Introduction

Innovation is broadly defined as ā€œthe intentional introduction and application of new ideas, processes, products or proceduresā€ (West, Farr, 1990, p. 9). It is a critical means by which organizations gain long-term competitive advantage (Lee et al., 2019). Leadership or leader behavior has long been recognized as a key driver of innovation (Afsar and Umrani, 2020; Rosing et al., 2011), such as transformational and transactional leadership (Bass, 1999), leader-member exchange (Gerstner and Day, 1997), and supervisor support (Tierney, Farmer, 2002).

Given that innovation is a complex process that can involve different activities (e.g., idea creation, promotion, and implementation), some researchers suggest that leaders should adopt different or even contradictory behaviors to meet the needs of different innovation situations (Rosing et al., 2011; Smith et al., 2012; Zhang et al., 2015). Rosing et al. (2011), for example, proposed an ambidextrous form of leadership, a combination of opening and closing leader behaviors. The basic principle of ambidextrous leadership is to either increase or decrease variation in employee behavior according to the needs of different innovation tasks (Rosing et al., 2011). Numerous studies have found a positive relationship between ambidextrous leadership and innovation performance (e.g., Deng et al., 2023; MascareƱo et al., 2021; Rosing and Zacher, 2017; Haider et al., 2023), even after controlling for the effect of transformational or transactional leadership (Yang et al., 2023; Zacher et al., 2016), employeesā€™ individual differences (Alghamdi, 2018), or team success (Zacher and Rosing, 2015).

Existing studies on the relationship between ambidextrous leadership and innovation have predominantly treated innovation as a unitary process and have not taken into account the heterogeneity of the innovation process. For example, these studies have failed to consider the distinct processes of idea generation and implementation (Alghamdi, 2018; Deng et al., 2023; Haider et al., 2023; Klonek et al., 2020; MascareƱo et al., 2021; Yang et al., 2023; Zacher and Rosing, 2015). This issue is of particular importance for ambidextrous leadership, as it suggests that innovation is a complex and dynamic process, and thus leaders should adaptively switch between different leader behaviors based on the ongoing innovation task or situation. Therefore, to accurately validate the effect of ambidextrous leadership, a process perspective is necessary. Furthermore, existing studies on ambidextrous leadership and innovation are predominantly confined to survey-based correlational research designs. This methodological limitation precludes the capacity to make causal interpretations and, consequently, hinders the validation of the unique impact of ambidextrous leadership, beyond the potential confounding factors.

This study uses a randomized controlled experiment to manipulate ambidextrous leadership behaviors in two simulated innovation tasks, aiming to test the causal effect of ambidextrous leadership on employee innovation performance in two different innovation processes (idea generation and idea implementation), relative to opening, closing, and transformational leadership styles. It contributes to research on ambidextrous leadership and innovation in two ways: a research perspective based on the heterogeneity of the innovation process and a research methodology based on simulated experiments.

Ambidextrous leadership and innovation

Existing theoretical models of innovation acknowledge that the innovation process encompasses two distinct categories of activities: exploration and exploitation (Amabile, 1988; Farr et al., 2003; West, 2002; Rosing et al., 2011). According to Rosing et al. (2011), ambidextrous leadership fosters exploration by adopting opening leadership behaviors to increase the variance in follower behavior and fosters exploitation by adopting closing leadership behaviors to decrease the variance in follower behavior. It emphasizes the temporal flexibility to switch between the two behaviors in order to meet the demands of different tasks in a dynamic innovation process. Recent empirical studies using correlational designs have shown that ambidextrous leadership, operationalized as the difference or multiplier between opening and closing leader behaviors, predicts innovation performance at both the team (Duc et al., 2020; Kung et al., 2020; Zacher and Rosing, 2015; Zuraik et al., 2020) and the individual level (Luo et al., 2018; Ma et al., 2019; Wang et al., 2021; Oluwafemi et al., 2020).

The ambidextrous leadership theory acknowledges the distinct processes of idea generation and implementation in innovation and further proposes that innovation is a non-linear and dynamic process that does not necessarily adhere to the traditional ā€œearly exploration-late exploitationā€ phase model (Rosing et al., 2011). In essence, exploration (and the concomitant opening leader behaviors) and exploitation (and the concomitant closing leader behaviors) may be indispensable in both the idea generation and implementation processes of innovation. Consequently, a contextual experiment employing simulated innovation tasks may offer a superior approach for studying the effect of ambidextrous leadership. This approach facilitates the identification of problems encountered by followers and the requisite leadership behavioral interventions in the real-time innovation process.

Innovation: idea generation and implementation

As previously discussed, innovation generally encompasses two distinct categories of activities: exploration and exploitation (Amabile, 1988; Farr et al., 2003; Rosing et al., 2011; West, 2002). Traditional views hold that exploration is linked to idea generation and entails ā€œthinking out of the box and experimentationā€, whereas exploitation is linked to idea implementation and requires ā€œefficiency, goal orientation, and routine executionā€ (Rosing et al., 2011, p. 965).

However, in actual innovation scenarios, idea generation may require not only exploration, but also exploitation, because ideas must not only be new but also useful (Rosing et al., 2011). While exploration generates new ideas, exploitation ensures the usefulness of new ideas by considering the fit between the new idea and actual conditions, which determines whether the idea is feasible or not. Exploitation is also needed to present new ideas in a more refined and structured way so that they are available for sharing and further consideration (Bain et al., 2001). The cognitive models of creativity offer a comprehensive explanation of individual behaviors in idea creation or generation, highlighting the parallel roles of exploitation and exploration. Wallasā€™s (1926) classic model, for example, identified four phases in the creative process, namely preparation, incubation, illumination, and verification. It appears that the preparation phase exhibits a certain degree of overlap with the idea exploitation phase, because it involves defining and refining the problem, gathering information surrounding the problem, and forming a body of knowledge through conceptualization and systematization. There are also commonalities between the verification phase and exploitation, as validation requires individuals to reflect on the elaboration of abstract new ideas and their usefulness throughout the creative process. The incubation and illumination phases are similar to exploration, as they involve the search for new solutions and a surge of creative awareness (Sadler-Smith, 2015).

Similarly, idea implementation may also require exploration. Since problems or difficulties not foreseen at the idea generation stage may arise during actual implementation, explorative efforts are needed to develop creative approaches or strategies to solve practical problems and facilitate the realization of the original idea (Rosing et al., 2011).

In summary, idea generation and implementation are not mutually exclusive, as innovation itself is a complex rather than linear or continuous process. Both idea generation and implementation may require exploration and exploitation. The transition between exploration and exploitation depends on the needs of the current innovation situation. In what follows, we discuss how ambidextrous leadership, particularly the flexible switching between different leader behaviors, may facilitate follower behavior and innovation performance in different innovation processes (idea generation and implementation).

Ambidextrous leadership, idea generation, and idea implementation

Leadership is a key driver of innovation owing to its role in guiding and supporting follower behavior. As previously discussed, ambidextrous follower behaviors (i.e., exploration and exploitation) are required at both the idea generation and implementation stages. Rosing et al. (2011) extended this ā€˜ambidextrousā€™ perspective to leadership behavior in innovation, with the aim of supporting follower ā€˜ambidexterityā€™. Central to the ambidextrous leadership theory is the temporal flexibility to switch between opening and closing leadership behaviors as the situation requires (Rosing et al., 2011).

Specifically, idea generation requires opening leader behaviors (e.g., encouraging experimentation and allowing errors) to increase follower variance and facilitate exploration. Meanwhile, it requires closing leader behaviors (e.g., setting guidelines and taking corrective action) to decrease follower variance and promote the exploitation of new ideas as needed to ensure their usefulness. Suppose an innovation team is asked to design a next-generation product. Since the product is completely unknown, the team members must take full initiative and come up with different ideas, which requires opening up leadership behaviors. However, some ideas, although novel, may not be realized due to existing constraints, so the leader needs to intervene in a timely manner to prevent such ideas from consuming too many resources and to focus the teamā€™s attention on ideas that are both novel and potentially feasible. Leaders, therefore, need to be highly sensitive to the creative process in order to flexibly switch between opening and closing behaviors to support followersā€™ ambidextrous behavior (i.e., exploration or exploitation) in idea generation. Our first hypothesis is thus proposed:

H1. Follower performance in idea generation will be higher when leaders show ambidextrous behavior.

Similarly, idea implementation requires closing leader behaviors to decrease follower variance and facilitate the exploitation of ideas. However, in some situations, opening leader behaviors are also required. For instance, when a team is implementing a creative idea, it may encounter unanticipated issues with the original concept. Consequently, a subsequent cycle of exploratory endeavors becomes imperative to identify novel solutions or more optimal concepts. This process necessitates the presence of opening leadership. Subsequent to this, the application of new solutions to the implementation of ideas will necessitate the reemergence of closing leadership. In short, closing and opening leader behaviors may alternate during idea implementation because even a seemingly optimal idea or a meticulously designed action plan does not ensure the absence of unanticipated practical challenges that may emerge during the implementation process. Thus, ambidextrous leadership can facilitate idea implementation. The second hypothesis is proposed:

H2. Follower performance in idea implementation will be higher when leaders show ambidextrous behavior.

The weighting of opening and closing leader behaviors

Above, we hypothesized that both opening and closing leader behaviors can be necessary in idea generation and implementation. However, given that ambidextrous leadership is based entirely on situational contingencies that arise during the innovation process (Rosing et al., 2011), it canā€™t assume the timing of using particular leadership behaviors before the emergence of the immediate situation. For this reason, some researchers argue that the concept of ambidextrous leadership is ā€˜vagueā€™ in timing and ā€˜outcome-basedā€™ (Klonek et al., 2020, p. 19). To be clear, it is only after achieving a high level of innovative performance that we know that appropriate leader behaviors must have been implemented at the right time (i.e., ambidextrous leadership). This temporal unpredictability clearly poses a challenge to the development of ambidextrous leadership.

In this section, we will present a hypothesis on the weighting of opening and closing leader behaviors, or the balance between these two behaviors in the two innovation stages (idea generation and implementation), in an attempt to provide some ex ante guidance on implementing ambidextrous leadership. While this is unlikely to fully resolve the timing issue, it will at least provide an indication of the compositional weighting of leadership behavioral interventions required for these two stages.

We suggest that idea generation or creativity is a comparatively endogenous process, driven primarily by auto-regulatory systems within creative individuals whose behavior is governed by inhibitory control. Inhibitory control is one of the core human executive functions. It refers to ā€œthe ability to inhibit or suppress salient thought processes or actions that are not relevant to the goal or task at handā€ (Carlson and Wang, 2007, p. 490). Researchers have also noted that inhibitory control encompasses not only the ability to inhibit dominant but irrelevant responses, but also the ability to initiate subdominant but adaptive responses, and even the ability to switch between activating and deactivating prepotent responses depending on environmental conditions (Carlson and Wang, 2007; Munakata et al., 2011). In an organizational setting, employeesā€™ inhibitory control over their attention, cognition, and action when generating ideas may arise from their understanding of existing knowledge, conditions, focused operations, or the organizational perspective. Employees use inhibitory control to regulate their creative behavior, for example, by inhibiting ideas that are irrelevant or inappropriate and promoting those that are more compatible with existing conditions or established plans. This process of self-regulation involves a considerable amount of exploitative thinking, with the clear aim of increasing the usefulness of the ideas generated.

Classical creativity models also suggest that a conscious creative process is largely driven by self-coherent mechanisms involving multiple steps, including information gathering, knowledge building, incubation, illumination, elaboration, and verification (Wallas, 1926; Ward et al., 1999). These activities demonstrate the role of exploitation and exploration working in parallel in ensuring that an idea generated is not only new but also useful. For example, when an individual generates an idea, he or she spontaneously tries to acquire information in the relevant field, even if there is little or no closing leader intervention. Having generated the initial idea, the individual realizes that it needs to be refined and systematized before it can be shared with the wider group. He or she is also aware of the need to exploit and validate it in order to achieve a good fit between the idea and organizational needs or conditions. In fact, these exploitative behaviors reflect individual adaptive and goal-directed behavioral regulation, driven by inhibitory control, and may not require much external stimulation (e.g., closing leader behavior). Numerous studies have summarized the different skills or resources required to maximize idea generation and implementation performance, suggesting that idea generation primarily requires intrinsic motivation, cognitive flexibility, and minimized external pressure, whereas idea promotion and implementation rely more heavily on extrinsic regulation and demands, planning, monitoring, and controlling (Byron and Khazanchi, 2012; Janssen, 2001; Mainemelis, 2010; Perry-Smith and Mannucci, 2017; Å kerlavaj et al., 2014). In sum, idea generation may not require as much external intervention (e.g., closing leader intervention) as internal regulation.

In contrast, idea implementation is a relatively exogenous process, driven mainly by external forces (Byron and Khazanchi, 2012; Janssen, 2001; Mainemelis, 2010; Perry-Smith and Mannucci, 2017; Å kerlavaj et al., 2014). In idea implementation, when a problem arises (the point at which re-exploration is required), followers are often unable to decide on their own whether to proceed with implementation without resolving it or re-explore the idea. This is because the re-exploration option often requires the overturning of previously established ideas or guidelines and involves a greater amount of work and a more unpredictable goal attainment, whereas proceeding with implementation without resolving the problem may not affect the goal attainment, but may also have negative consequences at a later stage. There appears to be no individual cognitive mechanism similar to inhibitory control that can help followers make subjective decisions on such risky issues. As noted above, the effect of inhibitory control on an individualā€™s creative behavior is to inhibit irrelevant responses and activate relevant ones, as a process of exploitation to ensure the usefulness of new ideas. However, when it comes to implementing ideas, both options (continuing or re-exploring) are relevant to the task at hand. Inhibitory control, therefore, does not regulate individual behavior in the same way as it does when ideas are being generated. This is where leader interventions are needed, for example, through opening behavior to initiate new explorations. Innovative tasks differ from general tasks in that there is a lack of drawing on previous experience, which often leads to problems in the implementation process. This requires leaders to be sensitive during implementation and to encourage openness whenever necessary. Therefore, closing leader behavior during idea generation may be given less weight than opening leader behavior during idea implementation. The third hypothesis is proposed:

H3. Closing leader behavior is given less weight in idea generation than opening leader behavior in idea implementation.

Methods

This study used two randomized controlled experiments to test the causal relation between ambidextrous leadership and innovation. We used a one-way between-subjects experimental design, with leadership (ambidextrous/opening/closing/transformational leadership) as the independent variable and innovation performance (assessed at different stages with different indicators) as the dependent variable.

Sample

A priori power analysis (Cohen, 1988) using G*Power (Faul et al., 2007) was conducted to determine the required sample size. On the one hand, we used a power of 0.80 and Ī±ā€‰=ā€‰0.05 with an estimated effect size of f2ā€‰=ā€‰0.25 for the one-way four-level analysis of variance. This resulted in a required sample size of 180 participants. On the other hand, previous research on ambidextrous leadership and innovation has reported effect sizes ranging from f2ā€‰=ā€‰0.052 (Alghamdi, 2018) to f2ā€‰=ā€‰0.140 (Zacher and Rosing, 2015), which requires a sample of 252 participants.

Therefore, we recruited 252 employees from different industries (63 per intervention condition) through the National Symposium on Research and Teaching of Innovation Methods in Colleges 2023 and the Cross-Strait Innovation Methods Education Forum 2023, as well as part-time MPA and MBA students. The recruited individuals all had a background in management studies or leadership practice, with appropriate theoretical understanding and practical experience of innovative activities and leadership styles, and were therefore able to understand and cooperate with the experimental tasks in the study. Two participants were excluded because they could not participate in both experiments. The final sample consisted of 250 participants (35.2% male and 64.8% female), mostly between the ages of 26ā€“35 (72.0%). 70% of the participants had a bachelorā€™s degree or higher, and 49.2% had 1ā€“5 years of work experience.

Study 1

Experimental task

We used a task from Girotra et al.ā€™s (2023) case study of an innovation program. The task explicitly captured participantsā€™ creativity or idea generation skills. The task was:

ā€œYou are a creative entrepreneur looking to generate new product ideas. The product will target college students in the United States. It should be a physical good, not a service or software. Iā€™d like a product that could be sold at a retail price of less than about USD 50.ā€ (Girotra et al., 2023).

The instructions for the task were also presented:

ā€œPlease think of 3 ideas and describe each in a paragraph of 40ā€“80 words. You have 30ā€‰min to complete this task. You will work with a supervisor. Every 10ā€‰min, you must report to your supervisor via WeChat in three areas: your innovation progress, your personal feelings and the problems you encountered. The supervisor will give you instructions accordingly. You must follow these instructions as closely as possible.ā€

Experimental interventions

We drew on Klonek et al.ā€™s (2020) study of leadership behavior interventions in the form of leaders sending behavioral instructions to employees, but unlike them, our interventions were based solely on participantsā€™ real-time reports. We developed a list of descriptions of the target leadership behaviors and used them as intervention instructions, based on conceptual accounts and examples provided in previous studies on ambidextrous leadership (Rosing et al., 2011), opening and closing leadership (Rosing et al., 2011; Zacher et al., 2016; Zacher and Rosing, 2015), and transformational leadership (Rosing et al., 2011; Zacher et al., 2016; Zacher and Rosing, 2015) (SI Appendix, Study 1 Experimental interventions).

Experimental procedures

Experimental procedures consisted primarily of participant check-in, grouping, task completion, and leader intervention (SI Appendix, Study 1 Experimental procedures).

Measurement of perceived and followed leadership behaviors

At the end of the experiment, participants rated the extent to which they perceived and followed all four leadership behaviors during the experiment. Scales from previous studies were used (SI Appendix, Study 1 Measurement of leadership behaviors).

Measurement of idea generation

Previous studies on idea generation or creativity (e.g., Hirt et al., 2008) have generally assessed idea generation along three dimensions: originality, fluency, and flexibility. Originality refers to the uniqueness of the idea, fluency refers to the number of ideas, and flexibility refers to the number of idea types. Since this experiment had a limit on the number of ideas (i.e., fluency was controlled), we used the other two dimensions (originality and flexibility) to assess idea generation. Based on previous studies (e.g., Hirt et al., 2008), originality was measured by one item (ā€˜This task outcome is very creativeā€™) and flexibility was measured by one item (ā€˜This task outcome is completely novel and does not at all rely on conventional solutionsā€™), which was adapted from the study by Rosing et al. (2017). Both items were scored on a 7-point Likert scale (1ā€‰=ā€‰strongly disagree, 7ā€‰=ā€‰strongly agree).

Idea generation performance was rated by four experts, including a creativity researcher, a professor of innovation methods, a director of corporate product design, and a research assistant with a background in psychology. We used a fully cross-rated design (Hallgren, 2012), whereby each of the four raters scored all 250 entries, resulting in an 8-item score (4ā€‰Ć—ā€‰2 items) for each task outcome. ICC (absolute, single measure)ā€‰=ā€‰0.62 based on single raterā€™s ratings indicated moderate reliability. ICC (absolute, mean measure)ā€‰=ā€‰0.75 based on all four ratersā€™ ratings indicated good reliability. Therefore, the mean values of the four ratersā€™ ratings were used as the idea generation score (Mā€‰=ā€‰5.91, SDā€‰=ā€‰0.53).

Study 2

Experimental task

We used Klonek et al.ā€™s (2020) innovation task, which asked participants to improve an organizationā€™s marketing materials to promote a 20-year longitudinal study. The task lasted 30ā€‰min. First, we showed participants an original marketing flyer that the organization intended to use for recruitment purposes. This marketing flyer was very poorly designed, with typographical errors, improper formatting, small fonts, etc. The participant was given suggestions for improvement in the task instructions:

(1) Add images to the existing marketing flyer; (2) Use different colors to highlight important information; (3) Remove redundant information; (4) Use layout balance, white space to improve readability and visibility; (5) Replace the background with one that is more relevant to the audience and has a higher market value.

Experimental interventions

The interventions used in Study 2 were the same as those used in Study 1 (SI Appendix, Study 1 Experimental interventions).

Experimental procedures

Experimental procedures consisted primarily of grouping, task completion, and leader intervention (SI Appendix, Study 2 Experimental procedures).

Measurement of perceived and followed leadership behaviors

At the end of the experiment, participants rated the extent to which they perceived and followed all four leadership behaviors during the experiment. Scales from previous studies were used (SI Appendix, Study 2 Measurement of leadership behaviors).

Measurement of idea implementation

Based on the innovation output measurement used in previous research (e.g., Klonek et al. (2020); Rosing et al. (2017); Zacher et al. (2016), we used two operational indicators that capture innovation implementation outcomes: applicability of innovation product (ā€œThis task outcome can readily be applied in the ā€˜real worldā€™ā€) and innovation quality ((ā€œThis task outcome exceeds the quality standardsā€). Similarly, participantsā€™ idea implementation performance was rated by the same four experts on a 7-point Likert scale (1ā€‰=ā€‰strongly disagree, 7ā€‰=ā€‰strongly agree). Thus, each task outcome was rated on an 8-item scale (4ā€‰Ć—ā€‰2 items). ICC (absolute, single measure)ā€‰=ā€‰0.65 based on single raterā€™s ratings, indicating moderate reliability. ICC (absolute, mean measure)ā€‰=ā€‰0.83 based on all four ratersā€™ ratings, indicating good reliability. Therefore, the mean values of the four ratersā€™ ratings were used as the idea implementation score (Mā€‰=ā€‰5.80, SDā€‰=ā€‰0.70).

Results

Study 1

Validity of leadership manipulations

Following He et al. (2004), we used an analysis of variance (ANOVA) to test whether, as expected, the leadership manipulations produced significant differences in perceived leadership behaviors between the four groups. The groups were named after the leadership manipulation condition to which they were assigned, i.e., Group A for the ambidextrous leadership group, Group C for the closing leadership group, Group O for the opening leadership group, and Group T for the transformational leadership group. Table 1 presents the results of the group differences in perceptions of the four leadership behaviors, all of which were significant (F-values ranging from 24.77 to 54.78, psā€‰<ā€‰0.001).

Table 1 Validity analysis of leadership manipulations (Study 1).
Full size table

To further examine the group differences in perceived leadership behaviors, we used the least significant difference (LSD) test. As shown in Table 1, the perceived leadership behaviors of all groups were consistent with the assigned leadership manipulations. For example, the perceived ambidextrous leadership behavior in Group A (Mā€‰=ā€‰33.189) was significantly higher than in Group C (diffā€‰=ā€‰2.124, psā€‰<ā€‰0.05), Group O (diffā€‰=ā€‰4.716, psā€‰<ā€‰0.001), and in Group T (diffā€‰=ā€‰8.595, psā€‰<ā€‰0.001). These results indicate that all four leadership manipulations were effective.

Descriptive and correlation analysis

Table 2 summarizes the means, standard deviations, and correlation coefficients for all variables. Idea generation was positively correlated with both ambidextrous leadership (rā€‰=ā€‰0.25, pā€‰<ā€‰0.001) and opening leadership (rā€‰=ā€‰0.18, pā€‰<ā€‰0.01), negatively correlated with closing leadership (rā€‰=ā€‰āˆ’0.41, pā€‰<ā€‰0.001), and not correlated with transformational leadership behavior (rā€‰=ā€‰āˆ’0.02, ns).

Table 2 The means, standard deviation, and correlation coefficient (Study 1).
Full size table

ANOVA

Hypothesis l focuses on the effect of ambidextrous leadership on idea generation. ANOVA was used to test the difference in idea generation performance between the four groups. Table 3 shows that idea generation differed significantly (F(3, 246)ā€‰=ā€‰19.79, pā€‰<ā€‰0.001; R2ā€‰=ā€‰0.194). Specifically, idea generation was highest in Group A (Mā€‰=ā€‰6.139), significantly higher than in Group C (diffā€‰=ā€‰0.599, pā€‰<ā€‰0.001) and Group T (diffā€‰=ā€‰0.252, pā€‰<ā€‰0.001), but not in Group O (diffā€‰=ā€‰0.066, ns). Thus, Hypothesis 1 was only partially supported.

Table 3 The effects of leadership behaviors on idea generation.
Full size table

Study 2

Validity of leadership manipulations

Similar to Study 1, ANOVA and LSD test were used to validate whether the leadership manipulations led to significant differences in perceived leadership behaviors between the four groups. Table 4 presents the results of the group differences in perceptions of the four leadership behaviors, all of which were significant (F ranged from 29.32 to 135.36, psā€‰<ā€‰0.001). Therefore, all four leadership manipulations were effective.

Table 4 Validity of leadership manipulations (Study 2).
Full size table

Descriptive and correlation analysis

Table 5 presents the means, standard deviations, and correlation coefficients for all variables. Idea implementation was positively correlated with ambidextrous leadership (rā€‰=ā€‰0.33, pā€‰<ā€‰0.001), negatively correlated with opening leadership (rā€‰=ā€‰āˆ’0.14, pā€‰<ā€‰0.05) and transformational leadership behavior (rā€‰=ā€‰āˆ’0.31, pā€‰<ā€‰0.001), and not correlated with closing leadership (rā€‰=ā€‰0.11, ns).

Table 5 The means, standard deviation, and correlation coefficient (Study 2).
Full size table

ANOVA

Hypothesis 2 addresses the effect of ambidextrous leadership on idea implementation. ANOVA was used to test the difference in idea implementation performance between the four groups. Table 6 shows that idea implementation differed significantly (F(3, 246)ā€‰=ā€‰17.83, pā€‰<ā€‰0.001; R2ā€‰=ā€‰0.179). Specifically, idea implementation was highest in Group A (Mā€‰=ā€‰6.139), significantly higher than in Group O (diffā€‰=ā€‰0.579, pā€‰<ā€‰0.001), Group T (diffā€‰=ā€‰0.784, pā€‰<ā€‰0.001), and in Group C (diffā€‰=ā€‰0.275, pā€‰<ā€‰0.05). Thus, hypothesis 2 was verified.

Table 6 The effects of leadership behaviors on idea implementation.
Full size table

Study 1 & 2: a joint analysis

The above results showed that ambidextrous leadership contributes significantly to both idea generation and implementation. Nevertheless, it is shown that there was no significant difference between Group A (Mā€‰=ā€‰6.139) and Group O (Mā€‰=ā€‰6.073) in idea generation (diffā€‰=ā€‰0.066, ns). This seemed to indicate that the closing behavior component in the ambidextrous leadership intervention did not play a significant role, or at least an equally significant role, in idea generation. Besides, Group A (Mā€‰=ā€‰6.208) was significantly higher (diffā€‰=ā€‰0.275, pā€‰<ā€‰0.05) than Group C (Mā€‰=ā€‰5.933) in idea implementation. This implies that both opening and closing behaviors in the ambidextrous leadership intervention played a significant role in idea implementation. These preliminary observations seemed to support Hypothesis 3, that closing leader behavior is given less weight in idea generation than opening leader behavior in idea implementation.

To go beyond these preliminary observations, we conducted a joint analysis combining data from both stages (i.e., Study 1 and Study 2). The data were first combined by standardizing the innovation performance scores of the two stages to form a common database of 500 cases containing individual IDs, stage-specific variables (1ā€‰=ā€‰stage 1, 2ā€‰=ā€‰stage 2), subgroup variables (including four dummy variables), and standardized innovation performance variables. Next, based on the results of the ANOVA and LSD tests, we analyzed the ā€œdifference in differencesā€. Specifically, we made a comparison between the innovation difference (standardized diffā€‰=ā€‰0.124, ns) between Group A (Stage 1) and Group O (Stage 1), the innovation difference between Group A (Stage 1) and Group C (Stage 1) (post-standardized diffā€‰=ā€‰1.131, pā€‰<ā€‰0.001), the innovation difference between Group A (Stage 2) and Group O (Stage 2) (post-standardized diffā€‰=ā€‰0.822, pā€‰<ā€‰0.001), and Group A (Stage 2) and Group C (Stage 2) (post-standardized diffā€‰=ā€‰0.391, pā€‰<ā€‰0.05). The comparison results are presented in Table 7.

Table 7 Comparison of differences in the effects of leadership behaviors on innovation.
Full size table

The comparison results that innovation in Group A (Stage 1) (Mā€‰=ā€‰0.434) was not significantly higher (diffā€‰=ā€‰0.124, ns) than that in Group O (Stage 1) (Mā€‰=ā€‰0.310), and that innovation in Group A (Stage 2) (Mā€‰=ā€‰0.582) was significantly higher (diffā€‰=ā€‰0.391, pā€‰<ā€‰0.05) than that in Group C (Stage 2) (Mā€‰=ā€‰0.191). These differences (i.e., Diff1 (A-O), Diff2 (A-C), as shown in Table 7) are visualized in Fig. 1. The difference between these two difference values was significant (F(1, 492)ā€‰=ā€‰5.03, pā€‰<ā€‰0.05). Furthermore, we compared innovation in Group C (Stage 1) (Mā€‰=ā€‰āˆ’0.697) with that in Group O (Stage 2) (Mā€‰=ā€‰āˆ’0.240), and the difference was significant (diffā€‰=ā€‰0.457, pā€‰<ā€‰0.01). These results showed that employees who received only closing leadership interventions during the idea generation stage were less innovative than employees who received only opening leadership interventions during the idea implementation stage. This further supports Hypothesis 3 that closing leadership behavior in idea generation is less necessary than opening leadership behavior in idea implementation.

Fig. 1
figure 1

Differences between intervention effects.

Full size image

Discussion

This study aims to replicate previous findings on the effect of ambidextrous leadership on innovation. Using a process perspective, we examined the effect of ambidextrous leadership on idea generation and implementation. Besides, this study used a randomized controlled experiment to validate causal inferences.

The results partially supported H1 (i.e., the effect on idea generation) that ambidextrous leadership was more effective than closing and transformational leadership (but not opening leadership) in facilitating employee idea generation. H2 (i.e., the effect on idea implementation) was fully supported, that ambidextrous leadership was the most effective of the four leadership interventions in facilitating employee idea implementation. Hypothesis 3 is corroborated that closing leader behavior is given less weight in idea generation than opening leader behavior in idea implementation. In what follows, we present some possible theoretical explanations for these findings and consider their practical implications.

First, ambidextrous and open leadership showed significant comparative advantages in idea generation. This finding attests to the critical role of opening leadership in idea generation, or the comparatively greater role of opening behavior (compared to closing behavior) in ambidextrous leadership in idea generation. One possible explanation is the internal regulation of creative employees driven by inhibitory control, which we discussed earlier. This internal regulation lies in suppressing unwanted or inappropriate creative ideas and promoting those that are more relevant and promising, thereby reinforcing the adaptive and goal-oriented nature of behavior. The self-regulation process suggests that employees are exploiting existing conditions or plans. That is, the cognitive control of creative employees in idea generation appears to be sufficient to ensure the necessary exploitation of new ideas with due regard to their usefulness. Accordingly, exploitation efforts driven by closing leadership (an external stimulus) may become less necessary at this stage.

Furthermore, the contextual and participant factors involved in this study may have contributed to the absence of significant differences between the effects of ambidextrous and opening leadership at the idea generation stage. On the one hand, the experimental taskā€™s contextual setting may have promoted participantsā€™ open-mindedness and self-directed exploratory behaviors, while the closing behavioral intervention in ambidextrous leadership may not have fully fulfilled its intended role. This may have led to the observed similarity in the effects of ambidextrous and opening leadership in promoting idea generation. On the other hand, the impact of leadership styles may vary among participants with different characteristics (e.g., age, professional background, work experience). For instance, Chinese employees who possess strong executive abilities but exhibit deficient creative thinking may be more adept at idea generation when guided by an open leader, potentially obviating the necessity for closing behaviors to further constrain or direct them. This phenomenon could potentially lead to a lack of significant discrepancy in the impacts of these two leadership styles at the idea generation stage. Additionally, the quality of the leader-employee relationship may have exerted a moderating effect. In teams where leaders exhibit high levels of trust and effective communication with their employees, employees may be more inclined to engage in idea generation autonomously with opening leadership, potentially reducing reliance on closing leadership within the ambidextrous leadership to provide guidance.

Second, ambidextrous leadership plays an important and unique role in idea implementation. Idea implementation requires not only closing leadership to unify employeesā€™ attention and action to facilitate exploitation, but also opening leadership to facilitate exploration to solve unanticipated problems. As mentioned earlier, during the idea implementation stage, when the situation changes (e.g., an unexpected problem arises), employees seem to be unable to make decisions through their personal cognitive mechanisms. This is because the potential choices faced by employees are all related to the current situation, such as continuing implementation without solving the problem or re-exploration. Additionally, these potential options may come with high costs or risks. Therefore, proactive external interventions (e.g., opening leader behaviors) are needed to guide employee behavior. This mechanism is described as the temporal flexibility to switch between the two leader behaviors to meet different task demands in a dynamic innovation process.

In summary, our findings fundamentally confirm the value of ambidextrous leadership, which emphasizes both opening and closing leadership, or exploration and exploitation, in both early and late innovation stages (Rosing et al., 2011), thus potentially refining the classical ā€˜early exploration-late exploitationā€™ innovation stage model (manifested in leadership as early opening-late closing) (Farr, 2003). A strict stage differentiation of innovation behavior may not be compatible with actual innovation activities, as it does not consider the complex and even iterative nature of the innovation process. In other words, the innovation process is not strictly linear; rather, it is characterized by the necessity of flexibility to transition between exploratory and exploitative behaviors at various stages. This highlights the critical importance of flexibility in leadership behaviors within a dynamic innovation process. Depending on the context, both idea generation and implementation may require exploration (opening leadership) and exploitation (closing leadership).

Practical implications

Our findings go a long way towards validating the adaptive value of ambidextrous leadership in innovation activities. An important implication is that innovation leaders should pay close attention to the innovation process, especially to changes in the situation and the needs they generate. At the idea generation stage, ambidextrous leadership is almost as effective as opening leadership. This requires leaders to give employees enough room for trial and error, to encourage new ideas, and to provide employees with adequate material and psychological resources. Meanwhile, leaders should have full confidence in employeesā€™ adaptive capacity (the ability to adapt their behaviors to the conditions and needs of the organization) rather than interfere too much with theirā€™ self-regulation of attention, cognition, and action.

The tougher test for leaders comes at the idea implementation stage. The findings showed that ambidextrous leadership significantly outperformed all other leadership styles. During implementation, the ambidextrous leader needs to be highly sensitive to the situation. First, the leader should use closing behaviors to develop a complete action plan, allocate personal resources, and monitor goal achievement. Second, when problems arise, leaders should intervene and make timely decisions, such as turning to exploration, to reduce employeesā€™ perceptions of potential risks or costs and increase their awareness of the need to explore new strategies or plans. Once the implementation plan has been refined, leaders need to refocus employee resources to move the idea forward. This cycle of flexible, demand-driven shifts continues until the idea is fully realized.

Although ambidextrous leadership appears to be applicable in a variety of contexts, its implementation must be meticulously adapted to each industry. In highly regulated sectors such as healthcare or aviation, where safety and compliance are paramount, the balance between opening and closing leader behaviors must be meticulously calibrated. In the healthcare sector, for instance, while opening behaviors are crucial for fostering patient care solutions, closing behaviors are equally vital to ensure that these innovative solutions adhere to stringent safety and ethical standards. Likewise, aviation leaders face the dual challenge of promoting technological and operational innovations while adhering to stringent safety protocols. In the technology sector, companies like Google have effectively utilized ambidextrous leadership to catalyze rapid innovation. These leaders foster an environment that encourages the exploration of novel technologies and business models (opening behaviors), while ensuring that these innovations are aligned with the companyā€™s strategic objectives and market demands (closing behaviors). In the financial industry, the regulatory landscape often necessitates a cautious approach to innovation. In this context, ambidextrous leaders must meticulously orchestrate the exploration of novel financial products or services, ensuring not only that they address customer needs but also that they comply with the regulatory requirements.

Limitations and suggestions

This study is not without limitations. Firstly, the findings may be partly explained by the fact that the specific innovative tasks used lent themselves to ambidextrous leadership interventions. Some studies suggest that ambidextrous leadership may not be applicable across industries. For example, in industries with high safety requirements, a shift to opening leadership should be implemented with caution, as opening behaviors may lead to excessive employee distraction and carelessness (Griffin and Hu, 2013), and excessive experimental activities may lead to a high incidence of production accidents. Therefore, replicating the experiment in various industrial contexts could assist in evaluating the stability and generalizability of the impact of ambidextrous leadership. Furthermore, it is imperative to consider specific antecedent conditions that enable a causal relation. From a realist perspective, the causal relation between leadership and innovation performance may not be purely mechanistic. Rather, it may be responsive to local contexts and human volition. For example, the applicability of ambidextrous leadership in different cultural (e.g., collectivist vs. individualist) and organizational (e.g., hierarchical vs. flat) contexts remains unclear. The leadership behaviors exhibited in these settings may be interpreted and responded to in divergent ways, thereby influencing the extent to which they impact innovation.

Additionally, the majority of the participants in this study were between the ages of 26 and 35, constituting 72% of the total sample. These individuals may possess a higher degree of innovativeness, flexibility, receptiveness to newness, and propensity to generate novel and distinctive ideas with the opening leadership style. Future research endeavors should prioritize the examination of the interaction between participantsā€™ individual difference factors (e.g., innovation self-efficacy, risk preference) and leadership styles. This can facilitate the elucidation of the intricate mechanisms through which leadership influences innovation performance. Moreover, the simulated innovation tasks employed in this study confirmed the presence of causality under controlled experimental conditions. However, innovation tasks in actual settings tend to be more intricate (e.g., cross-sectoral cooperation, coordination with external vendors), time-consuming, and influenced by numerous external factors (e.g., market changes). The replication of these factors in simulated tasks is often challenging, potentially compromising the ecological validity of the experiment. Future research could use more ecologically valid observation methods, such as field experiments in innovation teams. Finally, the relatively brief duration of the study precluded a comprehensive evaluation of the long-term impact of leadership on innovation. It is therefore recommended that subsequent research endeavors encompass the execution of prolonged studies, thereby facilitating a more profound comprehension of the enduring influence of leadership behaviors on innovation processes.