The missing link? Implications of internal, external, and relational attribution combinations for leader–member exchange, relationship work, self-work, and conflict

2.1 Attribution theories

“Attributions are the causal explanations that individuals use to interpret the world around them and adapt to their environment,” especially when attempting to understand events that are important, novel, unexpected, and negative (Eberly et al., 2011, p. 733). Scholars have remarked that “there is no ‘attribution theory’ … [rather] the term attribution theories applies to a broad group of theories that are concerned with causal reasoning” (Martinko & Thomson, 1998, p. 272). Attribution theories, therefore, focus on the processes through which people attempt to explain the causes of behaviors or events (Harvey, Madison, Martinko, Crook, & Crook, 2014; Heider, 1958; Kelley, 1967, 1973; Martinko, Harvey, & Dasborough, 2011; Weiner, 1985). Numerous scholars have extended Heider's (1958) initial work on attributions. Below, we describe several attribution theories that are most relevant to our model.

Kelley (1967, 1973) introduced the covariation model to describe how people make attributions using informational cues across multiple observations. He suggested that individuals consider three information criteria—distinctiveness, consistency, and consensus—related to a behavior or event when making attributions. Distinctiveness information involves within-person behavior that “compares the behaviors of the individual in other situations” (Martinko & Thomson, 1998, p. 273). Consistency information also involves within-person information, but it incorporates a longitudinal perspective that considers whether the behavior is similar or different across time (Kelley, 1973; Martinko & Thomson, 1998). Consensus information assesses whether behavior is common across individuals in similar situations; that is, it is a between-person appraisal that compares the focal party's behavior with the behavior of others (Harvey et al., 2014; Kelley, 1973). When consensus is high, it generally leads to an internal locus of causality, whereas low consensus information leads to an external locus of causality. When combined, Kelley's three informational criteria form patterns that individuals use to attribute causes to specific behaviors or events. Ability attributions generally occur when there is information that is high on distinctiveness, high in consistency, and low on consensus. Attributions related to a person's effort are likely when information is low on consensus, low on consistency, and high on distinctiveness. High consensus information leads to an external locus of causality. Information about the target that is high in consensus, consistency, and distinctiveness will likely result in attributing the outcome to the nature if the task. An outcome that is high on consensus and low in both consistency and distinctiveness will likely be attributed to luck. Thus, the consensus information affects locus of causality, and both distinctiveness and consistency information allow more nuanced attributions that go beyond the internal and external dimensions.

Whereas Kelley's (1973) covariation model focuses on how information is used to make attributions, Weiner's (1985) extensions to attribution theory emphasized three causal dimensions (i.e., the causal structure) and, importantly, the emotional and behavioral outcomes of attributions (i.e., the achievement motivation model). Weiner contends that the causal dimensions of attributions are locus of causality, stability, and controllability. Locus of causality “describes the extent to which an event is attributed to causes internal or external to the observer” (Harvey, Martinko, & Borkowski, 2017, p. 781). Stability refers to the extent to which an observed event is enduring (i.e., permanent) or temporary (i.e., variable), and controllability describes the event cause as one that is either “volitional or optional control” or uncontrollable (Weiner, 1985, p. 551). Weiner goes on to establish how the causal structure of attributions influences emotions and behavior. Emotional responses can include anger or frustration, commonly caused by external attributions, and guilt, commonly caused by internal attributions (Weiner, 1985, 1986). Behavioral reactions may consist of helping behaviors (Weiner, 1985, 1986), task performance (Thomas & Mathieu, 1994), and even deviant behaviors (Harvey et al., 2017). The achievement motivation model “therefore relates the structure of thinking to the dynamics of feeling and action” (Weiner, 1985, p. 548).

An important distinction between Kelley's and Weiner's research is the emphasis on intrapersonal or social attributions. That is, Kelley's (1967, 1973) covariation model primarily focused on providing explanations for social attributions (i.e., attributions for others), whereas Weiner's (1985) achievement motivation model primarily emphasized how individuals make intrapersonal (i.e., self) attributions. Given the scope of these different models, Martinko and Thomson (1998) integrated Kelley's and Weiner's research by noting a direct relationship between the consistency and consensus informational cues introduced by Kelley (1967) and Weiner's (1985) attributional dimensions of locus of causality and stability. They proposed a direct relationship between Kelley's consistency information and Weiner's stability dimension by arguing that within-person, consistent information is indicative of a stable attribution. In addition, they suggested that Kelley's consensus informational cues and Weiner's locus of causality dimension are related by noting that behaviors or events “high in consensus are attributed to the characteristics of the situation (i.e., external) and outcomes that are low in consensus are attributed to the internal characteristics of the actor” (p. 276). The integration of these models was an important contribution for the application of attribution theories to a range of contexts, including the leader–follower relationship in organizations.

In research aimed at specifically exploring attributions in leader–follower dyads, Martinko and Gardner (1987) built on this prior research to discuss combinations of leader and follower internal and external attributions. In their model, the leader first makes an attribution based on the follower's task-directed behavior. Then, on the basis of the leader's attribution and response, they suggested various ensuing member attributions and responses including helplessness, performance deficits, and conflict. Another important contribution of this theory was the integration with the literature on attributional biases. They suggested that followers will be more likely to attribute their own failure to external causes whereas leaders will be more likely to blame follower characteristics. Therefore, they argued that divergent attributions for follower failure will likely result in conflict.

More recently, Eberly et al. (2011) considered dyadic relationships in organizations to extend the theoretical research on attribution theory through the introduction of a third locus of causality category, relational attributions. In developing the theory of relational attributions, the authors advanced propositions to explain how the distinctiveness, consistency, and consensus information cues specified by Kelley (1973) elicit relational attributions at the dyadic level. Note that because negative achievement-related events are more likely to elicit attributional processing than are positive events (Martinko, Harvey, & Douglas, 2007; Weiner, 1990), Eberly et al. (2011) confined their model to attributions for the former. They also emphasized that relational attributions are based on high distinctiveness, high consistency, and low consensus information from a negative achievement-related event. For example, in the context of a leader–follower dyad, a follower who is not selected by her or his leader for an important client team may consider that (a) she or he had routinely been selected for important client teams by other supervisors (high distinctiveness); (b) she or he had yet to be selected for an important client team by this supervisor (high consistency); and (c) the majority of her or his colleagues were selected to be a part of this important client team (low consensus). With this information, the follower is likely to make a relational attribution. If we again consider the context of a leader–follower dyad, another example would be where a leader struggles to motivate a follower to achieve his or her goals. The leader may consider that (a) he or she has never struggled to motivate other followers (high distinctiveness); (b) he or she has struggled for several performance review cycles to motivate this follower (high consistency); and (c) other leaders have followers that respond positively to the motivation techniques that he or she employed (low consensus). Here again, with this information, the leader is likely to make a relational attribution.

Eberly et al. (2011) also asserted that relational attributions are more likely to prompt individuals to proactively repair their relationship by engaging in relationship work. Namely, relational attributions based on high distinctiveness, high consistency, and low consensus informational cues focus dyadic partners on addressing the relational causes of negative events and prompt individuals to engage in relationship work such as remedial voice behaviors (e.g., discussing the issue) or interpersonal citizenship behaviors (e.g., providing assistance with tasks or socioemotional support). In subsequent empirical research, Eberly et al. (2017) found that individuals spontaneously make internal, external, and relational attributions, and they discussed how leader–member interdependence increases the likelihood of a relational attribution. Additional empirical work has supported relational attribution theorizing and has examined the implications of relational attributions in the context of abusive supervision. Burton, Taylor, and Barber (2014) found that relational attributions for the behavior of abusive supervisors are likely when the subordinate believes he or she has not been treated this way by previous supervisors (high distinctiveness), the subordinate believes the abuse is happening frequently (high consistency), and the subordinate believes that the supervisor is not treating anyone else in the same way (low consensus).

Despite these advances in attribution theories and relational attribution research, there is still an incomplete understanding of how attribution combinations affect LMX quality. Therefore, we next provide an overview of LMX theory.

2.2 LMX theory

LMX theory focuses on the dyadic relationships between leaders and followers whereby relationships develop with varying levels of affect, loyalty, contribution to the relationship, and professional respect (Dienesch & Liden, 1986; Graen & Cashman, 1975; Graen & Scandura, 1987; Graen & Uhl-Bien, 1995; Liden & Maslyn, 1998). Followers with high-quality LMX relationships receive more information and attention, have greater influence, and experience more beneficial work outcomes than do followers in low-quality LMX relationships whose exchanges are more transactional and linked to formal job requirements (Dulebohn, Bommer, Liden, Brouer, & Ferris, 2012). Leaders in high-quality LMX relationships, on the other hand, enjoy relationships of high trust in members where they can delegate important work to members and have confidence it will be done well and on time, rely on members to look out for their best interests, and enjoy the satisfaction of providing support and in many cases mentorship to the member. In contrast, leaders in low-quality LMX relationships may become frustrated by a perceived lack of commitment from the member and an overall lack of trust that the member will perform work assignments well and look out for the leader's interests (Graen & Uhl-Bien, 1995; Liden, Sparrowe, & Wayne, 1997). Thus, high-quality LMX relationships are characterized by interdependent interactions that are mutually beneficial and focused on the long term, and low-quality LMX relationships are exemplified by high self-interest and immediacy and equivalence in exchange transactions (Uhl-Bien & Maslyn, 2003).

LMX quality is most commonly assessed from solely the follower's perception of the relationship, yet when it is also measured from the leader's perspective, there is often disagreement between the parties. Indeed, meta-analytic evidence indicates that the correlation between leader and follower reports of LMX is .29 (Gerstner & Day, 1997), and research has demonstrated that followers may overestimate the quality of the relationship, failing to recognize that the leader does not judge it to be equally beneficial (Cogliser, Schriesheim, Scandura, & Gardner, 2009). Such overestimation is associated with lower levels of leader-rated member job performance than is the case in balanced high-quality LMX relationships where both parties view the relationship positively. However, conceptually, LMX is a dyadic-level theory, and its measurement should include both leader and follower perceptions of relationship quality as both the leader and the follower must devote effort into the relationship for a high-quality relationship to develop (Maslyn & Uhl-Bien, 2001).

Why might individuals exert this effort? According to Martinko, Moss, Douglas, and Borkowski (2007), leader and follower attributions have a profound effect on the quality of LMX; therefore, the nature of the causal attributions they make for an important, novel, unexpected, and/or negative event is likely to influence the amount of relationship work by both the leader and the follower. As Eberly et al. (2011) suggested, when leaders and/or followers make relational attributions, they direct attention toward relational influences on events rather than on purely internal or external causes. That is, when leaders and followers make relational instead of purely internal or external attributions about events, they will subsequently either expend effort toward, or withhold effort from, the relationship, thereby causing increases or decreases to relationship quality. However, given the research on LMX agreement cited above, we contend that it is likely that leader and follower internal, external, and relational attribution combinations will frequently diverge. Thus, we present a model of attribution combinations, LMX quality, relationship work, self-work, and conflict, which considers the implications of convergent and divergent leader–follower attributions.

Before we elaborate on the elements of our proposed model and the associated propositions, we first follow Bacharach's (1989) recommendations for theory building by specifying the model's underlying assumptions and boundary conditions.

2.3 Underlying assumptions and boundary conditions

First, we recognize that attribution processes can be triggered by either positive or negative events. Nonetheless, as noted above, we follow the conceptual foundation of Eberly et al. (2011) and focus exclusively on negative achievement-related events in our model because they are more likely to “threaten goal accomplishment and motivate people to find underlying causes so they can avoid similar events in the future” (p. 735). This is consistent with prior research that asserts that not every event triggers attribution processes—they are most common when the event is important, surprising, and/or negative (Eberly et al., 2011; Weiner, 1986). Moreover, it is certainly possible that events may not trigger relationship work but instead cause the leader and/or follower to withdraw from the relationship (Eberly et al., 2011), become hostile or resentful (Ferris et al., 2009), or engage in avoidant or antisocial behaviors (Stafford & Canary, 1991). These reactions are also outside the scope of our model but warrant additional attention in future research.

Next, we note that both relational attribution theory and LMX theory are dyadic in nature. Therefore, we limit the focus of our model to the hierarchical relationship between a leader and a follower. Given this dyadic emphasis, we are focused on dyadic-level theory, analysis, and measurement. For example, when we discuss LMX quality, we are specifically referring to leader–member dyads with high agreement on LMX quality (i.e., both the leader and the member agree that the relationship quality is high or low). Moreover, we acknowledge that there are likely to be differential amounts of effort expended by the leader and follower to enhance LMX quality following a negative and/or unexpected event, and there are likely to be differential effects on LMX quality based on each party's capabilities and availability for relationship work. Indeed, Eberly et al. (2017) found that time or availability was a significant contextual moderator of the relationship between relational attributions and relational improvement behaviors. However, these differences are beyond the scope of our model.

Further, we acknowledge that dyads are nested within collectives, and therefore, the locus of causality attributions may be made that consider relationships at a higher level of analysis beyond the dyad (e.g., groups and organizations) as potential causes. Here again, however, our focus is exclusively on the dyadic level, consistent with the dyadic focus of relational attribution (Eberly et al., 2011) and LMX (Liden et al., 1997) theories. Also, with regard to social exchange processes, we do not consider higher level collective constructs such as team member exchange (Seers, 1989; Seers, Petty, & Cashman, 1995).

Finally, there are numerous contextual factors that may affect the relationships presented in our model. These could include the leader's span of control; the extent of the leader's and follower's task interdependence; the type of leader and follower work; and the organization's structure (Rousseau & Fried, 2001). These are also important considerations for future research which are not discussed in this manuscript.

3.1 Attributional biases

In attribution theory, there is an assumption that people use a naive version of the scientific method to make causal inferences. However, such inferences are subject to bias due to incomplete evidence and unknown or misleading contextual information (Kelley, 1973; Shaver, 1983); this is known as attributional bias. Although there are several types of attributional biases, we focus on two that are particularly salient within dyadic relationships: the self-serving bias and actor-observer bias.

The self-serving bias refers to the tendency for persons to attribute their successes to internal causes (effort, ability) while attributing their failures to external causes (Sedikides, Campbell, Reeder, & Elliot, 1998). The actor-observer bias involves a tendency for people to attribute their failures to external causes while attributing others' failures to internal causes (Jones & Nisbett, 1972; Ross, 1977). The implication for leaders and followers confronted with negative achievement-related events is that these biases will make external (person or situational) attributions more likely, and relational attributions less likely, because relational attributions partially implicate the attributor. For example, both of these biases would increase the likelihood that a leader who does not meet an important deadline (i.e., the negative event) attributes the cause to the follower's performance (i.e., external-person), rather than the relationship. Similarly, these biases would both increase the likelihood that a follower who fails a quality inspection (i.e., the negative event) will attribute the cause to the leader (i.e., external-person) or the situation (i.e., external-situational) rather than the relationship.

4.1 Convergent relational attributions

When both the leader and follower attribute a negative event (e.g., poor follower performance) to the relationship, the attributions converge (Tables 1 and 2, Cell 1: convergent relational attributions). Under these circumstances, relationship work by both the leader and follower is highly likely to occur, because both parties view some aspect of their relationship as causing the adverse event. When participants in a relationship make relational attributions, they both pay attention to the areas of the relationship that need improvement (Brewer & Gardner, 1996; Markus & Kitayama, 1991). Moreover, the leader and follower are both likely to experience feelings of guilt. Weiner (1986) considered guilt to be related to the controllability causal dimension, and he suggested that it commonly arises due to feelings of personal accountability (e.g., lack of effort). Therefore, in cases of convergent relational attributions for negative events, each party's guilt is likely to be directed toward effort to improve the relationship, and it will likely promote behaviors that address and provide retribution for the identified cause (Weiner, 1985). In subsequent research, Brees and Martinko (2015) elaborated on the connection between personal accountability and organizational citizenship behaviors. In other research, Harvey et al. (2017) described this as “adaptive guilt” because it prompts individuals to constructively address the adverse event to prevent it from happening again in the future (p. 783). In addition, the focus by both parties on improving the relationship is likely to increase LMX quality.

4.2 Divergent internal attributions

When either the leader makes a relational attribution and the follower makes an internal attribution (Table 1, Cell 2: divergent follower internal attributions) or vice versa (Table 1, Cell 3: divergent leader internal attributions), attributional divergence occurs. Here, it is important to note that internal attributions occur when the attributor identifies a personal quality as the proximal cause of the negative achievement-related event. That is, the leader or follower makes a self-attribution. For example, if a follower makes a self-attribution to effort, she or he is likely to feel guilty and increase effort in the future as a way to make amends (Allred, 1995; Weiner, 2004).

Both dyadic partners need to feel that they have the ability to improve, they are connecting with each other, and they have some control for them to be motivated to engage in relationship work (Deci & Ryan, 1985, 2000). However, because only one party is making a relational attribution, the relationship work will not be as successful in enhancing LMX quality as if both parties made the same attribution, resulting in stable or possible increases to LMX quality. Nevertheless, these divergent internal attributions are likely to result in leader or follower relationship work and some leader or follower self-work.

4.3 Divergent self attributions

If both the leader and the follower make internal attributions, they both attribute the cause to something about themselves (e.g., personal lack of effort or ability). As previously noted, the parties' attributions reflect divergence in this case because each party makes a self attribution and takes responsibility for the negative event (Table 1, Cell 4: divergent self attributions). Under these circumstances, no relationship work occurs, but each party engages in self-work because of guilt for their role in the adverse event. Indeed, Harvey et al. (2017) found that guilt was related to constructive workplace behaviors. Thus, the leader and follower may work to improve themselves by refining their skills to enhance future performance. Given that the leader and follower are engaged in self-work, LMX quality is unlikely to change unless the self-work facilitates better interpersonal exchanges.

4.4 Divergent external-person attributions

When the follower makes a relational attribution and the leader attributes the unexpected negative event to the follower (Table 2, Cell 4: divergent leader external-person attributions), the follower is likely to engage in relationship work, whereas the leader may respond to his or her attribution by initiating conflict. Following the logic above, the follower is likely to experience guilt and engage in relationship work behaviors to address the task or personal elements of the relationship that she or he perceives as being related to the negative event. Conversely, the leader is attributing the negative event to the follower and hence is likely to experience frustration and/or anger. Weiner (1986) considered anger (like guilt) to be related to the controllability causal dimension. He viewed anger as an other-generated emotion that occurs because of attributions of lack of effort from the other party. Hareli and Rafaeli (2008) described how feelings of anger are commonly mimicked or transferred, particularly when the angry person is in a position of higher power (i.e., the leader). Therefore, as a result of the leader's anger and the divergent attributions, leader-initiated conflict is likely.

In addition, LMX quality is likely to decrease (Martinko, Moss, et al., 2007). Previous research suggested that divergent attributions may lead to deteriorating relationships (Martinko, Moss, et al., 2007). For example, if the follower feels unfairly blamed for a problem that is perceived as arising from issues in the relationship, he or she is likely to perceive a decrease in relationship quality. Without the benefit of the relational attribution construct, Martinko and Gardner (1987) nevertheless considered why these types of divergent attributions can be so counterproductive. They noted that “if the leader attributes failure to a lack of effort and responds by punishing the subordinate, the member probably will view the leader's response as inappropriate and conflict could occur” (p. 242). Note that in the scenario they described, we view the punishment administered by the leader as a form of leader-initiated conflict, because the leader takes action that the follower is likely to view as unjust, given his or her relational attribution. The follower, in turn, is likely to experience resentment, thereby escalating the conflict and causing LMX quality to suffer (Martinko, Sikora, & Harvey, 2012). This is consistent with affective events theory, which suggests that if a leader expresses anger to a follower, their relationship will be harmed (Cropanzano, Dasborough, & Weiss, 2017).

In a similar manner, the leader could attribute the cause of the negative event to the relationship, whereas the follower makes an external-person attribution by blaming the leader for the negative event (Table 2, Cell 2: divergent follower external-person attributions). This scenario is likely to lead to leader relationship work and follower-initiated conflict, even if the latter is passive aggressive due to the power differential between the leader and follower. However, it may become particularly problematic if the leader becomes angry for being held responsible for the event (when he or she attributes the negative event to the relationship) and because of the divergence in attributions. Likewise, the follower may feel frustrated and angry that their counterpart thought something was wrong with the relationship, which is likely to result in an escalation of conflict.

If both the leader and follower make an external-person attribution for poor performance (Table 2 Cell 5: divergent external-person attributions), they attribute the cause to their dyadic partner rather than to themselves. That is, they blame the other party for the negative outcome. In this scenario, the leader and the follower are likely to be angry with each other, and there is likely to be mutually-initiated conflict. Here, stronger workplace outcomes are possible, such as aggression or deliberate mistreatment, particularly if the follower views the leader as an abusive supervisor (Harvey, Summers, & Martinko, 2010). Moreover, this has the potential to be most detrimental to LMX quality because each party holds the other accountable for the negative and unexpected event.

4.5 Divergent external-situational attributions

When one party makes a relational attribution while the other makes an external attribution to the situation (i.e., external-situational attribution; Table 2, Cell 7: divergent leader external-situational attributions, or Table 2, Cell 3: divergent follower external-situational attributions), the party making the relational attribution for the adverse event is likely to feel guilty and engage in task- or person-focused relationship work. Given that the other party attributed the negative event to the situation and not the other person, he or she is more likely to reciprocate the relationship work, signaling that he or she values the other party's effort and the relationship. Therefore, the increased effort devoted to the relationship is likely to cause LMX quality to remain stable or increase.

4.6 Divergent external-person/situational attributions

When one party makes an external-situational attribution while the other makes an external-person attribution to the dyadic partner (i.e., divergent external-person/situational attributions; Table 2, Cell 6 and 8), conflict is likely and LMX quality is expected to decrease. Given that either the leader or the follower feels that the other party is responsible for the negative and unexpected event, they are likely to be angry and hold the partner accountable. When the leader (follower) makes an external-person attribution while the follower (leader) makes an external-situational attribution, leader (follower)-initiated conflict is likely to ensure. Additionally, given that there is divergence in attributions and one party is angry, LMX quality is likely to decrease.

4.7 Convergent external-situational attributions

When the leader and follower both make attributions to the external situation (Table 2, Cell 9: convergent external-situational attributions), no relationship work, self-work, or conflict is expected, and LMX quality remains stable. This proposition follows from the fact that neither party views the relationship as the cause of the negative and unexpected event, so it is unlikely that either the leader or the follower will see a need for relationship work. Moreover, both parties attribute the cause to an external-situational factor (e.g., luck and task difficulty), so it is unlikely to elicit feelings of guilt or anger, which would prompt self-work or conflict.

7.1 Relational attribution theory implications

A central point of this paper was further integration of attribution theories with recent research on relational attributions and LMX theories. In integrating these literatures, we propose that as the intensity of self-serving or actor-observer attributional bias(es) experienced by leaders and/or followers increases, they become less likely to make relational attributions. Although our discussion of attributional biases was bounded by our discussion of the self-serving bias and the actor-observer bias, we acknowledge that there are numerous other attribution biases that could impact the propensity for leaders and/or followers to make relational attributions, and those warrant further theoretical integration in future research.

In addition, our discussion of emotions related to attributions focuses on the impact of guilt and anger. This led to two interesting insights. First, combinations of relational and internal attributions for the negative event are likely to elicit guilt, which is likely to prompt leader and/or follower self-work or relationship work. Conversely, combinations of relational and external-person/situational attributions are likely to elicit guilt or anger, which likely results in leader and/or follower relationship work or conflict. However, Weiner's (1986) research highlighted a range of emotions from positive (e.g., pleased, satisfied, proud, and confident) to negative (e.g., sad, pity, bitter, and disgust) that have yet to be integrated into relational attribution research. Therefore, future relational attribution research should explicitly examine how other emotions are connected to behavior.

Finally, we acknowledge that relational attribution research is still in its infancy given that the publication of the seminal theoretical piece by Eberly and colleagues occurred just 8 years ago (2011). Accordingly, there are relatively few empirical studies that have examined relational attributions. This is reflective of what Reichers and Schneider (1990) describe as the first stage of the model for the evolution of constructs. In this stage, they argue that it is important to introduce and theoretically elaborate a topic. Therefore, there are additional needs for theorizing on relational attributions, dyadic-level attribution combinations, and their potential impact on other social exchange quality variables (e.g., affective commitment and trust) and outcomes (e.g., job satisfaction and organizational citizenship behaviors).

7.2 Social comparison implications

When considering LMX and relational attributions, we must consider that the leader–member relationship is situated in an environment where the leader has relationships with all of her or his employees that vary in quality. Because employees all have distinct relationships with their leader, social comparisons regarding their relationship with the leader are likely to occur. Wood (1986) defined social comparison as “the process of thinking about information about one or more other people in relation to the self” (pp. 520–521). Therefore, employees are likely to compare their relationship with other employees and will judge the quality of LMX on the basis of this comparison. As such, the social comparison context is a vital factor to consider. Social comparison processes impact relational attributions and the motivation of leaders and followers to engage in relationship work. Social comparison theory states that people evaluate themselves by comparing their opinions and behaviors with those of other people, particularly those who are similar, as would be the case among coworkers (Festinger, 1954).

In a leader–follower relationship, followers compare their relationship with the leader to their coworkers' relationships with the leader. This between-person appraisal is a way in which followers seek consensus information (Kelley, 1973). Upon doing so, if an employee does not perceive that the leader is being consistent across relationships, the quality of the social exchange will be hindered (Colquitt et al., 2013; Masterson, Lewis, Goldman, & Taylor, 2000; Rupp, Shao, Jones, & Liao, 2014). This is particularly salient for leaders because employees may view the relationship as unfair when compared with another employee's relationship with the leader. If a follower does not perceive fairness in the relationship due to the leader's behavior, the follower is likely to blame the leader for relationship problems, rather than making a relational attribution, and is less likely to be motivated to engage in relationship work. In other words, depending on the participant's social comparison process, relationship work may be less likely to occur.

7.3 Time implications

As we have discussed, there are characteristics of the situation or context beyond interdependence that impact relational attributions. Specifically, for the leader–member relationship, future research should consider the history of the relationship between the leader and follower when analyzing the likelihood of relational attributions and relationship work. Of particular interest would be an examination of the type, frequency, and strength of negative and unexpected events in the relationship's history. For example, distinctiveness (i.e., comparing one's behavior across situations) and consistency information (i.e., comparing whether one's behavior changes across time) in response to negative or unexpected events could be used to explore whether relational attributions are more or less likely in long-tenured dyadic relationships. Moreover, consensus information (i.e., considering whether the other party's behavior is consistent with others) could be examined to determine whether in-group or out-group followers are less likely to make relational attributions.

According to Burger (1991), attributions change over time. Specifically, he or she found that the fundamental attribution error (a common attributional bias) diminished significantly over time to the point that research participants stopped utilizing personal information in making attributions about others when there was a time lag in making the attribution (Burger, 1991; Burger & Pavelich, 1994). When applying these findings to relational attributions and LMX, we can see how vital the change over time is to consider. In the beginning stages of the relationship, there may be a tendency for divergent external attributions due to attributional biases. However, as the relationship improves over time, these biases are likely to diminish, therefore changing how the participants view the negative achievement-related event. For example, if conflict occurs within the leader–follower relationship soon after the participants start working together, they may blame each other, therefore causing more conflict. However, based on Burger's (1991) research, after a period of time, the participants are likely to look back on that initial conflict and make less biased attributions. Furthermore, being able to reflect on these past conflicts may further enhance LMX quality, because this will enable them to learn from the experience. Therefore, as noted by previous researchers (Chong et al., 2015; Eberly et al., 2017), time continues to be an important aspect to consider in future research when examining the relational attribution–LMX relationship.

7.4 Impression management implications

Relationship work occurs when participants in a relationship want to improve or repair the relationship. In some cases, this may involve impression management. Previous research has suggested that impression management in the form of shaping others' attributions can influence others' behaviors (Bolino, Long, & Turnley, 2016; Gardner & Martinko, 1988; Jones & Pittman, 1982). In terms of relational attributions and LMX, this may mean that one party apologizes, uses ingratiation, or attempts to justify the negative event. Specifically, in the beginning of a leader–member relationship, this may be more likely than later, when LMX is of higher quality. The impression management literature may give some insights as to how impression management may change over time regarding relationship work; such changes should be considered in future research. For example, are subordinates more likely to use self-promotion as an impression management tactic to build a reputation for competence and make it less likely that they will be held responsible for negative events during early versus late stages in the relationship? We suspect motives for self-promotion decline as the relationship matures and followers (and leaders) grow more comfortable revealing their authentic selves (Gardner, Avolio, Luthans, May, & Walumbwa, 2005). Further, followers may be more comfortable exhibiting task- and person-focused voice behavior during the mature stages in the LMX relationship, particularly if the relationship is of high quality. Hence, future research that explores the extent to which the dyadic partners engage in strategic impression management as opposed to more authentic self-presentations is warranted.

7.5 Attribution training implications

For relationship work to occur, leaders and/or followers may need attributional training. Attribution theorists (Green & Mitchell, 1979; Martinko & Gardner, 1982, 1987) assert that attributional training is an effective tool for eliciting more constructive attributions from underperforming followers, as well as more objective attributions from leaders. In the social psychology literature, the situated attribution training technique has been used to help people overcome stereotyping by steering them away from dispositional attributions to more situational attributions for stereotype consistent behaviors (Stewart, Latu, Kawakami, & Myers, 2010). Support has been found for the efficacy of such training in the workplace. Researchers have used cognitive–behavioral training to elicit improvements in attributional style, job satisfaction, self-esteem, well-being, and productivity (Proudfoot, Corr, Guest, & Dunn, 2009). This research has also demonstrated that attribution training reduces employee turnover (Proudfoot et al., 2009). Therefore, attributional training may have utility in directing followers and leaders away from pessimistic attributions for poor performance toward more optimistic and realistic attributions pertaining to shortcomings in the relationship, which can be addressed through relationship work. This can occur by directing leaders and followers to consider whether there was high distinctiveness (e.g., similar or different personal behavior/outcomes in other situations), high consistency (e.g., similar or different personal behavior/outcomes across time), and low consensus (e.g., similar or different behavior from the dyadic partner) information following the poor performance event. Therefore, future research should consider attributional training and its effect on subsequent relationship work.