“Filter” or “Accelerator”? A Mixed-Methods Study of the Driving Mechanisms of Innovative Work Behaviour among Higher Vocational Education Teachers Under AI Ethics Constraints

Attribution Theory

Attribution Theory systematically examines how people explain the causes of their own and others’ behaviour, and how these explanations shape subsequent emotions, motivation, and behaviour (Hayat et al., 2022). In the early formulation of Attribution Theory, Heider (1958) classified the causes of behaviour into two broad categories: internal and external. Internal factors include personal traits, abilities, emotions, and effort, whereas external factors include task difficulty, luck, social pressure, and environmental conditions. Subsequent scholars further refined and extended this perspective. For example, Kelley’s (1967) covariation model argues that people rely on information accumulated across events and apply the covariation principle to resolve attributional questions under uncertainty. Weiner (1985) proposed that individuals make internal and external attributions mainly on the basis of locus of causality, stability, and controllability, thereby predicting and interpreting behaviour and events. The distinction between internal and external attribution has since become a cornerstone of attributional research and has been widely cited in the behavioural sciences (Soh & Kim, 2026), serving as an important theoretical framework for explaining complex human behaviour (Tavsanli, 2025).

Self-Determination Theory

Proposed by Deci and Ryan (1985a) and subsequently developed, Self-Determination Theory (SDT) is a macro theory of human motivation, personality development and psychological wellbeing. It goes beyond a simple dichotomy of intrinsic versus extrinsic motivation, offering a more comprehensive framework for understanding the drivers of individual behaviour (Ryan & Deci, 2000). SDT rests on the assumption that individuals possess innate growth tendencies and basic psychological needs (Ryan, 1995). It posits that three needs—autonomy, competence, and relatedness—determine autonomous motivation and constitute the core driving force for sustained, high-quality and effective behavioural outcomes (Ryan & Deci, 2017). Autonomy refers to the perceived sense of choice and control in action and decision-making; competence denotes the belief that one is capable of undertaking specific activities or completing tasks; and relatedness concerns the extent to which individuals feel accepted, understood and cared for by their social groups (Ryan, 1995). Over time, SDT has matured into a coherent framework comprising six sub-theories: Cognitive Evaluation Theory (Deci & Ryan, 1985a), Organismic Integration Theory (Ryan & Deci, 2000), Causality Orientations Theory (Deci & Ryan, 1985b), Basic Psychological Needs Theory (Ryan, 1995), Goal Contents Theory (Kasser & Ryan, 1993), and Relationships Motivation Theory (Ryan & Deci, 2017).

AI Involvement

Within social judgement theory, Sherif and Cantrill (1947) proposed the concept of “self-involvement” to capture the strength of the association individuals perceive between a situation or stimulus and the self. Krugman (1965) later introduced the notion of involvement into consumer-behaviour research, where it has since been widely adopted (Bruwer et al., 2017; Yang et al., 2023). Because the literature recognises multiple forms of involvement (Mou et al., 2020; Zhu et al., 2019), scholars have not reached a single agreed definition. Zaichkowsky (1986) distinguished product involvement, information involvement, and purchase-decision involvement, and defined involvement as “a person’s perceived relevance of an object based on inherent needs, values, and interests.” Accordingly, this study defines teachers’ AI involvement as the degree to which teachers attach importance to AI technology, or the perceived importance of AI technology to teachers’ work.

The double-edged nature of AI can enhance employees’ work efficiency and capabilities, but it may also trigger technology anxiety and, in turn, resistance to innovation. Accordingly, a key issue in promoting the digital innovation of teaching is how to heighten higher vocational education teachers’ recognition of the importance of AI and its relevance to their own work. From the perspective of both external and internal attributions, this study therefore examines the mediating role of AI involvement and its interplay with other factors in the proposed model.

Innovative Work Behaviour

Innovative work behaviour (IWB) refers to employees’ proactive generation, promotion and implementation of novel and useful ideas, methods, or solutions, driven by the goal of improving task performance (Anser et al., 2021; De Jong & Den Hartog, 2010; Liehr & Hauff, 2025; Scott & Bruce, 1994). At the individual level, many scholars adopt a process perspective (e.g., Bratianu et al., 2023; Kang & Lee, 2017; Lajçi et al., 2026). A classic three-stage process includes idea generation, idea promotion and idea application (Janssen, 2000). In response to diverse application scenarios, some have elaborated the process further—for instance, Bratianu et al. (2023) distinguish idea exploration, idea generation, idea championing, and idea implementation (Dong et al., 2025). Following Janssen (2000), we define teachers’ IWB in the present study as: vocational teachers’ proactive adaptation to new requirements for cultivating technically skilled talent by breaking through limitations of traditional teaching models, actively employing AI and other digital technologies, and creatively proposing, designing, advocating and implementing new pedagogical ideas, methods and assessments.

Perceived School Learning Support and AI Involvement

Perceived school learning support reflects institutional efforts to upgrade teachers’ TPACK competencies (Ng et al., 2021a, 2023) by creating safe and open learning environments, providing financial support, involving teachers in decision making processes, and providing rich and accessible learning opportunities to improve AI literacy (Chiu et al., 2024). In a safe, open campus environment, teachers can freely exchange digital knowledge such as AI, accelerating knowledge sharing and knowledge flows (Liehr & Hauff, 2025), and strengthening knowledge integration (Sondhi et al., 2024), thereby accurately recognising the value of AI literacy and its importance for teaching (Du et al., 2024). Moreover, financial support for AI-related training, encouragement to participate in decisions about smart-teaching reforms, and diverse training opportunities can bolster organisational commitment (Lee et al., 2020), thereby increasing the salience of AI knowledge and skills, and, in turn, AI involvement. Chiu (2022) confirmed that school learning support motivates teachers to invest more resources in integrating digital knowledge. Hence:

Basic Psychological Needs and AI Involvement

SDT posits that autonomy, competence and relatedness are key determinants of work motivation (Gagné & Deci, 2005; Ryan & Deci, 2000). The greater the satisfaction of these basic psychological needs, the more intrinsic motivation is activated, and autonomous motivation dominates behaviour (Deci & Ryan, 2008). Driven by autonomous motivation, vocational teachers’ behaviour is fuelled by interest, enjoyment and wellbeing associated with their professional mission (Deci & Ryan, 1985a). Confronted with new technologies such as AI, a positive professional attitude prompts teachers to enhance AI self-efficacy through greater technological involvement (Bergdahl & Sjöberg, 2025), thereby improving digital literacy (Du et al., 2024; Ng et al., 2023, 2024). According to Causality Orientations Theory, as need satisfaction increases, motivation becomes more internalised; teachers are more likely to respond to AI and smart-teaching requirements in line with professional interests and personal values (Deci & Ryan, 1985b), proactively increasing involvement with AI to strengthen their AI literacy framework (Long & Magerko, 2020). Need satisfaction is also motivating in itself and positively affects organisational commitment (Gagné et al., 2010), encouraging investment in AI involvement to achieve excellent performance. Similarly, Al-Mamary and Abubakar (2025) show that basic needs are positively related to task–technology fit and that individuals invest resources to learn tools such as ChatGPT. Therefore:

AI Involvement and Innovative Work Behaviour

AI involvement reflects the degree to which vocational teachers attach importance to AI technologies (Yang et al., 2023). As competence in AI increases, teachers, motivated by professional values and enjoyment, proactively identify problems, and innovate work methods and solutions (Cai & Tang, 2022). Involvement positively predicts domain-specific learning (Kyndt & Baert, 2013). As AI literacy improves, AI self-efficacy and self-esteem grow (Rodríguez-Ruiz et al., 2025), enhancing the willingness and ability to generate insights and propose creative solutions (Dong et al., 2025). In the face of technological change, high involvement signals a strong sense of responsibility and a positive work attitude; responsibility awakens a sense of ownership and concern for disciplinary viability and competitive advantage (Blanco-Gonzalez et al., 2020), fostering proactive transformation and more IWB. Importantly, higher AI involvement accelerates AI knowledge accumulation, enabling better adaptation to and mastery of smart-education innovations, promoting knowledge sharing and internal diffusion, raising team AI literacy, and stimulating pedagogical innovation. Hence:

The Mediation Role of AI Involvement

From H1 and H5, perceived school learning support creates richer learning opportunities, accelerates AI knowledge flows, and enhances perceptions of insider status and professional wellbeing (Chiu et al., 2024). Through AI involvement, teachers learn and share AI knowledge, and better implement IWB (Dong et al., 2025; Zhang et al., 2024), thereby fulfilling professional commitments. Prior research shows that external stimuli influence behavioural intentions via involvement (Mou et al., 2020; Yang et al., 2023). According to the S–O–R model (Mehrabian & Russell, 1974), school learning support (S) can influence IWB (R) via teachers’ AI involvement (O). Therefore:

Similarly, need satisfaction activates intrinsic motivation (Ryan & Deci, 2017). Autonomous motivation enhances professional identity and wellbeing (Tang et al., 2020) and organisational commitment (Lee et al., 2020), leading to greater emphasis on capability development (Yao et al., 2021). In response to AI, and to attain intrinsically valued goals that yield sustained wellbeing (Kasser & Ryan, 1993) and honour organisational commitments (Ćulibrk et al., 2018), vocational teachers increase AI involvement to drive digital-teaching reforms and innovation. Thus:

The Moderated Mediation Role of AI Ethics

AI ethics refers to the design, development and use of AI in ways that enhance human wellbeing while adhering to human values, moral norms, and social customs (Gümüş & Kukul, 2023; Kong et al., 2024; Ng et al., 2024; Pinski & Benlian, 2024; L. Zhao et al., 2022). Centred on human considerations, AI ethics typically encompasses five elements: fairness, accountability, transparency, ethical considerations, and safety (Ng et al., 2021a). Its goals are to develop AI that benefits society whilst preventing risks arising from misuse (Laupichler et al., 2022; Ng et al., 2021b). Ng et al. (2021a) argue that AI should be human-centred; AI literacy entails not only building citizens’ AI capabilities and enabling them to benefit, but also educating them to be responsible and ethical users. Jobin et al. (2019) list 11 ethical principles for citizens to observe (Beneficence, Non-maleficence, Trust, Transparency & Explainability, Freedom & Autonomy, Privacy, Justice, Fairness & Equity, Responsibility & Accountability, Dignity, Sustainability, Solidarity). In this study, higher versus lower AI ethics reflects individuals’ perceptions of differing degrees of ethical constraints associated with AI.

AI ethics requires balancing productivity with “AI for social good” (Chai et al., 2020; Jobin et al., 2019), minimising negative impacts while meeting users’ needs. AI bears responsibility for contributing to a sane social system, pursuing fairness and equality (Ng et al., 2021b; Ortega-Bolaños et al., 2024), and organisations should devise policies to address potential unemployment associated with AI diffusion (Jobin et al., 2019). Consequently, when perceived constraints of AI ethics are strong, teachers’ motivation to learn and share AI knowledge may be dampened, prompting them to act more cautiously and to engage in ongoing AI ethical reflection (Wang et al., 2025); they may lower AI involvement to avoid downsides and reduce disruptive pedagogical innovations in order to fulfil social responsibilities (Ortega-Bolaños et al., 2024). With tighter boundary-setting and risk control measures in place, innovation costs and cycles increase, creativity is supplanted by a “prudent creativity” approach, and the policy effectiveness of school learning support operating through AI involvement experiences a decline. Conversely, when perceived constraints of AI ethics are weak, AI’s spill-over effects are fully released with fewer uncertainty constraints (Le et al., 2024) and may even give rise to moral suspension (Chen et al., 2026); as the positive impact of school learning support is amplified and AI identity becomes more readily formed, AI involvement increases due to a sense of psychological entitlement, potentially accompanied by moral relativism (Cao et al., 2023) and a surge in educational innovation. Therefore:

By the same logic, when perceived constraints of AI ethics are strong, the level of motivation internalisation driven by autonomy, competence and relatedness needs diminishes; individuals perceive greater external pressure (Ryan & Deci, 2000), autonomous motivation shifts towards controlled motivation, and work anxiety (Gillet et al., 2016) and stress (Tremblay et al., 2009) rise, while job satisfaction and engagement fall. To adapt, teachers may reduce AI involvement, restrain their creativity, limit disruptive innovation, and avoid deviant innovation (Criscuolo et al., 2014). When perceived constraints of AI ethics are weak, external control lessens, need-driven internalisation strengthens (Deci & Ryan, 1985a; Mungra et al., 2024), autonomous motivation increases (Ryan & Deci, 2000) and positively predicts organisational citizenship behaviour (Lee et al., 2020; Sedlářík et al., 2024). In pursuit of excellent performance (Good et al., 2022), teachers will heighten AI involvement, enhance AI literacy and exhibit more IWB. Hence:

Based on the theoretical foundations and hypotheses above, we construct the conceptual model shown in Figure 1.

OPEN IN VIEWER

Figure 1.

Research model.

Perceived School Learning Support

Measured using the unidimensional scale revised by Chiu et al. (2024), comprising four items. Cronbach’s alpha = .881.

Basic Psychological Needs

Measured using the three-dimensional scale revised by Chiu et al. (2024), covering autonomy, competence and relatedness, with four items each. Cronbach’s alpha = .871, .863, and .921, respectively.

AI Involvement

Measured with the unidimensional scale developed by Ghali-Zinoubi and Toukabri (2019), with minor modifications to fit the educational AI context, comprising six items. Cronbach’s alpha = .901.

Innovative Work Behaviour

Measured with the unidimensional six-item scale used by Anser et al. (2021), with item wording adapted from Sary et al. (2023) to suit the teaching context. Cronbach’s alpha = .885.

AI Ethics

Measured using the five-item scale adapted from L. Zhao et al. (2022), theoretically developed by Ng et al. (2021a, p. 4). Cronbach’s alpha = .876. The items are listed in Appendix 1 .

Control Variables

Following previous studies, gender, age, educational level, academic title, and institute level were included as covariates.

Causal Analysis

To reduce collinearity effects, latent variable means were used in hypothesis testing (Variance inflation factors ranging from 1.030 to 2.371). With PROCESS 4.0 (95% confidence interval, 5,000 bootstrap resamples, Model 4), the results (see Table 3) revealed: Basic psychological needs were the primary driver of IWB (B = 0.481, β = .529, p < .001; Boot CI = [0.352, 0.605]). The effect of perceived school learning support on IWB was smaller but significant (B = 0.071, β = .119, p < .05; Boot CI = [0.001, 0.146]). Among the three needs, only competence significantly influenced IWB (B = 0.429, β = .517, p < .001; Boot CI = [0.322, 0.552]); autonomy (B = 0.047, β = .060, p > .1) and relatedness (B = 0.040, β = .051, p > .1) were not significant.

OPEN IN VIEWER

Table 3.

Mediation Effects of AI Involvement (N = 413).

Independent variables	Point estimate	Products of coefficients		Bootstrapping (5,000 samples)
		SE	Z	Lower	Upper
	Indirect effects
School learning support	−0.002	0.012	0.167	−0.026	0.024
Autonomy	0.050*	0.020	2.471	0.011	0.092
Competence	0.105***	0.026	4.086	0.058	0.159
Relatedness	0.011	0.021	0.549	−0.032	0.050
	Contrasts
School learning support vs. autonomy	−0.052*	0.023	−2.229	−0.097	−0.007
School learning support vs. competence	−0.107***	0.029	−3.737	−0.164	−0.050
School learning support vs. relatedness	−0.013	0.024	−0.537	−0.060	0.034
Autonomy vs. competence	−0.055	0.033	−1.677	−0.120	0.010
Autonomy vs. relatedness	0.039	0.029	1.345	−0.018	0.096
Competence vs. relatedness	0.094**	0.033	2.813	0.029	0.159

*

p < .05. **p < .01. ***p < .001.

Autonomy (B = 0.179, β = .190, p < .01; Boot CI = [0.046, 0.320]), competence (B = 0.372, β = .379, p < .001; Boot CI = [0.244, 0.504]), and education level (B = 0.144, β = .090, p < .05; Boot CI = [0.012, 0.272]) positively influenced AI involvement. In contrast, school learning support (B = −0.008, β = −.011, p > .1; Boot CI = [−0.094, 0.083]) and relatedness (B = 0.040, β = .044, p > .1; Boot CI = [−0.112, 0.183]) were not significant. AI involvement positively predicted IWB (B = 0.282, β = .334, p < .001; Boot CI = [0.186, 0.378]). Thus, H2, H3, and H5 were supported, while H1 and H4 were not.

Mediation Effect Analysis

Mediation effect was tested using PROCESS with 5,000 bootstrap iterations. Results (see Table 3) showed that AI involvement significantly mediated the relationships between autonomy, competence and IWB, with the strongest indirect effect in the competence → IWB path. Comparisons indicated that mediation effects for autonomy and competence were stronger than for school learning support → IWB. This finding further highlights basic psychological needs as the core driving force of AI involvement and IWB. Thus, H7a and H7b were supported, while H6 and H7c were not.

Moderated Mediation Effect of AI Ethics

Following the procedure proposed by Edwards and Lambert (2007), we tested the moderated mediation model as a whole. After setting AI ethics at one standard deviation above and below the mean and controlling for covariates, Mplus 8.3 was used to estimate the high and low AI ethics subgroups and to compare differences in the first stage, second stage, direct, and indirect path coefficients. As shown in Table 4, AI ethics negatively moderated the second stage of the mediation model (Difference = −0.106, p < .05), whereas its moderating effects on the four indirect paths were not significant. These findings suggest that AI ethics plays a filtering role in AI-driven IWB and helps safeguard responsible innovation. Taken together with the mediation results, the evidence provides only partial support for H9a and H9b, whereas H8, H9c are not supported.

OPEN IN VIEWER

Table 4.

Results of the Moderated Mediation Effect (N = 413).

Mediation path	Moderator	Stage		Effect
		First stage	Second stage	Direct effect	Indirect effect
PSLS → INVO → IWB	High AIET	−0.037	0.144**	0.067	−0.005
	Low AIET	0.039	0.250***	0.094	0.010
	Difference	−0.077	−0.106*	−0.028	−0.015
AU → INVO → IWB	High AIET	0.142*	0.144**	0.250***	0.020
	Low AIET	0.124	0.250***	−0.106*	0.031
	Difference	0.017	−0.106*	0.021	−0.011
CO → INVO → IWB	High AIET	0.238**	0.144**	0.376***	0.034*
	Low AIET	0.273***	0.250***	0.134*	0.068**
	Difference	−0.035	−0.106*	0.242**	−0.034
RE → INVO → IWB	High AIET	0.144**	0.144**	−0.073	0.021
	Low AIET	−0.055	0.250***	0.129*	−0.014
	Difference	0.199	−0.106*	−0.203**	0.034

Note. PSLS = Perceived school learning support; AU = Autonomy; CO = Competence; RE = Relatedness; INVO = AI involvement; IWB = Innovative work behaviour.

*

p < .05. **p < .01. ***p < .001.

Using the Johnson-Neyman technique recommended by Hayes (2018) and Preacher et al. (2007), this study further estimated the ranges of the moderator for which the conditional mediation effects were significant, together with their 95% confidence intervals. As shown in Figures 2 and 3, across the full observed range of AI ethics, the indirect effect of autonomy on IWB was not significant. When 4.744 ≤ AI ethics ≤ 6.705, the indirect effect of competence on IWB was significant and decreased as the moderator increased, thereby revealing the “filtering” role of AI ethics in shaping educational innovation among higher vocational education teachers.

OPEN IN VIEWER

Figure 2.

Johnson-Neyman plot of the moderated mediation effect of AI ethics (independent variable = autonomy).

OPEN IN VIEWER

Figure 3.

Johnson-Neyman plot of the moderated mediation effect of AI ethics (independent variable = competence).

Variable Selection

Correlation analysis revealed that perceived school learning support (r = .429, p < .01), autonomy (r = .476, p < .01), competence (r = .627, p < .01), relatedness (r = .458, p < .01), AI involvement (r = .577, p < .01) and AI ethics (r = .591, p < .01) were all significantly and positively correlated with IWB. Among control variables, academic title was significantly and negatively correlated with IWB (r = −.103, p < .05). Accordingly, perceived school learning support, autonomy, competence, relatedness, AI involvement, AI ethics and academic title were selected as conditions. The outcome variable was IWB.

Calibration

Continuous variables were calibrated in fsQCA 3.0 following Ragin’s (2008) direct method, with the fifth percentile representing full non-membership, the 50th percentile as the crossover point, and the 95th percentile as full membership. For the categorical variable academic title, values were assigned as follows: junior = 0, intermediate = 0.33, associate senior = 0.67, senior = 1 (Emmenegger, 2011).

Configuration 1: “Person–Organisation Goal Alignment”

When vocational colleges provide a supportive environment that couples organisational and personal goals, synergies emerge. With AI involvement and ethical governance in place, this leads to high IWB. This pathway indicates that schools should align institutional goals with teachers’ growth tendencies, satisfy autonomy and competence to trigger intrinsic motivation, and, coupled with AI involvement and ethical filtering, stimulate more sustainable and responsible pedagogical innovation.

Configuration 2: “Self-Determination”

When all three basic psychological needs are satisfied, teachers experience greater job satisfaction and autonomous motivation, voluntarily engaging in organisational citizenship behaviours. In combination with AI involvement and AI ethics, this generates professional value, career interest, and wellbeing, thereby enabling proactive, sustained and high-quality innovative behaviours.

Configuration 3: “Supply–Demand Alignment.”

This configuration reflects generally modest digital literacy among vocational teachers, alongside psychological dependence on institutional support. Driven by high competence needs, teachers aspire for training opportunities and financial support in AI and other digital technologies, to attain AI self-efficacy and achievement. With AI involvement and ethical regulation, supplemented by relatedness need satisfaction, innovation motivation is transformed, fostering more responsible educational innovation.