Uncertainty management, transformational leadership, and job performance in an AI-powered organizational context

ABSTRACT

This study coupled the theory of uncertainty management (TUM) with the notion of transformational leadership (TFL) to examine how the uncertainty over the adoption of artificial intelligence (AI) technologies affects employees. SEM analyses with two-wave data collected in Japan (N = 1318 employee–supervisor dyads) revealed that uncertainty is negatively associated and TFL is positively associated with employees’ job performance. In addition, consistent with TUM, the digital literacy of leaders was found to moderate the effects of TFL such that the positive association between TFL and job performance disappeared when employees simultaneously feel high uncertainty and find supervisors low on digital literacy. These findings are discussed with reference to the relevant literature.

KEYWORDS:

• Artificial intelligence (AI)
• digital literacy of leaders
• job performance
• theory of uncertainty management
• transformational leadership (TFL)

Acknowledgements

This research was partly funded by the faculty research grant of Kyushu University School of Economics and JSPS KAKENHI Grant-in-Aid for Scientific Research(C) (Grant Number: 21K01630). I thank the Editor, Dr. Paul Schrodt, and three anonymous reviewers for their insightful suggestions.

Notes

1 The service provider, Macromill, had the top share in the market research industry in Japan as of the current study’s data collection (Japan Marketing Research Association, 2019).

2 In the screening process, respondent candidates were asked about their occupation, and only those who were identified as full-time white-collar employees were included in this study.

3 Several measures were undertaken to ensure the data fidelity. First, attention checks were included at random places within all surveys. For example, one question specified that a particular response should be selected (“This is an attention check; please choose ‘strongly agree’ for this item.”) and the data provided by those who failed to follow the instruction were removed. Second, the cases where the time spent to complete the survey was either too short or too long (i.e., ±3SDs) were removed. Third, the IP address of every response was inspected. There was one case where the response to both for-supervisor and for-employee surveys had been sent from the same IP address; this case was removed. Fourth, at the end of the Wave-1 for-supervisor survey was included an invitation to a lottery to win 5000 JPY (approximately $50). To be included in the lottery, supervisors were asked to enter their name and e-mail address. Those e-mail addresses were inspected by the personnel of the research service used in this study; none matched with the respondents’ e-mail address registered to the service. Finally, from among those supervisors who entered corporate address for the lottery, about 10% were randomly selected and contacted via e-mail to confirm their identity.

4 All respondents were Japanese citizens. Most of them were employees of Japanese companies, whereas a fraction (approx. 10%) of them were working at Japan offices of international/multinational companies.

5 Initially, 1500 respondents were identified after the screening process, to whom an invitation for participation was sent. From this initial pool, 17 were removed by the data-fidelity checks detailed in Note 3. Another 92 dropped at Wave-1 as the response to the for-supervisor and/or for-employee surveys was not returned. Additional 73 dropped at Wave-2, yielding the final sample of 1318 member-supervisor dyads who completed both surveys at Waves 1 and 2.

6 This study represents part of a larger research project; respondents and supervisors provided data not used in this study at both waves (the questions tapping the focal constructs examined in this study represented about one-third of the entire questionnaire at each wave).

7 McDonald’s ωs were estimated based on CFA factor loadings computed with Mplus 8.5. Computations were run based on Hayes and Coutts’s (2020) procedure. The confidence intervals were estimated using the bootstrapping process with 10,000 iterations.

8 The impact of the demographic variables measured for the current study (Table 1) was examined by running an alternative SEM, in which the demographic factors were entered as control variables. The results suggested that the statistical significance and direction of the effects among the focal variables did not change. For the brevity’s sake, the results of the analysis without demographic variables are presented in this article. The details of the parameter estimates computed for the alternative model are available from the author upon request.

9 Both data on employee-rated TFL and DLL or supervisor-rated job performance of employees included minimum values of the 7-point scale; also, the skewness, kurtosis, and standard deviation of those data are found equivalent to those of the self-report uncertainty data.

Additional information

Funding

This work was supported by Kyushu University School of Economics.