Leader Member Exchange in Leaders’ Support for Voice: Good Relationships Matter in Situations of Power Threat

While previous research underscores the role of leaders in stimulating employee voice behaviour, comparatively little is known about what affects leaders’ support for such constructive but potentially threatening employee behaviours. We introduce leader member exchange quality (LMX) as a central predictor of leaders’ support for employees’ ideas for constructive change. Apart from a general benefit of high LMX for leaders’ idea support, we propose that high LMX is particularly critical to leaders’ idea support if the idea voiced by an employee constitutes a power threat to the leader. We investigate leaders’ attribution of prosocial and egoistic employee intentions as mediators of these effects. Hypotheses were tested in a quasi-experimental vignette study (N = 160), in which leaders evaluated a simulated employee idea, and a field study (N = 133), in which leaders evaluated an idea that had been voiced to them at work. Results show an indirect effect of LMX on leaders’ idea support via attributed prosocial intentions but not via attributed egoistic intentions, and a buffering effect of high LMX on the negative effect of power threat on leaders’ idea support. Results differed across studies with regard to the main effect of LMX on idea support.


INTRODUCTION
Employees' proactive contribution of ideas for constructive change is considered an invaluable asset for organizations as it contributes to individual performance (e.g., Grant & Ashford, 2008;Howell, Harrison, Burris, & Detert, 2015;Huang, Xu, Huang, & Liu, 2018;Whiting, Maynes, Podsakoff, & Podsakoff, 2012) and helps to improve organizational functioning and performance (e.g., Van Dyne & LePine, 1998;Maynes & Podsakoff, 2014;Morrison, 2011). This proactive type of work behaviour is referred to as voice behaviour (Van Dyne & LePine, 1998). Of particular relevance to organizations is "constructive" voice, which denotes "the voluntary expression of ideas, information, or opinions focused on effecting organizationally functional change" (Maynes & Podsakoff, 2014, p. 91). The majority of research on voice behaviour has focused on understanding when and why employees show voice (for reviews and meta-analytic evidence, see Chamberlin, Newton, & Lepine, 2017;Morrison, 2011;Tornau & Frese, 2013). This research implicitly builds on the assumption that voice behaviour will always be appreciated and supported in organizations. However, organizational practice shows that this assumption is wrong; by far not all ideas that are brought up by employees are actually welcomed and implemented (e.g., Baer, 2012;Burris, 2012). Thus, research on voice needs to expand its focus and take a wider perspective on voice-with the purpose to reveal under which conditions employees' voiced ideas are more likely to be implemented. The present research seeks to shed light on the chances of a voiced idea to be implemented. We will show that it makes a difference who voices an idea, what is voiced, and to whom it is voiced to predict which ideas will be endorsed by leaders and eventually be put into practice.
Central to our research is the notion that supervisors play a key role in voice implementation. The reason for this is that in organizations, dealing with employees' voiced ideas oftentimes is left to line management (Leach, Stride, & Wood, 2006). Thus, leaders can be a bottle neck to the implementation of employee voice because typically it is for them to support an idea and guide it through to implementation. So far, our knowledge on the factors that affect leaders' reactions to employee ideas and the subsequent leaders' willingness to support their implementation is limited. This research suggests that the extent to which a leader feels threatened by the voiced idea shapes their reluctance or willingness to support an idea. More specifically, pioneering work by Burris (2012) indicates that leaders show less support for a voiced idea when it is challenging the status quo rather than supporting it, because leaders' are more likely to perceive challenging voice as threatening their position and image. Also reflecting a response to threat, leaders low on managerial self-efficacy were found to be less likely to support the implementation of voiced ideas out of ego defensiveness (Fast, Burris, & Bartel, 2014). Research by Van Yperen (2015a, 2015b) shows that performance goal-oriented leaders were less willing to support employees' improvement-oriented input than mastery goal-oriented leaders. Again, leaders' perceived image threat constituted one mediating pathway of this effect (Sijbom et al., 2015b). Similarly, Urbach and Fay (2018) demonstrate that threatening leaders' power motivation reduces their willingness to support the implementation of employees' promotive ideas.
We want to extend this emerging stream of research. Because this research highlights the role of threat perceived by the leader in their idea support (Burris, 2012;Fast et al., 2014;Sijbom et al., 2015b;Urbach & Fay, 2018), we seek to identify variables that are capable of reducing leaders' perceived threat of employees' ideas. We propose that the relationship quality between leader and employee may function as a protective context factor. Specifically, we argue that high relationship quality sensu leader-member exchange (LMX) makes leaders' support for an idea or concern voiced by an employee more likely, and that high LMX is particularly important to obtain leaders' support when the consequences of implementing an idea challenges the status quo (Carnevale, Huang, Crede, Harms, & Uhl-Bien, 2017) and poses a potential threat to the leader. We suggest that when LMX is high, leaders' perceived risk associated with supporting an idea will be lower because leaders have higher levels of trust in the loyalty of the employee and the benevolence of their intentions behind the idea (e.g., Brower, Schoorman, & Tan, 2000;Graen & Uhl-Bien, 1995). This level of trust is particularly important in situations that are critical to the leaders' self. In this paper, we investigate the self-relevant situation that an idea threatens leaders' power motivation (Urbach & Fay, 2018). We propose that high LMX can buffer this power threat effect as it fosters "benefit-of-the-doubt" attributions of employees' behaviour (Bowler, Halbesleben, & Paul, 2010), and thereby reduces the likelihood that leaders refuse to support an idea. As an underlying process of these effects, we investigate the level of self-serving (egoistic) and otherserving (prosocial) intentions leaders attribute to an employee's idea proposal. We test these propositions in one quasi-experimental study and one field study.
We contribute to the literature on voice behaviour by investigating why and under which conditions employees' proactive idea proposals find their leaders' support. Previous research on the role of leadership for employee voice and related constructs (e.g., issue selling or organizational dissent) almost exclusively investigated how leaders can stimulate employee voice (for reviews, see Carnevale et al., 2017;Chamberlin et al., 2017;Chiaburu, Smith, Wang, & Zimmerman, 2014). The present research takes a different perspective and adds to our knowledge on leaders' role in the implementation of ideas for constructive change. Organizations can only exploit the full potential of employees' constructive voice if their voice is actually heard-first and foremost, by the leaders who decide whether an idea for improvement will actually be implemented, or if a voiced issue is considered serious enough to be addressed right away. Extending previous research on leaders' support for voiced ideas, we highlight that LMX is an important interpersonal context factor as it increases the likelihood that employees' ideas are heard and actually put into practice. Most importantly, we identify LMX as a protective factor to overcome the effects of threatening leaders' self or their position through voice, which has been identified as a reason why employees' voice behaviour may remain unheard in organizations (e.g., Burris, 2012;Urbach & Fay, 2018). We show that one mechanism behind these effects is that LMX affects leaders' attributions of employees' intentions behind their voice behaviour. These insights might inform organizational practice and future research on leaders' reactions to other forms of proactive work behaviour.

Leader-Member exchange and Leaders' support for employee Voice
Leaders and their subordinate employees develop unique dyadic relationships throughout their common work history. These dyadic relationships differ with regard to the exchange quality between leader and employee (called leader-member exchange, LMX), that is, their level of mutual trust, loyalty, and perceived ability (Graen & Uhl-Bien, 1995). Extant research suggests that employees are more likely to voice ideas to their supervisor when they perceive their relationship quality to be high rather than low (e.g., Carnevale et al., 2017;Chamberlin et al., 2017). Going beyond this research we propose that leaders are also more open to their employees' input and more likely to support employees' voiced ideas when LMX is high. From the perspective of a leader, lending support to employees' ideas for constructive change involves a certain level of risk that the implementation of an idea might not become a success (Lonergan, Scott, & Mumford, 2004). In addition, leaders may perceive upward voice as personal criticism, or as an act of undermining their judgement and status (e.g., Burris, 2012;Grant, Gino, & Hofmann, 2011;Kanter, 1988). The degree to which leaders perceive such risk associated with an employee's idea-and thus their willingness to support that idea-should depend on the level of LMX they have developed with the voicing employee. When LMX is high, leaders perceive employees as better performers (Dansereau, Graen, & Haga, 1975), and they have higher levels of trust in their abilities, their motivation, and the benevolence of their intentions (Brower et al., 2000) than when LMX is low. Thus, leaders should perceive lower levels of risk associated with an employee's idea if it was voiced by a high LMX employee. As the main reason for this we propose that leaders will have higher levels of trust that the employee's idea aims at benefitting the work group or organization. Previous research lends support to this proposition. Leaders provide employees with more autonomy, resources, and support when relationship quality is high rather than low (Liden & Graen, 1980 in making decisions (Scandura, Graen, & Novak, 1986). Moreover, leaders should be more willing to consider high LMX employees' input as they generally value their professional opinion (Liden & Maslyn, 1998). Research on innovative work behaviour-a different form of discretionary and changeoriented employee behaviour--suggests that the implementation of employees' innovative ideas is more likely when LMX is high (e.g., Clegg, Unsworth, Epitropaki, & Parker, 2002). Hence, we propose: Hypothesis 1: Leaders' intention to support an idea will be higher the higher their LMX with the employee voicing the idea is.

Leader-attributed Intentions and Leaders' support for employee Voice
Relationship quality affects social perception and attribution processes (Steiner, 1997). In essence, past research suggests that individuals attribute a target's behaviour to more positive, prosocial intentions when their relationship with the target is strong (Bowler et al., 2010). Research on proactive work behaviours has focused on two dimensions of intentions that observers may use when evaluating proactive behaviour (e.g., Grant & Ashford, 2008), that is, self-serving intentions (referred to as "egoistic" in the following), and other-serving intentions (referred to as "prosocial"). With regard to leaders' attributions of subordinate organizational citizenship behaviours (OCB, a different form of extra-role behaviour, e.g., helping others), Bowler and colleagues (2010, p. 312) argue that LMX fosters attribution in terms of "benefit of the doubt": When LMX is high, leaders expect the employee to be loyal, trustworthy, and to show positive behaviours because they have done so in the past. Given the assumption of stability of behaviour, ambiguous events are attributed in accordance with past behaviour, and thus in favour of the employee. Based on the positive set of mutual experiences and expectations, we propose that in dyads with high LMX, leaders attribute employees' proactive idea proposals to rather prosocial intentions, for example, that they intend to solve a problem within the group or organization, or that they speak up to help others. At the same time, leaders should attribute lower levels of egoistic intentions to high LMX employees' behaviour, for example, that they would only try to safeguard their own interests.
The level of attributed prosocial intentions and of egoistic intentions in turn affects the level of support given by the supervisor. We propose that leaders' attribution of prosocial intentions will positively relate to their idea support, while the attribution of egoistic intentions will negatively relate to idea support. In line with this, the literature on the general appreciation of proactive work behaviours suggests that these behaviours are more likely to be evaluated positively when they are seen as beneficial for others or for the organization; in case proactive behaviour is perceived as primarily serving the proactive individuals, it is evaluated less positively (for a review, see Grant & Ashford, 2008;Grant, Parker, & Collins, 2009). Based on these findings, we assume that a specific idea voiced by an employee should be more likely to find the leader's support if he/she attributes higher levels of prosocial intentions and lower levels of egoistic intentions to the employee's proposal. As outlined above, LMX might affect these attributions in favour of the employee. Thus, we propose: Hypothesis 2: The positive relationship of LMX and leaders' idea support is mediated by the level of prosocial and egoistic intentions attributed to the employees' idea.

Leader-Member exchange as a shield from power Threat
Above and beyond this general advantage for ideas that have been voiced by high LMX employees, we suggest that high LMX is particularly important to obtain leaders' idea support when the consequences of implementing an idea constitute a potential threat to the leader. A highly relevant threat for leaders represent ideas that imply a reduction of their influence or control over resources. In particular, we propose that LMX can reduce the negative effect of power threat on leaders' idea support.
Perceiving employees' upward voice as a threat to their self or their position (Burris, 2012;Fast et al., 2014;Kanter, 1988) is seen as a major reason why some leaders may deny their support for employees' ideas for constructive change. Extant research shows that threatening leaders' position-based power is likely to result in self-interested behaviour to secure their powerful position; this is particularly likely when leaders personally value being in power (Maner, Gailliot, Butz, & Peruche, 2007;Maner & Mead, 2010;Williams, 2014), that is, when their power motive is high. To protect their own power, leaders even engage in counterproductive behaviours, such as withholding important information from their team, or suppressing potential competition from subordinates (Maner & Mead, 2010). When the anticipated consequences of an idea imply some level of power threat, and at the same time, this idea is evaluated by a leader with a high power motive, the leader should perceive the idea as involving personal risk (Tett & Guterman, 2000). Under these conditions, the leader would most likely turn the idea down in order to protect their own need for power. Recent research by Urbach and Fay (2018) lends support to this notion. Past research suggests that leaders are more willing to delegate and to take risks when LMX is high (Bauer & Green, 1996;Brower et al., 2000;Tierney, 1999). In high-quality relationships, subordinates are perceived and expected to be loyal, and to share common goals (Graen & Uhl-Bien, 1995). Trust in the employee who voiced the idea may reduce the perception of personal risk involved in supporting a specific idea, or increase a leader's willingness to take that risk (Das & Teng, 2004). Yang, Long, and Chou (2010) found that the more risk leaders perceived related to delegation (in Yang et al.'s study, concerning task performance or the organization's benefit), the less they actually delegated to subordinates; this negative relationship was stronger the lower the quality of the dyad's LMX. Thus, high LMX may reduce the negative impact of perceived risk on leaders' delegation of authority to subordinates. We assume that the same mechanism applies to leaders' evaluation of employees' voiced ideas: If a potentially power-threatening idea was voiced by a high LMX employee, a leader high in power motivation mightdespite the power threat-conclude that the employee wants to contribute to shared goals by his/her idea rather than to threaten the leader. In contrast, if a potentially power-threatening idea was voiced by a low LMX employee, leaders high in power motivation may pay even more attention to the negative consequences of that idea, and are less willing to take the risk of losing their power.
In situations as ambiguous as this, the "benefit-of-the-doubt" attributions associated with high levels of LMX (Bowler et al., 2010) might become most important. If leaders perceive an idea to threaten their personal need for power, leaders are likely to attribute lower levels of prosocial intentions to the employee's idea because the leader would suffer negative consequences. As the source of power threat (i.e. the employee) is likely to be seen as a competitor striving for power him/herself (Maner & Mead, 2010), leaders might at the same time attribute rather egoistic intentions to the employee voicing a power-threatening idea. However, if LMX is high, leaders may be less aware of, or concerned about, the potential power threat and become more aware of the advantages of an idea for other beneficiaries. Consequently, LMX should buffer the power threat effects on idea support described above because it modulates leaders' attributions of employees' behaviour. When LMX is high, threatening leaders' personal power motive should result in less reduction of attributed prosocial and less elevation of attributed egoistic intentions; consequently, leaders' idea support should be higher than when LMX quality was low. Altogether, we assume: Leader Power Motive interaction) on their idea support is moderated by LMX, such that the effect will be weaker with increasing levels of LMX.
Hypothesis 4: The three-way interaction effect proposed in Hypothesis 3 is mediated by the level of prosocial and egoistic intentions attributed to the employee.
All hypotheses are summarized in the conceptual model presented in Figure 1.

Method
Participants. Data for this study were collected among working individuals holding leadership responsibility (i.e. having two or more direct reports). Participants were contacted through a psychology program at a German distance-learning university at which they were enrolled.
In total, 171 participants completed the survey. We excluded the responses of 11 participants who were identified as careless responders based on so-called long string responses (Johnson, 2005). These participants repeatedly selected the same response option to all items across the entire survey or on the same survey page, respectively, despite variation in item content and coding. The remaining 160 participants constitute the final sample of this study. Participants worked in various industries, such as health and social services (30.6%), retail and touristic services (17.5%), IT and media (13.1%), fIgURe 1. Conceptual model.

Idea consequences
Idea support research and education (10.6%), military, police and security services (9.4%), and others (18.6%). Their mean age was 37.0 years (SD = 9.9); 62 per cent were female. On average, participants had 8.5 years (SD = 7.5) of working experience, and held supervisory positions for 5.3 (SD = 5.5) years. Almost all participants held a vocational qualification (43.1%) or a university degree (50.0%). Half of the sample worked in organizations with more than 100 employees (50.6%); 68.1 per cent worked full time, 27.5 per cent part time (4.4% did not specify).
Design and Procedure. We conducted a quasi-experimental vignette study (e.g., Aguinis & Bradley, 2014) using written scenarios. In the scenarios, we described a male employee who voiced an idea to his leader, that is, to the study participant. Participants (who held a leadership position in real life) were asked to put themselves in the position of the leader described in the scenario, and to evaluate the idea proposed by the fictitious employee. The scenarios contained two experimental manipulations. First, we manipulated the level of relationship quality (sensu LMX) the participant had with the idea-presenting employee to be of low, medium, or high magnitude (Factor 1, see details below). Second, we manipulated the power-related consequences of the idea presented: In one condition, the consequences of the idea would involve power gain for the leader and thus satisfaction of his/her power motive; in the other condition, the consequences of the idea would involve a threat to leaders' power motive (Factor 2). This yields a 3 (LMX: low, medium, high) × 2 (idea consequences: power gain vs. power threat) factorial design with six experimental conditions; participants' power motive was treated as a continuous moderator variable.
The study was administered online. At first, participants completed a scale to assess their explicit trait power motive; to disguise the focus of the study on power motivation this scale was presented together with items measuring achievement and affiliation motivation. After that, participants were asked to put themselves in the situation of being a leader at a fictitious company. As we expected participants to work in various industries, we intended to establish a shared "organizational context" or mindset by embedding the experiment in a fictitious organizational setting. Due to the fact that all participants were in a leadership position in their real jobs, we expected them to immerse easily into this scenario. Following this introduction, participants were randomly assigned to one of the six experimental conditions (M = 27 participants per condition, Min = 21, Max = 35). Participants first read the description of a fictitious employee (including information on Factor 1, LMX), followed by an idea for change brought up by this employee (including information on Factor 2, idea consequences). Then, participants completed scales to assess the level of egoistic and prosocial intentions they attributed to the presenting employee, and their intention to support the idea. Last, participants completed manipulation check and demographic items. Participation was rewarded with course credits.

Manipulation of LMX.
We designed the descriptions of low, medium, and high relationship quality based on the definitions and measures of Vertical Dyad Linkage (Dansereau et al., 1975) and LMX (e.g., Graen & Uhl-Bien, 1995). Note that we left out those aspects that directly related to our hypotheses, that is, in high-quality relationships, employees make more suggestions to leaders, and leaders delegate more to employees. In the low LMX condition the idea-presenting employee was described as a subordinate who is difficult to handle in everyday work, with whom communication is scarce, and who would certainly not be of great assistance in "if the worst comes to the worst" situations. In the medium LMX condition the employee was presented as a "normal" subordinate, with whom work-related information is exchanged on a regular basis, and who would probably agree to help out in stressful situations. In the high LMX condition the employee was described as a faithful subordinate who is a good partner for discussing important issues, with whom cooperation is based on mutual trust and respect, and who would always help out when necessary. Across all conditions, participants were asked to assume they had been working with the employee for one and a half years (see Online Supplement for full scenarios).
We deliberately included the medium LMX category for two reasons: First, the majority of employees should have a more or less medium-quality relationship with their leader; thus, including the medium LMX condition instead of only investigating extreme forms of relationship quality is more representative and informative. Second, we seek to understand whether medium LMX follows the processes of high LMX or of low LMX (Liden & Graen, 1980), for example, whether only high LMX-and not medium LMX-has positive effects on leaders' support for employees' ideas.
Manipulation of Idea Consequences. The idea voiced by the employee dealt with employees' annual performance review, in particular, with leaders' gain or loss of power in this review. We described that in the past, this performance review has been based on both the leader's and the peers' ratings, while both ratings have been weighed differently. The employee now proposes that a change in the weighting of leader's versus peers' ratings is needed (see Online Supplement for full scenarios). To manipulate the consequences of implementing the employee's idea in terms of power threat versus power gain for the leader, we altered the weighting of the leader's rating (and thus the degree of their influence) in employees' performance reviews. In the power threat condition, we described a change from "60 per cent leader rating: 40 per cent peer rating" to "20 per cent leader rating: 80 per cent peer rating." Vice versa, in the power gain condition, we described a change from "40 per cent leader rating: 60 per cent peer rating" to "80 per cent leader rating: 20 per cent peer rating." Manipulation Checks. Participants rated two manipulation check items. One item assessed the level of LMX participants perceived to have with the employee described in the scenario (rated from 1 = very poor to 5 = very good). Another item assessed the degree to which "as a leader, one would lose influence" in case the idea was implemented (rated from 1 = not at all true to 5 = completely true).

Measures
Idea Support. We used four items to measure leaders' intention to support the employee's idea, which have been used in previous research (Urbach & Fay, 2018). These items capture supportive behaviours a leader can show in response to an employee's idea proposal. Sample items are "I will encourage the employee to take his idea further," and "I will seriously consider what has to be done to implement the idea." Participants assessed how likely they would show the respective behaviours in response to the described idea (from 1 = not at all to 7 = very much). The average of all items was used as an indicator of idea support.
Attributed Employee Intentions. To assess the degree of prosocial and egoistic intentions leaders attributed to the employee's idea proposal, we used four items each previously developed (Urbach, Fay, & Lauche, 2016) and adapted (Urbach & Fay, 2018). Participants were asked to think of the idea-presenting employee and consider whether the intentions described by each of the eight items were true for this person's idea proposal. Sample items are "The employee wants to help improving everyone's situation" (prosocial intentions) and "The employee wants to safeguard his own interests" (egoistic intentions). Ratings were made from 1 = not at all true to 5 = completely true. The means of the respective four items were used as measures for attributed prosocial and egoistic intentions.
Leader Power Motive. Participants' explicit power motive was captured by six items taken from a German leadership motivation inventory (Felfe, Elprana, Gatzka, & Stiehl, 2012). This measure was constructed based on established motive scales, that is, the PRF (Jackson, 1984). The items used in this study reflect the striving for influence component of the power motive, which measures the tendency to seek positive outcomes associated with power (e.g., appreciating to be in control). A sample is "I feel comfortable with being in control of what happens in my surroundings"; participants rated to what extent each item was true for them (from 1 = not at all true to 5 = completely true). Given that our theoretical considerations are based on threatening a leaders' striving for influence, we did not include the fear of losing control component of the power motive (i.e., the tendency to avoid negative outcomes of power) in our analyses.

Results and Discussion
Descriptive statistics, intercorrelations, and internal consistency reliabilities of all study variables are shown in Table 1.
We tested whether LMX had an indirect effect on leaders' idea support via attributed employee intentions (Hypothesis 2). To test this by means of multiple mediation analysis (Preacher & Hayes, 2008), we dummy-coded the categorical LMX variable into two dummy variables, that is, high LMX and medium LMX, using low LMX as reference category (Aiken & West, 1991). Results showed that there is a significant indirect effect of LMX on idea support via attributed prosocial intentions, ab = 0.39, SE = 0.18, 95% CI [.08; .81]. In line with our assumptions leaders attributed higher levels of prosocial intentions to the employee when LMX was described as high rather than low (B = 0.44, SE = 0.19, p = .018); attributed prosocial intentions were in turn positively related to leaders' idea support (B = 0.89, SE = 0.13, p < .001). There was no indirect effect via attributed egoistic intentions, ab = 0.02, SE = 0.08, 95% CI [−.15; .20]. In sum, Hypothesis 2 was partly supported.  Hypothesis 3 proposed a three-way interaction effect of idea consequences, leaders' power motive and LMX, such that the negative effect of threatening leaders' power motive through an idea-which is statistically expressed by the Idea Consequences × Leader Power Motive interaction-on their idea support will be weaker with increasing levels of LMX. We tested this three-way interaction in a hierarchical multiple regression analysis using a set of two three-way interaction terms, involving each of the LMX dummy variables (medium LMX and high LMX, see Table 2, Step 5). Since the additional proportion of variance explained by both three-way interaction terms together is significant, F(2, 141) = 3.96, p = .021, the three-way interaction effect is significant (see Aiken & West, 1991).
We conducted subgroup analyses to determine the nature of the threeway interaction by computing the two-way interaction of leaders' power motive and the power-related idea consequences for each level of LMX (low, medium, and high, respectively). Figure 2 shows the plots of these two-way interactions. When LMX was described as being low (graph on the left), the two-way interaction of leader power motive and idea consequences was significant (β = −.48, p = .003, ΔR 2 = .150; n = 54). Leaders with a high power motive (+1 SD) were far more willing to support the power gain as compared to the power threat idea, B = −2.09, t(47) = −4.26, p < .001; leaders with a low power motive (−1 SD) remained unaffected by the idea's consequences, B = 1.23, t(47) = 1.51, p = .137. When LMX was described as being medium (graph in the middle), the two-way interaction of leader power motive and idea consequences was not significant (β = −.25, p = .264, ΔR 2 = .024; n = 54); the same was the case when LMX was described as being high (graph on the right; β = .22, p = .259, ΔR 2 = .028; n = 48). Thus, when LMX was described as being medium or high, leaders' intention to support the described idea was not affected by the idea's potential to threaten their power motive. Accordingly, Hypothesis 3, that implies a buffering effect of LMX, was supported.
In Hypothesis 4, we predicted that the three-way interaction effect described in Hypothesis 3 is mediated by the level of prosocial as well as egoistic intentions attributed to the employee. We tested this mediated moderation hypothesis with conditional process analysis (Hayes, 2013). the power threat idea compared to the power gain idea, B = −0.56, t(47) = −1.68, p = .099; leaders low on power motivation (−1 SD) attributed more prosocial intentions to the employee voicing the power threat idea compared to the power gain idea, B = 1.71, t(47) = 2.11, p = .041. When LMX was described as being medium (β = −.21, p = .356, ΔR 2 = .016, n = 54) or high (β = .04, p = .804, ΔR 2 = .001, n = 48), the Idea Consequences × Leader Power Motive interaction was not significant. As reported above, attributed prosocial intentions were positively related to leaders' idea support intentions. In sum, Hypothesis 4 was supported for the mediator attributed prosocial intentions, but not for attributed egoistic intentions. Altogether, the results of Study 1 underscore the role of LMX in shaping leaders' attributions for employees' ideas, and leaders' response to ideas that may threaten their need for power. Specifically, irrespective of the content of the idea, LMX fostered the attribution of prosocial intentions which were in turn related to enhanced levels of supportive intentions in leaders. Moreover, LMX had the potential to buffer the negative effect of threatening leaders' power motive on their intention to support an idea (Figure 2), mediated by the level of prosocial intentions leaders attributed to the employee. Most notably, medium LMX was sufficient to achieve the buffer effect on the relationship between threatening leaders' power motive on idea support. We point out the practical implications of this result in the General Discussion. Interestingly, low LMX employees received reasonably high support for their idea when their idea resulted in a power gain for the leader and the leader was high on power motivation.
The major strength of this study is that the scenario-based, quasiexperimental approach allows for drawing causal inference on the effects of LMX on leaders' idea support and the intentions they attributed to the employee voicing the idea. Moreover, using scenarios meant that all participants evaluated standardized ideas. This way, the ideas did not differ with regard to aspects other than the intended, for example, the kind and scope of changes suggested. However, this approach also comes along with limitations of the external validity of our results (Aguinis & Bradley, 2014). We must fIgURe 2. Leader-member exchange buffers the power threat effect on leader's idea support (study 1). acknowledge that participants evaluated a fictitious idea outside their naturalistic organizational environment. To address this limitation, we sought to replicate our results using a field study design in Study 2.

sTUDY 2 Method
Participants and Procedure. Data were collected in an online survey. Participants were recruited through cold calling in various German organizations and a call for participation on the website of a German psychology journal. Participants were first informed about the purpose of the study and asked to participate only if they had leadership responsibilities in their organization. Participants first completed measures on their explicit personal motives, including the power motive (see Study 1). After this, participants were asked to recall a past event in which one of their subordinates approached them with an idea for change within their work unit or organization. This event was required to have happened less than 12 months ago. To facilitate participants' recall of and re-immersion into this past situation, we asked for a brief description of the event in free text format. Other than in Study 1, where idea consequences and LMX were manipulated variables, here participants rated the power-, achievement-, and affiliationrelated consequences the implementation of this idea would have had. We assessed different motive-related idea consequences to conceal the focus of the study on power (further details can be obtained upon request). Following this, participants rated their intention to support this idea at the very moment after the employee had voiced the idea-irrespective of whether the idea had been implemented meanwhile. Last, participants evaluated their relationship quality (LMX) with the employee, rated further control variables, and reported demographic information.
Altogether, 157 individuals completed the survey. One participant was excluded because he reported not to have any direct reports (i.e., no leadership responsibility). Data of further 21 individuals were excluded because they had given no or no useful idea description (e.g., a random character string), or because their reported event was longer than 12 months ago. Two individuals were excluded due to long string response patterns (Johnson, 2005) and multiple missing values. The final sample consisted of 133 working individuals with leadership responsibilities (31% female). On average, participants were 46.3 years old (SD = 10.9), had been working with their current organization for 14.5 years (SD = 11.2), and occupied a leadership position for 12.1 years (SD = 9.71). The majority of participants worked for larger organizations (>500 employees, 45%), followed by small (≤50 employees, 21%) and medium-sized organizations (>50 but <500 employees, 34%). Participants worked in a wide range of industries, such as transport and logistics (17%), government and administration (14%), IT and media (11%), the retail and service sector (8%), the financial sector (8%), and various other industries (42%). Participants were rather highly educated (university degree: 71%; vocational training: 12%; high school diploma: 17%).
Measures. Leaders' power motive was measured in the same way as in Study 1. The items to capture attributed employee intentions as used in Study 1 were adapted to past tense in order to make them applicable to a past event.
Participants were instructed to rate all constructs that were related to the idea (i.e. idea support, idea consequences, attributed employee intentions, control variables) as perceived at the very moment after the employee had voiced the idea-irrespective of whether the idea had been implemented meanwhile.
Idea Support. We used six items to measure leaders' idea support. We adapted three of the four items used in Study 1 to past tense. One item was dropped as it was considered too prone to social desirability to assess support intentions in retrospective (i.e., "I took my time to thoroughly listen to the employee's idea"). Moreover, we extended the measure by three items that more explicitly capture whether respondents had actually supported the reported idea. These items were adapted from Burris's (2012) idea endorsement scale, that is, "I thought the employee's idea should be implemented," "I agreed with the employee's idea," and "I supported the idea to make sure it gets implemented." Items were rated on a Likert-type scale from 1 = not at all true to 5 = completely true. A one-factor model of the items in a CFA proved good fit, χ 2 (9) = 10.24, p = .332, RMSEA = .032, CFI = .99, SRMR = .048. Idea Consequences. We developed a five-item scale to assess the powerrelated consequences of an idea for a leader. The items aimed at capturing the focal aspects of being in power, such as exerting influence over others, being in control, and having authority (e.g., McClelland, 1975). Participants assessed on a seven-point sematic differential scale to which degree the various aspects of power would 1 = increase to 7 = decrease, respectively, if the idea was implemented. Items were presented as follows: "My influence as a leader would… [increase-decrease]." The other four items captured "opportunities to manage people," "autonomy," "authority," and "control span" of the leader. The five items were averaged as an indicator of power threat, while higher values indicate higher levels of power threat of the idea. A CFA one-factor model fit the data well, χ 2 (5) = 6.54, p = .257, RMSEA = .048, CFI = .99, SRMR = .036. Leader-Member Exchange. Other than in Study 1, where LMX was manipulated, participants rated the quality of their relationship with the employee who presented the idea on the 12-item multidimensional LMX scale developed by Liden and Maslyn (1998). We used a validated German version of this scale by Paul and Schyns (2014) and adapted the items to a leader rating of LMX by changing the referent of the items from "my leader" to "this employee." Liden and Maslyn's (1998) scale measures LMX in terms of four dimensions, that is, affect, loyalty, professional respect, and perceived contribution. Sample items are "I like this employee very much as a person" (affect), and "I am impressed with this employee's knowledge of his/her job" (professional respect). Ratings were made on a Likert-type scale from 1 = not at all true to 7 = completely true. We computed one mean score representing overall LMX (e.g., as was done by Van Dyne, Kamdar, & Joireman, 2008).
Control Variables. To test the robustness of our results, we included additional control variables in Study 2, that is, implementation benefit, implementation cost, and idea presentation quality. It has been suggested that optimizing cost-benefit ratios is a major driver behind human decision making (e.g., Beach & Mitchell, 1978). From this perspective, one could argue that the main predictor of leaders' idea support should be the expected benefit of the changes proposed, while taking into account which costs the implementation of the idea would have. Moreover, previous research shows that leaders' evaluation of voice will depend on presentation quality, for example, whether employees offer a solution to the problem they identified (Whiting et al., 2012), or at least give a thorough justification for their proposal. We intended to show that the processes studied in this paper (i.e. the main and moderator effects of LMX on idea support) explain incremental variance in leaders' idea support, above and beyond these control variables.
We developed three items, each to assess perceived implementation benefit (i.e., "If the idea was implemented … we could make profits, e.g., gain money, time, or other resources/ … we could reduce losses, e.g., save money, time, or other resources/ … it would improve the way our work is organized"), perceived implementation costs (i.e. "Implementing the idea would … require many resources, e.g., time, money, or other resources/ … be costly/ … require coordination with many stakeholders"), and idea presentation quality (i.e. "The employee offered a logical justification for the idea./ The presentation of the idea was well structured./ The employee offered possible solutions how to implement the idea."). Items were rated from 1 = does not at all apply to 5 = applies fully; CFA supported the three-factor structure, χ 2 (24) = 23.49, p = .491, RMSEA < .001, CFI = 1.00, SRMR = .051.

Results and Discussion
Descriptive statistics, intercorrelations, and internal consistency reliabilities of all study variables are shown in Table 3. We conducted multiple regression analyses, mediation analyses, and conditional process analyses (Hayes, 2013) to test our hypotheses. We controlled for participants' age, gender, and tenure as a leader in all analyses.
Hypothesis 1 proposed a main effect of LMX on idea support. Results support this (see Table 4; β = .21, p = .017, ΔR 2 = .043). In Hypothesis 2 we stated that the main effect of LMX on idea support was mediated via attributed employee intentions. In line with this hypothesis, higher levels of LMX were related to higher levels of prosocial intentions attributed to the employee (β = .17, p = .044, ΔR 2 = .030). In turn, attributed prosocial intentions were positively related to idea support (β = .24, p = .008, ΔR 2 = .052; while controlled for attributed egoistic intentions). The indirect effect of LMX on idea support via attributed prosocial intentions was significant, ab = 0.03, SE = 0.02, 95% CI [<0.01; 0.08]. Contrary to our assumption, we found no indirect effect via attributed egoistic intentions, ab < 0.01, SE = 0.01; 95% CI [−0.01; 0.03], as LMX was not related to the level of attributed egoistic intentions (β = −.03, p = .773, ΔR 2 = .001), and attributed egoistic intentions were not related to idea support (β = −.11, p = .216, ΔR 2 = .011; while controlled for attributed prosocial intentions). The remaining direct effect of LMX on idea support reached only marginal levels of significance (β = .17, p = .058, ΔR 2 = .026). Taken together, Hypothesis 2 found partial support. The result pattern for this hypothesis is the same as in Study 1, such that the LMX-idea support linkage is mediated through attributed prosocial intentions, but not through attributed egoistic intentions.
Hypothesis 3 predicted that high levels of LMX could buffer the effect of threatening a leader's power motive through an idea on the leader's intention to support this idea. To test this hypothesis, we tested the three-way interaction effect of power-related Idea Consequences × Leader Power Motive × LMX (see Table 4). Results show that this three-way interaction effect on idea support was significant (β = .39, p < .001, ΔR 2 = .109). The plot of this interaction is depicted in Figure 3; graphs are plotted for +/-1 SD from the mean of leader power motive and LMX, respectively. When LMX was low (graphs with black end marks), leaders high on power motivation (solid line) were more likely to support an idea involving low rather than high power threat, B = −0.25, SE = −0.09, p = .008; there was no relationship of power threat with idea support when leaders were low in power motivation (dashed line), B = 0.04, SE = 0.11, p = .718. This is in line with Hypothesis 3. When LMX was high (graphs with white end marks), there was no relationship of power threat with idea support for leaders high in power motivation (solid line),  B = 0.20, SE = 0.12, p = .101. In other words, as predicted, high levels of LMX buffered the effect of power threat on idea support for leaders high on power motivation. Going beyond what was predicted in Hypothesis 3, leaders low in power motivation (dashed line, white end marks) responded to higher levels of power threatening idea consequences with reduced levels of idea support, B = −0.55, SE = 0.14, p < .001. We interpret this contradictory result from the perspective of psychological contract breach (Rousseau, 1989) in the General Discussion.
In Hypothesis 4 we proposed that the moderator effect of LMX described in Hypothesis 3 is mediated via attributed employee intentions. As summarized in Table 4, there was no significant three-way interaction effect of Idea Consequences × Leader Power Motive × LMX on attributed prosocial intentions (β = .00, p = .986, ΔR 2 < .001); the three-way interaction on attributed egoistic intentions was only marginally significant (β = −.17, p = .097, ΔR 2 = .019). Given that only attributed prosocial intentions were related to idea support (see Hypothesis 2), the indirect effects via prosocial Overall, the results of Study 2 replicate the results of Study 1 with regard to the most important process, that is, the buffering effect of LMX on threatening a leader's power motive through an idea on the leader's intention to support this idea (Hypothesis 3). In addition to the indirect effect of LMX on idea support via attributed prosocial intentions (Hypothesis 2) we also found a significant direct relationship of LMX and leaders' idea support in Study 2 (Hypothesis 1), which we will address in the General Discussion section. Other than in Study 1, we did not find support for a mediated moderation effect via attributed prosocial intentions here (Hypothesis 4).
To test the stability of our results, all analyses were re-run including the control variables implementation benefit, implementation cost, and idea presentation quality. Altogether, the result pattern reported in Table 4 remained the same albeit effect sizes were reduced. Implementation benefit (β = .45, p < .001) and idea presentation quality (β = .28, p < .001) were strongly related to idea support (implementation costs: β = −.08, p = .286; joint ΔR 2 =.359). This reduced the relationship of LMX with idea support to marginal levels of significance (β = .13, p = .068, ΔR 2 = .016), and the three-way interaction effect in effect size (β = .22, p = .011, ΔR 2 = .031). However, we would like to underscore that this complex interaction effect was significant (p = .011) even fIgURe 3. Three-way interaction effect of power-related idea consequences, leader power motive, and leader-member exchange on leader's idea support (study 2). after controlling for variables that should be major evaluation criteria for ideas in a naturalistic setting.

geNeRaL DIsCUssION
Making use of employees' ideas for improvement is considered to be an important success factor of today's organizations (e.g., Van Dyne & LePine, 1998;Frese & Fay, 2001;Grant & Ashford, 2008). While extant research has identified various predictors of employees' proactive voice (e.g., Morrison, 2011) and employee suggestion-making (Frese, Teng, & Wijnen, 1999), rather little is known about what facilitates the actual implementation of employees' ideas. A key step in the process from voicing an idea to implementing constructive change often is to obtain leaders' support for an idea. Previous research has shown that leaders might refuse to support employees' ideas for constructive change because an idea threatens their personal needs and orientations (Sijbom et al., 2015a(Sijbom et al., , 2015bUrbach & Fay, 2018). Thus, organizations might not yet exploit the full potential that employees' ideas offer. In this research, we make a case for leader-member exchange quality as an important facilitator of obtaining leaders' support for an idea. Across one quasi-experimental and one field study, our results show that when LMX was high rather than low, leaders attributed more prosocial intentions to employees' idea proposals, which in turn made their idea support more likely. Above and beyond, LMX functioned as a buffer for the negative effect of threatening a leaders' personal power motivation through an idea-a condition that is considered to be a major reason why voiced ideas may be turned down by leaders (Burris, 2012;Fast et al., 2014;Kanter, 1988;Urbach & Fay, 2018). The results of our two studies are inconclusive with regard to the direct effect of LMX on leaders' idea support, and to the role of attributed prosocial intentions as a mediator of the LMX buffer effect on power threat. Moreover, attributed egoistic employee intentions did not function as a mediator. In the following we discuss how future research could help to clarify these results.

Theoretical Implications
Our research contributes to the literature on LMX and voice behaviour by pointing out the role of LMX in leaders' receptivity to employee voice behaviour. Consistent with our assumptions, LMX was positively related to attributing prosocial intentions to employee voice and in turn being more likely to support an idea across both studies. Interestingly, we only observed a direct relationship of LMX and idea support in Study 2. A reason for this may be that in Study 2 leaders reported ideas that were voiced in a naturalistic context by an existing employee. Thus, their assessment of the relationship quality with this employee based on the Liden and Maslyn (1998)  to have captured a more detailed picture of the employee than the manipulation in Study 1 could have achieved. At the same time, leaders' assessment of LMX might have been biased by the voice event they recalled. Thus, in terms of reversed causality, remembering an employee's good idea might have augmented leaders' perceptions of LMX-particularly with regard to professional respect and perceived contribution, as a good idea may be regarded as evidence for employee ability and dedication. In this regard, the null effect of LMX on idea support based on the experimental design in Study 1 is more valid. Thus, further studies are warranted to test the direct effect of LMX on idea support, for example, by temporally separating the assessment of LMX and the voice event.
More consistently, both studies show an indirect effect of LMX on idea support via attributed prosocial employee intentions, but no indirect effect via attributed egoistic employee intentions. Higher levels of LMX were associated with higher levels of attributed prosocial intentions. Furthermore, in line with previous research on the evaluation of proactive work behaviour (e.g., Grant & Ashford, 2008;Grant et al., 2009), attributed prosocial intentions were positively related to idea support. On the contrary, attributed egoistic intentions did not predict idea support. The same result pattern has been reported in previous studies (Urbach & Fay, 2018;Urbach et al., 2016). This leads us to conclude that egoistic and prosocial intentions are not just the two sides of a coin but portray distinct processes. While the perception of prosocial intentions is generally beneficial for idea support, the perception of egoistic intentions is not uniformly negative, at least in the individualistic cultural context our studies were conducted in. An idea might serve the voicing employee, but this benefit is not necessarily perceived to come at the cost of the organization.
Moreover, we advance our knowledge on the evaluation of employee voice and other forms of proactive suggestion-making by shifting our research focus on leaders' support for specific acts of voice behaviour, that is, specific employee ideas. This shift in focus enables us to explicitly consider characteristics of ideas as predictors of leaders' idea support-in particular, the consequences that the implementation of an idea has for leaders themselves. Although employee voice behaviour may be perceived as generally powerthreatening (Burris, 2012), it is likely that the specific changes proposed in ideas differ in their power-threatening potential, and that leaders differ in their susceptibility towards such power threats (e.g., due to differences in their need for power). We argued that LMX is particularly important for facilitating idea support in the critical situation when an idea threatens a leader's high power motive. Across both studies, our results lent support to this assumption. This three-way interaction effect explained a considerable amount of variance in leaders' idea support in the quasi-experimental Study 1 (4.9%) as well as in the field in Study 2 (10.9%), which by far exceeded the predictive power of the direct effect of LMX in Study 2. However, in Study 2 the three-way interaction pattern revealed an effect that contradicts our initial theorizing: In high LMX relationships, leaders low on power motivation were less likely to support an idea the higher its power-threatening potential was. We had expected that leaders low on power motivation should not feel threatened by the fact that an employee's idea challenges their power, which should have resulted in a non-significant relationship between level of power threat and idea support. However, leaders low on power motivation might perceive such employee behaviour as a breach of their psychological contract, that is, leaders' implicit belief in their reciprocal obligation with an employee (Rousseau, 1989). Part of this psychological contract could be leaders' expectation that subordinates shall not question a leader's position or expertise. As a consequence, leaders might be less willing to support such an idea, because they are less willing to grant support to an employee who does not fulfil their obligations (Chen, Tsui, & Zhong, 2008). Moreover, individuals seem to feel particularly "betrayed" by a psychological contract breach when they have a high quality relationship (Restubog, Bordia, Tang, & Krebs, 2010). Given the comparatively high levels of trust and resources leaders grant employees when LMX is high, leaders' reciprocal expectations of employees' obligations towards the organization (including them as their leader) are higher than when LMX is low (Tekleab & Taylor, 2003). Applied to our study results, this can explain why leaders low on power motivation provide less support for a power threatening idea when LMX is high, but not when it is low. However, this explanation is highly speculative and warrants further empirical investigation.
Our data lent limited support to Hypothesis 4 that this complex three-way interaction effect is mediated by attributed prosocial and egoistic employee intentions. While there was evidence for the proposed mediation effect via attributed prosocial intentions in Study 1, this effect could not be replicated in Study 2. Similar to the results obtained on Hypothesis 2, attributed egoistic intentions did not play a role as a mediator here. Overall, our results suggest that LMX is particularly important in critical situations that involve risks for the leader. Consequently, we suggest leader trust as an alternative mediator of the buffer effect described in Hypothesis 3. Trust may act as a key mechanism as it reflects the willingness to take the risks in connection with the proposed changes (Mayer, Davis, & Schoorman, 1995). Future research needs to specify which aspect of trustworthiness is particularly important here, for example, the perceived ability of the employee to turn his/her idea into a success, or the benevolence of their intentions not to do harm to the leader.
Our study also contributes to the literature on proactive behaviour and individual performance, which shows that proactive work behaviour positively contributes to employees' general performance appraisals by their leaders (e.g., Grant & Ashford, 2008;Howell et al., 2015;Huang et al., 2018;Whiting et al., 2012). Our study helps to specify the process that links these two variables, that is, leaders' support for specific employee ideas should mediate the proactive behaviour-performance evaluation link. In line with this, Burris (2012, Study 2, r = .37) reports a positive relationship between leaders' endorsement of a specific idea and their general performance evaluation of the voicing employee. Our results and recent findings by Huang et al. (2018) suggest that voicers with high LMX might be generally rewarded more for showing voice than voicers with low LMX, both in terms of idea support and a more positive performance evaluation.

practical Implications
From a practical perspective, our research highlights that leaders through their own motives and their relationship with the employee are a critical factor for whether a voiced idea makes it or not. Thus, our findings build on the notion that leaders may become a bottle neck in making constructive change happen in organizations (Parker & Wu, 2014) by showing when this is more likely to happen. To a certain extent, leaders' support for employee voice behavior depends on the consequences voiced ideas have for the leader as well as the level of LMX the leader shares with the idea-presenting employee. In Study 2, this effect remained unchanged even after controlling for the potential benefits and costs leaders anticipated from implementing an idea. Thus, organizations need to be aware that they may not exploit the full potential of employees' proactive efforts because leaders could turn ideas down out of self-interest. On the positive side, high (and even medium) LMX quality proved to be a protective factor against such self-interested responses to employee voice. Thus, organizations should encourage leaders to build respectful and mutually satisfying relationships with their followers, for example, by investing in LMX training. Research shows that specifically employees in low LMX dyads benefit from such interventions (Scandura & Graen, 1984). Given that LMX develops over time, it seems particularly promising to sensitize newly appointed leaders to the importance of active relationship building (Manderscheid & Ardichvili, 2008).
From the employees' perspective, our results suggest that employees should establish a good relationship with their leader before voicing concerns or ideas their leader might perceive as a threat. It is most notable that even medium levels of LMX were sufficient to buffer the effect of threatening leaders' power motive on idea support. This supports results by Liden and Graen (1980) that not only high but also medium relationship quality can bring benefits for employees. Usually, research only highlights effects of low or high LMX on employee-related outcomes. However, if LMX follows a normal distribution across dyads, most dyads should be characterized by medium levels of relationship quality. Moreover, leaders seemed to be more likely to support ideas by low LMX employees if they satisfied their own motives (i.e., leaders high in power motivation receive a power gain). Leaders might not expect such behaviour expressing loyalty from a low LMX employee. For the employee, in the long run, positively evaluated proactive behaviour could be a means to improve their standing with the leader. In line with the idea that proactive behaviour could also affect LMX, results by N. Li, Liang, and Crant (2010) show that employees with a proactive personality, who should show proactive behaviours such as voice on a regular basis, establish higher-quality relationships with their leaders. Similarly, in the long run, proactive personality is associated with gains in leader support (W.-D. Li, Fay, Frese, Harms, & Gao, 2014). Likewise, a study by Cheng, Lu, Chang, and Johnstone (2013) found a positive relationship between employee voice and LMX, particularly when the leader attributed prosocial intentions to the voice behaviour. The latter is the case when a voiced idea serves to satisfy the leaders' motives (Urbach & Fay, 2018).
Our results further have implications for bottom-up influence tactics in teams or organizations (Yukl, Fu, & McDonald, 2003). In order to optimize the effectiveness of their upward influence attempts, employees should reflect upon the power-related consequences of an idea for their leader before bringing the idea to the leader's attention. Our results suggest that the best strategy to "sell" a potentially power-threatening idea to a leader would be to have it presented by a colleague that shares medium-or better high-levels of LMX with the leader. This way, a team could maximize their chances to benefit from their ideas.
Nevertheless, we need to consider possible boundary conditions of the effects studied. Although trust and LMX enhance risk-taking in relationships, this effect still is dependent on contextual factors, such as the level of risk perceived in a specific situation (Brower et al., 2000;Mayer et al., 1995), or the importance of the decision at hand (Leana, 1986). If the personal risk for the leader gets too high, it is possible that even high LMX will not help to reduce risk perception, such that the buffering effect of LMX might disappear.

Limitations and future Directions
While the different designs used in Study 1 and in Study 2 have their unique weaknesses, their strengths complement each other: The quasi-experimental design applied in Study 1 allows for drawing causal inference on the effects of the manipulated variables, which is not possible in Study 2. However, replicating the most important results of Study 1 in a naturalistic setting provides support for the external validity to our findings. A major limitation is that for some of the relationships under study we cannot draw causal inferences, that is, the relationships among attributed employee intentions and idea support. In both Study 1 and Study 2, these variables were captured by explicit, direct self-report measures. Thus, the rationale to conceptualize attributed employee intentions as preceding leaders' idea support was solely based on the theoretical argument that attributions inform leaders' responses to employees' behaviour (e.g., Bowler et al., 2010;Steiner, 1997). However, reverse causality does not seem plausible in this case. Moreover, in Study 2, all variables that were manipulated in Study 1 were measured through self-report at the same measurement occasion. The fact that we observed differential and even null relationships between variables speaks against a potential inflation of effects through common method variance (Spector, 2006). As common method variance cannot artificially produce interaction effects but would rather deflate them (Siemsen, Roth, & Oliveira, 2010), we further conclude that the results on the three-way interactions in Study 2 are reliable. The similarity of results to the experiment in Study 1 supports this conclusion.
Another potential limitation, particularly of Study 2, is that we observed a very high average idea support reported by leaders. This may reflect a positive recall bias resulting in a range restriction of events that can be obtained using a recall paradigm. In line with this notion, research on the recall of social interactions suggests that individuals tend to forget unpleasant events, and remember events in a more positive light than they were actually experienced (e.g., K. K. Li, 2013). To obtain a wider range of supported and unsupported employee ideas in future studies, researchers may consider guiding participants' recall using explicit instructions to think of positive versus negative events. Alternatively, relevant events could be sampled through daily or weekly diary studies, which should be less biased by memory effects due to the shorter time lag between event and recall.
Another limitation refers to our measurement of the dependent variable idea support. In Study 1 we have measured leaders' intentions to support an idea, not their actual response towards an employee. This assessment lacked alternatives due to the simulation-oriented nature of the vignette study. In Study 2, we went one step further and assessed leaders' support for the idea at the time of idea proposal; however, we do not know whether the implementation of the idea actually took place. There are two reasons why these types of measures are appropriate in the context of our research: First, the focus of our study is on identifying factors that help employees' ideas to pass the first "bottle neck" on their way to implementation, that is, obtaining their leader's support. Thus, leaders' behavioural intentions should be the best predictor of their behaviour (Ajzen, 1991), and reflect their individual needs and aspirations rather than situational constraints. Second, there are numerous factors that could have affected whether an idea had been implemented by the time of our study: Very recently voiced ideas might still have been in the process of implementation. Other variables than the leader's opinion might have determined the final decision on idea implementation (e.g., structural boundaries, an order of the leader's superior). Nonetheless, future studies should strive to assess the more immediate behavioural responses of leaders to employee voice in a naturalistic setting, for instance through the separated measurement of employees' ideas and leaders' response.
Future research should further take the gender of the voicing employee into account. In the vignettes in Study 1, we have constantly described the employee who voiced the idea as being male, because previous research suggests that ideas might be evaluated differently depending on the voicing employee's gender. Based on gender roles, women are expected to be more communal, while men are expected to be more agentic; accordingly, agency is not compatible with female role expectations (Eagly & Karau, 2002). Affiliative citizenship behaviours are considered to be communal in nature and thus are part of women's role expectations (Heilman & Chen, 2005). In contrast, constructively challenging extra-role behaviours such as employee voice are considered to be rather dominant and agentic (Grant & Ashford, 2008;Grant et al., 2011). Accordingly, men might find more leader support for their ideas than women as they are expected to voice. Recent research on the evaluation of innovative employee behaviour provides evidence for such gender bias (Luksyte, Unsworth, & Avery, 2017). Given the dyadic perspective of our research, investigating gender configurations of leader and member would be particularly interesting as behavioural expectations and thus the evaluation of others' behaviour might differ between male versus female versus mixed-gender dyads.