Methodology

Participants and Procedure

This study was performed in a large service company located in Ji Lin city, northeast China. This company has about 1000 employees and provides skiing services to tourists. The CEO and human resource department distributed research announcements to employees. The participants in this study comprised 300 full-time working adults who were selected at random from a list of employees in the company. The service industry presents some unique interactions between employees and customers that make it particularly suitable for our research purpose. Previous research has demonstrated that such a workplace environment is ideal for examining third-party oriented behaviors and UPB. Thus, we selected this sample to test our hypotheses by focusing on a single industry in the China context. In order to avoid common method bias, this study adopted a multi-period data collection method where participants were asked to answer the questionnaires throughout the whole of February in 2019. Previous research suggests that time separation between predictor and criterion variables can reduce standard method variance bias by decreasing consistency motifs and demand characteristics. Thus, we used three separate surveys with intervals of two weeks between each survey. A time lag of two weeks is consistent with published work examining the influence of leadership on UPB.

We collected data from eight departments and relied on convenience and snowball sampling. In the first survey (T1), with the help of the CEO and human resource manager (HR), we asked followers to fill in the questionnaire about gender, age, education, work tenure, and charismatic leadership. Two weeks later, at Time 2, participants rated psychological safety. Two weeks after this, at Time 3, respondents provided ratings on UPB. There were 276 respondents in the Time 1 survey (92% response rate), 232 participants in the Time 2 survey (84.05% reaction rate), and 226 respondents in the Time 3 survey (97.41% reaction rate). Among them, nine did not provide a complete response in the whole study. Furthermore, the respondents who incorrectly responded to the items and those from the three waves that could not be matched were removed from the analysis, with 214 effective participants remaining in the final sample. Among these final respondents, 62.62% were men, and 46.26% held a bachelor's degree. Most participants were aged 25 to 30 years (43.46%), with 37.85% having more than three years of work experience. The participants' departments were, respectively, customer reception (12.56%), ski marketing (25.61%), amusement park marketing (17.12%), security (1.51%), after-sale services (4.33%), membership card marketing (27.92%), transport services (5.08%), and ski training (5.87%). The sample demographic frequencies are shown in Table 1.

Table 1. Sample Demographic Frequencies.

Demographic Frequency Percent
Gender Male 134 62.62
Female 80 37.38
Age <25 years 92 42.99
26–35 years 93 43.46
36–45 years 18 8.41
46–55 years 11 5.14
Education high school 23 10.75
junior college 86 40.19
bachelor 99 46.26
master 3 1.40
Tenure <1 year 70 32.71
1–2 years 28 13.08
2–3 years 35 16.36
3–4 years 26 12.15
>5 years 55 25.70


Measurements

Charismatic leadership: Charismatic leadership was assessed by applying a 25-item scale proposed by Conger and Kanungo on a five-point Likert scale (1 = "Strongly disagree" to 5 = "Strongly agree"). The 25 items were divided equally into five subscales that measure the strategic vision and articulation, personal risk, sensitivity to the environment, sensitivity to member needs, and unconventional behavior. The sample items were "appears to be a skillful performer when presenting to a group" and "recognizes the abilities and skills of other members in the organization". Because this study focusses on the overall effect of charismatic leadership rather than the different effects of subdimensions, we summed the 25 items to arrive at an overall index of charismatic leadership following previous studies. The Cronbach's alpha was 0.90, and the KMO & Bartlett's test was 0.87 (p < 0.01).

Psychological safety: Psychological safety was measured using three items from Liang and Farh's scale. Items were rated on a five-point Likert scale (1 = "Strongly disagree" to 5 = "Strongly agree"). We selected these three items because they are typical cognitions in the service industry and have more representatives with a higher loading score (more than 0.60) than the other two items on the original scale. We also invited Ph.D. students to select items that suited the organizational context and assess the final scale to guarantee content validity. The pre-test with the 73 samples showed good reliability and construct validity (the loadings of the items ranged from 0.62 to 0.84, and the Cronbach's α was 0.71). The sample items were "I can express my true feelings regarding my job in my organization" and "I can freely express my thoughts in my organization". The Cronbach's α was 0.79, and the KMO & Bartlett's test was 0.70 (p < 0.01).  Unethical pro-organizational behavior: The respondents were required to assess UPB by applying Umphress et al.'s six-item scale. Items were rated on a five-point Likert scale (1 = "Strongly disagree" to 5 = "Strongly agree"). A sample item was "I exaggerated the truth about my company's products or services to customers and clients". The Cronbach's alpha was 0.74, and the KMO & Bartlett's test was 0.74 (p < 0.01).

Performance pressure: Performance pressure was examined with a three-item scale proposed by Rubin, Dierdorff, and Brown. Items were rated on a five-point Likert scale (1 = "Strongly disagree" to 5 = "Strongly agree"). Sample items were "There is a great deal of pressure to perform here". The Cronbach's α was 0.74, and the KMO & Bartlett's test was 0.70 (p < 0.01).

Control variables: Based on the recommendations of previous studies, demographic characteristics may exert influence on participants' propensity to engage in UPB. We used age, gender, education, and work tenure to control for their potentially spurious effects, as the findings of Kish-Gephart et al. revealed a week correlation between gender and age and unethical actions. Thus, information on followers' gender, age, education, and work tenure was controlled.


Data Analysis

In this study, we measured all constructs at the individual level and tested the hypothesized measurement model using Mplus 7.4. Next, the regression-based approach was obtained for testing each hypothesis in the theoretical model. Following Preacher et al., we examined the indirect effects and moderation hypotheses using a bootstrapping procedure with 5000 replications to derive CIs in SPSS 22.0 PROCESS.