Post-execution monitoring in dishonesty

Agents seem to dedicate considerable effort into monitoring the success of any of their responses (e.g., Jentzsch et al., 2007; Welford, 1952) though monitoring is likely to be especially pronounced for dishonest responses. An elaborate theory on lying, the Action–Decision–Construction–Action Theory (ADCAT), proposes that monitoring of dishonest responses serves two purposes: ensuring that own behavior unfolds as intended and inferring the believability of the lie from the recipient’s behavior (Walczyk et al., 2014). Whereas monitoring of the persuasiveness of a lie might be a deliberate, motivated process, the former type of monitoring may emerge automatically as a direct and immediate consequence of the cognitive operations mediating dishonest behavior. These cognitive operations are usually studied in highly controlled paradigms, where participants are prompted to respond honestly and dishonestly to simple questions as fast and accurately as possible via keypresses (e.g., Furedy, et al., 1988; Suchotzki et al., 2017).

To understand why monitoring could differ in extent between dishonest and honest responses, we propose to consider the underlying conflicting nature of dishonesty. Delivering a dishonest response comes with the obstacle of overcoming the initially, automatically activated truthful response in most situations (e.g., Duran et al., 2010; Walczyk et al., 2003). It seems that this initial activation of the truthful response is not only an unwelcome side effect, but rather an integral part of cognitive processing during dishonesty: The presentation of honest rather than dishonest response options as irrelevant distractors has been shown to facilitate honest and dishonest responding alike (Debey et al., 2014; Foerster et al., 2017a, 2017b). The activation of competing action plans is the defining feature of cognitive conflict which, in turn, activates cognitive control processes (e.g., Botvinick et al., 2001). According to conflict-monitoring theory, the detection of any conflict should thus trigger control adjustments to avoid or deal with similar conflicts in the future. Crucially, cognitive conflict is a necessary by-product of responding dishonestly in the face of an automatically activated truthful response, whereas honest responding does not come with a similar degree of conflict.

This perspective highlights striking similarities of honest vs. dishonest responding with the comparison of cognitive consequences of error commission: Committing an error should elicit a conflict between the intended and the delivered response, whereas no such conflict arises for typical correct responses. Because errors are known to trigger monitoring processes as a direct consequence of this conflict (accompanied by additional post-error adjustments; Jentzsch, & Dudschig, 2009; Steinhauser et al., 2017), it seems plausible to assume structural similarities for dishonesty (Foerster et al., 2018, 2019). That is, both dishonest responses and errors come with the parallel activation of competing action tendencies and the resulting conflict may trigger monitoring. These monitoring processes assess whether a response unfolds as intended, and they may lead to control adaptations that aim for more efficient and accurate (dishonest) responses in the future.

While response monitoring has been considered an integral part of information processing (Welford, 1952), little research has addressed the question which information processing stages are concerned with such monitoring (e.g., Pashler, 1994). Classical stage theory confines the processing between stimulus presentation and response execution, dividing it into a precentral, a central and a postcentral stage. The precentral and postcentral stages map mostly to perceptual and motor processes, respectively, and can operate in parallel with processing stages of another task. In contrast, the central stage is assumed to be capacity-limited and, amongst other things, devoted to response selection (for more fine-grained analyses, see Hommel, 1998; Janczyk et al., 2014). As monitoring is often proposed to be capacity-limited (Jentzsch & Dudschig, 2009; Jentzsch et al., 2007), central processes of another task should not be able to operate in parallel with ongoing monitoring.

Evidence for capacity-limited monitoring during dishonest responding comes from psychological refractory period methodology (Experiment 3 in Foerster et al., 2019). Participants responded honestly and dishonestly to yes/no questions about daily activities in a first task, and to the pitch of a tone in a second task, with a short or long stimulus-onset asynchrony between question and tone. The assumption that dishonest responding requires a longer central processing stage than honest responding predicts that the resulting response time (RT) costs in Task 1 could also affect processing of the nominally unrelated Task 2 by postponing its capacity-limited central stage (see Foerster et al., 2019, for a detailed argument). This was indeed the case, and this finding is consistent with the idea that overcoming the activated truthful response itself is capacity-limited. Crucially, the RT costs for dishonesty were even larger in Task 2 than in Task 1 if both tasks occurred in close succession (i.e., at a short stimulus-onset asynchrony). This latter finding seems to suggest that capacity-limited monitoring may be recruited in addition to the costs of overcoming the automatically activated truthful response in Task 1.

The described experimental setup with temporally overlapping tasks may reveal monitoring costs, but it cannot assess whether the assumed processes are still operating after response execution. The same pattern of results would emerge if either monitoring accompanied the postcentral stage of Task 1 but terminated at the time of responding, or if monitoring exceeded response execution. Here we propose to adopt a critical change in the experimental paradigm to distinguish between both theoretical possibilities: Presenting the same two tasks without temporal overlap, but rather in close temporal succession (see Fig. 1 for an illustration). Such paradigms delivered evidence for increased capacity-limited monitoring of erroneous responses after response execution (Jentzsch & Dudschig, 2009), whereas no such findings of prolonged monitoring emerged for dishonest responding (Experiment 4 in Foerster et al., 2019).

To conclude, tried and tested stage-oriented approaches allowed to examine monitoring of dishonest responding. Initial evidence from paradigms with overlapping tasks supports the notion of temporally extended monitoring processes with dishonest compared to honest responding but does not inform about the extent and point in time of these processes. Specifically, whether monitoring exceeds response execution can be investigated when tasks are presented in succession, but a first attempt did not show any differences in monitoring between honest and dishonest responding (Foerster et al., 2019).

Why did previous research fail to demonstrate extended monitoring of dishonest responses after response execution? A first possibility is that monitoring of dishonesty indeed terminates with response execution. Second, participants received explicit accuracy feedback in Task 1 in previous studies (Foerster et al., 2019). In case of a Task 1 error, no stimulus for Task 2 appeared, but a new trial started. The stimulus onset of Task 2 thus indicated that (dis)honest responding was successful and this feedback signal might have aborted any monitoring processes to determine the success of the response. Third, the monitoring process itself might be relatively short-lived. If it did exceed responding in Task 1 but not the precentral processing of Task 2, it could not have delayed the subsequent capacity-limited central processing of Task 2 either, and thus could not have showed up in the RT of Task 2.

The current experiment investigated whether dishonest responses are monitored more extensively than honest responses even after response execution has already finished. We assessed monitoring of (dis)honest responses of an Intention Task 1 via RTs in an unrelated Tone Task 2. We expected RT2 to be longer after dishonest than honest responses if Task 2 succeeded promptly after responding in Task 1 in blocks with relative uncertainty about response success in Task 1.

The obtained evidence is in line with the assumption of increased monitoring of dishonest responses after response execution. Dishonest responses were not only slower and more error-prone than honest responses in the Intention Task 1, but this influence propagated to responding in the Tone Task 2 at the short response–stimulus interval. This qualifies past results, which suggested the existence of capacity-limited monitoring of dishonest responses, but which did not support the notion of monitoring after response execution (Foerster et al., 2019).

A modulation of post-response monitoring by response uncertainty in the current experiment would have delivered a good reason for these diverging results. However, monitoring effects emerged equally in blocks with early and late error feedback for (dis)honest responses, although there was a descriptive trend toward smaller monitoring costs with immediate than late feedback. For one, this might suggest that either the absent monitoring effects in the literature is a beta error (Exp. 4 in Foerster et al., 2019), or the significant monitoring effects here is the result of an alpha error. The highly powered replication of the monitoring costs in Experiment 3 argues strongly against an alpha error here.

Future studies on monitoring should expand on the role of response uncertainty for post-response monitoring in general. In the current study, we could not explore the impact of response uncertainty via the timepoint of error feedback on monitoring after errors, because the early feedback condition increased the response–stimulus interval between tasks by 1500 ms. Any monitoring effects should have been completed in this extended period and, therefore, before Task 2 commenced. In addition, monitoring costs after the short RSI were small in our study. To address both issues, researchers might modulate response uncertainty via other modalities than the one that both tasks are presented in, for example, using keys that randomly do or do not move when being pressed. Such a manipulation of tactile instead of visual feedback would allow for an immediate presentation of Task 2 stimuli after responding to Task 1, enabling the assessment of monitoring effects. Furthermore, minimized proximal feedback by removing tactile response feedback, might boost monitoring processes. Finally, larger sample sizes would capture potentially smaller effects of response uncertainty on monitoring.

Although monitoring costs can become evident in error rates (Jentzsch & Dudschig, 2009; Steinhauser et al., 2017), they did not emerge here, nor in preceding investigations on dishonesty (Experiment 3 in Foerster et al., 2019). The absence of an effect in all these studies could be due to a floor effect, because error rates in Task 2 were consistently low. In any case, the resulting error rates do not suggest a speed–accuracy trade-off at the heart of monitoring effects.

Responses in the Tone Task 2 became more efficient with an increase in response–stimulus interval. These results were expected but not informative about our hypothesis. We had no predictions about the observed interactions in error rates and RTs of the Intention Task 1. Because these seem unproblematic for our hypothesized effects in the Tone Task 2, we would like to refrain from interpreting them.

We keep this section brief by reporting only aspects, where Experiment 2 deviated from Experiment 1.

A comparison of the mean RT of the tone (Experiment 1: M = 431 ms) and letter (Experiment 2: M = 476 ms) tasks for the late error feedback condition with a response–stimulus interval of 1000 ms suggests that we might have failed to find a robust monitoring effect, because we did not efficiently reduce precentral processing through using visual instead of auditory stimuli. Even though, mean response times represent the sum of precentral, central, and postcentral processes, this difference in RTs hints rather to an increase than to a reduction of the precentral stage.

The absent modulation of the elusive monitoring costs suggests that any potential aftereffects on the letter task might not only have been the result of early monitoring but also of late control adjustments (Jentzsch & Dudschig, 2009). Such a late mechanism of control adaptation and its transfer from (dis)honest responding to other (conflict) tasks needs further examination in the light of mixed results from this study (Experiments 1 vs. 2) and from published data (Foerster et al., 2017a, 2017b, 2018).

Experiment 3 was a high-powered replication of the experimental conditions that promised the highest likelihood to detect monitoring costs after response execution, and it did reveal such an effect. Taken together, we are confident to conclude that dishonest responses trigger monitoring even after response execution to a higher degree than honest responses.

We proposed conflict between honest and dishonest response activation to be at the heart of the prolonged monitoring processes of dishonest responses, and we employed a paradigm that allowed for an isolated examination of this difference (Furedy et al., 1988). This close tie of cognitive conflict and monitoring would predict that a reduction in conflict, for example, through practice, preparation of false alibis or recent dishonest responding (Foerster et al., 2017a, 2017b; Hu et al., 2012; Suchotzki et al., 2018; Van Bockstaele et al., 2012), also diminishes monitoring costs. Whenever honest responses are more conflicting than dishonest responses, e.g., because dishonest responses had been practiced or prepared thoroughly, honest responses should trigger stronger monitoring than dishonest responses. In contrast, rendering dishonest response selection more difficult by having more than two response alternatives might boost monitoring of the success of dishonest responses. Such manipulations of the complexity of dishonest responding will, therefore, allow for assessing the role of monitoring for dishonesty more comprehensively.

On the other hand, responding is also more difficult in the presence of cognitive conflict leading to less accurate and slower dishonest performance (Suchotzki et al., 2017). As such, increased monitoring of dishonest responses could either be a specific consequence of cognitive conflict, or it could relate to response difficulty in general (Foerster et al., 2019). A potential first step to approach this issue in dishonesty could be to correlate monitoring costs with conflict strength, measured via a tried and tested distractor procedure (Debey et al., 2014; Foerster et al., 2017a, 2017b). Speculatively, agents might explicitly notice that they often struggle to respond efficiently and accurately when being dishonest and this knowledge might then boost monitoring. This perspective calls for an examination of the history of the success in responding honestly and dishonestly on monitoring. In any case, the current study points to a greater role of internal conflict and feedback processes than of the incorporation of explicit external feedback for the emergence of extended monitoring.

A further open issue pertains to the functional significance of post-response monitoring. One interpretation would be that these monitoring efforts still aim at ensuring that a response had been delivered as intended, assuming that such processing simply takes longer for dishonest responses than for honest responses, because more conflicting action tendencies emerge in action planning of dishonest responses. An alternative interpretation would hold that post-response monitoring rather aims at detecting additional, response-contingent changes in the environment. This latter interpretation receives support from studies on action–effect monitoring which indicate that monitoring extends to the immediate consequences of one’s own actions (Kunde et al., 2018; Wirth et al., 2018). Such a mechanism would allow for efficient detection of unintended consequences of one’s own responses, which are of particular relevance in case of dishonesty. Work on action–effect monitoring also indicates that the monitoring of own responses and monitoring of their effects seems to rely on common processes (Steinhauser et al., 2018). Whether and how such automatic monitoring processes interface with strategic and deliberate attempts at monitoring the behavior of recipients of one’s own lie (e.g., Walczyk et al., 2014) is an open question. In any case, the motivation to lie (believably) likely boosts monitoring processes (DePaulo et al., 1988; Porter & ten Brinke, 2010; Walczyk et al., 2014). Investigations about this relation should clearly distinguish between effects of motivation on response monitoring that are specific to dishonesty and effects that apply to more general behavior as well. Motivation might affect monitoring of dishonest responses directly if people assume that dishonest compared to honest responding changes their demeanor markedly and detectably or if they anticipate that the recipient of their lie critically assesses the truth value of their statements. An indirect link could be that highly motivated liars might experience less conflict and this, in turn, could reduce monitoring requirements.

Both proposed mechanisms might entail the assessment of the strength of conflict (i.e., between behavioral tendencies or between expected and encountered consequences) and this perspective suggests that monitoring might be a precursor for adaptation to dishonest responding. Akin to adaptation to other behavioral conflicts as in the Stroop or the Eriksen task, differences between dishonest and honest responding are modulated by (dis)honesty in the preceding action episode (Foerster et al., 2018). In particular, these differences vanished or reversed after a dishonest response, indicating a transition to a processing mode that facilitated overcoming the initial honest response tendency, rendering dishonest responses faster and less error-prone. Re-analyses of the current experiments replicated that effects of (dis)honesty in Task 1 were smaller after a dishonest response than after an honest response in Task 1 of the preceding trial, despite the intervening Task 2. Accordingly, the strength of conflict between honest and dishonest response tendencies might affect the extent of monitoring, and these monitoring processes might in turn trigger adaptation. Alternatively, conflict strength might trigger both, monitoring and adaptation, without a direct link between the latter to processes.

Apart from monitoring, aftereffects of dishonest responding on Task 2 might reflect alternative processes. For example, mutual inhibition between response alternatives in dishonest responding might also delay Task 2. If inhibition processes are indeed involved in dishonest responding, they should probably only target the initial representation of the honest response that has to be overcome (e.g., Walzyck et al., 2014); however, there is no empirical evidence for this claim, yet. To the contrary, boosting the activation of the honest response actually facilitates dishonesty (Debey et al., 2014; Foerster et al., 2017a, 2017b). Furthermore, Task 1 and Task 2 always employed different sets of response keys, which would call for inhibition of all motor activity rather than mutual inhibition of response alternatives to explain slowing in Task 2. In the case of errors, prolonged capacity-limited processes emerged even if response sets were different between Task 1 and Task 2 (Steinhauser et al., 2017). Although other researchers proposed orienting processes or inhibition of all motor activity as an alternative interpretation for these effects, they deemed rareness of errors as a trigger for these processes (e.g., Notebaert et al., 2009; Wessel & Aron, 2017). We prompted honest and dishonest responses in an equal frequency here, however.

Finally, prolonged responses after being dishonest might reflect a shift toward a more conservative response criterion. For errors, initial monitoring does indeed seem to be followed by such a shift in response threshold (Jentzsch & Dudschig, 2009; Steinhauser et al., 2017). That is, these criterion shifts seem to take time and would emerge rather for long that for short response–stimulus intervals. Instead, effects of dishonest responding on Task 2 vanished (Experiment 1) or became descriptively smaller (Experiment 2) with the long response–stimulus interval here. Furthermore, we observed a more liberal response criterion in an unrelated task after a modest response–stimulus interval of 400 ms between the dishonest response in a preceding study (Foerster et al., 2017a, 2017b).¹ In a nutshell, we evaluate monitoring as the most feasible explanation for the observed effects considering all abovementioned findings from the literature.

People can lie by very different means, i.e., they can deliver information that is incongruent with the truth, or they can simply withhold true information. Withholding an initially activated response also qualifies as a cognitive conflict supporting the notion that lying by omissions should be monitored as thoroughly as lying by commissions in the current study. Addressing this question in traditional psychological refractory period paradigms is not trivial, because these rely on the localization of effects in RTs. RTs are, however, not available for response omissions (for a discussion of the pitfalls in examining such non-actions, see Weller et al., 2017).

Traditional psychological refractory paradigms and their predictions also come to a limit when the aim is to examine effects in the postcentral, motor stage, while capacity-limited response monitoring processes are present and exceed response execution as in the current study (Foerster et al., 2019). One of the core assumptions is that the postcentral stage can operate in parallel with all stages of another task (Pashler, 1994). As such, effects in this stage do not propagate from Task 1 to Task 2. Capacity-limited processes operating until or after response execution would cover such effects in Task 1 making it impossible to detect them.

The current experiments expand our knowledge of information processing in dishonest responding. Dishonest responses appear to be more thoroughly monitored than honest responses even past response execution. A prime candidate for the source of this extensive monitoring process is the parallel activation of both, honest and dishonest response tendencies. Although the employed paradigms aimed at an isolated examination of this conflict within dishonest responding, we cannot yet exclude that general response difficulty instead of conflict itself triggers extensive monitoring. Future examination should disentangle these aspects. We further recommend to explore monitoring of the effects of dishonest responses and its link to response monitoring.