When Trust in the Leader Matters: The Moderated-Mediation Model of Team Performance and Trust

This study contributes to the sport and team literature by exploring the conditions in which trust in a leader translates into trust in a team and subsequent team performance. Findings from 709 athletes on 74 basketball teams demonstrated that trust in the coach represents a critical antecedent of team trust, especially when the team's past performance has been poor. We also found a combined effect of the level and consensus in trust on team performance. Practical implications suggest that a coach needs to ensure that every player, rather than just some or even the majority of individual team members, trusts him or her and the team.

. Conceptual framework-moderated mediation trust model. a work unit is argued to shape team-level outcomes (Graen & Uhl-Bien, 1995). Sherony and Green (2002) empirically confirmed that LMX relations have an effect on the exchange quality between coworkers in dyads. Their analysis, based on Heider's balance theory, revealed that when coworker A's LMX rating was above average, higher levels of coworker B's LMX were associated with higher quality of their relationships, and disparity between LMX ratings for coworker A and coworker B negatively affected their relationships (Sherony & Green, 2002).
We propose that this effect will also occur for trust in the leader and trust in the team relationship. Similarly, the combination of low and high trust in the leader can create a context for higher (or lower) willingness to see team members as more trustworthy, more potent, and well intended. Indeed, high-quality LMX as assessed by the team members has been demonstrated to be positively related to organizational citizenship behaviors (Ilies, Nahrgang, & Morgeson, 2007) and negatively related to conflict (Boies & Howell, 2006). This suggests that trust in the leader, as a component of LMX, can enhance the "benevolence" dimension of trustworthiness among the teams' members, building overall team trust. Furthermore, trust in leadership leads to higher team performance because players are more willing to carry out the tasks and strategies set out by the coach, eschew any doubts about the team, and work toward a common goal (Dirks, 2000;Schaubroeck, Lam, & Peng, 2011). Therefore, the proposed mechanism in forming team trust suggests that trust in the leader acts as an antecedent, a prerequisite for trust in team members.
Thus, trust in teammates provides the condition for other processes to take place and affects team results, both directly and indirectly (Dirks, 1999;Jarvenpaa, Shaw, & Staples, 2004). Trust in teammates encourages cooperative behavior and leads to higher performance (De Jong et al., 2016;Hempel et al., 2009). Although trust in the leader affects team trust development, trust in teammates results in overall better performance (McAllister, 1995). This combined relationship suggests the presence of a mediation effect, where team trust represents a mechanism through which leaders can influence team performance. Therefore, the first hypothesis of our study was that team trust mediates the relationship between trust in the leader and team performance. This conceptual framework is depicted in Figure 1.

INTRATEAM TRUST CONSENSUS
Intrateam trust is frequently operationalized as the aggregate perception of trust that members have in their fellow teammates (e.g., Langfred, 2004) or in their direct manager (Dirks & Ferrin, 2002). However, the validity of the analysis of the trust level is conditioned by assumptions regarding a relatively low variability of perceived trust. Therefore, in addition to the additive model of trust proposed in the first hypothesis, we hypothesize that consensus will add to the explained variance in team performance.
In accordance with our theoretical argument based on trust transferability, lower trust in the leader implies lower trust in the team because it effectively eliminates the leader as an important source of team trust judgments. Just like a bad apple can spoil the bunch, an untrustworthy coach will fail to originate or maintain trust in the team. We expect that the development of consensual perceptions among teammates results in high quality within-group relations and, therefore, enhances goal accomplishment (Levine & Moreland, 1990;Lindell & Brandt, 2000). Indeed, a within-team LMX (averaged dyad scores) was found to be positively related to team potency via instilling team confidence (Boies & Howell, 2006), and, subsequently, we maintain that high levels and high consensus of trust in the leader positively relate to the team's perception of their capabilities, consequently enhancing the "ability" dimension of trustworthiness. Hence, we argue that groups with a high level of consensus about trust in their leader will also translate this shared perception into trust in their teammates and, consequently, into better performance. Notably, high consensus about low trust in leadership will only enhance the relationship, albeit working in the opposite direction. Subsequently, the second hypothesis proposed that the consensus of team members' perception about trust in the leader would moderate the positive relationship between levels of trust in the leader and trust in teammates such that, when consensus is high, the relationship between trust in the leader and trust in teammates is positive and strong. Conversely, when consensus is low, this relationship is weakened.
We turn now to shared trust among team members, a variable argued to have been virtually ignored by researchers (Lau & Liden, 2008). A growing number of scholars have argued that when treating dispersion as an error variance (Chan, 1998;Harrison & Klein, 2007), researchers overlook the potential importance of dispersion in predicting work-related outcomes (Chan, 1998;Kirkman & Shapiro, 2005). Dispersion of trust in the team may be understood as differences among members in their assessment of the level of trust in a team. Low dispersion (agreement) sets the context for establishing and maintaining a high level of mutual trust. When it happens to be out of balance-for example, when one team member experiences a considerably lower level of trust compared to other members-overall trust may quickly deteriorate, as no member of the team wants to be more vulnerable than another. These dispersion properties are thought to work as moderators that account for the differential effects of mean levels of trust on team performance (e.g., Colquitt, Noe, & Jackson, 2002;Gonzalez-Roma, Peiro, & Tordera, 2002). Therefore, for Hypothesis 3 we proposed that consensus in team trust moderates the positive relationship between levels of trust in the team and team performance such that, when consensus is high, the relationship is stronger.

PAST TEAM PERFORMANCE AS A KEY CONTEXTUAL FACTOR
Although there is consensus regarding the critical role that past performance plays in predicting the future performance (e.g., Dirks, 2000;Lau & Liden, 2008), traditional practice uses past performance as a control variable (e.g., Costa, 2003). Unfortunately, parceling out past performance's influence may also lead to numerous erroneous conclusions. Using past performance as a control variable can be an "urban legend" along with other frequently used controls such as demographics (Spector & Brannick, 2011). Rather, we see past performance as a situational constraint (Johns, 2006) that modifies the functional relationship between the focal variables. Previous research has already suggested the importance of past performance as a moderator. For example, Lau and Liden (2008) found that the relationship between a leader's trust in an employee and a team member's trust in this coworker was reinforced in a context of low performance and weakened the relationship when team performance was high. The authors explained that a sense of security obtained from previous high achievement may translate into less reliance on the leader's opinion about team members.
Based on similar logic, we argue that past performance represents an important context for trust in the coach and the team trust relationship. Specifically, trust in the coach is particularly important for the team experiencing high uncertainty about its performance due to inconsistent or poor past performance. This uncertainty limits teammates' capability and willingness to build their trust judgment based on their own observations. As such, trust in a third party becomes even more relevant. When team past performance is not perceived to be high, trust in the coach becomes an anchor for the team members' will and their ability to develop trust in their teammates.
Lack of agreement about a coach's good intentions multiplies the need to preserve one's own interest. Hence, we predict the combined effect of these moderators on the direct relationship between trust in the leader and trust in the team. Our fourth and final hypothesis predicts that the relationship between trust in a leader and team trust will be stronger among teams with poor past performance and higher consensus in trust in a leader.

Participants
The sample consisted of 709 players belonging to 74 teams in the basketball leagues in Catalonia, Spain. Fifty-two teams belonged to the men's league and 22 teams to the women's league. Of the study participants, 70.27% were male. The mean age of male players was 24.7 (SD = 4.7) and was 22.8 among female players (SD = 4.3). The average tenure on the same team was three seasons (SD = 3.4). The average seniority (years of sport experience) was 15.4 years (SD = 5.1). Participating players belonged to both professional (35%) and top amateur (65%) leagues. The response rate was 98.65%, as only one team chose not to participate in the study.

Trust
We used the Trust Questionnaire developed by McAllister (1995), adapted to sports settings by Dirks (2000) and validated 1 into Spanish (Mach et al., 2010), to assess the team members' perception of "Trust in the team" (eight item) and "Trust in the coach" (nine item).
We adapted it to the two distinct referents: "players" and "main coach," respectively. Respondents indicated their agreement with each statement on a 9-point Likert scale ranging from 1 (strongly disagree) to 9 (strongly agree). Sample items included "I have a sharing relationship with team players. I can freely share my ideas, feelings, and hopes with them"; "Players approach their job with professionalism and dedication"; "If I share my problems with my coach, I know he would respond constructively and caringly"; and "The coach approaches his job with professionalism and dedication." Team trust level. We operationalized this variable as the mean of all team members' responses. Higher scores represent a higher level of trust in the team or trust in the coach.
Cronbach's alpha for "Trust in the team" was 0.93; the interrater agreement was r WG(j) = 0.91, and the intraclass correlations were ICC(1) = 0.41 and ICC(2) = 0.87. Cronbach's alpha for "Trust in the coach" was 0.96; the interrater agreement was r WG(j) = 0.88, ICC(1) = .40 and ICC(2) = .87. The values of the indexes reached the accepted conventions for aggregation to the group level but not excessively high to preclude the analysis of the dispersion (Bliese, 2000).
Team trust agreement. This variable was operationalized as the standard deviation of the within-group, cumulating squared distances between pairs of individuals. This index is based on the dispersion or "separation" of data around the group-level means (Harrison & Klein, 2007) such that higher scores represent greater dispersion and zero implies that all team members' responses were identical.

Team performance
We collected the team performance measure from independent official sources (Basketball Federation League records) at two points in time: midseason and at the season's end.
Team future performance. The performance ratio captures the winning percentage of games played during the second half of the season. Specifically, this ratio represents the number of games won at the end of the season divided by the number of games played and multiplied by 100, minus the same ratio at the middle point of the regular season. Thus, the higher the percentage, the higher the team's effectiveness.
Team past performance. This variable was calculated similarly to the criterion variable, representing the ratio of won games to played games, but at the end of the previous season.

Control variables
Several contextual variables pertaining to team success in interactive sports were used as controls. At the team level, we controlled for the type of league, gender, player experience (seniority), seasons trained by the head coach, and perceived team effectiveness.
The coach's perception of team effectiveness. This was used to accommodate for the second assumption of trust transferability theory: A leader who trusts a team member influences other team members regarding the team member's trustworthiness. As previous research suggests, third-party trust does not have to be explicitly communicated (Wildman et al., 2012). Thus, we chose to control for the coach's perception rather than use it as a predictor or a moderator. We assessed this using four statements designed by the first author as a result of a focus group with basketball experts. On a 9-point Likert scale, the head coach provided an assessment of every team member. The results were aggregated to the team level. Cronbach's alpha for "coach's perception of team effectiveness" was 0.87; the interrater agreement was r WG(j) = 0.83, ICC(1) = 0.26, ICC(2) = 0.74. A sample item was "S/he is very effective and works together with teammates to accomplish team objectives."

Procedures
Two members of the research team collected data from teams after a training session. We aggregated the data at the team level and used it to calculate within-group distribution of members' responses (Chan, 1998;Harrison & Klein, 2007;Lindell & Bandt, 2000). Previously, the Catalan Basketball Federation had informed clubs about the research project and asked for their cooperation and consent to participate. The research team was also provided with a list of each team's players for that season. The first author or the research assistant then visited the 74 clubs and personally administered the survey at the end of a training session. To ensure confidentiality, completed questionnaires were submitted directly to the researcher by each player.
To avoid common method bias-when the correlation between predictors and outcomes is caused by the single source and common method of data collection-we sought predictors from multiple sources and gathered the data on team performance from the Basketball Federation's online archives. Furthermore, we collected data at different points: We collected predictor variables in the middle of the season and the criterion variable (objective performance) at the end of the season.

Data Analysis
Data analysis comprised three stages. First we generated descriptive statistics including means, standard deviations, and correlations at the team level of analysis (n = 74; see Table 1). Next, before testing our hypotheses, we checked the psychometric properties of the measures and their discriminant validity by performing confirmatory factor analyses. We confirmed that data aggregation at the team level was appropriate. Rather than simply assuming team members' perceptions to reflect a shared reality, we tested this expectation by examining the corresponding analyses of variance F statistic and associate ICC coefficients (Bliese, 2000). ICC(1) provides an indication of whether there is a team-level effect on the variable of interest; and ICC(2) provides an estimate of the reliability of the team-level mean (Bliese, 2000). Analyses of variance tests using the team as an independent factor, demonstrated that team members' ratings of trust in teammates and trust in the coach all differed significantly (p < .001) across teams.  Finally, we used the PROCESS macro for SPSS v.23 developed by Hayes, which assesses moderated mediation effects (Hayes, 2013). This macro estimates the coefficients of the models using ordinary least squared regression-based path analytical framework and employs the recommended bootstrapping methods (Mackinnon, Lockwood, & Williams, 2004) to probe the significance of the conditional indirect effect. A 10,000 bootstrap resample and a biascorrected 95% confidence interval (CI) at each level of the moderator was used for the analyses (Preacher, Rucker, & Hayes, 2007).

The Mediating Role of Trust in Team
The first hypothesis predicted that trust in the leader would influence team performance through trust in the team (simple mediation). Trust in the leader (Table 2) was found to be influenced by trust in the team (path a = 0.31), which in turn affected team performance (path b = 9.50). Based on the bootstrap analysis, a bias-corrected bootstrap CI for the indirect effect (ab = 2.90) was above zero [0.74, 6.11]. No evidence of a direct relationship between trust in leader and team performance was detected. Therefore, trust in team fully mediated trust in the leader-team performance relationship.

The Moderating Role of Trust Consensus
The second hypothesis stated that the consensus of team members on trust in the leader moderates the positive relationship between levels of trust in the leader and trust in the team (first path of the mediation). The interaction between the level and consensus of trust in the leader was significant in predicting trust in the team, corroborating the importance of consensus among members (Table 3). More specifically, this interaction between trust in the leader level and consensus was different from zero only in teams with high consensus about trust in the team; their CIs reported for conditional indirect effects did not contain a zero, suggesting that future team performance can be interpreted as a function of the interplay between the level of trust in the leader and team members' agreement about it, combined with trust in team among its members.
Of note is the growing number of researchers arguing that CI estimation is more useful and informative than the model p value, as it provides the estimation of how large the effect is rather than whether an effect exists or not (Cumming, 2011). Moreover, we ran an additional analysis (e.g., Cole et al., 2011;Dineen et al., 2007), where a more stringent approach (such as adding a square term into the regression) demonstrated no spurious relationship. This suggests that consensus was not dependent on the level of trust.
The third hypothesis predicted the moderating effect of team trust consensus on the team trust-team performance relationship. The trust in the team interaction was different from zero in few cases; the conditional indirect effect of trust was significant only in teams with high consensus among members (minus 1 standard deviation). The bootstrapping procedure further supported the finding that, when dispersion regarding trust in the team is low, trust in the team mediates the relationship between the antecedent trust in the leader and team performance.

The Intervening Effect of Team Past Performance
Finally, Hypothesis 4 predicted that the effect of trust in the leader (level) would depend on the consensus among team members when past performance was poor. The conditional indirect

Table 3 Conditional Effects-Indirect Effects of Moderated Mediation Model of Trust Foci
Consensus a

Moderated mediation model of trust in leader (X) and trust in team (M) on team performance (Y) at values of the moderators: Past performance and consensus (trust in the leader and trust in team)
[ effect corroborates this prediction (Table 4). When past performance was low or moderated, results provided evidence of a significant indirect effect. Specifically, an indirect effect of trust in the leader on team performance via trust in team in the context occurred where the relationship was moderated by the team's past performance and consensus in trust in the coach. Furthermore, when considering past performance interactions and trust in team consensus, a significant indirect effect occurred only when dispersion among members was low to moderate and past performance was poor. Taken together, these findings indicated that a team's future performance is a function of trust in the coach, level, and consensus, particularly in the context of poor team past performance, as well as the level and consensus in trust in the team.

DISCUSSION
This study tested the combined effects of trust in different targets (leader and team) and of teams' trust configurations (level and consensus) in predicting team performance. Many studies explicitly recognize employee trust as a key variable that influences the organization's effectiveness and performance (e.g., De Jong et al., 2016;Dirks & Ferrin, 2002;Lewicki, McAllister, & Bies, 1998;McAllister, 1995;Tzafrir, 2005). Despite the apparent connection between trust in the leader and trust in the team, to the best of our knowledge, no previous study has attempted to test this dynamic relationship. We corroborate in our study that these types of trust are tightly intertwined. Not only does trust in the leader shape trust in team members but also the sense of trust that coaches are able to instill in their players in turn allows a team to become more effective.
Specifically, we found empirical evidence to support the theoretical framework, explaining how trust in the leader affects team performance via trust in the team. The present study reinforces and extends previous findings (e.g., Dirks, 2000) that coaches are key in sport teams because trust in their decisions builds team trust and thus affects team performance. It also contributes to practice and theory in several ways.
First, our study has extended the conceptual and empirical work on trust transferability (Ferrin et al., 2006;McEvily et al., 2003) by demonstrating the criticality of the leader as a third party in facilitating trust transferability and, thus, in establishing and maintaining trust in the team. We corroborate that trust in the coach affects team performance because, when this sort of trust exists, team members are more open to building mutual trust within the team and willing to work toward a common goal. Trusting the leader's good judgment and benevolence helps to develop trust among members and allows them to cease wasting energy trying to impress the leader with individual results; consequently, the team becomes more capable of achieving high performance (Mayer & Gavin, 2005;Nienaber, Romeike, Searle, & Schewe, 2015;Yu, Meng, & Xie, 2009).
Our results are consistent with Mayer and Gavin's (2005) findings that team performance suffers when team members do not trust their coach, as more time is spent monitoring the leader's actions, as well as documenting personal performance to protect oneself, rather than performing important team tasks. Trust in the leader determines the dynamics of the team-member exchange, changing the ratio of self-serving versus collective good outcomes, supporting the second critical dimension of team trust-the belief that other team members will not jeopardize the individual's well-being and efforts in achieving the collective goal. Thus, in line with previous predictions (Chen & Wu, 2014), we found support for the critical role of trust in the team in facilitating the team's effectiveness and performance.
Second, our research has reinforced the importance of measuring consensus when assessing trust across levels of analysis, answering the call for models incorporating not only the average level of trust but also the consensus among team members regarding their trust in team and their leader (Bell, 2007;De Jong & Dirks, 2012;Lau & Liden, 2008). In line with recent research highlighting the importance of testing consensus (e.g., Cole et al., 2011;Dineen et al., 2007;Kirkman & Shapiro, 2005), we found that this can significantly enhance our understanding of team trust.
Consistent with previous findings (e.g., De Jong & Dirks, 2012;Tomlinson & Meyer, 2009), our results also corroborate that average levels of trust and consensus among members have an interactive effect on joint outcomes. Similar to Sherony and Green's (2002) findings on the role of LMX in coworkers' dyadic relationships, we demonstrate that trust in teammates increases when consensus about trust in the coach exists among team members and decreases when there is no consensus among members.
Furthermore, this type of trust in the leader influences team performance through its relationship with trust in teammates but engenders better team performance only with high levels of agreement about trust in the team. This finding can be explained via enhanced trust transferability, which occurs in teams in which trust is more uniform. In fact, the team trust development process can be put at risk if there is no trust in the critical third party. Team members who do not trust their leader act as a broken link in the chain, preventing the snowball effect of trust in the team. Furthermore, high levels of dispersion in members' trust in the leader may reflect disharmony or favoritism and may suggest the absence of a shared reality and, hence, an increased likelihood of misunderstanding and the subsequent decrease in team trust. High disagreement regarding trust in a coach may result in negative outcomes and weaken the effect between the aggregated level of trust in the coach and team trust. Hence, we extend trust research by providing empirical evidence of direct consensus models in the context of specific past performance within a single framework.
Last, we note that team past performance was operationalized in our study as a moderator rather than a controlling variable, allowing for preliminary findings regarding the role of trust in the context of poor past performance. Teams possessing a high level and low dispersion of team trust were able to achieve better performance results even when their past performance was not high.
Our findings corroborate that this effect becomes particularly pronounced when there is team consensus about trust in the leader as it allows for more effective trust transferability. Not only does trust in the leader serve as a key factor in deciding if the team is still trustworthy, but it can act as a motivator and even a perceived guarantor for future success. When poor past performance promotes enhanced self-serving behaviors and endangers psychological safety (Wildman et al., 2012), socio-emotional bonds and trust in the leader become particularly relevant as they attenuate these negative outcomes. It is well known that past performance is generally the best predictor of future performance (Costa, 2003;Dirks, 2000), regardless of the transmitting mechanism or indirect effect. The finding that team trust can moderate the relationship and change the relationship pattern between past and future performance can potentially explain the cases of extreme and positive turnaround for some teams acquiring a highly charismatic coach. Indeed, initiated uniform trust in the leader and shared trust in the team appear to be able to break the negative pattern of team performance.

Limitations and Areas for Future Research
Like any study, ours is not without limitations. First, our study sample consisted exclusively of basketball teams, limiting the generalizability to other sports or organizational settings. Although tested in a sports setting, findings might also apply to other contexts that closely mirror some of its features, as they also have well-established performance measures, a competitive environment, or a similar decision-making process to enhance short-and longterm success (Jones, 2002, Weinberg & McDermott, 2002Wright, Smart, & McMahan, 1995). Second, due to the self-reporting nature of several of the measures adopted, we cannot completely discard the possibility of common method bias. However, methodological remedies were taken to attenuate this problem, such as use of archival data for the criterion variable, collecting the predictor variables at different times in the season, and aggregating the data at the team level.
Third, we focused on one condition of trust transferability-the trustworthiness of the third party-while controlling for another-the leader's trust in others. However, the leader's trust and the followers' perceptions of the latter may also affect how trust develops in the team. Consequently, this needs to be studied in future research.
Finally, the role of team past performance in team processes and outcomes needs to be further investigated. Although we focused on the contextual role of past performance, it may also act as a variable that drives trust in the team, affecting the team's present and future performance. Future longitudinal research should further explore this interesting theoretical and practical question.

Practical Implications
Our findings have several important practical implications. First, if we are to raise coaches' awareness to achieve high-team performance, it is essential to focus on developing trust in the leader (Arnold, Fletcher, & Molyneux, 2012). Our findings provide further support for this notion that it serves as a prerequisite for trust in the team. At the organizational level, this can be achieved by emphasizing the coach's trustworthiness and, more specifically, his or her ability, integrity, and benevolence (Mayer, Davis & Schoorman, 1995). In highly competitive sports contexts where performance criteria are clear (Weinberg & McDermott, 2002), team members' perception about the coach's high competence has to be deliberately developed and continuously supported by the organization (Knoll & Gill, 2011). The importance of team members' perception of the coach's benevolence and compassion cannot be overrated and, thus, has to be worked on and continuously emphasized by both the leader and the organization. The players need to believe that their coach is always "on their side" and that he or she is honest and transparent in the decisions that affect the team. All three factors-ability, benevolence, and integrity-can also contribute to generate trust in the team (Schoorman, Mayer, & Davis, 2007).
We also found that both the absolute and relative levels of trust within the team matter. Based on the findings that high consensus regarding trust in the leader enhances team trust and team performance, we suggest that coaches need to ensure, as best as possible, that every team member, as opposed to just several players, trusts them. In other words, it is critical to have no distrusting "outliers." This can be achieved by early interventions performed from the coach, with a highly developed ability to recognize players' emotions and address any issues of procedural justice (Druskat & Wolff, 2001).

CONCLUSION
Our findings on the moderated mediation relationships advance our understanding of the team trust-performance dynamic by explaining how this relationship evolves across levels of consensus and across different foci of trust. Results of this study provide a better understanding of trust among group members and may open new avenues for the further development of sports team research and theory, as well as practical managerial techniques. To validate the trust scales into Spanish the standard procedures were followed: back-translation process, the internal reliability, and confirmatory factor analyses (Mach et al., 2010). Furthermore, trust measures were also validated through several focus groups, interviews, and pretests.