62 The Influences of Leaders’ Negative Implicit Followership Theories on Employees’Work Behaviors: A DualPathway Model research participants are nested in workgroups and each leader provides ratings for several focal employees (min = 1 and max = 5). By adopting the “Type = Complex” approach and the MLR estimator, we could address concerns about non-independence and test the significance of estimates with robust standard errors (Muthén and Satorra, 1995). For tests of mediating effects, statisticians have suggested that researchers apply a bootstrapping method to the estimation of the effects’ confidence intervals (CIs) (e.g., Shrout and Bolger, 2002). However, the conventional resampling bootstrapping method is inadequate for such an estimation if the data are clustered or non-independent in nature (Preacher and Selig, 2012). Therefore, we adopt the Monte Carlo simulation method recommended by Preacher and Selig (2012) to estimate the 95% CIs for the mediation effects. Specifically, our estimation involved R software (Version 4.0.2) with 20,000 simulations. 4. Results 4.1 Confirmatory Factor Analyses We perform a series of CFAs to verify the convergent and discriminant validity of our measures. First, we conduct CFAs for each of the main constructs to ensure whether all measurement items have significantly loaded onto their respective constructs or respective dimensions. Among the main constructs, leaders’ NIFTs, LMX, and psychological empowerment have multiple dimensions. Therefore, we conduct a second-order CFA for each of these constructs. The CFA results show that all items significantly have loaded onto their respective dimensions (the factor loadings ranged from .76 to .96, all p < .001), and all dimensions have significantly loaded onto their constructs (i.e., the general factors) (the factor loadings range from .63 to .91, all p < .001). As abusive supervision, negative mood, service performance, and altruistic behavior toward colleagues have only one dimension, we conduct a first-order CFA for each of these constructs. The CFA results show that all items significantly have load onto their respective constructs (the factor loadings range from .57 to .93, all p < .001). These results support the convergent validity of our measures. Second, we examine the hypothesized seven-factor model (i.e., Model A), which comprises leaders’ NIFTs, abusive supervision, LMX, negative mood, psychological empowerment, service performance, and altruistic behavior toward colleagues. Owing to the numerous items in our measures, we use an item-parceling strategy to maintain
RkJQdWJsaXNoZXIy ODg3MDU=