APA PsycNET Login

12 downloads 3400 Views 746KB Size Report
Login if you have a Gold, Gold plus or Platinum package; individual journal subscriptions; or an APA username and password. User ID or e-mail: Password:.
Journal of Applied Psychology 1984. Vol 69, No 3, 428-436

Copyright 1984 by the American Psychological Association Inc

Moderating Effects of Initial Leader-Member Exchange Status on the Effects of a Leadership Intervention Terri A. Scandura and George B. Graen Department of Management, University of Cincinnati In afieldexperiment involving 83 computer-processing employees of a large service organization, a leadership intervention based on the Leader-Member Exchange (LMX) model was tested against a control condition The effects of this intervention were hypothesized to depend on the initial quality of the LMX relationship Thus, employees having initially low LMXs with their immediate supervisor were compared to cohorts having initially higher LMXs It was hypothesized that employees having initially low LMX would respond more positively (after adjusting for regression effects) to the leadership intervention than those having higher quality relationships. Analysis of interaction effects indicates that comparing the leadership intervention condition to the control condition, the initially low-LMX group showed significant gams in productivity, job satisfaction, and supervisor satisfaction compared to the initially high-LMX group The initially low-LMX group also perceived significantly higher gains m member availability and support from their supervisors than the initially high-LMX group. The initial quality of LMX appears to moderate the leadership intervention effect in the hypothesized direction. The implications of these results are discussed

The dynamics of the processes that produce social structures (Weick, 1969) between persons in hierarchical organizations have been termed role making (Graen, 1976). Research on role making in leader-member dyads has indicated a consistent pattern characterized by leader-member transactions. In exchange for positional resources from a leader, the member commits himself or herself to higher degrees of involvement in the unit's functioning (Dansereau, Graen, & Haga, 1975). This source of influence can involve highly valued outcomes for both the superior and the member. The superior can offer outcomes of increased job latitude, influence m decision making, open communications, support of the member's actions, and confidence in and consideration for the member. The member can reciprocate with greater availability and comFundmg for this research was supplied in part by Grant DACH 1978 G-0012 to George Graen from the Army Research Institute of the Behavioral and Social Sciences The authors wish to thank Joan Graen, Mike Graen, Mike Novak, Anson Seers, Pat Sommerkamp, and Kathie Verderber for their research assistance and Ralph Katerberg and Ann Welsh for reading an earlier version of this article Requests for reprints should be sent to Tern A Scandura, Department of Management, Mail Location #20, University of Cincinnati, Cincinnati, Ohio 45221

mitment to the success of the entire unit or suborganization. This view of the organizational unit suggests that the unit becomes differentiated, based on leader-member dyads, into dyads with lowleader-member exchange (LMX) relations and dyads with higher LMX relations. Research on this vertical dyad linkage model has indicated that this differentiation process occurs in a predictable manner over time (Dansereau et al., 1975) and that the quality of these leader-member relations is related to productivity and satisfaction (Graen, Novak, & Sommerkamp, 1982) and employee turnover (Graen, Liden, & Hoel, 1982). The present study investigates the effect of the initial LMX status m dyads on the effectiveness of a dyadic leadership intervention. The leadership training was designed to enable and encourage supervisors to improve the dyadic exchange relations with their members through the use of a normative and behavioral model of dyadic exchange. To the extent that these relationships were modified from the traditional to the dyadic, the internal structure of the work-unit relationships would be affected. Members having initially low LMX can be expected to benefit the most from the new opportunity brought about by the change in

428

429

MODERATING EFFECTS OF INITIAL LMX

the leadership "game" because they were not very successful under the "old rules" of LMX. Therefore, the initial quality of LMX is hypothesized to moderate the effects of the leadership intervention after controlling for regression effects in the following manner. Those initially low in LMX will show more positive effects of the treatment (greater differences between experimental and control groups) than those initially higher in LMX. In sum, the leadership intervention should produce more positive effects in the low-LMX group than in the higher LMX group to the extent that the rules of the leadership game are changed. In other words, the initial "in group" and "out group" may become resorted during the change from one normative model of leadership (traditional, leader-group) to another (dyadic, leader-member). Thus, the leadership intervention may be an occasion signalling that a new "ball game" is in effect. This new game may be seen as an opportunity to change behavior, relationships, and attitudes. Although studies of moderator effects are numerous in the literature, debate still surrounds the issue of moderator effects (Cherrington & England, 1980; O'Conner, Rudolf, & Peters, 1980; White, 1978; Zedeck, 1971). Our understanding of moderator effects is limited by the frequent use of research designs that do not measure these effects over time (Arnold, 1982). Most of the designs examining moderator effects are cross sectional and thus produce results that pertain only to the relationship between moderator variables and employee responses at a single point in time. The only study that we know that used a longitudinal design was Seers and Graen (in press), which produced rather different results between single and multiple points in time. The present study investigates the quality of leadermember relationships and work outcomes over a period of 26 weeks, using a longitudinal design. Before we can draw inferences about moderator effects in a study, we must deal with regression effects over time. Regression effects may be present using fallible measures and employing initially extreme groups. If this is the case, the high scorers on the LMX scale should decline on subsequent measurements of leader-member exchange, whereas the low

scorers should rise. Thus,fromTime 1 to Time 2 these two groups should regress toward the same population mean as a function of regression. To control for these effects, a repeated measures analysis of variance (ANOVA) design was used with a leadership intervention group and a matched control group. The control group was used to estimate the effects of regression toward the mean. Hence, the moderator effect was the difference in mean gams over time (time one to time two) between the experimental group and the control group (estimated regression effect). Formally stated, the hypotheses are as follows: Hypothesis 1. Members having initially low-quality LMX will show more positive effects on overall job satisfaction due to the leadership intervention than their initially higher quality cohorts. Hypothesis 2. Members having initially low-quality LMX will show more positive effects on supervision satisfaction due to the leadership intervention than their initially higher quality cohorts. Hypothesis 3. Members having initially low-quality LMX will experience more positive effects on productivity due to the leadership intervention than their initially higher quality cohorts. Method Site and Participants This study was part of a larger research effort that was performed within a large government installation in the Midwest One entire department participated in the study. Participation was voluntary; however, 98% of the employees in the department took part in the study All participants had the same titles and same job description. The same equipment (computer terminals) and procedures were employed by all participants to process cases assigned randomly from a common batch. Most participants were female high school graduates over 40 years of age and were permanent employees who worked full time.

Procedure The design of the study was a before-after field experiment Questionnaires were administered to participants in small groups and to their immediate supervisors at Week 1 (before) and at Week 26 (after) Productivity was monitored over the full 26 weeks on a weekly basis. Participants were assigned to one of two conditions by work units Of the total participants at Time 1,35 were assigned to the leadership intervention condition and 65 were assigned to the control condition. During the 6 months of

430

TERRI A SCANDURA AND GEORGE B GRAEN

the experiment, 11 participants left the department Those who left were distributed as follows- treatment, 7, and control, 4 Unusable questionnaires accounted for additional losses treatment, 2, and control, 4 The resulting number of subjects was 83; with the treatment condition having 26 subjects and the control condition having 57 subjects Analysis of those who left or had unusable data at Time 2 and those who had complete data showed no reliable differences

Treatment The managers of the units receiving the LMX treatment received six 2-hour sessions (over a 6-week period) Training was done in a seminar setting and included lecture (material presented orally by the trainer), discussion (material was discussed among managers and trainer to make it more applicable to the particular work groups), and role modeling (managers role played the parts of supervisor and member during simulated one-on-one sessions) Topics covered consisted of a normative LMX model and how to use it (2 hours), active listening skill training (2 hours), exchanging mutual expectations (2 hours), exchanging resources (2 hours), and practicing one-on-one sessions (4 hours) The objective of the training was to get each manager to analyze thoroughly and to be prepared to act on major positive and negative components of his or her relationship with each subordinate During the training sessions the general structure of the conversations as well as the specific questions and techniques to facilitate the conversation were devised by the managers with the help of the trainer (a) the manager was to spend time asking about and discussing each person's gripes, concerns, and job expectations about the member's job, the supervisor's job, and their working relationship, (b) using "active listening" skills learned in the training, the manager was to be particularly attentive and sensitive to what issues were raised and how they were formulated by each subordinate, (c) the manager was to refrain from imposing his or her frame of reference about the issues raised, and (d) the manager was to share some of his or her own job expectations about his or her own job, his or her member's job, and their working relationship The actual treatment following this training, was a series of one-on-one conversations between unit manager and member that were to last 20 to 30 minutes (most were reported to last 30 to 40 minutes) The goal of the treatment was to increase the level of reciprocal understanding and helpfulness within dyads regarding job issues and behaviors Nine weeks after this treatment was begun, a manipulation check was conducted and all participants reported that these conversations occurred (90% took place during the first three weeks) In the placebo control condition, the managers of these units, along with all the above managers, participated in three 2-hour sessions of general input on job enrichment, performance evaluation, decision making, and communication

Instrumentation Moderator variable The Leader-Member Exchange (LMX) scale (Graen & Cashman, 1975, Liden & Graen, 1980) was employed to measure the quality of exchange between supervisors and subordinates. The member form of the LMX contained the following seven items

1 Do you usually feel that you know where you stand do you usually know how satisfied your immediate supervisor is with what you d o ' Always know where I stand (4), Usually know where I stand (3), Seldom know where I stand (2), Never know where I stand (1) 2 How well do you feel that your immediate supervisor understands your problems and needs 9 Completely (4), Well enough (3), Some but not enough (2), Not at all (I) 3 How well do you feel that your immediate supervisor recognizes your potential1' Fully (4), As much as the next person (3), Some but not enough (2), Not at all (I) 4 Regardless of how much formal authority your immediate supervisor has built into his or her position, what are the chances that he or she would be personally inclined to use power to help you solve problems m your work? Certainly would (4), Probably would (3), Might or might not (2), No chance (I) 5 Again, regardless of the amount of formal authority your immediate supervisor has, to what extent can you count on him or her to "bail you out" at his or her expense when you really need i t ' Certainly would (4), Probably would (3), Might or might not (2), No chance (1) 6 I have enough confidence in my immediate supervisor that I would defend and justify his or her decisions if he or she were not present to do so Certainly would (4), Probably would (3), Maybe (2), Probably not (1) 7 How would you characterize your working relationship with your immediate supervisor1' Extremely effective (4), Better than average (3), About average (2); Less than average (I) The seven items are summed for each participant resulting in a possible range of scores from 7 to 28 (Cronbach's alpha was 86 at Time 1 and 84 at Time 2)

Dependent Variables Productivity Productivity was measured by collecting weekly output records of individual performance from computer files These data were converted into individual quantity and quality measures Quantity of production was the total number of cases completed by an individual in a week divided by the total number of hours worked on the cases (alpha = 91) Quality of production was the total number of errors detected by a quality review process m a given week divided by the total number of cases reviewed that week (alpha = .70) Quantity and quality of production were calculated for each of the 26 weeks For productivity, before was Weeks 1 through 14 and after was Weeks 15 through 26 It should be noted that current productivity information was not available to supervisors either before or during the study Member's report Employee job attitudes were assessed using the JDS, the ROI, and the Hoppock Job Satisfaction Blank The Job Diagnostic Survey (JDS; Hackman & Oldham, 1975) containsfivesubscales that measure satisfaction with the following work dimensions' pay, security, social aspects, supervisor, and growth opportunities (alpha = .94 and 91) The Role Orientation Index (ROI) is a revision of the Job Descriptive Index (Smith, 1967) adapted to more effectively measure attitudes of supervisors (alpha = 76 and .80) The ROI contains four subscales containing bipolar descriptive adjectives that assess satisfaction with work, supervisor, promotions, and pay A measure of overall job satisfaction was the short form of the Hoppock Job Satisfaction Blank (Hoppock, 1935), which contains four

431

MODERATING EFFECTS OF INITIAL LMX items, each measuring a different aspect of job satisfaction: (a) affect, (b) duration, (c) social comparison, and (d) behavioral intention (alpha = .79 and 81). Member perceptions of other aspects of the working relationship with their supervisors were assessed using measures of member availability (MA) and superior support (SS) The member availability scale consists of three items that measure the degree to which the member is available to accept additional responsibility from the supervisor (alpha = 65 and .73) The superior support scale assesses the member's perception of the amount of support of member's suggestions and actions received from the supervisor (alpha = 82 and 94) Supervisor's report Overall job performance was measured by the Employee Rating Scale (ERS) developed by Graen, Dansereau, and Minami (1972) Aspects of performance encompassed by the measure include dependability, alertness, planning, know-how and judgment, overall present performance, and expected future performance The ratings (on a scale of 1, satisfactory to 7, outstanding) are unit weighted and summed to produce an overall performance rating (alpha = 92 and .92) Perceptions of the degree of member availability (willingness of a member to accept additional responsibility) were also reported from the supervisor's point of view (alpha = .65 and 73).

Analysis The overall experimental design of this study was 2 X 3 X 2 (a) two treatment conditions crossed with (b) three levels of LMX and with (c) before and after repeated measures The two treatment conditions were leadership intervention and control Leader-member exchange was tnchotomized based on deviation scores from the unit means The use of deviation scores has been shown to be an effective predictor of the withm-unit effects of LMX (Graen et al., 1982) These LMX deviation scores were tnchotomized to create three groups having approximately equal numbers of subjects low (n = 21), medium (n = 26), and high (n = 36) levels of deviation LMX The two treatment conditions (experimental and control) were crossed with the tnchotomized LMX deviation measure to produce a 2 X 3 X 2 repeated-measures design Multivanate analysis of variance (ANOVA) using repeated measures was used to statistically analyze hypothesized relationships (Bock, 1963) The particular a priori contrast of interest was lowLMX experimental minus control versus high-LMX experimental minus control A "protected" procedure was employed to test this contrast Only when the overall interaction was significant was the a priori contrast interpreted (Carmer & Swanson, 1973) As stated above, the purpose of this study was to investigate the contribution of initial LMX status on the effects of a leadership intervention after adjusting for regression effects The initially low-LMX groups were compared to the initially higher LMX groups for the effects of the leadership intervention The before-after design of the study using treatment and control groups made it possible to estimate and adjust effects due to the leadership intervention for the effects due to regression toward the mean The changes in the control groups from Time 1 to Time 2 were used to estimate regression effects Although this procedure may not be totally adequate to completely control for regression effects, it is offered as an approxi-

mation to sorting out moderator effects from regression effects The initial quality of LMX is hypothesized to moderate the response to the leadership intervention Analysis of this interaction was employed to determine if the moderating effect of initial LMX predicts productivity and satisfaction Cronbach (1975) suggests that analysis of interaction effects be employed to determine the moderating effects of situational variables Interaction effects indicate whether one variable moderates the form of the relationship between the independent and dependent variables (Arnold, 1982) Thus, analysis of interaction effects should indicate if the effects of the treatment on outcome variables was

Table 1 Internal Consistency Reliability (Cronbach Alpha) and Stability Estimates (Test-Retest Correlation) on Dependent Variables

Measure Leadership Leader-Member Exchange (LMX) Member's report Overall satisfaction (HOPPOCK)

Supervisor satisfaction (JDS) Supervisor satisfaction (ROI) Member availability (MA) Superior support (SS) Supervisor's report Job performance (ERS) Member availability (MA) Computer report Productivity Quantity (QUAN) Quality (QUAL)

Alpha No of items Before After

Beforeafter correlation

7

.86

.84

.67

4

79

.SI

.73

3

.94

.91

.59

15

76

.80

.62

3

65

73

.50

3

.82

.94

.41

6

.92

.92

.85

3

.65

.73

.68

8

.91 .70

.59 .39

26 26

a

Note Cronbach alpha coefficient is the average of all possible split-half correlations within a measure and serves as an estimate of homogeneity or internal consistency reliability Test-retest correlation was over a 6-month period (26-week) interval HOPPOCK = the Hoppock Job Satisfaction Blank, JDS = Job Diagnostic Survey; ROI = Role Orientation Index, ERS - Employee Rating Scale * Productivity was over 26 weeks

432

TERRI A. SCANDURA AND GEORGE B. GRAEN

Table 2

Mean Gams and Interaction p Values for the LMX Deviation and Leadership Intervention Groups Initial LMX deviation

Low Variable Member's report Overall satisfaction (HOPPOCK) Supervisor satisfaction (JDS) Supervisor satisfaction (ROI) Member availability (MA) Superior support (SS) Supervisor's report Job performance (ERS) Member availability (MA) Computer report productivity Quantity (QUAN) Quality (QUAL) Multivanate

High

Treatment

Control

Treatment

Control

P

2.11 4.11

-.80

1.50 -3.25 -1.50 -3.50

-1.11 -144 -.14

.05 .05 .01 .01 .05

1.89 2.33

50 .42 .08 .92

2 67 3.00

.83

.38

.36

-125

1.50

1.00

120 -40

-.40 -.40

.40

-2.80

1.20

-40

444

-88

-61 .21

ns .05 .01 ns 01

Note The p values denote interaction of Treatment X LMX X Time. Multivariate analysis of variance of 2 X 3 was run over all 8 dependent variables All interactions in full design were significant (p £ .01) except for the employee rating scale LMX = Leader-Member Exchange; JDS = Job Diagnostic Survey, ROI = Role Orientation Index

moderated by the initial quality of LMX To determine if the responses of the initially low-LMX groups were significantly different than those of the initially high-LMX groups, a priori interaction contrasts (Treatment X LMX X Time) were performed

Results Member's Report Table 2 shows the mean gains and interaction p values by LMX deviation and leadership intervention groups. Member attitudes toward the overall job situation as measured by the short form of the Hoppock Job Satisfaction Blank showed gains of 2.1 and 1.5 for the low- and high-LMX groups, respectively. The design of the study made it possible to isolate the treatment effect by removing the effect due to regression toward the mean. This procedure will be illustrated through the following example. The mean gain for the initially low treatment group is 2.11. The mean gain for the initially low control group is —.80 and is an estimate of the regression effect. Subtracting the control group mean gain from the treatment mean gain yields 2.91; this is the treatment effect for the initially low-LMX group after the regression effect has been removed. Similarly, the mean gain for the initially high treatment group is 1.50. Subtracting

the mean gain for the initially high control group (—1.11) yields 2.61, which is the treatment effect for the initially high-LMX group after the regression effect has been removed. The a priori contrast between the mean gains for the treatment effect on the initially low group (2.91) versus the treatment effects on the initially high group (2.61) indicated a significant (p < .05) interaction of Treatment X LMX X Time. The mean gainforthe mediumLMX group was —1.3 for overall satisfaction. Satisfaction with supervision as measured using both the JDS and the ROI subscales showed more positive treatment effects for the initially low-LMX group. The JDS measure showed a significant interaction (p < .05), and the ROI measure also showed a significant interaction (p< .01) on supervision satisfaction. The mean gains were 4.1 and 4.4 for the lowLMX and -3.3 and -1.5 for the high-LMX group for the JDS and ROI, respectively. The mean gains for the medium LMX group were .3 for the JDS and .7 for the ROI. The member availability and superior support scales both measure outcome dimensions of the quality of LMX occurring in a dyad. Members in the initially low-LMX group showed significantly more positive treatment effects on member availability (p < -.05) and

MODERATING EFFECTS OF INITIAL LMX

433

superior support (p < .01) than members in Productivity results (quantity) for the lowthe initially high-LMX group. The mean gains and high-LMX deviation groups are presented were 1.9 and 2.3 for the low-LMX group and in Figures 1 and 2, respectively. Figure 1 shows —3.5 and —.9 for the high-LMX group for the time series for productivity over the 26 member availability and support, respectively. weeks of the study for the initially low-LMX The mean gains for the medium-LMX group deviation group for both treatment and control were —.4 for member availability and .3 for conditions. As the figure shows, productivity support. increased significantly from before to after for the low-LMX group relative to the control group following the intervention (Week 15). Supervisor's Report Though the experimental group showed higher Job performance was assessed using the performance than the control group during Employee Rating Scale (ERS). Results of the the before-treatment period (Weeks 1 through interaction contrast for the initially low- versus 14), this group showed greater gains during high-LMX groups show no significant differ- the after-treatment period (Weeks 15 through ence in job performance, although the mean 26) than the control condition. Similarly, Figgains were 2.7 and .4 for the low- and high- ure 2 shows the time series for the initially LMX groups, respectively. The mean gain for high-LMX deviation group for the treatment the medium-LMX group was — 1.3. Because and control conditions. The figure shows that the supervisors ciid not have access to current the experimental group performed at about productivity information during the time pe- the same level as the control condition during riod of this study, this time lag in productivity the before period and that during the after information may explain why significant ef- period the experimental group maintained its fects on employee performance as measured production level, whereas the control condiby the ERS were not observed. In contrast, tions declined in performance. member availability from the supervisor's Taken together, Figures 1 and 2 demonstrate point of view did demonstrate a significant the timing and duration of the interaction ef(p < 05) interaction. The mean gains on supervisor's member availability were 3.0 and fect for productivity. Significant differences 1.5 for the low- and high-LMX groups, re- between the low- and high-LMX groups relspectively The mean gain for the medium- ative to controls are shown from before to after. The hypothesized "reversal effect" beLMX group was 2.8. tween the low- and high-LMX groups is illustrated by the trends shown m these figures. Computer Report The initial LMX deviation interacts signifProductivity was measured by collecting icantly (p < .04) with the treatment on gains weekly output records from computer files. in the quality of the exchange (LMX). Gains Quality demonstrated no significance but was in LMX deviation for the treatment groups uniformly high for both the treatment and over controls were 3.81 for the initially low-, control groups over the 26 weeks. The average .25 for the initially medium-, and —1.16 for number of errors per case was .06, or six errors the initially high-LMX groups. These results per 100 cases. Thus, any gains in quantity demonstrate that the initially low-LMX group were not made at the expense of quality. Re- responded more positively to the leadership sults of the interaction contrast show a sig- intervention than the initially higher LMX nificant effect (p < .01) on quantity and a groups after estimated regression effects were nonsignificant effect on quality for initial LMX removed. Thus, effects on the independent status crossed with treatment and control and variable (LMX) occurred concurrently with from before to after. The gains in hard pro- effects observed on the dependent variables ductivity for the treatment groups were 1.20 (satisfaction, productivity, member availability, for the initially low and 40 for the initially and support). high group In contrast, the gains for the reDiscussion spective control groups were —.40 and —2.80. The mean gain for the medium LMX group The results of this investigation support the was -.80. hypothesis that the initially low-LMX groups

434

TERRI A. SCANDURA AND GEORGE B. GRAEN

respond more positively than the initially higher LMX group to the leadership intervention after controlling for regression effects. Graen et al. (1982) reported other positive effects of this leadership intervention but did not examine the moderating effects being reported in the present study. Clearly, the increased quality of exchange in the initially lowLMX group is concomitant with a significant amount of the increased productivity and satisfaction resulting from the intervention. Moreover, because control groups were used, the design of the study made it possible to separate the effects of this differential effectiveness from those of regression toward the mean. The leadership intervention also resulted in significant increases in the degree of superior support and member availability perceived by the initially low-LMX group compared to the high group. In agreement, significant increases

in member availability were also perceived by the supervisors of these employees. The results of this study indicate that as the quality of the leaders' exchange relationships with these employees improved, hard productivity, dyadic outcomes, and satisfaction improved as well. It should be noted that members having initially low LMXs were not necessarily the poor performers in the work units. The initially lowLMX group clearly had the potential to consistently produce at higher levels, but it appears that they did not perceive higher performance as being worth the effort. After the one-onone leadership intervention, the initially lowLMX group responded more positively to the new opportunities than did their colleagues. The leadership intervention apparently produced significant changes in the dyadic relationship structure of the work groups examined. In the leadership intervention group, those having initially poor-quality exchange

.5OH

Low LMX Deviation Control Experimental

WEEK

25

Figure 1 Weekly quantity of production for low-Leader-Member Exchange (LMX) treatment and lowLMX control for before and after leadership intervention.

MODERATING EFFECTS OF INITIAL LMX

relationships became higher performers over the 26 weeks of the study. Thus, the leadership intervention most benefited members having initially poor-quality exchanges. Increases in hard productivity and satisfaction were the initially low LMX group members' response to their supervisors' increased attention following the leadership intervention. The hypothesis that the initial quality of LMX moderates the effects of the leadership intervention is supported. Analysis of interaction effects showed that members having initially low LMX responded to the leadership intervention more positively than those having initially higher LMX. The effects of initial LMX as a moderator of a leadership intervention were demonstrated on hard productivity, satisfaction, member availability, and superior support. We are led by these results to agree with Cronbach (1975) that more research should be done to find the interaction

435

effects of moderating conditions in our investigations. Given that behavior in organizations is the interaction of Actor X Behavior X Context (McGrath, 1982), we might expect that the relationship between any two of these might be moderated by the third. In the present study, we found that context (leadership intervention) effects on behavior (dependent variable) was moderated by actor (exchange status). The effects of a leadership intervention can be pervasive, but it appears that employees having poor-quality exchanges initially are most affected by the intervention These members responded to the improved LMX through increased productivity and satisfaction. As a practical matter, the leadership intervention produced a 19% improvement m hard productivity. This improvement resulted m an estimated annual cost savings of over $5 million. Thus, the improved quality of the ex-

5