臺大管理論叢第31卷第3期

63 NTU Management Review Vol. 31 No. 3 Dec. 2021 an adequate indicator-to-factor ratio (Landis, Beal, and Tesluk, 2000).4 For the multipledimensional constructs, we form 2 parcels for leaders’ NIFTs, 4 parcels for LMX, and 4 parcels for psychological empowerment on the basis of their theoretical dimensions. As abusive supervision and negative mood have numerous items, we also form parcels by combining every two items into one parcel. If a scale’s item number is odd, the extra item is placed in the last parcel. Eventually, we form 7 parcels for abusive supervision and 5 parcels for negative mood. For service performance and altruistic behavior toward colleagues, we use their original items as measurement indicators in CFA models. As shown in Table 1, the seven-factor model fits the data well (χ2 (474) = 928.15, CFI = .93, TLI = .92, RMSEA = .07, SRMR = .06). We then compare the hypothesized seven-factor model (Model A) with a number of alternative parsimonious models. As shown in Table 1, the hypothesized seven-factor model fits significantly better than any of the alternative models that we have examined (all p < .001). For example, it fits better than several sixfactor models, including Model B in which leaders’ NIFTs and abusive supervision are combined into one factor, Model C in which leaders’ NIFTs and LMX are combined into one factor, Model D in which abusive supervision and negative mood are combined into one factor, and Model E in which service performance and altruistic behavior toward colleagues are combined into one factor. It also fits better than the three-factor model (Model H), in which four self-rating constructs are combined into one factor and two coworker-rating constructs are combined into another factor. Taken together, these CFA results demonstrate that our measures have good discriminant validity. 4.2 Hypothesis Testing Table 2 presents the descriptive statistics, reliabilities, and correlations among the research variables. Table 3 and Figure 1 present the formal results of hypothesis testing 4 We also use original items as indicators and their corresponding subdimensions to run a 7-factor CFA model, which include 3 second-order constructs. Although the overall model fit was not ideal (χ2 (1984) = 3711.56, CFI = .86, TLI = .86, RMSEA = .07, SRMR = .07), all items and subdimensions have a standardized factor loading higher than .50. Furthermore, the 95% confidence intervals of all construct correlations do not include the value of 1.00 or -1.00. These results support the convergent validity and discriminant of our measures. In this 7-factor model, the number of estimated parameters is 226, which is greater than our sample size (N = 210). In such a situation, the estimation of parameters and their standard errors may not be trustworthy. Therefore, in our CFA model comparisons, we adopt an itemparceling approach to reduce the number of indicators (Little, Cunningham, Shahar, and Widaman, 2002).

RkJQdWJsaXNoZXIy ODg3MDU=