6.1. Descriptive Statistics ofEach Variable and the Relevant Coefficient
After this study excluded the questionnaire with over consistent filled in answers, obvious randomly filled in answers as well as the group agreement measure (rwgG)) is lower than .70, there were 81 questionnaires from managers and 243 from employees
The employee's sample statistic of population characteristic includes 54.3% for the male, 58.4% unmarried, the average age is 32 years old, 53.1% for the university colleges and universities education background, the team population are below 6 people (51.9%), the average period of service is 4.5 years, the time working together as colleagues with the manager rages from 1 year to 3 years (42.8%), the frequency of one year implementation of performance appraisal in team most1y are 2 times (including) below (88.9%), According to the correlation matrix in Table 1, that m吋 ority control variable assumes a remarkable relationship with satisfaction with the performance appraisal session and satisfaction with the performance appraisal system
The correlation coefficient of the main variable of individual-level L孔1X，
feedback-seeking behavior, team-level LMX, the interactive justice c1imate, the procedural justice climate and satisfaction with performance appraisal sessions and satisfaction with the performance appraisal systems are all being placed between .16 to .64, and the relevance all reaches the remarkable standard (p < .05), Hierarchical Linear Model analysis is used to confirm the hypotheses
Chiao Da Afanagωnent Review /lól. 30 No. 1, 2010
6.2.
The Result of This Study on The Performance Appraisal Satisfaction using The Hierarchical Linear Modelling Hierarchical Linear Modelling (HLM) analysis to confinn the hypotheses. First,the study performs null model analysis upon satisfaction with the performance appraisal session and performance appraisal system to understand whether the two variables are suitable to be carried out with the following analysis. Moreover, on the pa口 of confinnation on the mediating effect, it uses the random coefficient
144 Exploring A1u/tilevel Perspeclive of Leader-member Exchange relevanl
10 Performanc叫ppraisal Satiφclion
and the intercept forecast mix model are carried out for analysis. If the individual level intercept presents the remarkable variation (τ00 p < .05), then suitable to carry on the intercept model analysis, representing the intercept of variation at individual level can be remarkably explained by the team level variable (y01 p
< .05), nameIy presents the cross-leveI direct effect. Uses the mix modeI
confirmation of independent variable (individual-leveI LMX and team-leveI LMX) to have remarkable inf1uence upon the mediating variable (feedback-seeking behavior and team justice cIimate) (condition one); Next, independent variable (individual-level LMX and team-level LMX) as well as the mediating variable (feedback-seeking behavior and team justice cIimate) respectiveIy have remarkable inf1uence on he dependable variable (satisfaction with performance appraisal session and satisfaction with performance appraisal system) (condition two), Finally, after engaged the mediating variable (feedback-seeking behavior and team justice cIimate), independent variable (individual-level as well as team-level L扎直X) weakens the influence level of dependable variable (performance session satisfaction) (partly mediating effect) or presents not the obvious (completely mediating effect); If the intercept modeI assumes the cross-level direct effect (y01 p < .05) obviously and slope of individual level remarkable variation (τ11 P < .05), then it is suitable to ca口y out the slope prediction model analysis. Through the analysis of slope prediction modeI, the existence of disturbance outcome from the justice cIimate against the relationship between the individual-level LMX and satisfaction with performance session and performance appraisal system can be analyzed. If individual-level slope variation is explained by the team variable, then the cross-level disturbance effect is obviously assumed (y11 p < .05), representing the cross-level moderating effect.
6.2.1. Null Model Analysis
After the confirmation of null model, the group variance of performance interview and system of performance satisfaction (τ00) respectively is .14 (χ2=
175.96, df = 80,
P
< .001) and .14 (χ，2 = 144.19, d f= 駒，p< .001), representing the variation of satisfaction with the performance session and performance system under different team obviously is bigger than O. ICC(1) of satisfaction withChiao Da Alanagement Review 跆1. 30No. 1, 2010 145
perfonnance session and performance system respectively are .28 and .21, indicating that satisfaction with the performance session and performance system has 28% and 21%. This variation is possibly created by the team leve1. Hence, it is suitable for HLM analysis.
6.2.2. The Mixed Model Analysis of Random Coefficient and The Intercept Forecast Pattern
In the mixed pattem Model1, individual-level LMX has a remarkable impact upon the feedback-seeking behavior (ylO = .30, P < .001), Therefore hypothesis H3 was established, and team-level LMX does not present a remarkable impact upon the feedback-seeking behavior (yl0 = -.22, P > .05), showing that the feedback-seeking behaviour will be affected by the individual-level LMX. Therefore this conforms to condition one within the confinnation of mediating effect as propounded by Baron and Kenny (1986) since the variable affects the mediating variable. When the study confinns the impact of the individual level and team level upon the perfonnance appraisal satisfaction, individual-level LMX has the remarkable variation upon the gap between satisfaction with the performance session and with the perfonnance system(τOOp
< .05), This tallies with 伽 condition by Hofmann (1997) in confinning the
cross-level direct effect. When Model2-l was put to the perfonnance session satisfaction, individual-level LMX will obviously affect the performance session satisfaction (yl0 = .的，p < .001), therefore hypothesis H 1-1 was established. The team-level LMX has no remarkable impact upon the performance session satisfaction (yOl = -.07, P > .05), therefore hypothesis H2-1 wiII be untenable When Model2-2 was put to the system of performance satisfaction, individual-level LMX will obviously affect the performance system satisfaction (yl0 = .60,p < .001), hypothesis Hl-2 was established, but team level L1vαhas
no remarkable impact upon the system of performance satisfaction (yOl = -.11, P
> .05), therefore the hypothesis H2-2 wiII be untenable. Therefore individual-level
LMX has positive influence upon the performance appraisal satisfaction, hence, the hypothesis H1 was established, conforming to the condition two ofBaron and Kenny(1986) which confirms the mediating effect. The independent variable can
146
E中loringλ.fultilevelPerspective of Leader-member Exchange relevant
ω Performance Appraisal Satisfaction
Model1 Model2-1 Model3-1 Model2-2 R宜。de13-2
2.91*** 3.24*** 2.98*** 3.00*** 2.54***
affect the dependent variable, meeting the condition two. Team-level LMX has no remarkable influence upon the performance appraisal satisfaction, therefore
Chiao Da A1anagement Review I台1. 30 No. 1, 2010 147
hypothesis H2 is untenable and has not been able to confoffi1 to condition two of Baron and Kenny(1986) which confirms of the mediating effect. In ModeI2-1, individual-level LMX will affect satisfaction with the perfoffi1ance session ofwith the perfoffi1ance system. Therefore, the mediating variables of feedback-seeking behavior, the interactive justice c1imate and the procedural justice c1imate are put separately in Model3-1 and Model3-2. In 弘10del3 -1， after the feedback-seeking behavior being put, individual-Ievel L弘1x still assumed obvious relationships with the perfoffi1ance session satisfaction (p < .001), but Beta coefficient dropped from .65 to .61, and feedback-seeking behavior was remarkable upon the
perforτnance session satisfaction (yl0 = .18, P < .01), the hypothesis H4-1 was established, confoffi1s to condition three of Baron and Kenny(1986)which confiffi1S the mediating effect, therefore feedback-seeking behavior assumes partial mediating effect to individual-Ievel L弘1x and relationships of satisfaction with the performance session, therefore H5-1 was established; Interactive justice c1imate was not remarkable upon the perfoffi1ance session satisfaction (yOl = .02,
p> .05), therefore hypothesis H7-1 was untenable, however the procedural justice
c1imate wiII obviously affect the perfoffi1ance session satisfaction (y02 = .18, P
< .05), therefore hypothesis H7-2 was established. In Model3-2, after putting the
feedback-seeking behavior, the relevance and Beta coefficient between the individual-Ievel LMX and perfoffi1ance session satisfaction were reduced, and feedback-seeking behavior obviously did not intluence performance session satisfaction (yl0 = .18, P > .05), therefore hypothesis H4-2 was untenable. The feedback-seeking behavior has the mediating effect on individual-Ievel LMX and relationships of the performance session satisfaction; therefore H5-2 was untenable. According to 出e above confiffi1ation of hypo伽肘， since team level
L扎1x does not affect perfoffi1ance session satisfaction and the perfoffi1ance system satisfaction (p > .05), hypothesis H8 was untenable. As for the pa此 of
cross-Ievel moderating effect upon the justice c1imate, in ModeI2-1, individual-level LMX has not had the emarkable variation to the slope of the perfoffi1ance session satisfaction (τ11 = .05, P > .05), therefore the relationships between the interactive and the procedural justice c1ima
148 Exploring Alultilevel Perspective ofLeader-member Exchange relevant
的 PerformanceAppraisal Satis.，戶ction
moderation effect, hypotheses H9-1 and H9-2 were untenable. In Model2-2, individual-level LMX has not had the remarkable variation to slope of the performance system satisfaction (τ11 = .06, P > .05)， 出erefore the relationships between the interactive and the procedural justice climate to individual-level LMX and the performance session satisfaction has not had the cross-level moderating effect, therefore hypotheses H9-3 and H9-4 were untenable