Home Business & Finance
Moderation and Mediation Effects of Distance on the Leader-Follower Relationship
Situations in which research looks at the effects of leader characteristics resulting in specific work-related outcomes are often moderated (Yukl, 2013). The second sequence of hypotheses is thus concerned with moderating and mediating effects of physical distance, relationship quality, and interaction frequency on the influence of leadership behavior on follower self-leadership and performance. The research framework is illustrated in Figure 9.
Figure 9. Moderating and Mediating Influences of Distance
In the first step, the correlation matrix outlining the interrelatedness of variables in the model is presented. The correlations between leadership behavior, selfleadership, and performance have already been discussed. Table 22 now displays interrelatedness of FRL, self-leadership, performance and distance dimensions. Transformational leadership and physical distance show a weak negative but highly significant correlation (r = -.17, p < .001). Physical distance further indicates weak but significant positive relation with passive leadership (r = .15, p < .01). Correlations specify that physical distance might have the potential to influence leadership behavior in more than one direction. Transformational leadership and LMX reveal strong positive correlation (r = .81, p < .001). The relation with transactional leadership is smaller, yet interpreted as high according to Cohen (1988) with r = .54 (p
< .001). Correlations with passive leadership are expectedly negative (r = -.63, p
< .001). The matrix furthermore reveals positive interrelations of LMX with follower self-leadership (r = .21, p < .001) and individual performance (r = .19, p
< .001). Probably the most noteworthy piece appearing in the matrix is the negative significant correlation between LMX and physical distance (r = -.22, p < .001). This outcome indicates that the degree of LMX varies with leader-follower physical distance. Interaction frequency as index supported neither relationships with predictors, nor outcome variables.
Even if the interaction frequency index did not reveal any significant correlations with any other variables, it could be interesting to look at the usage of the different media channels. Table 23 displays the relevant intercorrelations of face-to-face, e?mail, telephone conference, videoconference, and chat interaction. Frequency of face-to-face interaction indicates significant correlation with follower gender (r = - .14, p < .01). The result suggests that leaders tend to have face-to-face meetings with female followers more frequently. The negative weak but significant correlation with follower tenure (r = -.10, p < .05) indicates that followers with less tenure reporting to a leader tend to see their supervisors more frequently face-to-face than those who have been reporting to a supervisor for a longer period. Followers who meet regularly with their leaders face-to-face do tend to use videoconferencing (r = -.15, p < .01) and chat (r = -.10, p < .05) less often. Expectedly, the frequency of face-to-face meetings correlates strongly negative with physical distance (r = -.61, p < .001). This is not surprising as large leader-follower physical distance makes personal face-to-face encounters challenging. Face-to-face interaction furthermore does correlate negatively with passive leadership (r = -.15, p < .01). Additionally, face-to-face encounters do negatively relate to ratings of self-leadership (r = -.14, p
< .01). A central revelation is that the number of face-to-face meetings does not correlate significantly with performance. This could indicate that the importance of face-to-face meetings is overrated and in turn this bears high potential for virtual interaction instead. In addition, face-to-face meeting frequency does relate positively to LMX (r = .20, p < .001). The frequency of e-mail exchanges between leader and subordinates does not show any significant correlations except with the usage of other communication channels. Respondents writing e-mails also tend to use the telephone (r = .32, p < .001), videoconferences (r = .12, p < .05), and chat software (r = .23, p < .001) frequently. Individuals using telephone calls or telephone conferences for interaction with their leaders do also tend to use videoconferencing (r = 19, p < .001) and chat software (r = .22, p < .001). While videoconferencing is mostly used when geographical distance is involved, it is not surprising that the frequency of videoconference interaction correlates positively and statistically significant with physical distance (r = .26, p < .01). Usage of chat is similarly related to physical distance (r = .23, p < .01). Those followers using chat are also more likely to communicate through videoconferences with their leaders (r = .47, p
< .001). Chat interaction is further negatively correlated with transactional leadership (r = -.15, p < .01).
Table 22. Intercorrelations of FRL, Self-Leadership, Performance, and Distance
Note, n = 372. * p < .05, ** p < .01, *** p < .001, TF = transformational leadership, TK = transactional leadership, PL = passive leadership, SL = global self-leadership, Perf = performance, PhyD = physical distance, LMX = leader-member exchange, IntF = leader-follower interaction frequency
Table 23. Intercorrelations of Full Range Leadership and Media Channels
to-face interaction, E-mail = frequency of e-mail interaction, Tele = frequency of telephone interaction, VidCo = frequency of videoconference interaction, Chat = frequency of chat interaction, TF = transformational leadership, TK = transactional leadership, PL = passive leadership, SL = global selfleadership, Perf = performance, LMX = leader-member exchange. Significant correlations are marked in bold.
Physical distance negatively moderates the influence of leadership behavior on follower self-leadership strategies.
Moderation occurs when the relationship between two variables changes with the change of another independent variable (Field, 2013, pp. 398-407; Hair, Black, Babin & Anderson, 2010). The regression calculation includes running an analysis where original predictor, moderator, and interaction term of predictor and moderator are expected to predict the outcome variable. Prior to computation, predictors need to be centralized. Significance of the interaction term indicates whether moderation is present. Moderation effects in this work are tested applying physical distance as potential moderating variable on the relationship between leadership behavior and follower self-leadership. Testing for moderation involves assessing a directed relationship, thus one-tailed p-values are calculated (Field, 2013, pp. 6566).
To test the first moderation hypothesis, the centralized value of transformational leadership is multiplied by the centralized variable of physical distance. Transformational leadership, physical distance, and the product term are then regressed on follower self-leadership. The same procedure is repeated for transactional and passive leadership behavior.
Calculations reveal that physical distance shows moderating effects for the influence of transformational leadership on follower self-leadership and it might play a role in the passive leadership/performance relation. However, physical distance shows no significant moderation for the influence of transactional leadership on follower self-leadership.
The analysis in Table 24 displays transformational leadership to have significant positive influence on followers’ self-leadership strategies (P = .24, t = 4.70, p < .001). With the interaction term of transformational leadership and physical distance taken into account, physical distance appears as moderator (P = .09, t = 1.67, p < .05). Therefore, cases are split into distance categories. Category one comprises all cases with leader-follower physical distance equal to 0 km. For those individuals very close to each other, transformational leadership predicts follower selfleadership (P = .19, t = 2.87, p < .01).
Concerning the more distant leader-follower pairings, when leaders and followers were working 1 to 10 km away from each other (P = .13, t = .68, n.s.), 11 - 100 km from each other (P = .22, t = 0.61, n.s.), and 101 - 1,000 km from each other (P = .12, t = .56, n.s.), transformational leadership did not predict follower selfleadership. Notably, for the very distant group indicating leader-follower physical distance of 1,000 km or more, statistics do again reveal significance (P = .49, t = 5.15, p < .001). For followers far away from leaders, transformational leadership even predicts follower self-leadership more strongly than when they are close.
Table 24. Influence of FRL on Self-Leadership: Moderating Effects of Physical Distance
Note. Dependent variable: Follower self-leadership strategies
Despite the fact that there was only a limited signal of a direct relationship between passive leadership and self-leadership in this model, the effect is still influenced by physical distance, indicated by the significance of the interaction term. Examining this effect more closely, direction and extent of influence are investigated. The data set is therefore split into categories, analogue to the prior procedure. For respondents who were very close to their leaders (0 km), passive leadership had no effects on follower self-leadership strategies (P = -.03, t = -.41, n.s.).
Self-leadership strategies of the follower groups working 1-10 km (P = .16, t = .86, n.s.), 11-100 km (P = -.52, t = -1.60, n.s.), and 101-1,000 km (P = .17, t = .89, n.s.) away from their leaders still appeared not to be influenced by passive leadership behavior. For those followers who were very distant from leaders (more than 1000 km), passive leadership suddenly predicted follower self-leadership negatively (P = -.24, t = -2.26, p < .05). Perceptions of passive leadership behaviors might even increase in physically distant leader-follower relationships. In other words, the neg?ative effect of passive leadership was even stronger in a distance work setting. For the diversity in findings for moderation effects of physical distance on the influence of FRL behavior on follower self-leadership, hypothesis 2.1 is partially accepted.
Physical distance negatively moderates the influence of leadership behavior on follower performance.
Similarly to the previous approach, physical distance was tested as a potential moderator influencing the effect of perceived leadership behavior on follower performance. Although transformational leadership appears to have significant influence on follower performance in this model, the product term of transformational leadership and physical distance reveals no statistically significant results. The outcome indicates that physical distance does not affect the influence of transformational leadership behavior on individual performance. For transactional and passive leadership moderation analyses show no statistical differences of physical distance either. Table 25 outlines findings of the moderation analysis for hypothesis 2.2.
Table 25. Influence of FRL on Performance: Moderating Effects of Physical Distance
Note. Dependent variable: Follower performance
The previous model shows direct effects of perceptions of transformational, transactional, and passive leadership on individual follower performance. Whereas transformational leadership expresses weak positive effects (в = .16, t = 3.01, p < .01), transactional leadership does not indicate significant influence (в = .08, t =
1.49, n.s.), and passive leadership reveals negative effects of low effect size (P = - .17, t = -3.30, p < .001). This outcome led to more in-depth analysis of the effects of leadership behavior at certain physical distances as it was assumed that the influence (even if it was small) of transformational and passive leadership could vary with physical distance. Therefore, correlations were computed for the different physical distance stages. At very close condition, transformational leadership predicted follower performance (r = .17, p < .05). This was also true for the very distant group when transformational leadership predicted follower performance even more strongly (r = .33, p < .01). Transactional leadership did not predict performance at any physical distance. Passive leadership did project negative performance at very close (r = -.17, p < .05) and very distant level (r = -.29, p < .01).
Despite the different effects of transformational and passive leadership on follower performance at very close and very distant condition, the effects are either too small or occur in a different manner when the contingent measure of physical distance is used (and not the five distance categories) to reveal moderation. For this reason, it can be concluded that physical distance appears to be not as relevant in the leader-follower relationship as presumed and hypothesis 2.2 must be rejected.
Physical distance does show negative effects on the quality of relationship.
The following hypothesis is concerned with the direct influence of physical distance on followers’ perceptions of relationship quality. Linear regression was carried out to determine the effect size of physical distance on relationship quality. Results reveal that leader-follower physical distance is a moderate negative predictor of relationship quality. In other words, if physical distance between leader and followers increases, relationship quality most likely decreases (P = -.22, t = -4.31, p < .001). The relation between the two variables is reflected in Figure 10.
Figure 10. The Influence of Physical Distance on Relationship Quality
Regression computations show that leader-follower physical distance does entail negative influence on followers’ perceptions of leader-member exchange and hence hypothesis 2.3 is accepted.
Relationship quality mediates the influence of leadership behavior on follower performance.
Studying the literature on Full Range Leadership and relationship quality the assumption arises that LMX acts as mediating variable affecting relationships to work-related outcomes (e.g., Davis & Bryant, 2010). To test mediating influences of LMX on FRL behaviors and their effect sizes, three different models are calculated.
In contrast to moderation, mediation expresses the relationship between two variables due to the relationship to a third variable (Field, 2013, p. 408). Accordingly, mediation occurs if the strength of direct relationship is reduced by including a third variable. Mediation effects are investigated using the procedure suggested by Baron and Kenny (1986). It is a frequently used practice to conduct mediation in social science (Birasnav, 2014; Fritz & MacKinnon, 2007; Wang et al., 2010). Four steps are required to confirm mediation. As a prerequisite, (1) the predicting variable should be significantly related to the outcome variable, (2) and to the mediator. At the same time, (3) the mediator must predict the outcome variable significantly. Once the mediator is controlled for, (4) the relationship between the predictor and outcome variable should approach zero (Baron & Kenny, 1986). In order to test the significance of the mediation effect, researchers recommend the Sobel test (1982).
The three-step regression analysis to test mediation proposed by Baron and Kenny (1986) was carried out for each of the three higher-order factors of Full Range Leadership and unstandardized regression coefficients were added to the respective paths in the model (Field, 2013, p. 409). For explanation, a represents the direct relationship between predictor and mediator, b describes the direct effect of the mediator, c stands for the isolated relationship between predictor and outcome variable, and c’ indicates the direct effect of predictor on the outcome variable if the mediator is included in the model (Field, 2013, p. 408). The indirect effect of ab is simply the difference of the total and the direct effect c - c’ (Hayes, 2009, p. 409). In contrast to Baron and Kenny (1986), who suggest monitoring the reduction of effect sizes to determine mediation, Field (2013, p. 410) explains that mediation occurs if the direct relationship between the predictor and outcome variable is significant, yet approaches zero if a third variable is entered in the model.
Bootstrapping is recommended by Hayes (2009) and Williams and MacKinnon (2008) as it is not only more valid but also more powerful in testing intervening effects. Figure 11 visualizes the mediation for Model 1, assessing impending mediating effects of relationship quality.
Figure 11. Model 1: Transformational Leadership and Mediating Effects of Relationship Quality
Model 1 shows paths indicating the relationship between predictor, mediator, and outcome variable. The linear effect (a) of transformational leadership on relationship quality is highly significant (b = .81, t = 26.69, p < .001). The second path b directing from mediating to output variable also reveals a significant effect (b = .18, t = 2.06, p < .05). In addition, the third path c indicating the isolated relationship of predictor on outcome variable shows statistical significance (b = .16, t = 3.01, p < .01). Hence, all preconditions for mediation according to Baron and Kenny (1986) are fulfilled. Represented by path c’ the prospective mediator is now added to the model. In order for a mediation to occur, the effect of the predictor on the outcome variable should be reduced when the mediator is introduced in the model and at best, path c’ should become insignificant (Field, 2013). Visualized in Figure 11, path c’ is no longer statistically significant (b = .01, t = .10, n.s.). To test whether the outcome is statistically significant, the Sobel test is applied, which reveals whether the indirect effect of the predictor on the outcome variable via the mediating variable significantly differs from zero (Preacher & Hayes, 2004, p. 718). Outcomes of the Sobel test affirm the statistical significance of the mediation (z = 2.05, p < .05). The findings show that relationship quality fully mediates the relationship between transformational leadership and follower performance.
For the first mediation model, the indirect effect is computed with bab = .146, whereas the direct effect is specified with bc = .155. Interpretations according to
Urban and Mayerl (2006, p. 3) quantify the total effect at bab+c = .301. In other words, relationship quality accounts for 30.1% of the effect of transformational leadership on follower performance.
The second model is calculated using a similar approach. Three regressions were calculated and unstandardized coefficients were retrieved. Figure 12 shows the visualization of Model 2 with respective coefficients for paths a, b, c, and c’.
Figure 12. Model 2: Transactional Leadership and Mediating Effects of Relationship Quality
Path a shows a highly significant result of direct relation between transactional leadership and relationship quality (b = .54, t = 12.25, p < .001). A smaller but still highly significant effect is retrieved for path b (b = .21, t = 3.38, p < .001). The isolated relationship between transactional leadership and follower performance c reveals no statistical significance (b = .08, t = 1.49, n.s.). With c not being statistically significant a vital precondition for mediation is violated (Field, 2013, pp. 409410). Yet, Hayes (2009, pp. 413-415) outlines the heavy criticism on the causal steps approach by Baron and Kenny (1986), arguing that indirect effects should still be taken into consideration for analyses if a and b are significantly different from zero, even if c is not. The researchers attest that in this case, the term indirect effect should be used rather than mediation. For mediation effects of relationship quality on the influence of transactional leadership on follower performance, the Sobel test revealed statistically significance for prior model (z = 3.25, p < .01). This indicates the indirect effect to be significant with the product of ab: bab = .110, and the direct effect bc = .077. The total effect is thus calculated at bab+c = .008, which may be interpreted that relationship quality accounts for only 0.8% of the influence of transactional leadership on follower performance.
Aiming to test for potential mediating effects of relationship quality on the influence of passive leadership on follower performance, the third model was calculated and unstandardized coefficients were computed as illustrated in Model 3 in Figure 13.
The effect of passive leadership on relationship quality is indicated in path a with a highly negative statistically significant outcome (b = -.63, t = -15.68, p < .001). Path b aims at explaining the influence of relationship quality on follower performance which results in a weak but positive direct effect (b = .13 t = 2.03, p < .05). The initial direct relationship between passive leadership and follower performance c is confirmed by a highly significant negative result hence of rather low strength (b = -.17, t = -3.30, p < .001). Described as the best way to detect mediation (Field, 2013), adding the mediator to the model, the initial relation between passive leadership and follower performance c’ becomes insignificant (b = -.09, t = -1.29, n.s.).
Figure 13. Model 3: Passive Leadership and Mediating Effects of Relationship Quality
Testing for statistical significance of the mediation model, the Sobel test is significantly different from zero (z = -2.01, p < .05). Yet, mediation does occur. The indirect effect, taking relationship quality into account, reveals a value of bab = .0084; the direct effect is computed with bc = -.169. The total effect accounts for bab+c = -.161. As the result turns out to be negative, interpretation includes that relationship quality accounts for 16.1% hindering the negative influence of passive leadership on follower performance. A numeric summary of the mediation tests is provided in Table 26.
Reviewing prior mediation analyses and in response to hypothesis 2.4 it can be concluded that followers’ perceptions of relationship quality have the potential to mediate effects of leadership behavior on follower performance.
Table 26. Influence of FRL on Performance: Mediating Effects of Relationship Quality
Note. Dependent variable: Follower performance
Interaction frequency positively moderates the influence of transformational leadership and transactional leadership behavior on follower performance.
For the third distance dimension assessed in this work, it was hypothesized that interaction frequency has positive moderating effects on the influence of leadership behavior on follower performance.
To respond to the stated hypothesis, moderating effects are tested. For this reason, different models are calculated for transformational and transactional leadership. As passive leadership is identified as non-leadership and those leaders are known to interact with their workforce infrequently by nature, passive leadership is ignored at this stage of analysis. Table 27 presents a summary of the computations.
Table 27. Influence of FRL on Performance: Moderating Effects of Interaction Frequency
Note. Dependent variable: Follower performance
Findings of the analysis reveal moderation effects of interaction frequency on the influence of transformational leadership on follower performance. Transformational leadership reports significant but weak direct effects on follower performance (P = .16, t = 3.01, p < .01). Including the product term of transformational leadership and interaction frequency (P = .62, t = 3.06, p < .001) the moderator appears significant. To assess how the effect influences the relationship, the data file is split at a value of -.013, representing the standardized mean score. The first data set represents low interaction frequency. For this set of responses transformational leadership does not show significant effects on follower performance (P = .07, t = 1.01, n.s.). For the second data set, representing candidates interacting frequently with their leaders, transformational leadership suddenly reveals highly significant effects of low to medium strength (P = .23, t = 3.25, p < .001). Moderation of interaction frequency on the relationship between transformational leadership and follower performance is thus confirmed.
For effects of transactional leadership on follower performance, interaction frequency does not intervene. For this reason, the last hypothesis 2.5 is partially accepted.