The Dynamic Nature of Interpersonal Conflict and Psychological Strain in Extreme Work Settings

Abstract

Humanity will mount interplanetary exploration missions within the next two decades, supported by a growing workforce operating in isolated, confined, and extreme (ICE) conditions of space. How will future space workers fare in a closed social world while subjected to persistent stressors? Using a sample of 32 participants operating in ICE conditions over the course of 30-45 days, we developed and tested a dynamic model of conflict and strain. Drawing on conservation of resources (COR) theory, we investigated reciprocal relationships between different forms (i.e., task and relationship) of conflict, and between conflict and strain. Results demonstrated evidence for a resource threat feedback loop as current-day task conflict predicted next-day relationship conflict and current-day relationship conflict predicted next-day task conflict. Additionally, results indicated support for a resource loss feedback loop as current-day relationship conflict predicted next-day strain, and current-day strain predicted next-day relationship conflict. Moreover, we found that job conditions affected these associations as current-day relationship conflict was more associated with next-day task conflict when next-day workload was high, but not when next-day workload was low. Similarly, current-day relationship conflict was more associated with next-day strain when next-day workload was high; however, this association decreased when next-day workload was low. Therefore, the results suggest that workload plays a critical role in weakening the effect of these spirals over time, and suggests that targeted interventions (e.g., recovery days) can help buffer against the negative impact of relationship conflict on strain and decrease the extent that relationship conflict spills over into task disputes.

Commercial space exploration is projected to be a trillion-dollar industry by 2040 (CNBC, 2017) and federal agencies expect manned exploration of Mars to begin in the early 2030s (Golden, Chang, & Kozlowski, 2018). As federal, international, and commercial organizations continue to push the boundaries of space exploration and development, a substantial number of work teams will soon operate in near-Earth orbit and beyond. These job environments will involve isolated, confined, and extreme (ICE) work conditions that require interpersonal interactions and coordination to complete complex and high-stakes tasks (Bell, Fisher, Brown, & Mann, 2018; Salas, Tannenbaum, Kozlowski, Miller, Mathieu, & Vessey, 2015). In most job environments, interpersonal stressors prove more problematic for worker health than any other form of stress (Almeida, 2005). The impact of such stressors likely increases within ICE conditions as workers operate in fully immersive, closed social systems with limited access to broader sources of social support. Under these conditions, individuals work and live together in the same setting over long periods of time, and this environment can amplify disputes over task (task conflicts) and personal (relationship conflict) characteristics (Jehn, 1995; Golden et al., 2018). Given that such conflicts can place a considerable amount of demand on individual health and well-being (De Dreu & Weingart, 2003; De Dreu, 2008), we must understand how the effects of these interpersonal stressors manifest over time, and which job conditions may buffer or exacerbate their harmful impact.

However, the literature on ICE teams currently suffers from conceptual and methodological gaps that hinder this objective. In a recent review, Golden and colleagues (2018) found that relatively few ICE studies examined within-person stressor-strain dynamics, and how these stressors may reciprocally relate to one another over time. Consequently, the authors advocated for increased research on stressor-strain dynamics in ICE environments to better reduce the harmful effects of interpersonal stressors such as conflict from propagating over time. To that end, our investigation examined the consequences of different forms of conflict on strain over time. Drawing on conservation of resources (COR) theory, we identified temporal and contextual elements influencing the conflict-strain process to advance a more dynamic understanding of stress in ICE settings. In doing so, we provide a number of novel contributions to the conflict and stress literatures.

First, we clarify the association between task and relationship conflict. Although there exists widespread recognition that relationship conflict negatively impacts individual (e.g., affect, psychological strain) and group (e.g., cohesion, performance) outcomes, the effects of task conflict remain more equivocal. Some researchers have found that task conflict can prove beneficial by increasing innovation (De Dreu & West, 2001) and decreasing groupthink (Schulz-Hardt, Brodbeck, Mojzisch, Kerschreiter, & Frey, 2006), which ultimately results in better decision-making and group performance (de Wit, Greer, & Jehn, 2012). However, others suggest that the benefits of task conflict are overstated as it generally holds negative implications for individual well-being (De Dreu & Weingart, 2003; De Dreu, 2008). The lack of consensus regarding task conflict may occur because researchers have generally examined the two types of conflict in isolation of each other and do not consider the temporal components of conflict. Both task and relationship conflict can feed into one another over time (Rispens, 2012), and this spillover effect may be especially pronounced in ICE conditions, as these conditions can tax the cognitive resources needed to separate work from personal disagreements (Jehn & Chatman, 2000). We take a dynamic perspective and conceptualize task and relationship conflict within a conflict feedback loop in which the two stressors reciprocally relate to each other (Cronin & Bezrukova, 2019), and consider relationship conflict as a mechanism through which task conflict may negatively relate to individual well-being.

Second, we extend COR theory by distinguishing between different types of spiraling effects. COR theory posits that individuals are motivated to conserve their resources (Halbesleben et al., 2014; Hobfoll et al., 2016). When stressors cause resource loss, individuals can end up in a loss spiral, as they seek to conserve resources instead of using those resources to cope with the stressor effectively (Halbesleben et al., 2014). As a result, individuals become increasingly susceptible to the stressor over time. In this case, the strain (i.e., the outcome of conflict) would increase the likelihood of individuals experiencing conflict in the future. However, COR theory also suggests that individuals often face multiple co-occurring stressors. Integrating these notions, we suggest that individuals in stressful environments face dual feedback loops in the form of resource loss and resource threat spirals. The former occurs when the effects of resource loss (i.e., strain) leads to greater exposure to the stressor (i.e., task and relationship conflict), whereas the latter occurs when co-occurring stressors (i.e., task and relationship conflict) reinforce one another. Relatedly, we distinguish the concept of positive feedback or self-reinforcement from spiraling behavior, by evaluating whether conflict and strain converge to an equilibrium or spiral upwards. Distinguishing between these patterns of change is also important because endless spiraling would suggest that conflict and strain inescapably increase over time, whereas observing a convergence to an equilibrium would suggest that the negative effects of conflict can be controlled and mitigated. By assessing these dynamic properties, we provide temporal specificity to COR theory and clarify the trajectory that conflict and strain take over time.

Furthermore, we identify job characteristics that can serve as moderating forces within our hypothesized dual feedback loops. Namely, we conceptualize task contribution as a source of empowerment that can buffer conflict-strain relationships by replenishing individuals’ resources (Demerouti, Bakker, Nachreiner, & Schaufeli, 2001). On the other hand, higher workloads should exacerbate strain by further reducing the available resources that individuals can mobilize to cope with the stressor (Spector & Jex, 1998). Job characteristics have the potential to alter the extent to which interpersonal conflicts affect individual well-being; however, research in this area remains limited (De Dreu, van Dierendonck, & Dijkstra, 2004). While prior research tends to focus on individual differences as buffers of conflict-strain relationships (e.g., Ilies Johnson, Judge, & Keeney, 2011; De Dreu & Weingart, 2003; Dijkstra, van Dierendonck, Evers, & De Dreu, 2005), examining contextual buffers can provide new opportunities to help reduce the harmful effects of conflict. More specifically, these contextual buffers may have the potential to decrease the equilibrium of stressors and strain over time.

Finally, a number of researchers have suggested that findings from workers in non-ICE settings may not generalize to workers operating in ICE conditions (Bell, et al., 2018; Golden, et al., 2018). Therefore, we provide ecological validity for our model by testing our hypotheses on a sample of individuals operating in ICE conditions. By using a sample of individuals “in the wild” (Maynard, et al., 2018, p. 695), we can better predict the psychological health risks that future ICE workers will face, and present strategies that may help them overcome these risks.

To organize our presentation, we start with an overview of our conceptual model and how it builds upon and extends COR theory. Next, we discuss the concept of resource threat spirals and how task and relationship conflict may reinforce each other over time. Importantly, we distinguish resource threat spirals from resource loss spirals, and how the effects of conflicts (i.e., strain) may ultimately reinforce the conflicts themselves. With both spirals, we discuss how job conditions (i.e., lower workload and greater task contributions) serve to attenuate the association between task and relationship conflict, between relationship and task conflict, and between conflict and strain. Finally, we test our model (Figure 1) on a sample of 32 individuals operating under ICE conditions over the course of 30-45 days.

Figure 1.

Theoretical model

Conceptual Development

Conservation of resources (COR) theory argues that individuals are motivated to conserve resources needed for survival (Hobfoll, Tirone, Holmgreen, & Gerhart, 2016). Although researchers have operationalized resources in a number of different ways, Halbesleben, Neveu, Paustian-Underdahl, and Westman (2014) explicitly state that “resources of interest [are] those which [are] central to survival or major goal attainment” (p. 113). In other words, although resources can refer broadly to many different constructs, individuals only wish to conserve the resources that directly impact their chances of survival and goal attainment, and experience strain when these critical resources are threatened (Hobfoll, 1989; Hobfoll, et al., 2016). Specifically, in group settings, social support acts as a crucial interpersonal resource that allows individuals to cope with their job demands and work together to achieve interdependent tasks (Golden et al., 2018; Hobfoll et al., 2016). However, interpersonal conflict can threaten social support by creating rifts in the social bonds that hold groups together (Jehn, 1995). This can be particularly challenging in the closed social world that characterizes ICE settings. In typical non-ICE settings, individuals can reduce tensions by limiting their social interaction with the conflicting team member at work and break from the conflict at the end of the workday when they go home (Marcinkowski, Bell, & Roma, 2021). Neither approach is possible in ICE settings, as individuals are confined together in close proximity both during and after their workday, which inhibits their ability to distance themselves from the conflict. Moreover, since these teams are isolated from the outside world, the team serves the role of both a work and a quasi-family unit (Kanas et al., 2009). As such, social support among team members becomes relevant not only for completing tasks, but also coping with the harsh physical environment and psychologically demanding work conditions. Therefore, when individuals engage in conflict, it can generate strain by threatening an individual’s interpersonal resources (Halbesleben et al., 2014).

Mitigating the negative effects of resource loss can prove difficult because threats to resources often display complex and interconnected relationships with each other over time. Conserving resources is often difficult, because individuals do not tend to experience single, isolated stressors. Instead, individuals tend to face “risk factor caravans” (Hobfoll et al., 2016, p. 67), which refer to highly correlated and co-occurring stressors. As individuals invest resources to protect or recover from resource losses generated by one stressor, they have fewer resources to reduce other stressors (Halbesleben et al., 2014). As a result, a resource threat spiral can occur whereby exposure to one stressor can increase the levels of another stressor (Halbesleben, 2010). Effectively, this process produces a positive feedback loop between the various stressors, wherein co-occurring stressors within the risk factor caravan positively reinforce one another.

Specifically, task and relationship conflicts have the potential to reinforce each other (Rispens, 2012; Cronin & Bezrukova, 2019) and such resource threat spirals imply increasingly deleterious effects as the positive feedback between different conflicts increasingly drain an individual’s resources. Moreover, as an individual loses resources, they may be unable to cope with these stressors. Due to an inability to cope, the threats will have an increasing impact on an individual such that their resources are further drained. As a result, the resource threat spiral can generate a resource loss spiral, wherein the strain generated from conflict predicts future increases in conflict levels.

However, job conditions can mitigate or amplify the negative effects of these harmful spirals. Hobfoll and colleagues (2016) note that environmental circumstances known as “caravan passageways” can support or undermine individuals’ ability to conserve resources (p. 66). Notably, these passageways often occur as a matter of circumstance or job context rather than personal choice. In line with this perspective, ICE conditions are likely to serve as a caravan passageway that facilitates resource loss because the hazardous conditions and isolation from broader sources of social support should make it difficult for individuals to conserve resources. As a result, individuals may face greater exposure to risk factor caravans, leading to the emergence of loss spirals. In the context of conflict, engaging in high-stakes tasks in isolation from broader sources of social support can lead to exposure to co-occurring conflict types that reciprocally influence one another (Rispens, 2012; Golden, et al., 2018).

Aside from environmental conditions, other contextual factors also have the potential to serve as passageways that undermine or help individuals conserve resources. Specifically, Demerouti and colleagues (2001) suggest that individuals’ workload has the potential to undermine the aggregation of resources, whereas the ability to contribute to tasks has the potential to help individuals maintain or gain resources. Although researchers have often focused on the buffering or amplifying effects of individual difference variables on the relationship between conflict and strain outcomes, identifying how job conditions moderate the conflict-strain relationship can prove useful for designing organizational strategies that help ICE workers. Therefore, we focus on the impact of workload and task distribution as potential points for interventions and mitigation of conflict’s negative consequences.

Resource Threat Spirals

Task Conflict to Relationship Conflict.

Task conflict and relationship conflict are highly correlated and often co-occurring (Rispens, 2012; DeChurch, Mesmer-Magnus, & Doty, 2013; Cronin & Bezrukova, 2019). For instance, Jehn and Chatman (2000) note that when individuals are exposed to a given type of conflict, it takes a significant amount of cognitive resources to distinguish between whether the conflict is driven by task versus interpersonal disagreement. When individuals do not have the cognitive resources needed to separate between person-related and task-related stimuli, disagreements regarding the task can be mistaken for criticisms regarding an individual’s competence (Tidd, McIntyre, & Friedman, 2004) and personality (Greer, Jehn, & Mannix, 2008; Choi & Cho, 2011). As a result, individuals in a dispute with others over task procedures can attribute the reasons for the task dispute to incongruences regarding personal characteristics (Brehmer, 1976). In this fashion, task conflict can increase relationship conflict. Presenting empirical evidence for this relationship, Mooney, Holohan, and Amason (2007) found that task conflict was positively related to subsequent relationship conflict. Similarly, Yu and Zellmer-Bruhn (2018) found that task conflict was predictive of social undermining through relationship conflict. In effect, these studies suggest as individuals invest resources to deal with one risk factor (i.e., task conflict), the reduced resources increase the likelihood of experiencing another co-occurring risk factor (i.e., relationship conflict).

Hypothesis 1A: Current-day task conflict relates positively to next-day relationship conflict.

Relationship Conflict to Task Conflict.

Similarly, other researchers have suggested that relationship conflict can increase task conflict. Specifically, relationship conflict can create negative attributions between conflicting parties regarding each other’s personal characteristics, making it difficult to objectively evaluate the ideas of the opposing party (Rubin, Pruitt, & Kim, 1994; Choi & Cho, 2011). For instance, Pelled (1996) suggested that when individuals engage in relationship conflict, their personal differences tend to bleed over into the task (see also, Pelled, Eisenhardt, & Xin, 1999; Jehn, 1995). Others have referred to this notion as a negative halo effect bias, whereby individuals make judgments of others using global impressions in order to conserve cognitive resources (Goldman, Cowles, & Florez, 1983). In this case, individuals who have personal disagreements can find fault with each other’s ideas due to their negative global impression of the other individual (Eisenhardt & Bourgeois, 1988). In other words, relationship conflict can reduce the cognitive resources needed to distinguish personal from task differences, which leads individuals to evaluate each other’s ideas using negative personal impressions, thereby increasing the likelihood of formulating a negative reaction towards the ideas themselves. Indeed, researchers have generally found that relationship conflict can be positively related to subsequent task conflict (e.g., Humphrey, Aime, Cushenberry, Hill, & Fairchild, 2017; Choi & Cho, 2011).

Hypothesis 1B: Current-day relationship conflict relates positively to next-day task conflict.

Moderators.

In this representation, task and relationship conflicts are mutually reinforcing threats to strain. Specifically, the existence of either conflict reduces individuals’ ability to distinguish between task- and person-related stimuli. This error in judgment due to the occurrence of one form of conflict will increase the levels of another form of conflict. However, job conditions such as workload provide one avenue to reduce this spiral. For instance, Strayer, Watson, and Drews (2011) found that when individuals had a high workload and split attention over multiple tasks, they were more likely to make errors in judgment (see also Castro, Strayer, Matzke, & Heathcote, 2019). However, when their workload was reduced, they were less likely to make such errors. Similarly, higher workload is associated with higher levels of cognitive strain, and subsequently greater errors in decision making tasks compared to low workload levels which display a reduced relationship with decision making errors (Gonzalez, 2005). In general, individuals must expend high levels of cognitive resources to successfully meet the strenuous workload, which could in turn hinder their ability to separate person-related from task-related stimuli (Brehmer, 1976; Jehn & Chatman, 2000). Nevertheless, when individuals have lower workloads, they are generally better able to make judgments and attributions. In this case, the lower workload should reduce the likelihood that task conflict predicts increases in relationship conflict (and vice-versa).

Hypothesis 2A: Workload moderates the relationship between current-day task conflict and next-day relationship conflict, such that this positive relationship is stronger when workload is higher versus lower.

Hypothesis 2B: Workload moderates the relationship between current-day relationship conflict and next-day task conflict, such that this positive relationship is stronger when workload is higher versus lower.

Although an increased workload can drain individuals of resources, the collective nature of the team tasks enables individuals to parcel the workload across team members. Consequently, ICE teams need to ensure that the workload is fairly distributed across the team, and that each member contributes their share to the task (Kanas et al., 2009). When individuals do not complete their share of the task – or engage in social loafing – they may view the task as unimportant (Karau & Hart, 1998), or they may consider themselves disconnected and ostracized from the task (Williams & Sommer, 1997). This can be a demoralizing emotional experience, breeding feelings of apathy and resentment (Kanas et al., 2009) that drains mental resources needed to distinguish between types of conflict. However, when individuals are able to complete their fair share can indicate the opposite sentiments (i.e., that the team member is engaged in the task and feels connected to the group). Contributing to the group can help reduce the likelihood that individuals view task disputes as challenges to their personal characteristics (e.g., competence; Tidd et al., 2004). Similarly, when individuals are able to contribute towards the task in a meaningful manner, they can perceive a greater sense of trust and respect with their teammates (Cronin, Bezrukova, Weingart, & Tinsley, 2011), which in turn has the potential to decrease both the effect of task on relationship conflict (Simons & Peterson, 2000), as well as relationship on task conflict (Choi & Cho, 2011). In this fashion, task contributions can serve as a buffering mechanism that reduces the extent the two types of conflict reinforce each other.

Hypothesis 3A: Task contribution moderates the relationship between current-day task conflict and next-day relationship conflict such that this positive relationship is weaker when task contribution is higher versus lower.

Hypothesis 3B: Task contribution moderates the relationship between current-day relationship conflict and next-day task conflict such that this positive relationship is weaker when task contribution is higher versus lower.

Effects of Resource Threats

Task Conflict to Strain.

Although some studies of task conflict have found positive effects with respect to performance-related outcomes (e.g., creativity, communication, and performance; de Wit et al., 2012; De Dreu & West, 2001; Schulz-Hardt et al., 2006), most reviews of conflict (e.g., De Dreu & Weingart, 2003; De Dreu, 2008) have suggested that all forms of conflict are negatively related to well-being related outcomes. As individuals experiencing task conflict argue for their perspective to be included and integrate differing opinions, the cognitive and emotional load of this process can theoretically result in a loss of resources that generate strain. In addition to these direct effects on well-being, task conflict is also associated with reduced team trust, satisfaction, commitment, and identification (de Wit et al., 2012). These are all sources of resource gain, which can be mitigated by the existence of task conflict, and therefore also hinder individuals from accumulating resources to combat other stressors. The potential downsides of task conflict related to these outcomes are likely to be heightened in ICE conditions wherein alignment around task roles and team identification and trust prove critical to mission success. Therefore, while others have suggested that task conflict may have benefits for broader performance-related outcomes, we posit that task conflict can have negative effects on well-being outcomes by increasing strain.

Hypothesis 4A: Current-day task conflict relates positively to next-day strain.

Relationship Conflict to Strain.

The research on relationship conflict is far more conclusive and demonstrates an almost uniformly negative relationship between relationship conflict and individual well-being. For instance, interpersonal (i.e., relationship) conflict has been noted as one of the most important workplace stressors (Keenan & Newton, 1985), and has been associated with strains such as anxiety and depression (Spector & Jex, 1998). Other studies have found that conflict positively relates to exhaustion and burnout, and can even result in psychosomatic strain (e.g., stomach cramps, rapid heart rate; for reviews, see De Dreu, van Dierendonck, & De Best Waldhober, 2002; De Dreu & Weingart, 2003). Although most of these studies examine between-person relationships, similar patterns have been observed in intraindividual designs. For instance, Ilies, et al. (2011) conceptualized negative affect as an indicator of strain and found that individuals’ negative affect increased moments after experiencing a conflict. However, the effects of conflict can persist beyond the momentary level. For instance, Wickham, Williamson, Beard, Kobayashi, and Hirst (2016) found that interpersonal conflict was negatively associated with an individual’s well-being on the next day.

Hypothesis 4B: Current-day relationship conflict relates positively to next-day strain.

Moderators.

A high workload can serve as a passageway that undermines resource conservation by limiting the number of resources individuals have to cope with stressors. As a result, a high workload can exacerbate resource loss associated with a given stressor as individuals have fewer resources to cope with the stressor and recover (Sonnentag & Fritz, 2015). Empirical evidence supports the notion of this exacerbation effect, whereby environmental conditions can increase stressor-strain relationships. For instance, Lavee, McCubbin, and Olson (1987) initially found that facing simultaneous stressors increases the effect of a specific stressor. More specifically, van Woerkom, Bakker, and Nishii (2016) showed that higher workload increased the association between emotional demands and absenteeism. However, the authors also suggested that a reduced workload would decrease the association between stressors and strain outcomes. These exacerbation effects suggest that workload has the potential to increase the resource drain caused by another stressor. Specifically, we expect that individuals will experience greater strain when they have conflict with each other during periods of high workload. Conversely, we expect that individuals experience less strain when they experience conflict during periods of low workload.

Hypothesis 5A: Workload moderates the relationship between current-day task conflict and next-day strain such that this positive relationship is stronger when workload is higher versus lower.

Hypothesis 5B: Workload moderates the relationship between current-day relationship conflict and next-day strain such that this positive relationship is stronger when workload is higher versus lower.

Additionally, as previously discussed, task contributions can offset the negative effects of conflict. Team members that contribute their fair share to group tasks may feel more connected to the group as a whole (Karau & Hart, 1998), which should limit the effect that conflict has on threatening the social bonds among team members. By bolstering the connection to the group, task contributions can lead to social support which will attenuate the impact of conflict on individual well-being. Alternatively, when team members are unable to perform their fair share, they may feel alienated and isolated from the group (Williams & Sommer, 1997). In this fashion, those who are less able to contribute may suffer greater harm when conflict occurs. Therefore, the more that individuals feel they can contribute towards the task, the less that conflict should generate strain.

Hypothesis 6A: Task contribution moderates the relationship between current-day task conflict and next-day strain such that this positive relationship is weaker when task contribution is higher versus lower.

Hypothesis 6B: Task contribution moderates the relationship between current-day relationship conflict and next-day strain such that this positive relationship is weaker when task contribution is higher versus lower.

Resource Loss Spiral

As individuals experience resource loss due to stressors, they tend to focus efforts on conserving their remaining resources (Hobfoll et al., 2016). In this manner, they focus on preventing future resource loss rather than gaining resources. As a result, individuals tend to become more susceptible to future resource losses (Halbesleben, 2010; Whitman, Halbesleben, & Holmes IV). This can cause a resource loss spiral, wherein the outcomes of the stressor generate increased levels of the experienced stressor.

In the context of conflict, individuals experiencing strain tend to have less ability to self-regulate emotions and behavior (Baumeister, Zell, Tice, 2007), which makes them more likely to express negative affect (Grandey, 2000) and engage in aggressive behavior (DeWall, Baumeister, Stillman, & Gailliot, 2007). These findings are consistent with the notion of a loss spiral, because they suggest that individuals experiencing strain seek to conserve resources and therefore allocate less resources towards self-regulation and conflict resolution. As a result, individuals can engage in “tit-for-tat” (Rubin, Pruitt, & Kim, 1994; Brett, Shapiro, & Lytle, 1998) or respond-in-kind (Weingart, Thompson, Bazerman, & Carroll, 1990; Weingart, Prietula, Hyder, & Genovese, 1999; Putnam & Jones, 1982) approaches, which are characterized by successive retaliations that increase the intensity of conflicts to a point of intractability. As a result, conflicts can escalate substantially when individuals lack self-regulatory resources and engage in retaliatory behaviors. Given that both types of conflict have the potential to increase strain, based on the logic of COR loss spirals, strain has the potential to increase both forms of conflict.

Hypothesis 7A: Current-day strain relates positively to next-day task conflict.

Hypothesis 7B: Current-day strain relates positively to next-day relationship conflict.

Summary

Based on our COR framework, we would expect that task and relationship conflict positively relate to one another over time. Additionally, we would expect that relationship conflict and strain positively relate to one another over time. Finally, we would expect that job conditions (i.e., lower workload and greater task contributions) serve to reduce the association between task to relationship conflict, relationship to task conflict, and relationship conflict to strain. This model is presented in Figure 1 and represents the interplay between a resource threat spiral and resource loss spiral based on COR theory.

Spirals and Stability of Resource Threats and Loss

There are many types of trajectories that a construct can take over time and COR theory suggests that resources and resource threats will generally exhibit a spiraling behavior (Hobfoll et al., 2016; Halbesleben et al., 2014). Although COR theorists have not specifically articulated the nature of loss and resource threat spirals, in the broader literature on dynamic phenomena, spirals are generally represented as the level of a construct endlessly increasing or decreasing over time (Dishop, Olenick, & DeShon, 2020). With respect to the constructs of interest in our study, a resource threat spiral implicitly suggests that relationship and task conflict both endlessly increase over time, and a resource loss spiral suggests that strain will endlessly increase over time as well (in the absence of any intervention). However, spirals are only one possible trajectory that a construct can display (Dishop et al., 2020; Monge, 1990) and there is little empirical evidence that these constructs will increase or decrease boundlessly. Indeed, some researchers suggest the opposite, and posit that conflict (e.g., Vallacher, Coleman, Nowak, & Bui-Wrzosinska, 2010) and strain (e.g., Ganster & Rosen, 2013) will instead converge towards an equilibrium point. In this case, the levels of the constructs (i.e., strain, relationship conflict, and task conflict) will trend upwards or downwards towards a stable amount, at which they will remain over time.

The topic of equilibria and stability has obvious implications for COR theory as it would contradict the proposition that resource threats and resource losses exhibit spiraling behavior. Instead, providing evidence for an equilibrium point would suggest that resource threats and losses reach a point of intractability. This point of intractability could represent an individual’s floor with respect to their resources, or their ceiling with respect to the value that their stressors (e.g., conflict) will take. Although there are many properties of equilibria worth exploring, one property of relevance is that once a construct reaches its equilibrium level, it can be relatively insensitive to shocks such as organizational interventions or stressful experiences (DeShon, 2012a, 2012b). In other words, although interventions can increase or decrease the level of the construct after they are applied, the construct will return back to its equilibrium point. This property has obvious relevance to the topic of stressor-strain relationships, because it suggests that isolated interventions may do little to reduce an individual’s strain over time, and the same can be said for conflict. Alternatively, a more positive implication of equilibrium is that shocks to the individual which generate spikes in strain or conflict will also have little lasting impact over time. Instead, finding evidence for an equilibrium would suggest that reducing strain and conflict involves changing the underlying aspects of the system in which these constructs are embedded (Olenick, Blume, & Ford, 2020). Specifically, one way to change these aspects can involve changing the level of another time-varying construct (e.g., workload or task contributions) that influences the focal construct (i.e., conflict, strain; Meadows, 2008).

Information regarding the equilibrium of a construct is represented in the magnitude of the autoregressive (AR) coefficients in a dynamic model. More specifically, DeShon (2012a, 2012b) provides simulation evidence that dynamic constructs with AR coefficients of absolute value less than 1 converge towards an equilibrium point. On the other hand, coefficients with absolute values greater than 1 will demonstrate explosive growth (i.e., neverending upwards or downwards spirals). When these AR coefficients are positive, the change over time occurs smoothly and when the AR coefficients are negative, the change occurs in an oscillatory fashion. Although COR somewhat suggests that our focal constructs should display an autoregressive value greater than 1, there exists little empirical evidence explicitly examining the autoregressive values of these constructs. Additionally, COR tenets are not very specific with respect to the operationalization of spirals, making it unclear whether the term spirals as used in COR refers to a specific type of trajectory that constructs display over time. Therefore, we explored this pattern of change as a research question.

Research Question 1: Do the constructs of task conflict, relationship conflict, and strain converge to an equilibrium point over time?

Method

Setting and Participants

We tested our model on a sample of 32 participants who engaged in the Human Exploration Research Analog (HERA), which is a simulated crew capsule designed to emulate an asteroid transit mission. The HERA habitat is vertical cylinder (19.3 feet in diameter). It is comprised of three work levels, including a hygiene module and simulated air lock, with a half-level loft on top for sleeping quarters. It is approximately the size of a small two-room apartment, with 636 square feet space of space. Figure 2 provides a general graphic of the habitat and a virtual tour of the HERA habitat can be found on the HERA webpage (https://herastudy.jsc.nasa.gov/).

Figure 2.

Cutaway diagram of the HERA habitat.

Note: see https://www.nasa.gov/sites/default/files/atoms/files/2019_hera_facility_capabilities_information.pdf for more detail.

Participants were recruited through online solicitation on the NASA website and underwent health screening prior to entering HERA (https://herastudy.jsc.nasa.gov/apply). For mission simulations, individuals were grouped into teams of four to live within the HERA habitat for 30-45 days. Over a seven-day period, participants adhered to a tightly regimented schedule for 5.5 consecutive days with 1.5 consecutive days reserved for off-duty activities. Every day lights turned on at 7 A.M. and off at 11 P.M. As a result, participants were awake for 16 hours (no napping was permitted), during which they completed team-oriented, space mission relevant tasks. Participants also engaged in daily exercise and had weekly housekeeping chores. A sample schedule is provided in Appendix A.

Participants were provided three meals per day, based upon caloric requirements calculated using the Mayo Clinic calculator. Meals consisted of food packaged for space travel as there is no refrigeration inside of HERA. Finally, participants had no access to internet, social media, television, radio, or telephone. They were also continuously monitored during isolation, excluding sleep quarters, hygiene module, and private conferences.

In summary, the HERA habitat represents a replica of the vessel in which astronauts would live while in outer space. The tasks that the participants went through were structured in the same fashion as the tasks that astronauts would complete, and their confinement and isolation from broader society was directly comparable to the conditions experienced by astronauts. Therefore, the conditions that the participants experienced and tasks they completed were highly representative of an isolated and confined work environment and provide high psychological fidelity to life in an ICE occupation (Landon, Slack, & Barrett, 2018; Nasrini et al., 2020).

Sample

Our sample comprised 32 participants working together in teams of four for 30-45 days in HERA. Half of the missions were 30-day missions and half were 45-day missions. Previous research has shown that this habitat provides a high-fidelity setting for ICE environments (Landon et al., 2018). Out of the participants, 19 (59.4%) were male, 12 (37.5%) were female, and one did not disclose their gender identity (3.1%). In terms of race, 23 (74.2%) were non-Hispanic White, two (6.5%) were Hispanic White, two (6.5%) were Asian, two (6.5%) were Black American, and three (6.5%) did not state their cultural/ethnic group. We administered surveys at the end of the individuals’ operational day. We had a 96% response rate at the day-level; missing responses were deleted listwise. After accounting for 50 day-level missing responses, our final sample had 1191 daily (Level 1) and 32 person-level (Level 2) observations.

Measures

Conflict.

Conflict was measured using separate task conflict (TC) and relationship conflict (RC) measures. Both task conflict and relationship conflict were assessed using a one-item measure on a 6-point Likert scale (1 = “Strongly Disagree”, 6 = “Strongly Agree”). The relationship conflict item is “I had conflicts with other people on my team that were personal in nature” and the task conflict item is “I had conflicts with other people on my team about how to perform tasks”.

Strain.

Strain was measured using a single-item measure on a 4-point Likert scale (1 = “Not at All”, 4 = “To a Large Extent”). The item is “To what extent did you experience health difficulties today?”.

Workload.

Workload was measured using a six-item measure on a 5-point Likert scale (1 = “Very Low”, 5 = “Very High”). An example item is “How hurried or rushed was the pace of your work?” (Hart & Staveland, 1988). Reliability of this scale was α = .77 calculated as the average alpha score across measurement occasions.

Task Contributions.

Task contribution was measured using a one-item measure on a 6-point Likert scale (1 = “Strongly Disagree”, 6 = “Strongly Agree”). The item is “I completed a fair share of work with regard to what other people on my team were doing.”

Scale Validation

Most of our variables were measured using single-item scales due to the nature of the data collection. Specifically, our study was part of a larger data collection effort which required participants to take an assessment battery every night over the course of the mission. To reduce the likelihood of response biases associated with respondent fatigue, we administered shorter surveys. In this regard, previous authors have noted that single-item measures can serve as a useful approach to gathering intensively longitudinal data (Gabriel et al., 2019; Fisher & To, 2012; Uy, Foo, & Aguinis, 2010). However, to ensure that our items captured the constructs in question, we performed a scale validation study by correlating the single items against existing scales.

Sample.

Our validation sample comprised 271 participants from a large midwestern university. Out of the participants, 53 (19.6%) were male, 217 (80.1%) were female, and 1 did not disclose their gender identity (.3%). In terms of race, 203 (74.9%) were non-Hispanic White, 36 (13.3%) were Asian, 15 (5.5%) were Black American, 1 was Native American (.4%) and 16 (5.9%) did not state their cultural/ethnic group.

Procedure.

We asked participants to reflect on their most recent experience working as part of a team or work group. Then, we asked them to describe the positive and negative aspects of this experience to prime memory. Finally, we asked them to answer questions on measures of task and relationship conflict, psychological and physiological symptoms, social loafing, task distribution, and competence needs. Note that we expected positive correlations between the single-item task conflict, relationship conflict, and strain measures with the validated measures of task and relationship conflict (Bendersky & Hays, 2012), and strain (Spector & Jex, 1998). However, we expected a negative correlation between our item of fair contributions and social loafing (George, 1992), because social loafing involves individuals not completing their fair share of the task. Full scales are presented in Appendix B.

Results.

Correlations between our single-item measures and the validated scales from the literature are presented in Table 1, and the results showed strong positive correlations between our single-item measures and their respective validated counterparts except for social loafing and fair contributions which displayed a small negative correlation. To provide further support for construct validity, we estimated two measurement models. The first model was a freely estimated confirmatory factor analysis in which the items loaded onto latent factors based on the scale they were drawn from. In this model, the single-item measures were loaded onto the same latent factor as their validated scale item counterparts (Figure 3). Next, a second measurement model was estimated in which the loadings and intercepts for the single-item measures were constrained to zero. Comparing the fit statistics enabled us to determine whether the original scales would have better or worse fit if the single-item measure was not included. Change in model fit (Δχ²(df) = 382(4), p < .001) suggested that the freely estimated model provided a substantially better fit. Additionally, the loadings of the single-item measures on their corresponding latent factors were all significant (Figure 3). Together, these results provided evidence for construct validity for our single-item measures.

Table 1.

Validation study means, standard deviations, and correlations with confidence intervals

Note. M and SD are used to represent mean and standard deviation, respectively. Values in square brackets indicate the 95% confidence interval for each correlation. The confidence interval is a plausible range of population correlations that could have caused the sample correlation (Cumming, 2014). “Single-item” subscript denotes the single-item measure used in our study; “validated” subscript denotes the validated scale drawn from the literature.

^†indicates p < .10.

^*indicates p < .05.

^**indicates p < .01.

Figure 3.

Measurement models for scale validation study.

Note: *indicates lab-created item. TC = task conflict. RC = relationship conflict. S = strain. SL = social loafing.

Analytic Strategy

To test our hypotheses, we used random-coefficient modeling (RCM) to account for the nested data structure, and person-mean centered the predictor variables (Bliese & Jex, 2002; Wang & Maxwell, 2015). In our model, we regressed next-day task conflict onto current-day strain and relationship conflict, next-day relationship conflict onto current-day strain and task conflict, and next-day strain onto current-day relationship conflict and task conflict. We also computed two-way interaction terms to identify the moderation effects of workload and task contributions. All models were estimated using the lavaan package in R (Rosseel, 2012) and simple slopes were computed for significant interaction terms using the emmeans package in R (Lenth, Singmann, Love, Buerkner, & Herve, 2018). Note that since our hypotheses were fully directional, we utilized one-tail tests. However, when we explored the simple slopes of our interaction terms, we utilized two-tail tests since we did not have any directional hypotheses regarding these slopes.

The ICCs for task conflict [ICC(1)_person = .43; ICC(1)_team = .13], relationship conflict [ICC(1)_person = .39; ICC(1)_team = .13], and strain [ICC(1)_person = .28; ICC(1)_team = .05] suggested that there was substantial between-person variance, but there was little variance between-teams. Additionally, we performed a variance decomposition to parse the amount of variance explained in our variables across time between individuals and between teams. The results of the variance decomposition are presented in Table 2 and suggest that between-team differences did not account for variance in the variables over and above the variance accounted for by within- and between-person differences. Finally, to further ensure that between-team differences did not influence the results of our model, we ran a separate model with fixed effects for team membership (Verbeek, 2008). The results of the analysis did not change and are included in Appendix C.

Table 2.

Results of variance decomposition.

Note: F scores are calculated incrementally so F_individual = MS_individual / MS_Residual, and F_team = MS_team / MS_individual. As such F_individual statistics test whether person-level differences account for a significant portion of the variance beyond within-person residual variance, and F_team statistics test whether team membership accounts for a significant portion of the variance in data beyond individual variation and within-person residual variance.

^***p < .001

Results

Unit Roots Test

Stationarity is a methodological prerequisite to analyzing longitudinal data with an ordinary least-squares (OLS) or maximum likelihood (ML) estimator. These are the primary estimators for RCM and will generally produce biased and/or spurious estimates when data are nonstationary (Kuljanin, Braun, & DeShon, 2011; Braun, Kuljanin, DeShon, 2013; Beal, 2015). Therefore, we tested task conflict, relationship conflict, and strain variables independently for non-stationarity by following the Im-Pesaran-Shin panel roots test (IPS: Im, Persaran, & Shin, 2003). In this approach, stationarity is assessed by averaging independent Dicky-Fuller test statistics for each participant. The IPS test was conducted using the plm package for R (Croissant & Millo, 2008). Results indicated that strain (W_tbar = −15.47, p < .001), relationship conflict (W_tbar = −10.07, p < .001), and task conflict (W_tbar = −10.28, p < .001) were stationary at a single-lag length. By establishing stationarity, we were able to proceed with further RCM analyses.

Primary Analyses

Descriptive statistics are presented in Table 3, and the results for the multi-level path model are presented in Table 4. Simple slopes for significant interactions are presented in Table 5. In testing resource threat spirals, we found support for Hypothesis 1A as task conflict predicted next-day relationship conflict [γ(SE) = .15(.03), p < .001], but not for Hypothesis 1B as relationship conflict did not predict next-day task conflict [γ(SE) = −.003(.04), p = .47]. Additionally, we found support for Hypotheses 2A and 2B as workload moderated the associations from task to next-day relationship conflict [γ(SE) = .18(.04), p < .001], as well as from relationship to next-day task conflict [γ(SE) = .21(.05), p < .001]. Specifically, through simple slopes analyses (Figure 4), we found evidence for a cross-over interaction as relationship conflict was negatively related to next-day task conflict when workload was lower [γ(SE) = −.11(.05), p = .005], but positively related to next-day task conflict when workload was higher [γ(SE) = .10(.04), p = .005].

Table 3.

HERA data means, standard deviations, and both within and between person correlations.

Note: M and SD represent mean and standard deviation, respectively. Values in square brackets indicate 95% confidence interval. Value above the diagonal represent between-person correlations based on the within-person mean scores. Values below the diagonal represent within-person correlations (Bakdash & Marusich, 2017).

^†indicates p < .10.

^*p < .05.

^**p < .01.

^***p < .001.

Table 4.

Results of multilevel path analysis.

Note:

^†p < .10.

^*p < .05.

^**p < .01.

^***p < .001 based on one-tail tests, t = current-day and t+1 = next-day measurement. CFI = .88; RMSEA = .08; SRMR_within = .01.

Table 5.

Results of simple slopes analyses.

Note:

^†p < .10.

^*p < .05 based on two-tail tests. CI = 95% confidence interval. All simple slope estimates reflect slopes at 1 SD above and below the mean. t = current-day and t+1 = next-day measurement.

Figure 4.

Simple slopes for the interaction of current-day relationship conflict and next-day workload on next-day TC.

Note: relationship conflict is person-mean centered and task conflict is not centered.

Similarly, using simple slopes (Figure 5), we found that task conflict was not related to next-day relationship conflict when workload was lower [γ(SE) = .05(.04), p = .09], but was more positively related to next-day relationship conflict when workload was higher [γ(SE) = .24(.04), p < .001]. However, we did not find evidence of task contributions moderating the effects of task to next-day relationship conflict [γ(SE) = .01(.06), p = .45], or relationship to next-day task conflict [γ(SE) = −.05(.09), p = .27]; therefore, Hypotheses 3A and 3B were not supported. Together, the results suggest that workload can play a critical role in the extent that task and relationship conflict relate to one another, and that the amount of work has more impact on individual well-being rather than the amount that individuals contribute to the work.

Figure 5.

Simple slopes for the interaction of current-day task conflict and next-day workload on next-day RC.

Note: task conflict is person-mean centered and relationship conflict is not centered.

When evaluating the impact of conflict on strain, we did not find support for Hypothesis 4A as task conflict was not related to next-day strain [γ(SE) = −.03(.02), p = .06]. However, we did find support for Hypothesis 4B as relationship conflict was positively related to next-day strain [γ(SE) = .07(.02), p < .001]. Therefore, it appears that strain is largely a function of interpersonal rather than task disagreements. However, workload did not moderate the associations between task conflict [γ(SE) = −.01(.03), p = .42] or relationship conflict [γ(SE) = −.01(.04), p = .40] and next-day strain. Additionally, task contributions did not moderate the associations between task conflict [γ(SE) = .004(.04), p = .47] or relationship conflict [γ(SE) = −.05(.06), p = .18] and next-day strain either. Finally, in testing the presence of resource loss spirals, we found that strain was not related to next-day task conflict [γ(SE) = .05(.05), p = .19], but was positively related to next-day relationship conflict [γ(SE) = .09(.05), p = .03].

Together, our findings indicate the presence of a double feedback loop. Specifically, this feedback loop suggested that when workload is high. task conflict leads to relationship conflict and relationship conflict in turn leads to higher task conflict. At the same time, relationship conflict leads to strain which can lead to relationship conflict. However, the results of our analyses also suggest this double feedback loop may not emerge under certain job conditions, as task conflict did not predict relationship conflict and relationship conflict was negatively related to next-day task conflict when workload was low. Therefore, these findings provide evidence for the notion of resource threat and resource loss spirals as discussed by COR under conditions of high workload. However, to determine whether these feedback loops generate spiraling behavior in strain and conflict, we conducted equilibrium analyses.

Equilibrium Analyses

We explored the equilibria of task conflict, relationship conflict, and strain and calculated points of stability using Equation 1, wherein y_e refers to the equilibrium point of the construct, b₀ refers to the intercept, b₁ refers to the value of the AR coefficient, and x refers to any covariate with a coefficient b₂. The derivation for this equation is provided in Appendix D, and we refer interested readers to Wainwright (2005) for a more thorough treatment of this topic.

$$y_e=\frac{b_0+b_{2}x}{1-b_1}$$ (1)

We found that the coefficients for the AR terms were relatively small in relation to the intercept values, which suggested that equilibrium was largely determined by their intercepts. Moreover, we also found that the equilibrium for each of the constructs increased or decreased concomitantly with their covariates. For instance, using the results of our model, individuals’ average strain equilibrium can be calculated using Equation 2, which suggests that the equilibrium of strain (S) will shift upward slightly as the level of previous-day relationship conflict (RC) and current-day workload increases (calculations for the equilibrium of each construct are provided in Appendix D). Similarly, our results suggest that the equilibrium of task conflict (TC) increases concomitantly with current-day workload and the interaction of relationship conflict and workload (Equation 3). On the other hand, relationship conflict did not relate to itself over time. However, task conflict was positively related to relationship conflict, and relationship conflict was positively related to task conflict. This pattern of results suggests that relationship conflict can influence itself through task conflict. For instance, when task conflict reaches an equilibrium (i.e., Equation 3), its equilibrium value will equal its value at all points in time. In other words, the lagged task conflict term will equal its equilibrium value. In this case, the equilibrium of relationship conflict (Equation 4) can be calculated as a function of itself by inserting the equation for TC_e in place of TC_t–1.

$$S_e=1.72+0.9R{C}_{t-1}+.16workloa{d}_t$$ (2)

$$T{C}_e=1.98+.003R{C}_{t-1}workloa{d}_t+.18workloa{d}_t$$ (3)

$$R{C}_e=1.59+.08S_{t-1}+.16workloa{d}_t+.15T{C}_{t-1}+.18T{C}_{t-1}workloa{d}_t$$ (4)

In fact, an interesting finding regarding this system of equations is that when all constructs reach equilibrium, the terms of Equations 2 and 3 can respectively be substituted into Equation 4. A notable property of this system of equations emerges when Equation 3 is substituted into the task conflict terms of Equation 4. Then, by factoring out the relationship conflict term of the resulting equation and further simplifying, the equilibrium for relationship conflict can be written entirely as a function of workload (Equation 5). By substituting Equation 5 into the relationship conflict terms of Equations 2 and 3, the equilibria for task conflict and strain can also be written as a function of workload (Equations 6 and 7). Therefore, our analyses suggest that workload effectively calibrates the stable levels of conflict and strain.

$$R{C}_e=\frac{2.03+.46workloa{d}_t+.032workloa{d_t}^2}{(.99-.005workloa{d_t}^2-.008workloa{d}_t)}$$ (5)

$$S_e=1.72+.09\left(\frac{2.03+.46workloa{d}_t+.032workloa{d_t}^2}{.99-.005workloa{d_t}^2-.008workloa{d}_t}\right)+.16workloa{d}_t$$ (6)

$$T{C}_e=1.98+.003\left(\frac{23+.46workloa{d}_t+.032workloa{d_t}^2}{.99-.005workloa{d_t}^2-.008workloa{d}_t}\right)+.18workloa{d}_t$$ (7)

To provide a more intuitive illustration of the implications of our equilibrium analyses, we have provided a simulation of the long-run levels in relationship conflict, task conflict, and strain, based on the regression estimates in our dynamic model (Figure 6). Using these estimates, we simulated the trajectory of each variable across varying conditions of workload. For each condition, we instituted a shock that drove the value of the variable upwards or downwards. The results of our analyses indicate that across all conditions, experiencing a shock did little to change the conflict and strain levels of the individual over time, as the levels returned quickly back to an equilibrium. Additionally, they highlight the importance of reducing workload, as workload reductions drove the equilibrium of each variable downward.

Figure 6.

Construct levels over time based on regression equations.

Note: w = average workload level on a scale of 1 to 5 (1 = low workload, 5 = high workload).

Exploratory Analyses

Although we did not hypothesize a particular directionality for task conflict and relationship conflict, we found that current-day task conflict directly predicted next-day relationship conflict, but not subsequent strain, and that current-day relationship conflict directly predicted subsequent strain. These results suggested a mediational pathway whereby task conflict may be related to subsequent strain through relationship conflict. We tested the indirect effect of task conflict on strain by regressing a two-day lag of strain on a one-day lag of relationship conflict, and the one-day lag of relationship conflict on current-day task conflict. The indirect effect was estimated using the mediate function in R, which calculates the indirect effect using the Sobel method and bootstrapped standard errors (5000 samples). Our results are presented in Table 6 and indicate a significant indirect effect of task conflict on strain [b(SE) = .01(.01), p = .03, 90% CI = (.0004, .02)]. We present an updated model based on the results of our analyses in Figure 7.

Table 6.

Mediation analyses of current-day task conflict to two-day lag strain through one-day lag relationship conflict.

Note:

^†p < .10.

^*p < .05.

^**p < .01.

^***p < .001 based on two-tail tests. t = current-day, t+1 = next-day measurement, and t+2 = day after the next-day measurement.

Figure 7.

Updated model based on the results of our study.

Discussion

Our study found evidence for complex and dynamic relationships between conflict and strain in ICE teams. Specifically, we found evidence for a spillover effect as task conflict was related to subsequent strain through its positive association with relationship conflict. Moreover, we found evidence for dual feedback loops as task and relationship conflict were positively associated with each other, and relationship conflict and strain were positively associated with each other when workload was high. Furthermore, we found that the presence of resource threat and resource strain feedback loops did not result in spiraling behavior, as both forms of conflict and strain converged to an equilibrium point that increased and decreased concomitantly with the amount of workload an individual experienced at a given point in time. These results present implications for organizational theory and practice.

Theoretical Implications

The results of our study have several implications for COR theory and conflict research. First, consistent with previous research, we found that relationship conflict served as a stressor that negatively affected individuals’ well-being (De Dreu & Weingart, 2003). However, we found that task conflict affected strain through relationship conflict, implying that task conflict will only negatively affect individuals’ well-being when accompanied by relationship conflict. Indeed, previous meta-analyses (e.g., de Wit et al., 2012) have suggested that task conflict is not necessarily harmful and, in some conditions, can even be productive for other important team outcomes such as performance and decision-making. Although these studies did not focus on well-being outcomes, our results complement previous findings by further suggesting that task conflict is not necessarily harmful to individual well-being and health as long as it is decoupled from relationship conflict. In this regard, we found that while contributing towards a task did little to decouple task conflict and relationship conflict, reducing workload did have a substantial impact in mitigating the negative effects of conflict. Theoretically, our findings provide evidence for the notion that reducing sources of resource loss (i.e., workload) has more potential for mitigating the negative effects of conflict than increasing sources of resource gain (i.e., task contributions) (Halbesleben, et al., 2014; Hobfoll, et al., 2016).

Relatedly, our results also provide evidence for dual feedback loops, as all constructs were predictive of each other during periods of high workload. Specifically, our findings of the task and relationship conflict feedback loop are consistent with the notion of loss spirals and highlight how stressors can mutually reinforce each other. Additionally, the relationship conflict and strain feedback loop suggests that the outcomes of resource loss can reinforce the cause of resource loss, as strain was positively associated with relationship conflict. From a COR perspective, the two types of feedback loops portray different forms of loss spirals. The task and relationship conflict feedback loop displays a loss spiral characterized by mutual reinforcement between co-occurring stressors. In this type of spiral, the levels of the stressor increase each other, likely due to an inability to cope with either single stressor (Halbesleben, 2010; Hobfoll, et al., 2016). However, the relationship conflict and strain feedback loop is consistent with a loss spiral in which a stressor lowers the resources that an individual has available to cope with the stressor, leading to defensive and aggressive behavior by the individual to maintain their remaining resources with a consequent increase in their interpersonal conflict (Halbesleben, et al., 2014).

In sum, the results are consistent with the notion that resource loss can occur as part of a dynamic system in which the effects of one resource drainer (i.e., conflict) permeate to other connected variables within its nomological network. Therefore, although traditional investigations of conflict and stress have examined the effects of resource loss through unidirectional models, a systems model that incorporates bidirectionality may provide a more appropriate description of an individual’s resource loss process (Cronin & Bezrukova, 2019). Although the feedback loops suggest that resource loss can spread in multiple ways, it also implies that there are multiple points of intervention that may reduce the effects of the stressor on individuals’ well-being. In particular, reducing the association between strain and relationship conflict can limit the extent to which such spirals occur. Similarly, reducing workload could make the association between relationship conflict and task conflict nonsignificant. In this case, relationship conflict would not provide positive feedback to task conflict and therefore limit the extent to which the effects of conflict propagate over time.

Finally, our results found strong evidence for stressors and strain levels converging towards an equilibrium. This finding is not consistent with COR and suggests that resource threat and loss “spirals” do not endlessly intensify, but rather converge towards a stable point. The concept of stability provides further temporal precision to the notions presented in the COR framework. Although our research found evidence for reciprocal relationships between relationship conflict and task conflict, as well as between relationship conflict and strain, these relationships did not generate true spiraling behavior. This is important as COR theorizing suggests that these positive feedback loops will generate upward spiraling behavior. However, we find instead that the levels of these constructs converge towards an equilibrium. It appears then that positive feedback and spiraling behavior are two separate concepts, and the description of resource threat and loss spirals reflect positive feedback loops rather than true spirals. Moreover, incorporating stability into the COR framework is important because it implies the impermanence of single, event-based effects on the levels of stressors and strain over time. In other words, finding evidence for stability suggests that experiencing either a stressful or a resource gaining event can provide substantial impact in the short run, but the effects of this event will not persist over time.

The finding is good news as it suggests that individuals do not experience an unending increase in strain as they lose resources and provides more concrete direction for intervention efforts. Specifically, our findings suggest that standalone interventions may do little to decrease individuals’ levels of strain, but that changing job conditions by reducing workload at key points can help reduce the equilibrium levels of stressors and strain. The results of our equilibrium analyses also suggest that individuals may not experience lasting strain from an isolated event, as they will converge back towards an equilibrium after experiencing a spike in conflict or strain.

Practical Implications

Individuals operating in ICE conditions face extreme environmental pressures and require successful social interaction to ensure mission success and survival. Our findings provide actionable recommendations for teams seeking to mitigate the harmful effects of conflict while operating in these conditions. First, our results suggest that merely encouraging individuals to contribute towards their group goal will not necessarily provide positive benefits in terms of buffering the negative effects of conflict. This points to the importance of conducting more research in ICE environments. Findings from prior research on individuals in non-ICE conditions suggest that an equitable distribution of task workload is a resource gain because it provides a sense of fairness and implies that individuals are integrated within the team. However, for individuals in ICE environments, this mitigating effect seemingly does not hold. Ultimately, our findings suggest that the amount of workload – rather than its perceived distribution – dictates how conflict affects individuals.

Indeed, the effects of relationship conflict on task conflict depend on the amount of workload, such that relationship conflict was negatively associated with task conflict when individuals had lower workload. Similarly, task conflict was not associated with relationship conflict when workload was low. This suggests that when one form of conflict occurs, reducing workload can prove useful to limiting the possible spillover between task and relationship conflict. Furthermore, the results of our mediation analysis also indicated that task conflict was harmful to individual well-being through increased relationship conflict. Therefore, reducing workload may decouple task conflict and relationship conflict, which could prove useful not only for mitigating the negative effect of conflict on team members’ well-being, but also in maximizing the potential benefits of task conflict related to innovation and decision-making (De Dreu, 2008).

In fact, workload proved critical for not only governing the relationship between relationship conflict and task conflict, but also shifting the equilibrium levels of all three variables. This suggests that proactive workload management strategies could be useful for mitigating strain and reducing conflict feedback loops. Such strategies could involve facilitating periods of lower workload after relationship conflicts arise or providing activities that can generate short-term recovery. Such activities can range from fairly simple actions (e.g., taking micro-breaks; Zacher, Brailsford, & Parker, 2014) to more extensive initiatives (e.g., virtual reality escapes; Ahmaniemi, Lindholm, Muller, & Taipalus, 2017). Nevertheless, any of these practices have the potential to facilitate psychological recovery that reduces the extent that workload reinforces a given conflict.

However, our results also suggest that merely encouraging individuals to contribute towards their group goal will not necessarily provide positive benefits in terms of buffering the negative effects of conflict. Other researchers have suggested that learning climates and other techniques that integrate individuals by encouraging mutual trust and respect can lead to the reduction of intragroup division (Harvey, Leblanc, & Cronin, 2019; Cronin, et al., 2011). This may further help bridge differences in task representation without incurring negative personal attributions that can occur during task disagreements. Such approaches may hold similarly positive implications for reducing strain.

Limitations and Future Directions

Our study was limited due to the unique challenges of collecting data in ICE environments. Although the data collection duration was long, the number of individuals and teams observed in HERA was smaller than one would typically find in more traditional team studies. That limited sample size reduced statistical power and the ability to detect between-person and between-team differences in within-person variability. Although our focus rested on understanding how strain and conflict changed within individuals over time, and evaluating these longitudinal patterns against COR theory, future research should identify boundary conditions that determine whether these temporal patterns systematically differ across individuals. If so, it will enable more precise intervention efforts. Relatedly, ICE teams may differ based on the type of task that they perform and the individuals they select for these tasks. Differences in the composition and structural nature of the team could result in team-level differences that mitigate or facilitate the impact of conflict over time (Park, Mathieu, & Grosser, 2020). Therefore, examining team-level differences in conflict-strain patterns could prove to be a fruitful direction for future research.

Additionally, our investigation focused primarily on day-level relationships. Unfortunately, there simply does not exist much research that specifies the optimal length of time over which to observe the constructs in our study. Nevertheless, there are studies which suggest that the effects of conflict can be felt across days (Wickham et al., 2016; see Cronin & Bezrukov, 2019 for a review) and that its effects (i.e., strain) can be observed at the daily level. Moreover, we focused on conflict at the daily level because of the context of our sample. Previous research has strongly suggested that individuals in ICE environments are likely to feel the consequences of stressors to a great extent due to the already stressful nature of their physical environment (e.g., Golden et al., 2018). This is especially true for social stressors such as conflict, because the individuals are confined together, and cannot restore lost resources at the end of the workday by psychologically distancing themselves from their job and the people in the context as they can in a non-ICE setting. Therefore, based on previous research, and on the unique environmental context of our sample, we reasoned it likely that the effects of conflict would manifest across days rather than over longer time windows. We also thought it more informative to examine conflict at the day level because each day in an ICE environment is important for team performance and for individual well-being and survival. Finally, given that we found that the data were stationary, we are confident that the relationships observed at the day level would remain constant even if we extended the observational window. Therefore, we do not feel that a change in temporal intervals would substantively change the findings. Nevertheless, future research should examine whether the relationships between stressors and strain differ across different time scales to better bound the results of our study.

Due to the intensive longitudinal nature of the data collection and time constraints set by habitat work schedules, single-item measures, which have unknown reliability and assume no measurement error, were utilized to capture the constructs of interest. However, these challenges are common when studying ICE teams and are not easily overcome (Bell et al., 2018; Golden et al., 2018; Beal, 2015). Relatively few ICE teams exist compared to the types of teams typically studied in organizational research. This relative rarity and uniqueness represent one reason ICE teams are interesting and important to study, but it also makes samples small. Moreover, studying ICE teams is difficult because they are difficult to access, making data collection a time and resource intensive process for researchers and participants. Nevertheless, these challenges provide future research directions for ICE researchers.

First, emphasizing the importance of ICE research will enable researchers to build databases and accumulate a larger body of research. Further research on conflict in ICE teams over time can provide opportunities for verification (or refutation) for the results found in the present study (e.g., meta-analyses) and can better identify when conflict affects individuals the most over the course of their mission. Additionally, larger data collection efforts would enable more complex studies that examine multi-level boundary conditions in within-person variability. In relation to our study, such efforts may focus on how individual differences in personality traits (e.g., optimism, neuroticism) and preferred conflict resolution methods (e.g., avoidance, compromise) affect the relationships between conflict and well-being over time. Similarly, team-level factors such as group composition may also help identify how potential faultlines emerge, and which types of faultlines produce the most harmful conflicts. Additionally, we were not able to model all of the cognitive and perceptual processes that may be at play. Therefore, future research would benefit from examining the more micro-processes (e.g., negative impressions and reactions) through which task and relationship conflict are thought to affect one another. Such research would provide more specificity regarding the mechanisms through which conflict spillover occurs. To that end, unobtrusive data collection through remote sensors may help overcome measurement and access limitations associated with ICE research by reducing the time and effort participants face in completing daily self-reports in intensive longitudinal designs. This approach has the potential to provide greater behavioral fidelity and capture a wider range of individual and group behaviors that are not easily evaluated using questionnaire style assessments (Kozlowski, 2015; Kozlowski, Chao, Chang, & Fernandez, 2015). In this manner, unobtrusive measures can better measure how conflicting individuals interact and how the conflict ripples across the team.

Conclusion

As organizations continue to push the boundaries of the human experience, workers will increasingly engage in long-duration missions within highly immersive ICE work environments. The unique challenges these workers face will dramatically increase the level of their interpersonal interactions. Conflict will almost inevitably occur as these workers seek to complete group objectives under life-or-death conditions in a closed social system. Our results suggest that conflict and strain affect each other in a mutually reinforcing fashion and converge towards an equilibrium that shifts in response to changing workload. Parsing through these dynamics will help ICE teams design more effective stress reduction strategies and improve worker well-being.

Appendix

Appendix A: Sample HERA Schedule

Time	Activity
7:00 AM	Post-sleep period, includes morning meal
8:30 AM	Conferences and work preparation
10:30 AM	Work (scheduled research and operations tasks)
2:30 PM	Mid-day meal
3:30 PM	Work (scheduled research and operations tasks)
6:30 PM	Exercise period
9:30 PM	Pre-sleep period, includes evening meal
11:00 PM	Sleep

For more details, see: https://www.nasa.gov/sites/default/files/atoms/files/2019_hera_facility_capabilities_information.pdf

Appendix B: Measures used in Scale Validation Study

Instructions: Based on the team experience you described above, indicate the extent to which you agree with the following statements.

Task Conflict:

• My team members experienced conflict of ideas.

• My team members frequently had disagreements about the task we were working on.

• My team members often had conflicting opinions about the task we were doing.

Relationship Conflict:

• My team members experienced relationship tension that was not related to the task.

• My team members often got angry while working in this team.

• My team members experienced emotional conflict.

Social Loafing:

• I deferred to others the responsibilities I should have assumed

• I put forth less effort on the job when others were around to do the work

• I did not do my share of the work

• I spent less time helping when others were present

• I put forth less effort than other members of my work group

• I avoided performing housekeeping tasks as much as possible

• I left work for others which I should have really completed

Spector and Jex (1998)

Instructions: During the team experience you described previously, how often did you experience the following symptoms?

Physical Strain:

• An upset stomach or nausea

• A backache

• Trouble sleeping

• Headache

• Acid indigestion or heartburn

• Eye strain

• Diarrhea

• Stomach cramps (Not menstrual)

• Constipation

• Ringing in the ears

• Loss of appetite

• Dizziness

• Tiredness or fatigue

Appendix C: Results of Alternate Multilevel Models

Model with Fixed Effects for Team Membership

	Task Conflict (t+1)	Relationship Conflict (t+1)	Strain (t+1)
Predictors	γ (SE)	γ (SE)	γ (SE)
Intercept	−.576 (2.669)	.688 (2.158)	3.030 (1.251)*
Main Effects
Task Conflict (t)	.086 (.033)**	.148 (.030)***	−.029 (.019)
Relationship Conflict (t)	−.003 (.036)	.026 (.032)	.068 (.021)***
Distress (t)	.045 (.051)	.086 (.046)†	.249 (.029)***
Workload (t+1)	.258 (.051)***	.159 (.046)***	.122 (.029)***
Task Contribution (t+1)	.013 (.052)	−.030 (.047)	−.027 (.030)
Two-Way Interactions
Task Conflict (t) x Workload (t+1)		.185 (.044) ***	−.009 (.035)
Task Conflict (t) x Task Contribution (t+1)		.006 (.061)	−.051 (.056)
Relationship Conflict (t) x Workload (t+1)	.209 (.053)***		−.008 (.032)
Relationship Conflict (t) x Task	−.055 (.086)		.003 (.044)
Contribution (t+1)
Fixed Team Intercepts
T2	.303 (.448)	−.176 (.363)	−.406 (.210)†
T3	−.384 (.448)	−.437 (.362)	−.355 (.210)†
T4	.355 (.448)	.480 (.362)	.003 (.210)
T5	−.062 (.446)	−.075 (.360)	−.165 (.208)
T6	.424 (.453)	.277 (.367)	−.258 (.213)
T7	.451 (.446)	−.006 (.360)	−.207 (.208)
T8	1.043 (.446)*	.746 (.360)*	−.162 (.208)

Note: t = current-day and t+1 = next-day measurement.

^†p < .10.

^*p < .05.

^**p < .01.

^***p < .001

Appendix D: Calculating Equilibria

Deriving the General Formula for Calculating Equilibrium

An autoregressive model for a construct y can be written as follows:

$$y_t=b_0+b_1{y}_{t-1}$$ (1)

When y reaches its equilibrium (y_e), its value will equal any future value (Wainwright, 2005), which can be represented as follows:

$$y_e=y_t=y_{t+1}$$ (2)

Therefore, when y reaches its equilibrium, we can calculate the value of the equilibrium by assuming that all y terms (past and future) are equivalent:

$$\begin{gathered} y_e=b_0+b_1{y}_e \\ \to y_e-b_1{y}_e=b_0 \\ \to y_e(1-b_1)=b_0 \\ \to y_e=\frac{b_0}{1-b_1} \end{gathered}$$ (3)

Note that this equation can be extended to include covariates:

$$\begin{gathered} y_e=b_0+b_1{y}^0+{\mathrm{b}}_{2}x \\ \to y_e-b_1{y}^0=b_0+{\mathrm{b}}_{2}x \\ \to y_e(1-b_1)=b_0+{\mathrm{b}}_{2}x \\ \to y_e=\frac{b_0+{\mathrm{b}}_{2}x}{1-b_1} \end{gathered}$$ (4)

Further note that that the covariate x is a broad representation for any covariate which may have lagged or concurrent effects, and may be static or vary over time. Additionally, this representation can be further generalized to include multiple covariates which will be added onto the numerator term.

$$\begin{gathered} y_e(1-b_1)=b_0+b_2{x}_1+b_3{x}_2+\cdots b_n{x}_{n-1} \\ \to y_e=\frac{b_0+b_{2}x+b_3{x}_2+b_n{x}_{n-1}}{1-b_1} \end{gathered}$$ (5)

Calculating Equilibria of Strain, RC, and TC

Using the general equation provided above, we can calculate the equilibria of strain (S_e), task conflict (TC_e), and relationship conflict (RC_e).

Strain Equilibrium

$$\begin{gathered} S_e=\frac{1.29+.07R{C}_{t-1}+.12workloa{d}_t}{1-.25} \\ =1.72+.09R{C}_{t-1}+.16workloa{d}_t \end{gathered}$$ (6)

TC Equilibrium

$$\begin{gathered} T{C}_e=\frac{1.82-.003R{C}_{t-1}workloa{d}_t+.26workloa{d}_t}{1-.08} \\ =1.98+.003R{C}_{t-1}workloa{d}_t+.18workloa{d}_t \end{gathered}$$ (7)

RC Equilibrium

$$\begin{gathered} R{C}_e=\frac{1.59+.08S_{t-1}+.16workloa{d}_t+.15T{C}_{t-1}+.18T{C}_{t-1}workloa{d}_t}{1-0} \\ =1.59+.08S_{t-1}+.16workloa{d}_t+.15T{C}_{t-1}+.18T{C}_{t-1}workloa{d}_t \end{gathered}$$ (8)

By substituting the strain and task conflict equations into the relationship conflict equation:

$$\begin{gathered} R{C}_e=1.59+.08(1.72+.09R{C}_{t-1}+.16workloa{d}_t)+ \\ .16workloa{d}_t+(.15+.18workloa{d}_t)T{C}_{t-1} \\ =1.59+.08(1.72+.09R{C}_{t-1}+.07workloa{d}_t)+ \\ .07workloa{d}_t+(.15+.18workloa{d}_t)(1.98+.003R{C}_{t-1}workloa{d}_t+.18workloa{d}_t) \end{gathered}$$ (9)

By distributing and adding like terms, we get the simplified equation:

$$\begin{gathered} R{C}_e=.005R{C}_{t-1}workloa{d_t}^2+.032workloa{d_t}^2+.001R{C}_{t-1}workloa{d}_t \\ +.46workloa{d}_t+.01R{C}_{t-1}+2.02 \end{gathered}$$ (10)

However, note that when relationship conflict reaches its equilibrium, all values of relationship conflict are the same. In other words, the equilibrium value of relationship conflict will be the same as its lagged value (and future values). Therefore, we can rewrite the previous equation as follows:

$$\begin{gathered} R{C}_e=.005R{C}_{e}workloa{d_t}^2+.032workloa{d_t}^2+.001R{C}_{e}workloa{d}_t \\ +.46workloa{d}_t+.01R{C}_e+2.02 \end{gathered}$$

$$\begin{gathered} \to R{C}_e-.01R{C}_e-.005R{C}_{e}workloa{d_t}^2-.001R{C}_{e}workloa{d}_t=2.02+.46workloa{d}_t \\ +.032workloa{d_t}^2 \\ \to R{C}_e(.99-.005workloa{d_t}^2-.008workloa{d}_t)=2.03+.46workloa{d}_t \\ \to R{C}_e=\frac{2.03+.46workloa{d}_t+.032workloa{d_t}^2}{(.99-.005workloa{d_t}^2-.008workloa{d}_t)} \end{gathered}$$ (11)

Alternative task conflict and relationship conflict equations

Finally, when relationship conflict reaches its equilibrium, we can substitute the previous equation into the initial task conflict and strain equations, and rewrite their equilibria entirely in terms of workload:

$$S_e=1.72+.09\left(\frac{2.03+.46workloa{d}_t+.032workloa{d_t}^2}{.99-.005workloa{d_t}^2-.008workloa{d}_t}\right)+.16workloa{d}_t$$ (12)

$$T{C}_e=1.98+.003\left(\frac{23+.46workloa{d}_t+.032workloa{d_t}^2}{.99-.005workloa{d_t}^2-.008workloa{d}_t}\right)+.18workloa{d}_t$$ (13)