臺大管理論叢第31卷第3期

61 NTU Management Review Vol. 31 No. 3 Dec. 2021 Service performance. We measure employees’ service performance by using a 7-item scale, which have originally been developed by Borucki and Burke (1999) and adapted by Liao and Chuang (2004). Sample items are “[He/she is] friendly and helpful to customers” and “[He/she] asks good questions and listens to find out what a customer wants.” The scale’s alpha reliability is .95 in this study. Altruistic behavior toward colleagues. We measure employees’ altruistic behavior toward colleagues by using a 4-item scale developed by Farh, Earley, and Lin (1997). Sample items are “[He/she is] willing to help colleagues solve work-related problems,” and “[He/she is] willing to cover work assignments for colleagues when needed.” The scale’s alpha reliability is .95 in this study. Control variables. We control for three demographic variables—employees’ sex, age, and organizational tenure—when examining the hypotheses, because these demographic variables have been found to be related to service performance and altruistic behaviors (e.g., Chattopadhyay, 1999; Raub and Liao, 2012; Van Dyne and LePine, 1998). Furthermore, we control for leaders’ PIFTs. Although leaders’ PIFTs (i.e., prototypes) and NIFTs (i.e., antiprototypes) are assumed to be relatively orthogonal (Sy, 2010; Whiteley et al., 2012), we control for leaders’ PIFTs in order to more accurately estimate the unique effects of leaders’ NIFTs. Leaders’ PIFTs are measured with the nine positive attributes from the Leaders’ Implicit Followership Theories (LIFTs) scale developed by Sy (2010). The scale has three positive dimensions (industry, enthusiasm, and good citizenship), each of which consists of three items. The alpha reliability for this 9-item measure is .95 in this study. 3.3 Data Analyses We employ Mplus Version 8.2 (Muthén and Muthén, 2017) to conduct confirmatory factor analyses (CFAs) and test the hypothesized relationships. In all of the CFAs, we use the “Type = General” approach with the estimation option of ML (i.e., maximum likelihood). In our tests of the main hypotheses, we adopt the “Type = Complex” cluster approach and the estimation option of MLR (i.e., maximum likelihood estimation with robust standard errors).3 Although all of our main research variables are individual level, 3 Our model does not contain level-2 constructs, and none of the explained variables have significant level-2 variances, so we adopt a “Type = Complex” instead of a “Type = Twolevel” analytical approach to address the non-independence problem of the data (Muthén and Muthén, 2017).

RkJQdWJsaXNoZXIy ODg3MDU=