The impact of moral injury-related content on reasoning and its neural correlates: Data from the Canadian Armed Forces (CAF)

Abstract

Recently, there has been growing interest in understanding the causes and consequences of moral injury—defined as the functionally impairing psychological, biological, spiritual, behavioural, and social impact of perpetrating, failing to prevent, bearing witness to, or being a victim of acts that transgress deeply-held moral beliefs and expectations. Neuroimaging studies have revealed that moral injury is associated with functional alterations in regions that underlie emotions, somatosensory processing, internally-oriented thoughts, and cognitive control. However, to date, no study has examined the impact of moral injury on how people reason, or its neural correlates. We hypothesized that content referencing moral injury themes would reduce reasoning accuracy, and engage structures associated with memory and/or emotion. We tested this hypothesis by administering structurally identical arguments that included neutral content or content referencing salient moral injury outcomes (e.g., shame, anger, trust violations) to Canadian Armed Forces members in the fMRI scanner. As predicted, relative to neutral content, reasoning accuracy was reduced on arguments with moral injury themes, particularly in participants who surpassed clinical thresholds on the Moral Injury Outcome Scale (≥31) and the Posttraumatic Stress Disorder Checklist for DSM-5 (≥33), suggesting that reductions in reasoning accuracy might be driven by elevated moral injury symptoms and psychological distress. Furthermore, reasoning on arguments with moral injury-related content engaged the right posterior parahippocampus (BA 19). Given this region's role in representing contextual associations in episodic memory, this suggests that content with moral injury themes might trigger contextual associations that interfere with the reasoning system.

Highlights

• •Impact of moral injury (MI) on reasoning and brain function remains underexplored.
• •Military members reasoned about syllogisms with and without MI-related themes.
• •Reasoning accuracy was reduced on arguments with MI themes.
• •Reduced reasoning accuracy was linked to higher PTSD and MI symptom severity.
• •MI reasoning engaged the right posterior parahippocampus, linked to memory context.

1. Introduction

Recently, there has been growing interest in understanding the causes and consequences of moral injury (MI)—typically defined as a potential clinical problem with psychosocial, spiritual, and behavioural impacts following the commission, omission, witnessing, or being a victim of acts that transgress deeply-held moral beliefs (Litz et al., 2009; Litz and Walker, 2025). Perhaps nowhere is this growth more apparent than in the proliferation of self-report instruments that have been developed to attempt to measure this construct. For example, Houle et al.’s (2024) systematic review and content analysis of existing instruments that measure MI and the related construct of moral distress—a condition that arises “when one knows the right thing to do, but institutional constraints make it nearly impossible to pursue the right course of action” (Jameton, 1984, p. 6)—revealed no fewer than 42 unique scales currently in circulation. These scales have provided researchers with a rich array of measurement tools to examine specific aspects of moral stressors—with practitioners in the nursing and healthcare fields focusing mostly on moral distress, and military behavioural health researchers focusing primarily on MI—as evidenced by six scales designed specifically to measure MI within military populations (Houle et al., 2024; see also Vermetten et al., 2023). Yet, the proliferation of measures comes with some potential drawbacks, such as conceptualization differences that can confuse the expression of MI with potential MI exposure (Houle et al., 2024).

Still, the relatively greater interest in the military domain on MI is not surprising, given that this condition is observed frequently among military members and Veterans who have been deployed and/or exposed to combat (e.g., Griffin et al., 2019; Litz et al., 2009; Shay, 1994, 2014). In the US, Jordan et al. (2017) found that approximately 25 % of active-duty Marines reported perpetration and/or betrayal items on a measure of MI, with similar results among combat Veterans (Wisco et al., 2017). In a study focusing on a sample of Royal Australian Air Force personnel with previous deployment to a war or war-like zone and without a diagnosis of posttraumatic stress disorder (PTSD), approximately 62 % reported at least one potentially morally injurious event (PMIE)—considered a necessary trigger for moral distress and MI—due to witnessing transgressions, perpetrating transgressions, or feeling betrayed by allied forces or their own leaders (Hodgson and Carey, 2019; Hodgson et al., 2022; for review see Jamieson et al., 2023). Research in Canada has also demonstrated the existence, prevalence and etiology of MI among Canadian Armed Forces (CAF) personnel (e.g., Easterbrook et al., 2022, 2023; Nazarov et al., 2018, 2020). In one study, over half of a sample of CAF personnel who had deployed to Afghanistan endorsed at least one PMIE, and those exposed to PMIEs demonstrated a greater likelihood of having past-year PTSD and major depressive disorder (MDD) (Nazarov et al., 2018). More recent research has documented the incidence of moral distress and MI among CAF personnel who deployed as part of Op LASER—CAF's mission to provide support to civilian staff at long-terms care facilities in Ontario and Quebec during the COVID-19 pandemic—and the persistence of the associated symptoms across the nine months following the mission (Thompson et al., 2022). This work suggests that military members are at elevated risk of developing MI following morally conflicting or distressing situations, and that its effects can persist over time.

As its definition suggests, MI tends to arise in the aftermath of events which violate deeply-held moral beliefs and expectation, such as a sense of betrayal on the part of a trusted leader, peer or institution (Litz et al., 2009; Terpou et al., 2022). Furthermore, MI is understood to be a multidimensional syndrome rather than a discrete disorder, that can impact individuals across multiple domains, including cognitive, emotional, behavioural and spiritual functioning (Barnes et al., 2019). Its core impacts include alterations in perceptions of self and others, moral reasoning, relationships/belongingness, moral emotions, self-harm, and changes in beliefs about life's purpose (Litz et al., 2022; Litz and Walker, 2025). PMIEs may simultaneously meet the threshold to be considered a PTSD Criterion A traumatic event, creating significant symptom overlap between PTSD and MI, which may include anger, mood alterations, anxiety, insomnia, nightmares, hypervigilance, and social withdrawal (Vermetten et al., 2023). This substantial overlap and divergence between MI and PTSD can complicate diagnosis and treatment, highlighting the need for careful assessment and interventions that address both trauma-related fear and moral-existential suffering. Further explorations of psychological and biological mechanisms are necessary to shed light on the unique aspects of the etiology of MI, including its relationship with outcomes such as suicidality (Bryan et al., 2018; Litz et al., 2018; Magues et al., 2012; Nichter et al., 2021).

1.1. Neural bases of MI

Despite the historical issues with the operationalization and measurement of MI referred to earlier, a handful of studies have recently attempted to explore the mechanisms and processes that underlie MI by probing its neural correlates. Based on a sample of US Veterans, Sun et al. (2019) investigated the association between PTSD symptoms and MI (as measured by the Moral Injury Event Scale [MIES], Nash et al., 2013) with patterns of spontaneous neural activity as measured by the amplitude of low frequency fluctuations and functional connectivity during resting-state functional Magnetic Resonance Imaging (fMRI) (see Houle et al. (2024) for conceptual issues with the MIES). They found that the amplitude of low frequency fluctuations in the left inferior parietal lobule was correlated positively with the MIES subscale of transgressions but negatively with the subscale of betrayals, whereas functional connectivity between the left inferior parietal lobule and bilateral precuneus was correlated positively with PTSD symptoms and negatively with the MIES total scores. The inferior parietal lobule is a major hub for integrating multisensory information for comprehension and manipulation (Tomasi and Volkow, 2011), and along with the precuneus forms part of the default-mode network (DMN) which contributes to internally-generated cognition (Andrews-Hanna et al., 2014), as well as theory of mind (ToM) regions that underlie our inferences about the mental states of others (Mar, 2011). The DMN is a well-established system for self-related processing that underlies autobiographical memory. Its engagement suggests that moral emotions related to recollecting a PMIE might be driven in part by a system that underlies self-relevant processing in the brain. These findings are important not only because they suggest that MI and PTSD are associated with different patterns of intrinsic neural activity, but also because the differences implicate brain structures involved in processing information about one's sense of self and others, and social cognition and emotion more broadly (Sun et al., 2019).

Lloyd et al. (2021) collected fMRI data from CAF members and police or corrections officers—all of whom had a primary diagnosis of PTSD—as well as PMIE-exposed civilian controls. Building on earlier autobiographical memory studies of trauma, participants were prompted to recall neutral and PMIEs in the fMRI scanner, the results of which demonstrated that the recollection of PMIEs in the military and public safety-related PTSD group was associated with greater activation in the salience network (i.e., posterior insula, dorsal anterior cingulate cortex, dACC) that underlies processing sensory, visceral and arousing content (Uddin et al., 2017), as well as regions that involve top-down control of emotion in the prefrontal cortex (i.e., dorsolateral prefrontal cortex, DLPFC) (Clarke et al., 2020). These results suggest that in those with PTSD, there is greater engagement of bodily sensation as well as regulatory control during the recollection of PMIEs. Extending the design used by Lloyd et al. (2021), Terpou et al. (2022) administered a script-driven memory recall of MI task to three groups of individuals in the fMRI scanner: Participants with military- or law enforcement–related PTSD, participants with civilian-related PTSD, and healthy controls exposed to a PMIE. They examined group differences in functional connectivity, with an additional focus on brainstem, midbrain and cerebellar systems that are not always included in connectivity analyses despite their role in processing moral emotions such as shame, guilt and betrayal. The results showed that participants with civilian-related PTSD exhibited stronger precuneus functional connectivity across the DMN than participants with military- and law enforcement–related PTSD, and there was stronger functional connectivity involving the midbrain periaqueductal grey and the cerebellar lobule in participants with civilian-related PTSD than healthy controls. These results involving the DMN reinforce the role of regions that underlie processing information about one's self and emotion in trauma-related cognition.

Also using a script-driven memory retrieval paradigm involving PMIEs, Kearney et al. (2023) examined the functional connectivity between the sensorimotor network (SMN) and the posterior DMN network in individuals with PTSD and healthy controls. The researchers focused on the SMN due to a hypothesized link between MI and disrupted motor planning (i.e., failure to act in accordance with one's moral norms). The findings demonstrated alterations in functional connectivity between SMN and posterior DMN, including hyperconnectivity between the two networks, increased functional connectivity and expansion of the SMN, and increased recruitment of other regions of the brain into both networks—but only among individuals with PTSD during episodic recall of PMIEs. Behaviourally, the results also revealed that there was a positive correlation between PTSD severity and subjective re-experiencing intensity ratings after episodic recall of PMIEs. Drawing on classic work in psychology (Janet, 1889), the authors argued that successful (motoric) action is necessary for the integration of sensorimotor traces into self-referential memory. In cases where such integration fails to occur, as might be the case in certain instances of MI (i.e., failure to act in accordance with one's moral norms), one can expect that the sensorimotor traces will not be integrated into memory, ultimately resulting in alterations in functional connectivity in neural systems involved in self-referential (i.e., DMN) and sensorimotor (SMN) processing. More recently, Fulton et al. (2024) also focused on the relationship between somatosensory processes and MI, but used a different experimental paradigm to study it. Specifically, focusing on a sample of civilians who had completed the Moral Injury Exposure and Symptom Scale – Civilian (MIESS-C) (Fani et al., 2021), they administered the affective Stroop task which includes both threat-relevant and distractor stimuli in the fMRI scanner. They found that MIESS-C scores were correlated positively with functional connectivity between the right amygdala and left postcentral gyrus/primary somatosensory cortex, and that this was moderated by race such that it was observed in Black but not White participants. These results indicate that MI is likely associated with emotion-somatosensory connectivity, and that the experience of PMIEs such as racial trauma could be related to a distinct neural activation pattern.

In turn, Andrews et al. (2023) reasoned that in persons who have experienced MI, feelings of shame may become particularly salient in the presence of direct eye contact. In turn, this will likely engage psychological processes that support ToM (i.e., one's ability to make inferences about the mental states of others). To test this hypothesis, they used a script-driven memory retrieval paradigm involving virtual reality, in the course of which persons with PTSD and healthy controls were instructed to recall PMIEs or neutral memories, and exposed to an avatar with direct or averted gaze. The results demonstrated that after recalling PMIEs and only while exposed to an avatar with direct gaze, there was greater activation in the right temporoparietal junction (TPJ) in persons with PTSD compared to healthy controls. Furthermore, there was a positive correlation between feelings of distress and activation in right medial frontal gyrus in persons with PTSD. Given the engagement of both TPJ and medial frontal gyrus in ToM and self-referential processes (de Gelder et al., 2011; Liddell et al., 2004; Senju and Johnson, 2009; Van Overwalle et al., 2014), it appears that in persons with PTSD the recall of PMIEs may be accompanied by feelings of shame that prompt avoidance of the gaze of others. Here it is important to note that there can be differences in neural activation when one perceives actions performed by a human, a robot or an android (Saygin et al., 2011), and as such the patterns of neural activity observed using virtual reality may not necessarily extend to other paradigms.

In summary, this relatively small set of neuroimaging studies has produced a remarkably consistent set of results, demonstrating that a set of systems and structures involved in internally-generated cognition, sensorimotor processing, emotions and cognitive control are implicated in MI. This suggests that MI is associated with functional alterations in networks underlying these processes, which could contribute to the emotional, cognitive and physiological symptomatology of this specific moral stressor. However, less is known about how MI influences how people reason, and it is our interest in this question that underlies our focus on the relationship between MI and reasoning.

1.2. Neuroscience of reasoning

There are a number of different ways in which one can assess a person's ability to reason—defined as “the cognitive activity of drawing inferences from given information. All arguments in reasoning involve the claim that one or more propositions (the premises) provide some grounds for accepting another proposition (the conclusion)” (Grafman and Goel, 2002, p. 875). The type of reasoning that will be the focus of the present investigation is deductive reasoning, a key feature of which is that the premises provide absolute grounds for determining the validity of the argument. In other words, accepting the arguments as true forces the reasoner to also accept the conclusion as true. For example, consider the following deductive argument that consists of two premises followed by a conclusion—referred to as a “syllogism” (Evans et al., 1993):

Premise 1: All men are mortal.

Premise 2: Socrates is a man.

Conclusion: Therefore, Socrates is mortal.

Accepting premises 1–2 as true forces one to also accept the conclusion as true. Critically, validity is a function of the logical structure of the argument rather than the content of the sentences, meaning that the terms within the argument (e.g., Socrates, men) could be replaced with any other content, but the validity of the argument would still hold.

There is now a growing literature on the neural bases of reasoning, including a focus on brain regions that support deductive reasoning (for reviews see Goel, 2007; Prado et al., 2011). This work has shown that different neural systems can be engaged based on the features of the argument (i.e., syllogism). For example, when reasoning about arguments that have semantic content, a left hemisphere temporal system is recruited; in turn, when reasoning about formally identical arguments that lack semantic content, a parietal system is recruited (Goel et al., 2000). This pattern is consistent with the idea that reasoning involving semantic content engages a linguistic system, whereas reasoning about abstract content engages a visuospatial system. Interestingly, this work has shown that although determining the validity of an argument should be driven strictly by its structure and not by its content, people's performance suggests that the specifics of the content do interfere with people's reasoning processes. Of particular interest here is content that is emotionally salient. For example, several studies have shown that accuracy decreases when the content of an argument is negative rather than neutral (Blanchette, 2006; Blanchette and Richards, 2004; Lefford, 1946; for review see Blanchette and Richards, 2010). Goel and Dolan (2003a) explored the neural correlates of the effect of emotion on syllogistic reasoning by presenting participants with otherwise structurally identical arguments in the fMRI scanner with either neutral (Some Canadians are not children; All Canadians are people; [Therefore] Some people are not children) or emotional (Some wars are not unjustified; Some wars involve raping of women; [Therefore] some raping of women is not unjustified) content. Their results demonstrated that the brain exhibits sensitivity to variation in emotional content: When content was neutral, there was greater activation in lateral and dorsolateral prefrontal cortex (L/DLPFC) and suppression of activation in ventromedial prefrontal cortex (VMPFC); in contrast, when content was emotional, there was greater activation in VMPFC and suppression of activation in L/DLPFC. These results are consistent with the well-established roles of VMPFC and L/DLPFC in the experience of emotion (Barrett et al., 2007; Barrett and Wager, 2006) and executive functions (Stuss and Levine, 2002), respectively. Overall, the behavioural and neural evidence suggests that people exhibit sensitivity to content when they reason, and that this is apparent both in their behavioural performance as well as neural activity in relation to emotional content.

1.3. Present study

The literature on MI has demonstrated that this condition is associated with alterations in emotional processing. For example, Houle et al.’s (2024) content analysis of MI measures revealed that the domain of moral emotions (e.g., shame, guilt) is a consistent target across scales, demonstrating its centrality in our understanding of this construct along with behavioural, cognitive, spiritual, social, and other factors (see also Nazarov et al., 2015). Furthermore, our own review of the neuroscientific studies presented above also revealed that neural systems that underlie the processing of emotions are reliably implicated in MI, and form an important aspect of our understanding of its biological underpinning (see Fulton et al., 2024; Lloyd et al., 2021). However, what has not been examined before is whether MI influences how people reason—an important question because establishing this relationship would support the notion that the impact of MI extends beyond emotions such as distress, and can influence one's cognitions. Extending the literature from the psychology of reasoning (Blanchette, 2006; Blanchette & Richards, 2004, 2010; Lefford, 1946), we hypothesized that compared to structurally identical arguments with neutral content, participants would exhibit lower accuracy on arguments that reference MI-related content (Hypothesis 1). Such content would reflect a lack of trust in other people as well as institutions and persons in positions of power that one feels betrayed by (see Shay, 2014). Indeed, well-known scales for measuring the outcomes of MI such as the Moral Injury Outcome Scale (MIOS, Litz et al., 2022) include several items that capture a sense of betrayal and lack of trust in others (e.g., I lost trust in others; I have trouble seeing goodness in others; I have lost faith in humanity). In turn, and building on previous findings from the neuroscience of reasoning (Goel and Dolan, 2003a), we hypothesized that compared to reasoning about structurally identical arguments with neutral content, reasoning about arguments that reference MI-related content would engage regions of the brain that underlie the experience of emotions (Barrett et al., 2007; Barrett and Wager, 2006) and/or episodic memory (Eichenbaum, 2017; Moscovitch et al., 2016) (Hypothesis 2). This is because the examination of content involving such adverse MI-related outcomes (e.g., betrayal, mistrust) will likely engage (negative) emotions as well as episodic memories triggered by self-reflection on past experiences associated with similar features and contexts.

2. Methods

2.1. Participants

The protocol for this study was approved by Defence Research and Development Canada's Human Research Ethics Committee. All participants consisted of CAF members (n = 47) who gave written informed consent to take part in the study. The average age of the sample was 26.52 years (SD = 6.03). On average, the sample had 5.51 (SD = 5.57) years of service in the CAF. Their demographics appear in Table 1. No participant had formal training in reasoning. Handedness was not recorded.

Table 1.

Demographics, history of trauma, and psychological stress and moral injury scores among Canadian Armed Forces (CAF) participants (n = 47).

Demographics	Number (%)	Mean (SD)	Range
Age	–	26.52 (6.03)	19–48
Sex, male	37 (80.4)	–	–
Status	–	–	–
Regular Force	37 (80.4)	–	–
Primary Reserve	9 (19.6)	–	–
Rank	–	–	–
NCM	41 (89.1)	–	–
Senior NCM	2 (4.3)	–	–
Junior Officer	3 (6.5)	–	–
Education	–	–	–
High school diploma	24 (52.2)	–	–
College diploma	15 (32.6)	–	–
Undergraduate degree	6 (13.0)	–	–
None of the above	1 (2.2)	–	–
CAF years of service	–	5.51 (5.57)	1–28
Trauma history	–	–	–
Exposure to PMIE	–	–	–
Yes	41 (89.1)	–	–
No	5 (10.9)	–	–
Type of PMIE exposurea	–	–	–
Action/inaction by self	27 (58.7)	–	–
Witnessing	27 (58.7)	–	–
Betrayal	25 (54.3)	–	–
NA	5 (10.9)	–	–
Mental health scores	–	–	–
PCL-5 score	–	20.17 (14.15)	0–52
MIOS score, total	–	18.72 (8.97)	0–37
MIOS score, trust violation	–	10.46 (5.09)	0–19
MIOS score, shame	–	8.26 (5.26)	0–19

Notes. SD = standard deviation; NCM = Non-commissioned members (i.e., a CAF member who is not an officer or officer cadet); PMIE = Potentially morally injurious event; PCL-5 = Post-Traumatic Checklist for PTSD according to the Diagnostic and Statistical Manual of Mental Disorders, 5th edition (DSM-5); MIOS = Moral Injury Outcome Scale.

Missing responses, N (%): Sex 1 (2.1 %), Status 1 (2.1 %), Rank 1 (2.1 %), Education 1 (2.1 %), Years of service 1 (2.1 %), PMIE exposure 1 (2.1 %), PCL-5 1 (2.1 %), MIOS 1 (2.1 %).

^aSome participants reported more than one PMIE type.

We conducted a post-hoc power analysis using the effect size derived from our primary comparison of interest (i.e., accuracy when presented with arguments that reference MI content vs. neutral content). Based on the paired-samples effect size (Cohen's d = .65), 80 % power at α = .05 [two-tailed]) requires ∼21 paired observations. The number of participants assessed for accuracy (n = 46) therefore provides adequate power to detect the observed effect.

2.2. Materials and procedure

The data were collected in two separate sessions. One session consisted of the collection of individual differences measures outside of the fMRI scanner, using paper-and-pencil measures (intelligence, symptoms of MI and PTSD) and computerized tasks to measure simple and complex working memory span. At the outset the participants completed the MIOS, which is a psychometrically valid scale for the assessment of MI in relation to three types of PMIE (i.e., events involving action/inaction by self, witnessing, and betrayal). Also as part of the MIOS, respondents completed a series of 14 questions concerning the range and intensity of MI outcomes associated with the event that they have experienced. Anticipating that not everyone will have a PMIE in this convenience sample of non-clinical participants, the version of the MIOS we used included an option to continue to those 14 questions even if a PMIE was not endorsed. Instead, they were asked to answer the questions in relation to the “worst, most currently distressing event.” Sample items of MI outcomes include “I lost trust in others” and “I have trouble seeing goodness in others.” Responses are collected using a 5-point scale ranging from 0 (Strongly disagree) to 4 (Strongly agree). Higher scores indicate greater levels of current MI. A score of ≥31 was used as the threshold for identifying clinically significant symptoms of MI. Then, to assess symptomatic criteria for psychological distress, we administered the 20-item Post-Traumatic Checklist (PCL-5) for PTSD according to the Diagnostic and Statistical Manual of Mental Disorders, 5th edition (DSM-5) (Weathers et al., 2013). Sample items include “In the past month, how much were you bothered by repeated, disturbing, and unwanted memories of the stressful experience?” and “In the past month, how much were you bothered by repeated, disturbing dreams of the stressful experience?” Responses are collected using a 5-point scale ranging from 0 (Not at all) to 4 (Extremely). Higher scores indicate greater levels of current psychological stress. A score of ≥33 was used as the threshold to indicate probable PTSD.

We administered measures of intelligence and cognitive ability (i.e., simple and complex working memory span) to examine their correlations with reasoning performance, as well as to explore whether MIOS and PCL-5 scores might be correlated with reduced cognitive performance. Specifically, we suspected that participants with higher MIOS (and perhaps also PCL-5 scores) might be more sensitive to the presence of MI-related content during reasoning, which could serve to amplify its effects on accuracy. In turn, there is a large literature that has shown that individual differences in intelligence and cognitive ability are correlated with reasoning ability, and that working memory in particular is a strong predictor of performance on such tasks (De Neys, 2006; Kyllonen and Christal, 1990; Süß et al., 2002). We therefore wished to explore whether the benefits of greater intelligence and cognitive ability would extend to reasoning on arguments that reference MI-related content. Our measures of crystallized and fluid intelligence consisted of the Vocabulary (10 min) and Block Patterns (10 min) subsets of the Shipley-2, which were in turn standardized into full-scale intelligence scores (Shipley et al., 2009). Next, we administered two measures of simple working memory span (i.e., short-term memory tasks), modeled after Harrison et al.’s (2013) simple working memory span tasks (see Vartanian et al., 2016, 2021). For word (verbal) span, four-letter monosyllabic words were presented one at a time on a monitor. After each block of words, participants were prompted by the software to recall the words they saw in the order in which they were presented. Blocks ranged from three to nine words. For spatial (matrix) span, participants were presented with a 4 × 4 matrix where one square (out of 16) appeared in red and the rest in white. At the end of each block of matrices, participants were instructed to recall the locations of the red squares in the order in which they were presented. Blocks ranged from three to nine matrices. The computer application provided a detailed description of each task. Before beginning the first trial, the experimenter reviewed the instructions and provided an example in each case to the participants. Note that both the verbal and matrix span are simple working memory span tasks that primarily tax short-term memory storage capacity (e.g., Harrison et al., 2013; Cowan, 2008; Unsworth and Engle, 2007). In turn, our measure of complex working memory span consisted of the short version of the Operation Span Task (Foster et al., 2015). This task uses letters as the to-be-remembered items, and simple math problems as the distractor task (Kane et al., 2004; Unsworth et al., 2005). Specifically, on each trial participants must first solve a math problem, then see a letter, then solve another math problem, and see another letter. This math-letter sequence is repeated from three to seven times for each trial, with an unpredictable length each time. After each math-letter sequence, the participants are instructed to recall, in the order in which they were presented, the preceding letters. Scores are calculated by summing the number of letters correctly recalled in the correct order (i.e., partial score, Turner and Engle, 1989).

Prior to entering the fMRI scanner, the experimenters explained the concept of logical validity to the participants. Critically, it was explained to them that an argument is considered valid if and only if accepting the conclusions as true forces one to also accept the conclusion as true. Else, it is invalid. Participants were then administered examples of valid and invalid arguments as practice, and given the opportunity to ask any questions they might have about the task. Then, using a laptop, they were shown trials as they would appear in the fMRI scanner to familiarize them with the timing of the task. In the fMRI scanner we administered the reasoning task. The task included 96 trials. Half of the trials included content that was neutral whereas the other half included content that referenced potentially morally injurious content (Table 2). Critically, the arguments with neutral content and arguments that referenced MI-related content were balanced in several ways. First, they both included the same number of structurally identical argument types (i.e., figures = IA4, EI2, AA1, EI1, AE4, EI3, AE2, EA2, AO2, AI2, OE1) (Evans et al., 1993). Second, they were identical in terms of syllogism difficulty in relation to those argument types (55 %–95 %, Dickstein, 1978). Third, they were identical in terms of congruency. Specifically, arguments can vary in terms of whether the conclusion of the argument is consistent or inconsistent with one's beliefs. In turn, this feature impacts whether or not the arguments are accepted as valid by participants, although validity should only be driven by the structure of an argument rather than its content. For example, the following valid argument with a believable conclusion is accepted as valid 96 % of the time (Evans et al., 1983; Goel and Dolan, 2003b):

Table 2.

Study design with representative arguments (i.e., syllogisms).

	Content
	Moral Injury-Related	Neutral
Reasoning arguments	All victims are angry.	All athletes are fit.
	All sufferers are victims.	All Olympians are athletes.
	Therefore, all sufferers are angry.	Therefore, all Olympians are fit.
	All people trust institutions.	All miners are tall.
	Some people are leaders.	Some miners are men.
	Therefore, some leaders do not trust institutions.	Therefore, some men are not tall.
Control	All citizens have trust issues.	All jury members are citizens.
	No residents have trust issues.	No judges are citizens.
	Therefore, some offenders are not guilty people.	Therefore, some popes are not Catholics.

Notes. There were 96 trials in total. There were 48 reasoning arguments and 48 control trials (see Methods). Control arguments involve two premises followed by an unrelated conclusion; they are always invalid (i.e., conclusions do not follow necessarily following the acceptance of the premises as true). In turn, half of the reasoning arguments are valid (top row) whereas the other half are invalid (bottom row).

No cigarettes are inexpensive.

Some addictive things are inexpensive.

Therefore, some addictive things are not cigarettes.

In contrast, a logically identical argument with an unbelievable conclusion is accepted as valid only 46 % of the time:

No addictive things are inexpensive.

Some cigarettes are inexpensive.

Therefore, some cigarettes are not addictive.

The difference in acceptance rate is driven by the fact that people are less likely to agree with the conclusion “some cigarettes are not addictive” than with the conclusion “some addictive things are not cigarettes,” despite identical structure of the arguments that they are embedded in. As such, in the present study, arguments that reference MI-related content and arguments with neutral content were balanced in terms of congruency (four variants: valid arguments with believable conclusions, valid arguments with unbelievable conclusions, invalid arguments with believable conclusions, invalid arguments with unbelievable conclusions). Furthermore, the arguments in both conditions were also balanced in terms of their number of constituent letters, to equate them in terms of reading difficulty. Examples of valid and invalid arguments can be found in Table 2.

Finally, as in prior neuroimaging studies of logical reasoning (Goel and Dolan, 2003b; Goel et al., 2000), for each of the arguments regardless of its content, our experimental design also included a control argument that consisted of a syllogism that included two premises paired with an unrelated conclusion (e.g., All athletes are fit; All Olympians are athletes; Therefore, no dropouts are ambitious). Because by definition these control arguments do not necessitate the engagement of the reasoning system to be deemed invalid, they can be contrasted with regular arguments to elucidate the neural systems that underlie reasoning.

In the fMRI scanner all 96 trials were administered using the following structure (Fig. 1): Each trial began with a fixation sign (+) presented for 1s, followed by premise 1 (3s), premise 1 and premise 2 (3s), premise 1, premise 2 and the conclusion (6s) during which the participant was required to enter a validity judgment (i.e., press a response button to indicate “valid” or “invalid”). This was followed by a variable inter-trial interval (ITI) with a 1–4s range.

Fig. 1.

Structure of a trial in the fMRI scanner.

Notes. ITI = inter-trial interval (variable); RT = reaction time associated with motor response; variable ITI functioned as jitter in the design.

2.3. fMRI acquisition and processing

Magnetic resonance images were acquired on a Siemens MAGNETOM Prisma Fit 3 T system (Erlangen, Germany). We obtained T1-weighted anatomical images with the following parameters: repetition time = 2300 msec, echo time = 2.62 msec, and voxel size = 1 × 1 × 1 mm³, for a total of 192 axial slices covering the whole brain. For functional imaging, T2∗-weighted gradient-echo images were acquired with the following parameters: repetition time = 1630 msec, echo time = 30 msec, flip angle = 66°, field of view = 240 × 240 mm², matrix = 120 × 120 voxels, and voxel size = 2 × 2 × 2 mm³, for a total of 72 contiguous 2-mm thick axial slices positioned to cover the whole brain. The first six volumes were removed to account for T1 equilibration effects. In total, 510 vol were acquired.

2.4. Statistical analysis of participant characteristics, accuracy, cognitive function, psychological stress, and MI

Descriptive statistics were used to summarize demographic and mental health variables, with categorical variables reported as frequencies and percentages and continuous variables as means with standard deviations (SD). To examine differences in accuracy across conditions—neutral (arguments with neutral content), neutral control (arguments with neutral content with unrelated conclusion), MI (arguments that reference MI-related content), and MI control (arguments that reference MI-related content with unrelated conclusion)—a repeated measures analysis of variance (ANOVA) was conducted, followed by paired t-tests for post-hoc comparisons. ANOVA and paired comparisons were conducted in accordance with the relevant statistical assumptions. Specifically, for paired-samples t-tests, the normality of the difference scores was evaluated using Q–Q plots. For the repeated-measures ANOVA, the normality of residuals was inspected, and Mauchly's test of sphericity was performed. When the assumption of sphericity was violated, Greenhouse–Geisser (GG) corrected estimates were reported. Results of paired t-tests were presented alongside false discovery rate (FDR)-adjusted p-values derived using the Benjamini-Hochberg method. Associations between demographic characteristics, PCL-5 and MIOS scores, and cognitive function measures with accuracy across experimental conditions were analyzed using linear mixed-effects models. Natural splines with knots at the 33rd and 67th percentiles were applied to assess potential non-linear relationships. Interaction effects between condition (MI vs. neutral) and PCL-5 and MIOS scores on accuracy were tested. Additionally, a sensitivity analysis was performed for ANOVA and paired comparisons, restricting the dataset to participants who endorsed PMIE exposure. All statistical analyses were conducted using R Studio (v. 4.4.1).

2.5. Statistical analysis of fMRI data

Data were analyzed using Statistical Parametric Mapping (SPM12; www.fil.ion.ucl.ac.uk/spm/). Prior to realignment, we checked data quality to ensure that (a) there was no missing data in any of the runs, and (b) head movement was within 3 mm in all cases. All functional volumes were spatially realigned to the first volume. A mean image created from realigned volumes was spatially normalized to the MNI EPI brain template using nonlinear basis functions. The derived spatial transformation was applied to the realigned T2∗ volumes and spatially smoothed with an 8-mm FWHM isotropic Gaussian kernel. Time series across each voxel were high-pass filtered with a cut-off of 128 s, using cosine functions to remove section-specific low-frequency drifts in the BOLD signal. Condition effects at each voxel were estimated according to the general linear model, and regionally specific effects were compared using linear contrasts. The BOLD signal was modeled as a box-car, convolved with a canonical hemodynamic response function. We applied a voxel-level correction within SPM12 for determining statistical significance. Specifically, reported activations survived a voxel-level threshold of p < .05, corrected for multiple comparisons (whole-brain FWE: Family-wise error).

Using an event-related design, in the first level, we specified regressors corresponding to the following time points in the problem structure (Fig. 1): (1) fixation point, (2) neutral arguments, (3) control neutral arguments, (4) arguments that reference MI-related content, (5) control arguments that reference MI-related content, (6) motor response, (7) inter-trial interval (ITI, rest). The analyses of interest involved regressors 2–5. Regressors corresponding to the fixation point, motor response and ITI were given weights of “0” and modeled out of the analyses.

3. Results

3.1. Demographics and measures of psychological stress and MI

Table 1 presents participant characteristics, history of exposure to PMIEs, and scores on the PCL-5 and MIOS scales. Among the 47 CAF participants, the majority were male (N = 37/46, 80.4 %), Regular Force members (37/46, 80.4 %), and Non-Commissioned members (41/46, 89.1 %). Most participants reported exposure to at least one type of PMIE (41/46, 89.1 %), with some indicating exposure to multiple types. Fig. S1 illustrates the overlap of PMIE categories, including events involving action/inaction by self, witnessing, and betrayal.

3.2. Measures of accuracy and cognitive function

Table 3 presents the mean accuracy and cognitive function scores. Accuracy was highest under the neutral control condition, with a mean (SD) of 74.9 % (26.5) and a range of 25.0 %–100 %. In contrast, accuracy was lowest under the MI condition, with a mean (SD) of 52.3 % (12.4) and a range of 29.2 %–75.0 %. Dispersions of scores by condition are displayed in Fig. S2. Importantly, average accuracy on “standard” (i.e., neutral reasoning) trials was 60.51 % (SD = 12.03). This is comparable to what has been observed in several other neuroimaging studies involving the presentation of syllogisms under neutral (range = 63–66 %) and emotional conditions (Goel and Dolan, 2003a; Smith et al., 2014, 2015). As such, no participant who performed poorly was excluded from the analysis. A repeated-measures ANOVA revealed a significant main effect of condition on accuracy, F(3, 135) = 14.67, GG-adjusted p < .001, indicating that accuracy varied across conditions after adjusting for sphericity violations. The generalized eta squared (ges) for the main effect of condition was .25, signifying a large effect size (Bakeman, 2005). This suggests that approximately 24.6 % of the variance in accuracy was attributable to the experimental condition. Pairwise comparisons (Table 4), with FDR-adjusted p-values, indicated that accuracy was lowest under the MI condition, with significant differences observed when compared to both the neutral reasoning and neutral control conditions.

Table 3.

Accuracy and cognitive measures among Canadian Armed Forces (CAF) participants.

	Mean (SD)	Range
Measures of accuracyǂ	–	–
Neutral accuracy, mean (SD)	60.51 (12.03)	29.20–87.50
Neutral control accuracy, mean (SD)	74.90 (26.53)	25.00–100.00
MI accuracy, mean (SD)	52.27 (12.37)	29.20–75.00
MI control accuracy, mean (SD)	66.65 (30.66)	12.50–100.00
Measures of cognitive function	–	–
Simple verbal WM span (STM)	9.52 (1.86)	5.65–13.05
Simple matrix WM span (STM)	8.57 (2.15)	4.50–14.67
Complex WM span (partial score)	18.09 (5.97)	3.00–25.00
Shipley-2 Block Patterns (standardized score)	18.23 (4.46)	11.00–26.00
Shipley-2 Vocabulary (standardized score)	30.70 (3.93)	15.00–37.00

Notes. SD = standard deviation, MI = moral injury, WM = working memory, STM = short-term memory. The simple WM span measures were calculated based on partial-credit unit scoring (see Conway et al., 2005).

Missing responses, N (%): Accuracy measures 1 (2.1 %), Simple verbal WM span 1 (2.1 %), Simple spatial WM span 1 (2.1 %), Complex WM span 3 (6.4 %), Shipley-2 Block Patterns 4 (8.5 %), Shipley-2 Vocabulary 3 (6.4 %).

ǂIn the control conditions, arguments were presented with unrelated conclusions (see Methods).

Table 4.

Pairwise comparisons involving the four conditions in the experimental design.

Comparison	Mean difference	P-value∗
MI accuracy – MI control accuracy	−14.383	<.001
MI accuracy – Neutral accuracy	−8.239	.025
MI accuracy – Neutral control accuracy	−22.624	<.001
MI control accuracy – Neutral accuracy	6.143	.084
MI control accuracy – Neutral control accuracy	−8.241	.025
Neutral accuracy – Neutral control accuracy	−14.385	<.001

Notes. In the control conditions, arguments were presented with unrelated conclusions (see Methods). ∗ FDR-adjusted.

3.3. Relationships between demographics, PCL-5 and MIOS scores, and cognitive measures with accuracy across conditions

Table 5 presents estimates of the effects of demographic factors (age, sex, and years of service); PCL-5 score; MIOS total score, trust violation, and shame subscales; and cognitive measures (simple working memory: verbal and spatial; complex working memory span: partial; Shipley-2 Block Patterns; Shipley-2 Verbal) on accuracy across experimental conditions, using linear mixed-effects models. To account for potential non-linear relations, splines were applied to MIOS total score, PCL-5 score, years of service, and age.

Table 5.

Associations of demographics, PCL-5 and MIOS scores, and cognitive measures with accuracy by condition.

	βǂ (95 % CI)	P-value
Demographic	–	–
Age (spline 1, k = 23,27)	20.60 (−3.67, 44.88)	.104
Age (spline 2, k = 23,27)	−9.84 (−46.51, 26.84)	.602
Age (spline 3, k = 23,27)	−13.87 (−39.41, 11.67)	.293
Sex, female	−13.41 (−24.98, −1.84)	.028a
Years of service (spline 1, k = 3,5)	37.42 (5.04, 69.81)	.029a
Years of service (spline 2, k = 3,5)	10.43 (−15.00, 35.87)	.426
Years of service (spline 3, k = 3,5)	−6.86 (−32.16, 18.45)	.598
Mental health scores	–	–
MIOS, total (spline 1, k = 17.07,23.86)	11.68 (−8.97, 32.33)	.274
MIOS, total (spline 2, k = 17.07, 23.86)	−8.36 (−47.50, 30.78)	.678
MIOS, total (spline 3, k = 17.07, 23.86)	−7.64 (−30.59, 15.32)	.518
MIOS, trust violation	−.19 (−1.15, .77)	.702
MIOS, shame	.31 (−.62, 1.24)	.515
PCL-5, total (spline 1, k = 11.07,23.86)	22.48 (1.57, 43.39)	.041
PCL-5, total (spline 2, k = 11.07,23.86)	18.76 (−27.24, 64.76)	.429
PCL-5, total (spline 3, k = 11.07,23.86)	−22.87 (−40.60, −5.14)	.015a
Measures of cognitive function	–	–
Simple verbal WM span (STM)	2.09 (−.50, 4.68)	.121
Simple matrix WM span (STM)	.80 (−1.43, 3.03)	.484
Complex WM span (partial score)	.85 (.04, 1.65)	.046
Shipley-2 Block Patterns (standardized score)	1.47 (.45, 2.50)	.007a
Shipley-2 Vocabulary (standardized score)	1.77 (.62, 2.93)	.004a

Notes. SD = standard deviation, PCL-5 = Posttraumatic Checklist for PTSD according to the Diagnostic and Statistical Manual of Mental Disorders, 5th edition (DSM-5), MIOS = Moral Injury Outcome Scale, WM = working memory, STM = short-term memory, k = knots.

Missing responses, N (%): Accuracy measures 1 (2.1 %), Simple verbal WM span 1 (2.1 %), Simple spatial WM span 1 (2.1 %), Complex WM span 3 (6.4 %), Shipley-2 Block Patterns 4 (8.5 %), Shipley-2 Vocabulary 3 (6.4 %).

ǂEstimate of accuracy across conditions based on linear mixed models.

^aP-value <.05 after FDR adjustment.

After FDR correction, female sex remained inversely associated with accuracy, controlling for condition (β −13.41, 95 % CI: 24.98, −1.84). Standardized scores of Shipley-2 Block Patterns and Vocabulary were positively associated with accuracy across conditions (Patterns: β 1.47, 95 % CI: .45 to 2.50; Vocabulary: β 1.77, 95 % CI: .62 to 2.93). In contrast, relationships between simple or complex working memory span and accuracy were statistically non-significant after FDR adjustment. The difference between how measures of intelligence vs. working memory were correlated with reasoning performance is broadly consistent with the idea that accuracy might be influenced particularly strongly by general knowledge, whereas working memory capacity determines processing speed (Kyllonen and Christal, 1990). Future studies that also measure processing speed could test that relationship. Additionally, accuracy was higher among individuals with fewer years of service (<3 years) and lower among those with elevated PCL-5 scores (≥23.86), both modeled using splines. However, wide confidence intervals indicate substantial uncertainty in these estimates (years of service spline 1: β 37.42, 95 % CI: 5.04 to 69.81; PCL-5 spline 3: β −22.87, 95 % CI: 40.60 to −5.14).

Interactions between MIOS total score, MIOS subscales (shame and trust violation), and PCL-5 score with condition (MI vs. neutral) were tested for their impact on accuracy. Of these, only the shame subscale showed an interaction with condition, indicating that shame moderates the effect of condition on accuracy (−.89, 95 % CI: 1.55 to −.23). Specifically, the negative impact of content referencing MI-related content on reasoning accuracy was more pronounced among individuals with elevated shame scores.

Scatter plots with Locally Estimated Scatterplot Smoothing (LOESS) curves illustrate trends between MIOS scores vs. neutral accuracy and neutral control accuracy measures (Fig. 2a); MIOS scores vs. MI accuracy and MI control accuracy measures (Fig. 2b); PCL-5 scores vs neutral and neutral control accuracy measures (Fig. 2c); and PCL-5 scores vs MI accuracy and MI accuracy measures (Fig. 2d). The plots show a decline in accuracy for both neutral and MI conditions around the clinical thresholds for PCL-5 (≥33) and MIOS (≥31), suggesting potential reductions in task performance associated with elevated psychological stress and MI symptoms, irrespective of condition. Accuracy measures by condition were compared between individuals with PCL-5 scores below (<33) at or above (≥33) the clinical threshold for PTSD (Figure S3 a–d). Across all conditions, accuracy was lower among those with PCL-5 scores ≥33 compared to those with PCL-5 scores <33. These differences were statistically significant under the neutral control condition (Mean: PCL-5 <33 = 80.22 %, PCL-5 ≥33 = 53.75 %, p < .01) and the MI control condition (Mean: PCL-5 <33 = 74.15 %, PCL-5 ≥33 = 37.92 %, p < .01).

Fig. 2.

Scatter plots with Locally Estimated Scatterplot Smoothing (LOESS) curves illustrating accuracy measurements recorded under experimental conditions in relation to MIOS and PCL-5 scores.

Fig. 2a–d. (a) Top left: MIOS scores and accuracy under neutral and neutral control conditions; (b) Bottom left: MIOS scores and accuracy under MI and MI control conditions; (c) Top right: PCL-5 scores and accuracy under neutral and neutral control conditions; and (d) Bottom right: PCL-5 scores and accuracy under MI and MI control conditions. Dotted lines indicate the clinical thresholds for PCL-5 and MIOS. Among the 45 individuals assessed for accuracy and who completed both scales, 10 (22.2 %) met the clinical threshold for PTSD (PCL-5 ≥33) and 3 (6.7 %) met the threshold for moral injury (MIOS ≥31).

3.4. Sensitivity analyses

When restricting to participants who endorsed exposure to PMIEs (n = 41), a repeated-measures ANOVA continued to demonstrate a main effect of condition on accuracy, F(3, 117) = 9.367, GG-adjusted p < .001, ges = .194. Results for paired comparisons are displayed in Table S1. Accuracy remained lowest in the MI condition; however, the magnitude of mean differences between conditions was reduced compared to the main analysis. Notably, the accuracy difference between the MI and neutral conditions was no longer statistically significant (p = .057). This pattern suggests a broader impairment in reasoning accuracy across conditions among participants who endorsed exposure to PMIEs.

3.5. fMRI results

The contrast of reasoning about arguments that reference MI-related content vs. reasoning about arguments with neutral content activated a large bilateral cluster (k_E = 5621, p < .001) with a peak of activation centered on the right posterior parahippocampus (BA 19, X = 44, Y = −44, Z = −2, T = 6.05, p = .013)—both corrected for multiple comparisons (whole-brain FWE: Family-wise error) (Fig. 3). Within the same cluster, two additional peaks of activation reached statistical significance, located in the lingual gyrus (BA 18, X = −6, Y = −76, Z = 6, T = 5.66, p = .036) and the cerebellum (X = −10, Y = −64, Z = −12, T = 5.60, p = .042).

Fig. 3.

Reasoning about moral injury-related vs. neutral content.

Notes. Reasoning on trials that included moral injury-related content activated the right posterior parahippocampus (BA 19). SPM rendered into standard stereotactic space and superimposed onto sagittal MRI in standard space. Bar represents the corresponding T score.

4. Discussion

We conducted this study to test two hypotheses. First, drawing on the literature from the psychology of reasoning (Blanchette, 2006; Blanchette and Richards, 2004; Lefford, 1946), we hypothesized that participants would exhibit lower accuracy on arguments that reference MI-related content compared to structurally identical arguments with neutral content (Hypothesis 1). Our data supported this prediction. Specifically, whereas accuracy on neutral arguments was greater than 60 %, performance on arguments with MI-related content did not differ from chance (Fig. S2). This finding has a number of important implications. To begin with, it suggests that the presence of MI-related content can reduce reasoning accuracy, much like has been shown with the presence of emotionally salient content in past research (Blanchette, 2006; Blanchette and Richards, 2004; Lefford, 1946). This interpretation is supported by two additional findings. First, shame-related MI outcomes were shown to moderate the effect of condition on accuracy, indicating that such manifestations following exposure to PMIEs play an important role in impacting reasoning performance. Second, there was a decline in accuracy among participants who surpassed clinical threshold of PCL-5 (≥33) and MIOS (≥31), suggesting that potential reductions in reasoning accuracy might be associated with elevated psychological distress and MI symptoms. Having said this, it remains to be seen whether the effect will be amplified in individuals impacted by clinically significant severity of MI, assuming that they will be affected more by the presence of MI-related content compared to the convenience sample assessed in the present study. Third, and more broadly, we suspect that the effect of MI-related content on reasoning could impact other constructs. For example, it is known that some forms of MI could stem from perceived moral betrayal by a trusted individual or organization in power (Shay, 2014; Schorr et al., 2018). In this sense, it is possible that MI-related content may impact interpersonal trust, by influencing the reasoning process during interpersonal interactions. We hope that such downstream effects of reasoning on other constructs can be pursued in future studies.

Second, building on previous findings from the neuroscience of reasoning (Goel and Dolan, 2003a), we hypothesized that compared to reasoning about structurally identical arguments with neutral content, reasoning about arguments with MI-related content would engage regions of the brain that underlie the experience of emotions (Barrett et al., 2007; Barrett and Wager, 2006) and/or episodic memory (Eichenbaum, 2017, Moscovitch et al., 2016) (Hypothesis 2). We made this prediction because we reasoned that the examination of MI-related content would likely engage (negative) emotions associated with those feeling states, as well as episodic memories and self-reflection triggered by past experiences associated with similar features and contexts. This hypothesis was also supported. Specifically, reasoning about MI-related content vs. neutral content engaged a large cluster that centered on the right posterior parahippocampus (Fig. 3). Historically, the medial temporal lobe (MTL) has been known to play a central role in episodic memory (Eichenbaum et al., 1992; Scoville and Milner, 1957). Regarding the parahippocampal cortex (PHC) specifically, there is converging evidence to suggest that it plays a critical role in contextual associations—defined as the “link between objects strongly associated with the same context, or the spatial relation between items, or the configuration associated with a context” (Aminoff et al., 2013). Such contextual associations contribute to establishing the ‘contextual frame’ of our episodic memories, which are the network of associations that define their contexts. As noted by Aminoff et al. (2013) and Diana et al. (2007), both the neuroanatomical location and connectivity of the PHC facilitate its contribution to the formation of contextual associations. In terms of location, PHC encompasses a large portion of the MTL, and is located at the junction between brain regions that are essential to both memory formation (e.g., the hippocampus) and high-level visual processing (see Aminoff et al., 2013). In terms of connectivity, PHC receives input from multimodal cortical areas that convey integrative information about the context in which objects or events are encountered, and this information is relayed to the hippocampus where this and other types of information converge to form episodic memories (Diana et al., 2007). According to the ‘binding of item and context’ (BIC) model (see Diana et al., 2007; Eichenbaum et al., 2007) “contextual information in the PHC provides the input to the hippocampus in order to bind to new memories and link the memory of that particular episode within a larger network” (Aminoff et al., 2013). The hippocampus is in turn involved in binding the target of an episode with the context in which it is encountered. Drawing on this literature, we believe that the activation of the right posterior parahippocampus while reasoning about MI-related content could be related to the greater engagement of contextual associations triggered by MI-related content relative to neutral content. In turn, the engagement of those contextual associations in episodic memory might interfere with the reasoning process, leading to lower accuracy on such trials.

Interestingly, there is also evidence to suggest that along with the visual cortex (BA 18–19), the right posterior parahippocampus (BA 19) contributes to the experience of intrusive memories during recall of vivid autobiographical memories—a finding that is consistent with its role in contextual associations in episodic memory. Specifically, Daniels et al. (2012) recruited persons who had been acutely traumatized due to a motor vehicle accident, workplace accident, physical assault, or another traumatic event, and administered to them a script-drive memory task in the fMRI scanner in the course of which they were instructed to recall traumatic or neutral events. The investigators were interested in examining the extent to which PTSD symptomatology and its associated neurobiology are impacted by peritraumatic dissociation—characterized as perceptual alterations, emotional detachment from one's surroundings, and a loss of agency. Their results demonstrated that peritraumatic dissociation predicted PTSD diagnostic status at both 5–6 weeks and three months after the traumatic event, and was correlated positively with brain activation in early and late visual cortex and the right posterior hippocampal gyrus (BA 19). The authors argued that these results are consistent with a model in which “intrusions constitute involuntary memory recall from an imagery-based memory system in the temporal-occipital region storing the sensory aspects of an event” (Daniels et al., 2012, p. 420; see Conway, 2009). We agree with that inference, and believe that such sensory aspects (e.g., sounds, scenes) constitute part of the larger contextual associations that episodic memories are embedded in (Aminoff et al., 2013), and as such are part of the explanation for the engagement of the right posterior hippocampal gyrus (BA 19) in relation to peritraumatic dissociation during episodic recall of traumatic memories.

There are a number of considerations that must be taken into account while interpreting the results of the present study. To begin with, we administered our paradigm to a convenience sample of CAF members to examine whether the presence of MI-related content can impact reasoning performance. On average, this group is relatively young, with fewer than 5 years of military service. Although the sample was characterized by a range of MIOS scores, it included only a small number of individuals who exceeded the clinical thresholds for PCL-5 ( ≥ 33) and MIOS ( ≥ 31). Given these characteristics, future research with more individuals with higher PCL-5 and MIOS scores, along with a varied nature of PMIEs (e.g., commission, omission, betrayal, witnessing), would be beneficial to validate the trends observed in this study and allow for generalization to clinical samples. Second, when we restricted our participants to those who endorsed any exposure to PMIEs (n = 41), we continued to observe a main effect of condition on accuracy. Although accuracy remained lowest in the MI condition as before, the magnitude of mean differences between conditions was reduced compared to the main analysis. A possible explanation for this observation is that trauma may influence accuracy in both MI and neutral conditions. Indeed, we saw declines in accuracy scores for both MI and neutral conditions with higher scores of PCL-5 and MIOS. Future work should examine the extent to which a certain degree of posttraumatic stress/MI can cause general reductions in reasoning accuracy that extends beyond potentially morally injurious content specifically. Third, while we assessed the effects of various covariates, the wide confidence intervals indicate a high degree of uncertainty. This is due to the relatively small size of subgroups, as well as the overall study sample. For example, with only eight female participants, our findings suggest that female sex may be associated with lower accuracy across conditions. However, the high uncertainty in the confidence interval underscores the need for caution in interpreting this result. Replicating this study with larger and more diverse groups would enhance the reliability of these findings and provide a more robust understanding of the observed effects. Fourth, unlike paradigms that have employed script-driven memory recall of MI (Kearney et al., 2023; Lloyd et al., 2021; Terpou et al., 2022), our arguments did not include content that was person-specific (i.e., was designed specifically to probe personally-relevant traumatic memories). Again, our expectation would be that such a design would serve to amplify the observed effects, especially if it includes participants with elevated symptoms of MI. Finally, although previous research has shown that race can moderate functional connectivity patterns in relation to MI (Fani et al., 2021), we did not collect any demographic data in relation to race or other cross-cultural factors for this study. Going forward, we recommend that such data be collected in order to provide a more comprehensive picture of individual differences in brain activity in relation to MI. For example, it is known that there are cross-cultural differences in proneness to shame (Sznycer et al., 2012). Given the central role that shame plays in MI (Nazarov et al., 2015), it would be valuable to explore the extent to which individuals from different cultures reason about content that references MI themes that reflect shame, among others.

5. Conclusion

Our results demonstrated that compared to reasoning about neutral content, reasoning about content that references MI themes resulted in poorer performance. Furthermore, not only did symptoms of shame-related MI outcomes moderate the effect of condition on accuracy, but there was also a decline in accuracy among participants that surpassed clinical thresholds of PCL-5 (≥33) and MIOS (≥31), suggesting that potential reductions in reasoning accuracy might be associated with elevated psychological distress and MI symptoms. This extends earlier behavioural findings from the reasoning literature that have shown that emotionally salient content results in lower accuracy (for review see Blanchette and Richards, 2010), perhaps because emotionally salient content interferes with the reasoning process. In addition, our findings also showed that compared to reasoning about neutral content, reasoning about MI-related content is correlated with greater activation in the right posterior parahippocampus (BA 19). Given this structure's role in contextual associations that accompany episodic memories, we suspect that reflecting on MI-related content may serve to trigger contextual associations in memory to a greater extent than does exposure to neutral content, a process that can serve to reduce reasoning accuracy. Future studies can aim to tease apart the unique contributions of PTSD and MI to reasoning performance involving MI-related content, especially in populations that are at greater risk of exhibiting clinically-relevant symptoms.

CRediT authorship contribution statement

Oshin Vartanian: Writing – review & editing, Writing – original draft, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Anthony Nazarov: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Methodology, Investigation, Formal analysis, Conceptualization. Timothy K. Lam: Writing – review & editing, Writing – original draft, Validation, Supervision, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Erin Collins: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Investigation, Formal analysis. Megan M. Thompson: Writing – review & editing, Writing – original draft, Validation, Supervision, Methodology, Investigation, Conceptualization. Shawn G. Rhind: Writing – review & editing, Writing – original draft, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Data curation, Conceptualization. Stacey Silins: Writing – review & editing, Methodology, Investigation, Conceptualization. Maria Shiu: Writing – review & editing, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Data curation. Elaine Maceda: Writing – review & editing, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Data curation. Kristen King: Writing – review & editing, Supervision, Project administration, Methodology, Investigation, Formal analysis, Data curation. Janani Vallikanthan: Writing – review & editing, Validation, Supervision, Project administration, Methodology, Investigation, Formal analysis, Data curation. Maitri Lad: Writing – review & editing, Resources, Methodology, Investigation, Data curation.

Declaration of Competing interest

We have nothing to declare.

Appendix A. Supplementary data

The following is the Supplementary data to this article.

Data availability

The authors do not have permission to share data.