and the Zurich Center for Integrative Human

Physiology (ZIHP).

Author contributions

Azade Seid-Fatemi (A.S.), Felix Heise (F.H.), Alexander F. Wagner (A.F.W.), Carmen Tanner (C.T.),

Rajna Gibson (R.G.) and Philippe N. Tobler (P.N.T) designed research; A.S. and F.H. performed

26

research; A.S., Yosuke Morishima (Y.M.), F.H., A.F.W., and P.N.T. analyzed data; A.S., Y.M., C.T.,

R.G., A.F.W. and P.N.T. wrote the paper.

Competing financial interests

The authors declare no competing financial interests.

Figure Legends

Figure 1. Experimental design and behavioral results. (A) In each trial of each task, participants first

viewed a fixation cross for a variable ITI of 2 – 6 s followed by the presentation of a cue (1 s) that

indicated which kind of task participants had to perform. The first, variable option was then shown for

3 s. In the truthtelling task the true earning (31 cents per share) was presented, in the effort task the

option for solving one problem was shown and in the valuation task the option for the low value

project (project ZEM). Below the option the CEO compensation was indicated together with the

corresponding participant payoff in parentheses. The payoff of the first options varied between 1 and 5

CHF. After an interstimulus interval (2 - 6 s) the second, constant option (5 CHF) was presented

together with the first option. The second option was the false report (35 cents per share) in the

truthtelling task, the option for performing five calculations in the effort task and the high value

project in the valuation task (project XIR). Upon presentation of the second option, participants had 2

s to indicate their choice by performing a button press. When subjects pressed a button, the color of

the written text on the screen changed from white to yellow to indicate that a response had been

recorded. (B) The difference in the percentage of truthtelling between individuals with stronger and

weaker honesty-related values increased as a function of cost. When the economic costs of truthfulness

were high, participants with stronger honesty-related values (upper tercile) were more honest

compared to those with weaker honesty-related values (lower tercile), suggesting that honesty-related

values are more important in determining truthful decisions when the costs of truthfulness are higher.

Error bars indicate SEMs.

27

Figure 2. DLPFC-IFG and DMPFC-IFG interactions increase with truthtelling costs particularly in

individuals with stronger honesty-related values. Functional connectivity between the honesty-related

seed regions DLPFC and DMPFC (shown in smaller scale) and the IFG (DLPFC ↔ IFG: -46, 28, 30;

t(30) = 4.56; DMPFC ↔ IFG: -44, 28, 28; t(30) = 5.12; both whole-brain FWE cluster-level corrected)

was increased in high compared to low cost conditions as honesty-related values increased. Voxels in

red show correlation with the DLPFC seed, voxels in blue show correlation with the DMPFC seed,

and voxels in purple show the overlap between the correlations of the two seed regions. Throughout,

activations are displayed at the uncorrected level of p < 0.001 for visualisation purposes. The left

graph shows individual connectivity strengths for the DLPFC-IFG interaction and the right graph

shows individual connectivity strengths for the DMPFC-IFG interaction

Figure 3. DLPFC-IFG interaction patterns are specific for the truthtelling task. (A) The increased

connectivity between the DLPFC and the IFG for high compared to low cost conditions related more

strongly to honesty-related values in the truthtelling task as compared to the two control tasks (-46, 24,

16; t(60) = 4.52; p < 0.05, whole-brain FWE cluster-level corrected). (B) The plot shows the average

relationship (lines ± SEM as shaded area) between the cost-related DLPFC-IFG connectivity strength

in the truthtelling (red), the effort (blue), and the valuation (green) task and honesty-related values. (C)

Voxels in red show functional connectivity between the honesty-related seed region DLPFC and the

IFG (shown in figure 2A), voxels in green show DLPFC-IFG interactions specific for the truthtelling

task (shown in figure 3A) and voxels in yellow show the overlap (between figure 2A and figure 3A).

Figure 4. DCM models and BMS results. (A) Illustration of 3 alternative model families with changes

in connectivity modulated by high cost. Red arrows represent potential modulation of coupling

between the DLPFC and IFG. Models within a family differed in terms of the coupling between

DMPFC and IFG (grey dashed line: modulatory effect influenced the connection either from DMPFC

to IFG, the IFG to the DMPFC or bidirectionally) and driving input region (grey dashed arrows:

driving inputs could be either into the DLPFC, the DMPFC, the IFG or any combination of these

28

regions). (B) Bar graph showing the exceedance probability for each model. The labels on the x-axis

correspond to the numbers assigned to each model family in A.

29

Figure 1

Figure 2

30

Figure 3

Figure 4