Will Employees Still Speak up Under Algorithmic Management? The Differential Effects of Distinct Algorithmic Functions—Evidence from the Meituan Platform in China

Abstract

Employees’ voice is an important source of organizational learning and adaptive change. As algorithmic management is increasingly applied across organizational management processes, an urgent practical question arises: Does it affect employees’ participation in organizational improvement through voice? To address this challenge, drawing on signaling theory, this study examines the differential effects of distinct dimensions of algorithmic management on voice, while also considering work locus of control as a key moderating variable. We collected one-to-one matched data from 351 employees and their supervisors in a large Chinese platform-based enterprise. We tested the hypothesized theoretical model using structural equation modeling and bootstrapping procedures. The results show that algorithmic feedback enhances employees’ felt responsibility for constructive change, which in turn promotes employees’ voice. In contrast, algorithmic directing, algorithmic scheduling, and algorithmic monitoring undermine employees’ felt responsibility for constructive change and thereby inhibit voice. In addition, work locus of control moderates these relationships: employees with an external work locus of control strengthen the negative effects of algorithmic directing, algorithmic scheduling, and algorithmic monitoring, whereas employees with an internal work locus of control strengthen the positive effect of algorithmic feedback. These findings deepen our understanding of how different dimensions of algorithmic management shape voice and offer practical insights for fostering voice in contexts characterized by algorithmic management.

1. Introduction

Employees’ voice is typically defined as a constructive, change-oriented behavior through which employees voluntarily express work-related opinions, suggestions, or concerns with the aim of improving the organization or their work unit [1,2,3]. Because voice helps organizations identify problems, correct deviations, optimize processes, and promote continuous improvement, it has long been regarded as an important source of organizational learning and adaptive change [1,2]. Prior research has shown that employees’ willingness to speak up does not occur randomly; rather, it is jointly shaped by a range of factors, including leadership behaviors, psychological safety, fairness perceptions, individual motivations, and expectations regarding the consequences of speaking up [4,5]. In other words, whether employees choose to voice their suggestions depends on whether the context in which they are embedded leads them to believe that “speaking up is both valuable and worthwhile” [1,3]. However, as algorithmic systems increasingly become involved in core managerial activities such as task allocation, process control, behavioral supervision, and performance evaluation, a new question has become increasingly salient: under algorithmic management, are employees still willing to proactively speak up and express suggestions aimed at promoting organizational improvement [6,7,8]?

Algorithmic management is typically defined as a management approach through which organizations use data-driven software systems, algorithmic models, and artificial intelligence technologies to automate or augment managerial functions that were traditionally performed by human managers, such as task allocation, scheduling, behavioral monitoring, performance evaluation, and feedback [6,7]. It has profoundly changed how employees experience work coordination, supervision, evaluation, and control [9]. Scholars have generally focused on the possibility that algorithmic management may intensify surveillance, constrain autonomy, and reduce employees’ discretionary space at work [9,10]. Recently, however, some scholars have also begun to recognize its potential bright side. For example, algorithmically developed generative artificial intelligence can create new space for individual reflection by providing information, feedback, and behavioral references, thereby helping employees understand work requirements, identify problems, and form more evidence-based judgments [11,12]. Thus, algorithmic management is not merely a neutral technological tool; rather, it is a managerial mechanism that simultaneously embodies both enabling and controlling properties [9].

Although scholars have shown why leadership behaviors, interpersonal climate, psychological safety, fairness judgments, and individual traits influence whether employees speak up in traditional organizational contexts [4,5], these explanations do not directly address the question of voice under algorithmic management. This is because algorithmic management changes not only managerial tools, but also the ways in which organizations communicate expectations, define roles, and guide employee behavior [6,7]. Existing research on algorithmic management has focused more on issues such as control, well-being, fairness, and performance, whereas evidence remains limited regarding whether algorithmic management affects employees’ voice and, if so, through what pathways it does so [8,9]. Therefore, whether algorithmic management shapes employees into “rule executors” or “improvement participants,” and further, when it constrains employees’ autonomous space and strengthens control over them versus when it supports employees’ reflection, judgment, and preparation for voice, has become an important question that urgently needs to be addressed [7,8].

Building on these limitations, this study draws on signaling theory and argues that the relationship between algorithmic management and voice should be reexamined from a responsibility-based perspective. Signaling theory suggests that, under conditions of information incompleteness, organizational management practices not only serve coordination and control functions but also convey signals to employees about organizational intentions, role expectations, and behavioral orientations. Employees, in turn, use these signals to infer “what role the organization expects me to play and what behaviors it expects me to enact” [13]. Accordingly, algorithmic management is not merely a technical arrangement, but rather an organizational mechanism that continuously emits managerial signals. Different types of algorithmic management define the boundaries of employees’ roles in different ways and shape whether employees perceive promoting improvement as part of their own responsibility [6,8].

This study further argues that whether employees engage in voice depends not only on whether they notice problems or have suggestions, but also on whether they believe they should assume responsibility for constructive change. Felt responsibility for constructive change captures precisely this subjective sense of responsibility that “I am responsible for helping the organization become better,” and has been shown to be an important psychological foundation for change-oriented proactive behavior [14,15]. Although the psychological pathways through which algorithmic management influences voice may not be limited to this mechanism—factors such as psychological safety, perceived autonomy, voice efficacy, and perceptions of algorithmic fairness may also play a role—this study focuses on felt responsibility for constructive change because it more directly explains how employees translate organizationally conveyed role expectations into responsibility judgments about whether they should proactively promote improvement. Moreover, voice itself is a voluntary, improvement-oriented proactive behavior. Thus, employees with a stronger felt responsibility for constructive change are generally more likely to speak up rather than remain silent [5,16].

Building on this logic, this study further introduces work locus of control as a moderator. Work locus of control reflects employees’ stable belief about whether outcomes in the work context are primarily determined by their own actions. Prior research has shown that employees with a stronger internal locus of control tend to place greater emphasis on personal agency, responsibility taking, and their ability to influence the environment [17,18]. Therefore, in the context of algorithmic management, employees with a stronger internal work locus of control may be more sensitive to the constraint and discipline signals conveyed by algorithmic directing, algorithmic scheduling, and algorithmic monitoring, and may consequently perceive that their space for proactive improvement within the organization has been compressed. By contrast, employees with a weaker internal work locus of control may be more likely to regard external feedback as a source of action guidance and behavioral reference, thereby developing felt responsibility for constructive change [17,18]. On this basis, this study incorporates work locus of control into the model and examines its boundary role in the process through which algorithmic management influences felt responsibility for constructive change. The specific model is presented below (see Figure 1).

In summary, this study makes three main contributions. First, it introduces algorithmic management into research on the antecedents of voice, thereby addressing the limitation of existing voice research, which has focused primarily on traditional antecedents such as leader behavior, psychological safety, fairness perceptions, and individual motivation, while paying insufficient attention to algorithmic management factors [4,5]. More specifically, this study distinguishes among four dimensions of algorithmic management—algorithmic directing, algorithmic scheduling, algorithmic monitoring, and algorithmic feedback—and shows that different algorithmic functions convey different role-expectation signals to employees. In doing so, it extends research on the antecedents of voice in the context of algorithmic management. Second, drawing on signaling theory, this study uncovers the mechanism through which algorithmic management influences voice. Prior related research has mainly explained the relationship between algorithmic management and voice from perspectives such as fairness cognition [19]. By contrast, this study introduces felt responsibility for constructive change on the basis of signaling theory and argues that whether employees engage in voice depends not only on whether they detect problems or have suggestions, but also on whether the use of algorithmic management leads them to believe that they themselves ought to take responsibility for constructive change [14,15]. In this way, the theoretical explanation of how algorithmic management affects voice is advanced from fairness judgments to responsibility perceptions. Third, this study identifies work locus of control as a key boundary condition. Employees do not interpret the signals conveyed by algorithmic management in the same way; their control beliefs influence whether they understand algorithmic management as requiring “compliance and execution” or as inviting “participation in improvement,” which in turn shapes the process through which algorithmic management affects voice via felt responsibility for constructive change [17,18]. This finding further enriches our understanding of the boundary conditions of the relationship between algorithmic management and voice.

2. Theory and Hypotheses

This section develops the theoretical framework and hypotheses of the study. We first introduce signaling theory as the overarching theoretical lens, then explain how algorithmic management may influence voice. Next, we discuss felt responsibility for constructive change as the mediating mechanism and work locus of control as the boundary condition. Finally, we propose the moderated mediation hypotheses.

2.1. Signaling Theory

Signaling theory suggests that, under conditions of incomplete information, individuals rely on observable signals to infer the intentions, expectations, and likely responses of other actors [13]. In organizations, management practices not only coordinate work but also convey signals about what the organization values, what roles employees are expected to play, and what behaviors are considered appropriate [13]. Therefore, employees’ behavioral responses depend not only on the objective features of management practices, but also on how they interpret the signals embedded in those practices.

In the context of algorithmic management, the organization and its algorithmic management system jointly act as the signal sender. Although algorithmic systems operate through technical rules and data-based procedures, they are introduced and used by organizations to perform managerial functions such as directing, scheduling, monitoring, and feedback provision [6,7]. These four algorithmic functions can be understood as signal vehicles through which organizations communicate role expectations to employees. Specifically, directing, scheduling, monitoring, and feedback differ not only in their managerial functions, but also in what they imply about employees’ expected roles in the work system [20,21].

The signal content conveyed by algorithmic management mainly concerns employees’ role expectations. Algorithmic directing signals that employees are expected to follow prescribed procedures [21,22,23]; algorithmic scheduling signals that employees should respond to system-arranged workflows [24,25]; algorithmic monitoring signals that employees are expected to remain visible, measurable, and correctable [26,27,28]; and algorithmic feedback signals that employees should use performance information to adjust and improve their work [29,30]. From this perspective, algorithmic management functions differ not merely in what they technically do, but in what they imply about whether employees are expected to act as executors of system-defined arrangements or as participants in constructive improvement.

Employees are the signal receivers in this process. They interpret algorithmic management signals to understand their role boundaries, behavioral discretion, and responsibility within the work system. This signaling process is particularly relevant to voice because voice is a discretionary and change-oriented behavior through which employees express ideas, suggestions, or concerns intended to improve the organization or work unit [1,2]. When algorithmic management signals that employees should primarily comply with system instructions and predefined procedures, employees may be less likely to view speaking up as consistent with their role. In contrast, when algorithmic management signals that employees are expected to use information to improve their work, employees may be more likely to regard voice as an appropriate way to contribute to organizational improvement.

Importantly, signaling theory also suggests that signals do not produce uniform effects automatically; rather, their effects depend on how receivers interpret and respond to them [13]. Because work locus of control reflects employees’ beliefs about whether work outcomes are mainly determined by their own actions or by external forces, it may shape whether employees interpret algorithmic management signals as external constraints or as actionable information for improvement [17,18]. Thus, signaling theory provides the theoretical foundation for explaining not only how different algorithmic functions shape voice, but also why employees may respond differently to the same algorithmic management signals.

2.2. Algorithmic Management Functions and Voice

Algorithmic management refers to a digital form of management in which organizations use algorithmic systems, data-driven technologies, and artificial intelligence tools to direct, schedule, monitor, and provide feedback to employees in the work process [21]. Compared with traditional management, algorithmic management is more continuously embedded in employees’ daily work. It can specify how tasks should be performed, determine the sequence and timing of work, track employee behavior, and provide feedback based on performance data [7,20]. Therefore, algorithmic management is not only a technical means of improving efficiency, but also a managerial system that shapes how employees understand their roles and behavioral space.

Algorithmic management can be understood as comprising four core functions: algorithmic directing, algorithmic scheduling, algorithmic monitoring, and algorithmic feedback. Algorithmic directing refers to the algorithmic system’s specification of employees’ work methods, work processes, and work outcomes in advance. Algorithmic scheduling refers to the algorithmic system’s dynamic arrangement of employees’ task sequence, time allocation, work pace, and execution progress. Algorithmic monitoring refers to the algorithmic system’s continuous tracking, recording, and evaluation of employees’ behavioral trajectories, work processes, and performance outcomes. Algorithmic feedback refers to the algorithmic system’s provision of timely and data-based information about work results, deviations, and possible directions for improvement [21].

Employees’ voice refers to employees’ discretionary expression of work-related ideas, suggestions, or concerns with the intention of improving the organization or work unit [1,2]. Because voice is not a formally required behavior, employees usually need to judge whether speaking up is appropriate and consistent with their role before they engage in it [3,5].

Algorithmic management may influence this judgment by conveying different role-expectation signals. Some algorithmic functions may signal compliance and control by emphasizing system instructions and standardized work arrangements, whereas others may provide performance information and improvement cues. Therefore, whether algorithmic management suppresses or supports voice depends on the specific function performed by the algorithmic system and the role-expectation signal it conveys. Building on this logic, the following section further explains how different functions of algorithmic management influence employees’ felt responsibility for constructive change. Specifically, we argue that algorithmic management affects voice because it shapes whether employees perceive themselves as responsible for promoting constructive change in the workplace.

2.3. Algorithmic Management Functions and Felt Responsibility for Constructive Change

Building on the signaling theory, we further argue that these four algorithmic management functions are likely to have differentiated effects on employees’ felt responsibility for constructive change. Felt responsibility for constructive change captures employees’ subjective sense that they are personally responsible for bringing about beneficial change in the organization or work unit. Such responsibility is more likely to emerge when employees perceive that the work context provides room for initiative, cues for improvement, and expectations for constructive change [14,15]. In algorithmic management contexts, directing, scheduling, monitoring, and feedback convey different role-expectation signals and therefore may shape this responsibility cognition in different ways.

First, algorithmic directing specifies work methods, work processes, and work outcomes in advance [21,22]. When employees’ ways of performing tasks are largely prescribed by the system, their perceived discretion to adjust work practices or initiate improvements is likely to be compressed. Under such conditions, employees may be less likely to feel that they are personally responsible for changing or improving existing work arrangements, because the system has already defined how tasks should be performed and what outcomes should be achieved [20]. Thus, algorithmic directing may weaken employees’ felt responsibility for constructive change by reducing their perceived responsibility to initiate work-related improvements.

Second, algorithmic scheduling dynamically arranges task sequence, time allocation, work pace, and execution progress [24,25]. When employees must follow system-arranged workflows and temporal rhythms, they have limited discretion over how to organize their work and when to make adjustments. In this situation, employees may perceive that the coordination and optimization of work processes are mainly handled by the algorithmic system rather than by themselves [31,32]. As a result, their sense of personal responsibility for bringing about constructive change is likely to decline.

Third, algorithmic monitoring continuously tracks, records, and evaluates employees’ behaviors and performance [29,30]. Continuous visibility and evaluation may direct employees’ attention toward meeting metrics, avoiding deviations, and complying with standards. When employees are constantly observed and evaluated by the system, they may be less likely to regard themselves as responsible for questioning or improving existing work arrangements [33,34]. In this sense, algorithmic monitoring may weaken employees’ felt responsibility for constructive change by reinforcing a compliance-oriented understanding of their work role.

By contrast, algorithmic feedback conveys a different role-expectation signal. Rather than primarily prescribing, arranging, or observing employees’ behavior, feedback provides employees with information about work outcomes, performance gaps, deviation alerts, and possible directions for improvement [29,30]. Such information can make improvement needs more visible and help employees understand which aspects of their work may require adjustment [35]. Therefore, algorithmic feedback may strengthen employees’ felt responsibility for constructive change by making them more aware that work improvement is relevant to their own role and responsibility. Therefore, while algorithmic directing, scheduling, and monitoring are expected to weaken employees’ felt responsibility for constructive change, algorithmic feedback is expected to strengthen it. Accordingly, we propose:

Hypothesis 1: 

Algorithmic management functions are differentially related to felt responsibility for constructive change.

Hypothesis 1a: 

Algorithmic directing is negatively related to felt responsibility for constructive change.

Hypothesis 1b: 

Algorithmic scheduling is negatively related to felt responsibility for constructive change.

Hypothesis 1c: 

Algorithmic monitoring is negatively related to felt responsibility for constructive change.

Hypothesis 1d: 

Algorithmic feedback is positively related to felt responsibility for constructive change.

2.4. Felt Responsibility for Constructive Change and Voice

We focus on felt responsibility for constructive change as the key psychological mechanism linking algorithmic management functions to voice. Voice is not a formally required behavior; rather, it involves employees voluntarily expressing suggestions, concerns, or ideas to improve the organization or work unit [1,3]. Therefore, employees are more likely to speak up when they believe that improving existing work practices is something they are personally responsible for, rather than something outside their role [14,15]. Thus, when employees experience stronger felt responsibility for constructive change, they are more likely to translate their concerns and improvement ideas into voice.

First, felt responsibility for constructive change strengthens employees’ sense of obligation toward organizational improvement, making them more likely to view voice as “something I ought to do” rather than “something I may or may not do.” Because voice is not a required in-role behavior, but rather a discretionary, extra-role, and somewhat risky proactive behavior [3], employees are more likely to break their silence and proactively offer ideas and suggestions only when they believe they should take responsibility for constructive change. Second, felt responsibility for constructive change motivates employees to move beyond merely “recognizing problems” toward “taking action.” It leads employees not only to perceive that there is room for improvement in the status quo, but also to believe that they are responsible for advancing such change [14]. Prior research has shown that employees’ felt responsibility for constructive change can significantly promote voice [36]. In addition, felt responsibility for constructive change enhances employees’ proactive orientation, making them more inclined to engage in promotive behaviors. Existing studies suggest that responsibility-related psychological states can activate proactive behavior and encourage employees to exhibit more promotive voice [5,16]. Therefore, when employees experience a stronger felt responsibility for constructive change, they are more likely to fulfill that responsibility through voice. Accordingly, we propose:

Hypothesis 2: 

Felt responsibility for constructive change is positively related to voice.

2.5. The Moderating Role of Work Locus of Control

Work locus of control reflects employees’ stable belief about whether work outcomes are mainly determined by their own actions or by external forces. Employees with an external work locus of control tend to view work outcomes as shaped by external arrangements, institutional rules, and system control, whereas employees with an internal work locus of control are more likely to believe that outcomes can be influenced by their own effort, judgment, and actions [17,18]. Because work locus of control shapes how employees interpret control, rules, and feedback cues in the workplace, it may influence how employees understand the role-expectation signals conveyed by algorithmic management [37,38].

For algorithmic directing, when work methods, processes, and outcomes are specified by algorithmic systems, externally oriented employees are more likely to view these prescriptions as requirements to follow system-defined rules. Relative to employees with an internal work locus of control, they may have a weaker sense that they are personally responsible for adjusting or improving existing work arrangements. In this way, an external work locus of control may amplify the negative relationship between algorithmic directing and felt responsibility for constructive change [39,40]. A similar logic applies to algorithmic scheduling. When task order, time allocation, and work pace are arranged by algorithmic systems, employees with an external work locus of control are more likely to see the work process as determined by the system rather than shaped by their own judgment. Because these employees tend to attribute work outcomes to external arrangements, system-driven scheduling may further reduce their sense of personal responsibility for improving how work is organized. Accordingly, an external work locus of control may strengthen the negative relationship between algorithmic scheduling and felt responsibility for constructive change [31,32].

Algorithmic monitoring may also have a stronger negative effect among employees with an external work locus of control. When employee behavior is continuously tracked, recorded, and evaluated, employees with an external work locus of control are more likely to interpret such monitoring as evidence that they are objects of supervision and correction. This interpretation makes compliance with system standards more salient than personal responsibility for constructive change. Hence, the negative relationship between algorithmic monitoring and felt responsibility for constructive change is expected to be stronger for employees with an external rather than an internal work locus of control [41,42]. By contrast, algorithmic feedback was more positively related to felt responsibility for constructive change among employees with an internal work locus of control. Employees with an internal work locus of control tend to believe that work outcomes can be improved through their own effort, judgment, and adjustment. When algorithmic systems provide timely and improvement-relevant feedback, these employees are more likely to regard such feedback as actionable information within their own sphere of responsibility. This interpretation can make them more likely to feel personally responsible for constructive change. Thus, the positive relationship between algorithmic feedback and felt responsibility for constructive change is stronger for employees with an internal work locus of control [43,44]. Accordingly, we propose:

Hypothesis 3: 

Work locus of control moderates the relationships between algorithmic management functions and felt responsibility for constructive change.

Hypothesis 3a: 

Work locus of control moderates the relationship between algorithmic directing and felt responsibility for constructive change; specifically, the negative relationship is stronger for employees with an external work locus of control than for those with an internal work locus of control.

Hypothesis 3b: 

Work locus of control moderates the relationship between algorithmic scheduling and felt responsibility for constructive change; specifically, the negative relationship is stronger for employees with an external work locus of control than for those with an internal work locus of control.

Hypothesis 3c: 

Work locus of control moderates the relationship between algorithmic monitoring and felt responsibility for constructive change; specifically, the negative relationship is stronger for employees with an external work locus of control than for those with an internal work locus of control.

Hypothesis 3d: 

Work locus of control moderates the relationship between algorithmic feedback and felt responsibility for constructive change; specifically, the positive relationship is stronger for employees with an internal work locus of control than for those with an external work locus of control.

2.6. Moderated Mediation

Building on Hypotheses 1, 2, and 3, and following the logic of signaling theory, we further propose that work locus of control moderates the indirect effects of algorithmic management functions on voice through felt responsibility for constructive change. Specifically, for employees with an external work locus of control, algorithmic directing, scheduling, and monitoring are more likely to be interpreted as externally imposed signals of rules, arrangements, and evaluation. These interpretations should strengthen the negative effects of these algorithmic functions on felt responsibility for constructive change and subsequently weaken voice. By contrast, for employees with an internal work locus of control, algorithmic feedback is more likely to be interpreted as actionable information that can be used for work adjustment and improvement. This interpretation should strengthen the positive effect of algorithmic feedback on felt responsibility for constructive change and subsequently promote voice. Accordingly, we propose:

Hypothesis 4: 

Work locus of control moderates the indirect effects of algorithmic management on voice through felt responsibility for constructive change.

Hypothesis 4a: 

The indirect negative effect of algorithmic directing on voice through felt responsibility for constructive change is stronger for employees with an external work locus of control than for those with an internal work locus of control.

Hypothesis 4b: 

The indirect negative effect of algorithmic scheduling on voice through felt responsibility for constructive change is stronger for employees with an external work locus of control than for those with an internal work locus of control.

Hypothesis 4c: 

The indirect negative effect of algorithmic monitoring on voice through felt responsibility for constructive change is stronger for employees with an external work locus of control than for those with an internal work locus of control.

Hypothesis 4d: 

The indirect positive effect of algorithmic feedback on voice through felt responsibility for constructive change is stronger for employees with an internal work locus of control than for those with an external work locus of control.

3. Method

This section describes the empirical design of the study. We first introduce the research context, participants, and data collection procedure, and then present the measurement of the key variables and the analytical strategy used to test the proposed model.

3.1. Participants and Procedure

This study surveyed employees working at Meituan service stations in Beijing and Weihai, China. As a representative platform enterprise, Meituan relies extensively on algorithmic systems for task allocation, work arrangement, process supervision, performance evaluation, and feedback provision. Before the formal survey, we conducted preliminary field visits and interviews with station managers and employees to understand how algorithmic management operated in this context and to confirm the fit between the research design and the survey setting. The participants included delivery workers, service station employees, station operation personnel, and administrative staff involved in station-level operations. Although their job duties differed, they were all directly or indirectly connected to algorithmically managed platform work, such as task assignment, performance feedback, workflow support, and communication with managers. Despite relatively strong algorithmic management, employees still had opportunities to express work-related suggestions about station workflows, performance feedback and coordination problems through work meetings, feedback mechanisms, and suggestion boxes. Therefore, the company’s institutional arrangements and the characteristics of the employee sample provide an appropriate empirical setting for examining how algorithmic management shapes voice.

During the implementation stage, with the assistance of station managers, the research team organized employees to complete the questionnaires collectively during morning meetings. Before the survey began, the researchers explained the purpose of the study, the instructions for completion, and relevant precautions to the participants. They also emphasized that the study was intended solely for academic analysis, that participation was voluntary, and that all information would be processed anonymously and kept strictly confidential. To reduce potential common method bias, we used a three-wave, two-source matched survey design, with a two-week interval between adjacent waves. This design introduced temporal separation among key measurements and helped reduce respondents’ tendency to provide consistent answers across waves. In addition, voice was rated by supervisors, which further helped alleviate single-source bias. The two-week interval was selected to balance two considerations: ensuring sufficient temporal separation between measurements and minimizing respondent attrition as well as unrelated changes in the work context during the survey period [45,46].

To ensure accurate matching of responses across the different waves, the research team assigned each participant a unique identification code and asked them to enter the same code each time they completed a questionnaire so that the responses could later be matched across waves. All questionnaires were administered in paper form and were distributed and collected in a coordinated manner by station managers. During the survey process, research assistants and company staff were present on-site to help address participants’ questions regarding item interpretation or the completion procedure. As a token of appreciation, participants received RMB 20 for each completed wave of the survey.

In the first wave, employees reported their perceptions of algorithmic directing, algorithmic scheduling, algorithmic monitoring, algorithmic feedback, algorithm trust, algorithm transparency, algorithm explainability, and algorithm effectiveness when using the platform, along with their individual work locus of control and demographic information. In the second wave, employees who had completed the first-wave survey were invited to complete a questionnaire assessing their felt responsibility for constructive change. In the third wave, their direct supervisors were asked to evaluate employees’ voice. A total of 419 matched questionnaires were distributed to employees and their immediate supervisors. After three rounds of matching and data screening, cases with unsuccessful matching, substantial missing information, or invalid responses were excluded, resulting in 351 valid matched questionnaires and an effective response rate of 83.8%.

Among the final employee sample, 181 were men (51.6%) and 170 were women (48.4%). In terms of education, 19 participants (5.4%) had a high school education or below, 235 (67.0%) had an associate degree, 51 (14.5%) had a bachelor’s degree, 15 (4.3%) had a master’s degree, and 31 (8.8%) had a doctoral degree. The average age was 34.79 years (SD = 8.17), and the average work tenure was 17 months (SD = 13.78).

3.2. Measures

We followed a rigorous translation and back-translation procedure to adapt the original English questionnaire for use with the Chinese sample [47]. First, two researchers proficient in both Chinese and English translated the English items into Chinese. Then, two bilingual scholars with expertise in organizational behavior translated the Chinese items back into English. All authors subsequently reviewed and evaluated the Chinese translation. Unless otherwise noted, all variables were measured using a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree).

Algorithmic directing

We used the four-item scale developed by Röttgen et al. (2025) to assess algorithmic directing in employees’ work processes [21]. A sample item is: “The algorithmic system determines which approaches or methods I should use to complete my work.” (Cronbach’s α = 0.929).

Algorithmic scheduling

Algorithmic scheduling in employees’ work processes was measured using the three-item scale developed by Röttgen et al. (2025) [21]. A sample item is: “The algorithmic system determines when I am allowed to take a break or pause my work.” (Cronbach’s α = 0.894).

Algorithmic monitoring

The two-item scale developed by Röttgen et al. (2025) was employed to assess algorithmic monitoring in employees’ work processes [21]. A sample item is: “The algorithmic system continuously collects and analyzes data (e.g., operational records, online behavior, or video data) to evaluate my work performance.” (Cronbach’s α = 0.885).

Algorithmic feedback

To assess algorithmic feedback in employees’ work processes, this study drew on the two-item scale developed by Röttgen et al. (2025) [21]. A sample item is: “The algorithmic system provides me with evaluations or feedback based on my work outcomes.” (Cronbach’s α = 0.876).

Work locus of control

Employees’ individual work locus of control was assessed with the 16-item scale developed by Spector et al. (1988) [18]. In this study, higher scores on work locus of control indicate a stronger external work locus of control, whereas lower scores indicate a stronger internal work locus of control. Accordingly, in the following interpretation of simple slopes, high and low levels of work locus of control are treated as corresponding to employees with an external work locus of control and employees with an internal work locus of control, respectively. A sample item is: “Whether one gets the job one wants depends largely on luck.” (Cronbach’s α = 0.959).

Felt responsibility for constructive change

For felt responsibility for constructive change, we used the five-item scale developed by Morrison and Phelps (1999) [15]. A sample item is: “I feel a personal sense of responsibility for bringing about change at work.” (Cronbach’s α = 0.922).

Voice

Voice was measured with the six-item scale employed by Van Dyne and LePine (1998) [48]. A sample item is: “I develop and make recommendations concerning issues that affect my work group.” (Cronbach’s α = 0.918).

Control variables

In the analyses, we controlled for gender, age, education level, and work tenure. In addition, because employees’ basic perceptions and evaluations of algorithmic systems may shape their overall responses to algorithmic management and thereby confound the relationships among the focal variables, we also controlled for employees’ perceived algorithm transparency, algorithm explainability, algorithm trust, and algorithm effectiveness. Prior research has shown that algorithm transparency, explainability, and trust constitute important foundations for individuals’ understanding and evaluation of algorithmic systems, and that these factors influence individuals’ acceptance of algorithmic systems, their judgments about them, and their subsequent behavioral responses [49]. At the same time, individuals’ perceptions of a system’s effectiveness in terms of capability and job performance can also affect their overall evaluation of the work system and their subsequent behavioral responses [50]. Therefore, when examining the effects of algorithmic management on employees’ felt responsibility for constructive change and voice, controlling for the above variables helps reduce the possibility that employees’ general cognitive evaluations of algorithmic systems will confound the findings.

In terms of operationalization, age and work tenure were measured in actual years. Gender was coded as a dummy variable, with 1 = “male” and 2 = “female.” Education level was coded on a five-point scale ranging from 1 = “high school or below” to 5 = “doctoral degree.” Specifically, perceived algorithm transparency, algorithm explainability, and algorithm trust were each measured using three-item scales developed by Shin [49]. A sample item for perceived algorithm transparency is: “The evaluation criteria and operating rules of the algorithmic system in use should be publicly disclosed and understandable to people” (Cronbach’s α = 0.870). A sample item for algorithm explainability is: “I believe that the services provided by the algorithmic system are explainable and understandable” (Cronbach’s α = 0.873). A sample item for algorithm trust is: “The service outcomes provided by the algorithmic system are trustworthy” (Cronbach’s α = 0.889). Algorithm effectiveness was measured using the five-item scale developed by De Jong [50]. A sample item is: “The algorithmic system can effectively handle various problems that may arise in the course of work” (Cronbach’s α = 0.919). In addition, in the robustness tests, the main findings did not change substantively after these control variables were excluded.

4. Results

This section reports the empirical results. We first present descriptive statistics and correlations, then assess the measurement model and common method bias, and finally test the direct, mediating, moderating, and moderated mediation hypotheses.

4.1. Analytical Strategy

This study used IBM SPSS Statistics 27 for data analysis and employed the PROCESS macro to test the research hypotheses [51]. First, multiple linear ordinary least squares (OLS) regression analyses were conducted to examine the effects of the four dimensions of algorithmic management on employees’ felt responsibility for constructive change, as well as the effect of felt responsibility for constructive change on voice. Second, to test the moderating role of work locus of control and the moderated mediation effects, algorithmic directing, algorithmic scheduling, algorithmic monitoring, and algorithmic feedback were entered separately as the independent variable, and PROCESS Model 7 was used for the analyses. This model is suitable for examining whether work locus of control moderates the effects of different dimensions of algorithmic management on employees’ felt responsibility for constructive change, and whether such moderation further extends to voice. To assess the significance of the indirect effects, conditional indirect effects, and the differences between them, this study used a bias-corrected bootstrap procedure with 5000 resamples and constructed 95% confidence intervals [52]. If the confidence interval did not include zero, the corresponding effect was considered statistically significant.

4.2. Correlation Analysis

As shown in

Table 1. Means, standard deviations, and correlations.

	Mean	SD	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
Transparency	3.99	1.57	0.870
Explainability	3.93	1.56	0.037	0.873
Trust	4.06	1.56	0.118 *	0.069	0.889
Efficacy	4.01	1.55	−0.033	0.105 *	−0.090	0.919
Gender	1.48	0.50	0.079	−0.099	0.008	−0.003	--
Age	34.79	8.17	−0.057	0.015	0.038	−0.029	−0.113 *	--
Education level	2.44	0.99	0.042	0.002	0.017	−0.017	−0.018	0.201 **	--
Tenure	17.96	13.78	−0.048	0.049	0.030	−0.003	−0.081	0.859 **	0.108 *	--
Algorithmic Directing	4.01	1.72	0.027	0.014	−0.017	0.020	0.022	−0.038	−0.064	−0.003	0.929
Algorithmic Scheduling	3.96	1.72	0.073	−0.047	−0.047	0.062	−0.014	−0.069	−0.093	−0.044	0.582 **	0.894
Algorithmic Monitoring	3.69	1.79	0.015	0.010	−0.032	0.040	0.080	−0.010	−0.041	−0.009	0.492 **	0.471 **	0.885
Algorithmic Feedback	3.84	1.77	−0.001	−0.019	0.061	−0.056	−0.048	−0.011	0.124 *	0.006	−0.218 **	−0.287 **	−0.314 **	0.876
Work Locus of Control	4.03	1.53	0.057	0.065	−0.068	0.004	−0.019	−0.085	−0.026	−0.091	−0.073	−0.009	−0.105	0.113 *	0.959
Felt Responsibility for Constructive Change	4.00	1.64	0.000	0.048	0.071	0.005	−0.059	0.036	0.010	0.011	−0.681 **	−0.614 **	−0.563 **	0.366 **	0.115 *	0.922
Voice	4.21	1.60	−0.003	0.027	0.078	−0.038	−0.037	−0.073	−0.023	−0.056	−0.401 **	−0.370 **	−0.364 **	0.281 **	0.119 *	0.441 **	0.918

Newcomer’s Gender: 1 = female; 2 = male. Newcomer’s Education level: 1 = middle school and below; 2 = high school; 3 = associate degree; 4 = bachelor’s degree; 5 = master’s degree or above. The bold italic values on the diagonal indicate the Cronbach’s α coefficients for each variable. ** p < 0.01, * p < 0.05 (two-tailed tests).

, the correlations among algorithmic directing, algorithmic scheduling, algorithmic monitoring, algorithmic feedback, work locus of control, felt responsibility for constructive change, voice, and the control variables are presented. All correlation coefficients were significant at the 0.01 level, providing preliminary support for the proposed hypotheses. Specifically, algorithmic directing was negatively correlated with felt responsibility for constructive change (r = −0.681, p < 0.01), providing initial support for Hypothesis 1a. Algorithmic scheduling was negatively correlated with felt responsibility for constructive change (r = −0.614, p < 0.01), providing initial support for Hypothesis 1b. In addition, algorithmic monitoring was negatively correlated with felt responsibility for constructive change (r = −0.563, p < 0.01), providing initial support for Hypothesis 1c. Algorithmic feedback was positively correlated with felt responsibility for constructive change (r = 0.366, p < 0.01), providing initial support for Hypothesis 1d. Finally, felt responsibility for constructive change was positively correlated with voice (r = 0.441, p < 0.01), providing initial support for Hypothesis 2.

4.3. Confirmatory Factor Analysis

Before testing the hypotheses, we conducted a confirmatory factor analysis (CFA) using Mplus 8.3 to ensure that the seven core variables—algorithmic directing, algorithmic scheduling, algorithmic monitoring, algorithmic feedback, work locus of control, felt responsibility for constructive change, and voice—demonstrated adequate discriminant validity. The results of the measurement model indicated good discriminant validity among the variables. As shown in

Table 2. Confirmatory factor analysis.

Model	χ2	df	χ2/df	CFI	TLI	RMSEA	SRMR
Seven-factor Model: DI, SCH, MO, FB, FRCC, VO, WLC	798.779	644	1.510	0.976	0.974	0.026	0.032
Six-factor Model: DI+SCH, MO, FB, FRCC, VO, WLC	1061.782	650	1.517	0.936	0.931	0.042	0.043
Five-factor Model: DI+SCH+MO, FB, FRCC, VO, WLC	1244.585	655	1.518	0.909	0.902	0.051	0.049
Four-factor Model: DI+SCH+MO+FB, FRCC, VO, WLC	1460.146	659	1.518	0.876	0.868	0.059	0.058
Three-factor Model: DI+SCH+MO+FB+FRCC, VO, WLC	2103.771	662	1.524	0.777	0.763	0.079	0.079
Two-factor Model: DI+SCH+MO+FB+FRCC+VO, WLC	2287.133	664	1.530	0.748	0.734	0.083	0.082
One-factor Model: DI+SCH+MO+FB+FRCC+VO+WLC	4898.874	665	1.513	0.344	0.306	0.135	0.239

N = 351. DI = Algorithmic Directing; SCH = Algorithmic Scheduling; MO = Algorithmic Monitoring; FB = Algorithmic Feedback; FRCC = Felt Responsibility for Constructive Change; WLC = Work Locus of Control; VO = Voice.

, the seven-factor model exhibited the following fit indices: χ² = 798.779, df = 644, CFI = 0.976, TLI = 0.974, RMSEA = 0.026, and SRMR = 0.032. All of these values met commonly accepted thresholds for good model fit and were significantly better than those of the alternative models, suggesting that the study variables demonstrated satisfactory discriminant validity at the statistical level.

4.4. Hypothesis Testing

Next, we tested the four hypotheses proposed in this study. First, we examined the relationships between the four algorithmic management and employees’ felt responsibility for constructive change (see

Table 3. Results of path analyses.

Variable	Felt Responsibility for Constructive Change								Voice
	Model	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9
Gender	−0.129 (0.130)	−0.232 (0.140)	−0.015 (0.148)	−0.098 (0.166)	−0.114 (0.127)	−0.194 (0.138)	−0.014 (0.147)	−0.105 (0.165)	−0.069 (0.156)
Age	0.005 (0.016)	0.009 (0.017)	0.023 (0.018)	0.030 (0.020)	0.001 (0.016)	0.004 (0.017)	0.019 (0.018)	0.032 (0.020)	−0.029 (0.019)
Education level	−0.066 (0.067)	−0.091 (0.073)	−0.044 (0.076)	−0.091 (0.086)	−0.038 (0.066)	−0.079 (0.071)	−0.022 (0.076)	−0.089 (0.086)	−0.009 (0.081)
Tenure	−0.002 (0.009)	−0.007 (0.010)	−0.011 (0.010)	−0.014 (0.012)	0.001 (0.009)	−0.002 (0.010)	−0.009 (0.010)	−0.016 (0.012)	0.007 (0.011)
Transparency	0.016 (0.041)	0.052 (0.045)	0.005 (0.047)	0.001 (0.053)	0.005 (0.041)	0.049 (0.044)	−0.002 (0.047)	−0.001 (0.052)	−0.014 (0.050)
Explainability	0.050 (0.042)	0.004 (0.045)	0.052 (0.048)	0.052 (0.053)	0.040 (0.041)	−0.004 (0.044)	0.048 (0.047)	0.052 (0.053)	0.004 (0.050)
Trust	0.060 (0.042)	0.044 (0.045)	0.053 (0.047)	0.049 (0.053)	0.056 (0.041)	0.053 (0.044)	0.060 (0.047)	0.056 (0.053)	0.050 (0.050)
Efficacy	0.021 (0.042)	0.052 (0.045)	0.032 (0.047)	0.030 (0.053)	0.019 (0.041)	0.058 (0.044)	0.020 (0.047)	0.035 (0.053)	−0.043 (0.050)
Algorithmic Directing	−0.650 (0.038) ***				−0.224 (0.108) *
Algorithmic Scheduling		−0.595 (0.041) ***				−0.248 (0.120) *
Algorithmic Monitoring			−0.515 (0.041) ***				−0.231 (0.112) *
Algorithmic Feedback				0.345 (0.047) ***				0.661 (0.134) ***
Work Locus of Control					0.460 (0.103) ***	0.419 (0.109) ***	0.313 (0.105) **	0.346 (0.116) **
Felt Responsibility for Constructive Change									0.431 (0.047) ***
Algorithmic Directing× Work Locus of Control					−0.093 (0.022) ***
Algorithmic Scheduling× Work Locus of Control						−0.075 (0.025) **
Algorithmic Monitoring× Work Locus of Control							−0.064 (0.024) **
Algorithmic Feedback× Work Locus of Control								−0.074 (0.029) *
R2	0.473	0.391	0.326	0.148	0.503	0.418	0.343	0.170	0.208

Note: N = 351. Without control variables, the model was also supported. *** p < 0.001, ** p < 0.01, * p < 0.05.

). In Model 1, the maximum VIF was 4.079, below the commonly used threshold of 10, indicating that multicollinearity was not a serious concern [53,54]. Algorithmic directing was negatively related to employees’ felt responsibility for constructive change (β = −0.650, p < 0.001), supporting Hypothesis 1a. In Model 2, the maximum VIF was 4.074, below the threshold of 10. Algorithmic scheduling also showed a significant negative relationship with felt responsibility for constructive change (β = −0.595, p < 0.001), supporting Hypothesis 1b. In Model 3, the maximum VIF was 4.068, suggesting no problematic multicollinearity. Algorithmic monitoring was negatively related to felt responsibility for constructive change (β = −0.515, p < 0.001), supporting Hypothesis 1c. Finally, in Model 4, the maximum VIF was 4.088, still below the threshold of 10. Algorithmic feedback was positively related to felt responsibility for constructive change (β = 0.345, p < 0.001), supporting Hypothesis 1d. Then, the maximum VIF for Model 9 was 4.077, suggesting that multicollinearity was not a serious concern in this model. Employees’ felt responsibility for constructive change was positively related to voice (β = 0.431, p < 0.001), supporting Hypothesis 2.

To further examine the conditional effects, we used bootstrapping procedures to estimate the direct and indirect effects of algorithmic management functions at different levels of work locus of control. The results are presented in

Table 4. Summary of direct and indirect effects at different values of work locus of control.

Level	Effect	S.E.	95% CI
Conditional direct effects at different values of Work locus of control
(Algorithmic Directing → Felt Responsibility for Constructive Change)
−1 SD	−0.456	0.059	[−0.572, −0.340]
+1 SD	−0.741	0.043	[−0.826, −0.655]
(Algorithmic Scheduling → Felt Responsibility for Constructive Change)
−1 SD	−0.436	0.065	[−0.564, −0.309]
+1 SD	−0.667	0.047	[−0.760, −0.575]
(Algorithmic Monitoring → Felt Responsibility for Constructive Change)
−1 SD	−0.391	0.060	[−0.509, −0.273]
+1 SD	−0.587	0.050	[−0.685, −0.489]
(Algorithmic Feedback → Felt Responsibility for Constructive Change)
−1 SD	0.476	0.072	[0.336, 0.617]
+1 SD	0.249	0.058	[0.136, 0.363]
Conditional indirect effects at different values of Work locus of control
(Algorithmic Directing → Felt Responsibility for Constructive Change → Voice)
−1 SD	−0.136	0.039	[−0.219, −0.065]
+1 SD	−0.222	0.055	[−0.329, −0.113]
Difference	−0.085	0.031	[−0.156, −0.032]
(Algorithmic Scheduling → Felt Responsibility for Constructive Change → Voice)
−1 SD	−0.144	0.037	[−0.224, −0.078]
+1 SD	−0.220	0.049	[−0.320, −0.127]
Difference	−0.076	0.034	[−0.150, −0.017]
(Algorithmic Monitoring → Felt Responsibility for Constructive Change → Voice)
−1 SD	−0.132	0.033	[−0.200, −0.072]
+1 SD	−0.199	0.042	[−0.287, −0.120]
Difference	−0.066	0.030	[−0.133, −0.017]
(Algorithmic Feedback → Felt Responsibility for Constructive Change → Voice)
−1 SD	0.182	0.037	[0.116, 0.259]
+1 SD	0.095	0.036	[0.030, 0.169]
Difference	−0.087	0.041	[−0.171, −0.010]

N = 351.

.

As shown in Table 3, the interaction term between algorithmic directing and work locus of control had a significant negative effect on employees’ felt responsibility for constructive change (β = −0.093, p < 0.001). The maximum VIF for Model 5 was 4.094 suggesting that multicollinearity was not a serious concern in this model. To present this moderating effect more intuitively, we further used the bootstrap method to estimate the conditional effect of algorithmic directing at different levels of work locus of control. In Table 4, the results showed that when work locus of control was at a relatively low level (−1 SD, indicating a more internal work locus of control), algorithmic directing had a significant negative effect on employees’ felt responsibility for constructive change (β = −0.456, 95% CI [−0.572, −0.340]). When work locus of control was at a relatively high level (+1 SD, indicating a more external work locus of control), this negative effect became even stronger (β = −0.741, 95% CI [−0.826, −0.655]). At the same time, the simple slope analysis shown in Figure 2 also indicated that, compared with employees with an internal work locus of control, those with an external work locus of control were more likely to experience a reduction in felt responsibility for constructive change when facing algorithmic directing. Thus, Hypothesis 3a was supported.

Second, we examined the boundary role of work locus of control in the relationship between algorithmic scheduling and employees’ felt responsibility for constructive change. As shown in Table 3, the coefficient for the interaction term between algorithmic scheduling and work locus of control was significantly negative (β = −0.075, p < 0.01), indicating that work locus of control significantly strengthened the inhibiting effect of algorithmic scheduling on employees’ felt responsibility for constructive change. For Model 6, the maximum VIF was 4.115; no problematic multicollinearity was detected. In Table 4, Further conditional effect analyses showed that when work locus of control was at a relatively low level (−1 SD, indicating a more internal work locus of control), algorithmic scheduling had a significant negative effect on employees’ felt responsibility for constructive change (β = −0.436, 95% CI [−0.564, −0.309]). When work locus of control was at a relatively high level (+1 SD, indicating a more external work locus of control), this negative effect became more pronounced (β = −0.667, 95% CI [−0.760, −0.575]). The simple slope results shown in Figure 3 further confirmed this pattern: compared with employees with an internal work locus of control, those with an external work locus of control were more likely to show a decline in felt responsibility for constructive change in the context of algorithmic scheduling. Therefore, Hypothesis 3b was supported.

Third, the results for the moderating effect of algorithmic monitoring showed that the interaction term between algorithmic monitoring and work locus of control had a significant negative effect on employees’ felt responsibility for constructive change (β = −0.064, p < 0.01). This indicates that, compared with employees with an internal work locus of control, those with an external work locus of control were more likely to experience a weakening of felt responsibility for constructive change when exposed to algorithmic monitoring. The collinearity diagnostics for Model 7 showed a maximum VIF of 4.094, indicating that the model did not suffer from serious multicollinearity. Specifically, when work locus of control was at the −1 SD level (i.e., a more internal work locus of control), algorithmic monitoring had a significant negative effect on employees’ felt responsibility for constructive change (β = −0.391, 95% CI [−0.509, −0.273]); when work locus of control was at the +1 SD level (i.e., a more external work locus of control), this negative effect became even stronger (β = −0.587, 95% CI [−0.685, −0.489]). The simple slope plot in Figure 4 likewise shows that, compared with employees with an internal work locus of control, the negative relationship between algorithmic monitoring and employees’ felt responsibility for constructive change was steeper for employees with an external work locus of control. Therefore, Hypothesis 3c was supported.

Finally, we examined the moderating role of work locus of control in the relationship between algorithmic feedback and employees’ felt responsibility for constructive change. In Model 8, the results showed that the interaction term between algorithmic feedback and work locus of control was significantly negative (β = −0.074, p < 0.01), indicating that work locus of control weakened the positive effect of algorithmic feedback on employees’ felt responsibility for constructive change. The maximum VIF reached 4.097, still lower than the standard threshold of 10, suggesting that collinearity was unlikely to bias the estimation results. Further analyses showed that when work locus of control was at a relatively low level (−1 SD, i.e., a more internal work locus of control), algorithmic feedback had a significant positive effect on employees’ felt responsibility for constructive change (β = 0.476, 95% CI [0.336, 0.617]); when work locus of control was at a relatively high level (+1 SD, i.e., a more external work locus of control), this positive effect remained significant but was weaker in magnitude (β = 0.249, 95% CI [0.136, 0.363]). Combined with the simple slope analysis in Figure 5, the results suggest that, compared with employees with an external work locus of control, those with an internal work locus of control were more likely to interpret algorithmic feedback as guidance for participating in improvement, thereby developing a stronger felt responsibility for constructive change. Therefore, Hypothesis 3d was supported.

Furthermore, we used the bootstrap method to test the moderating effect of work locus of control on the indirect effects of the different dimensions of algorithmic management on voice through employees’ felt responsibility for constructive change. As shown in Table 4, when work locus of control was at a relatively low level (−1 SD, i.e., a more internal work locus of control), the indirect effect of algorithmic directing on voice through employees’ felt responsibility for constructive change was significantly negative (β = −0.136, 95% CI [−0.218, −0.064]); when work locus of control was at a relatively high level (+1 SD, i.e., a more external work locus of control), this indirect negative effect became stronger (β = −0.222, 95% CI [−0.329, −0.113]). Moreover, the difference between the indirect effects under the two conditions was significant (Δβ = −0.085, 95% CI [−0.156, −0.032]). These results indicate that, compared with employees with an internal work locus of control, algorithmic directing had a stronger indirect negative effect on the voice of employees with an external work locus of control through employees’ felt responsibility for constructive change. Thus, Hypothesis 4a was supported.

Similarly, the conditional indirect effect of algorithmic scheduling on voice through employees’ felt responsibility for constructive change also differed significantly across levels of work locus of control. Specifically, when work locus of control was at a relatively low level (−1 SD, i.e., a more internal work locus of control), the indirect effect of algorithmic scheduling on voice through employees’ felt responsibility for constructive change was significantly negative (β = −0.144, 95% CI [−0.224, −0.078]); when work locus of control was at a relatively high level (+1 SD, i.e., a more external work locus of control), this indirect negative effect was stronger (β = −0.220, 95% CI [−0.320, −0.127]). Further comparison showed that the difference between the two was significant (Δβ = −0.076, 95% CI [−0.150, −0.017]). Therefore, compared with employees with an internal work locus of control, algorithmic scheduling exerted a stronger indirect negative effect on the voice of employees with an external work locus of control through employees’ felt responsibility for constructive change, providing support for Hypothesis 4b.

With respect to algorithmic monitoring, the results likewise showed that its indirect effect on voice through employees’ felt responsibility for constructive change was significantly moderated by work locus of control. When work locus of control was at a relatively low level (−1 SD, i.e., a more internal work locus of control), the indirect effect of algorithmic monitoring on voice through employees’ felt responsibility for constructive change was significantly negative (β = −0.132, 95% CI [−0.200, −0.072]); when work locus of control was at a relatively high level (+1 SD, i.e., a more external work locus of control), this indirect negative effect became even stronger (β = −0.199, 95% CI [−0.287, −0.120]). In addition, the difference in the indirect effects across levels of work locus of control was significant (Δβ = −0.066, 95% CI [−0.133, −0.017]). These results indicate that, compared with employees with an internal work locus of control, those with an external work locus of control were more likely to reduce their voice in the context of algorithmic monitoring because of a decline in felt responsibility for constructive change. Thus, Hypothesis 4c was supported.

Finally, we examined the moderating effect of work locus of control on the indirect effect of algorithmic feedback on voice through employees’ felt responsibility for constructive change. The results showed that when work locus of control was at a relatively low level (−1 SD, i.e., a more internal work locus of control), the indirect effect of algorithmic feedback on voice through employees’ felt responsibility for constructive change was significantly positive (β = 0.182, 95% CI [0.116, 0.259]); when work locus of control was at a relatively high level (+1 SD, i.e., a more external work locus of control), this indirect positive effect remained significant, but its magnitude was weaker (β = 0.095, 95% CI [0.030, 0.169]). At the same time, the difference between the indirect effects under the two conditions was significant (Δβ = −0.087, 95% CI [−0.171, −0.010]). This means that, compared with employees with an external work locus of control, algorithmic feedback had a stronger indirect positive effect on the voice of employees with an internal work locus of control through employees’ felt responsibility for constructive change. Therefore, Hypothesis 4d was supported.

4.5. Supplementary Analyses

To assess the robustness of our findings, we conducted a comparative analysis between two nested models: (1) the full model incorporating all hypothesized predictors and control variables, including gender, age, education, tenure, employees’ perceptions of algorithm transparency, algorithm explainability, algorithm trust, and algorithm effectiveness; (2) the reduced model excluding all control variables. The results showed substantial consistency across the two models. Specifically, after excluding the control variables, the direction, magnitude, and statistical significance of the focal relationships remained largely unchanged. Hypotheses H1–H4 were consistently supported in both models, these results suggest that our main findings are robust regardless of whether the control variables are included [55].

5. Discussion

Against the backdrop of increasing digitalization in organizational management, algorithmic management has not only changed how organizations allocate tasks, control processes, monitor behavior, and provide performance feedback, but has also profoundly influenced how employees understand their roles within the organization and whether they are willing to proactively contribute to organizational improvement. Existing research on voice has mainly explained why employees speak up from traditional perspectives such as leader behavior, psychological safety, fairness perceptions, and individual motivation. However, research remains limited regarding how algorithmic management affects voice, through what psychological mechanisms it does so, and under what conditions these effects occur. To address these issues, drawing on signaling theory, this study developed a moderated mediation model linking algorithmic management to voice and conducted a field survey among employees of Meituan, a Chinese platform organization.

In Figure 6, the results showed that algorithmic directing negatively affected employees’ felt responsibility for constructive change (H1a), algorithmic scheduling negatively affected employees’ felt responsibility for constructive change (H1b), algorithmic monitoring negatively affected employees’ felt responsibility for constructive change (H1c), and algorithmic feedback positively affected employees’ felt responsibility for constructive change (H1d). Felt responsibility for constructive change positively affected voice (H2). Work locus of control moderated the relationship between algorithmic directing and felt responsibility for constructive change, such that the negative effect of algorithmic directing on felt responsibility for constructive change was stronger for employees with an external work locus of control than for those with an internal work locus of control (H3a). Work locus of control also moderated the relationship between algorithmic scheduling and felt responsibility for constructive change, such that the negative effect of algorithmic scheduling on felt responsibility for constructive change was stronger for employees with an external work locus of control than for those with an internal work locus of control (H3b). Likewise, work locus of control moderated the relationship between algorithmic monitoring and felt responsibility for constructive change, such that the negative effect of algorithmic monitoring on felt responsibility for constructive change was stronger for employees with an external work locus of control than for those with an internal work locus of control (H3c). In contrast, work locus of control moderated the relationship between algorithmic feedback and felt responsibility for constructive change, such that the positive effect of algorithmic feedback on felt responsibility for constructive change was stronger for employees with an internal work locus of control than for those with an external work locus of control (H3d).

Furthermore, work locus of control moderated the indirect effect of algorithmic directing on voice through felt responsibility for constructive change, such that the indirect negative effect of algorithmic directing on voice was stronger for employees with an external work locus of control than for those with an internal work locus of control (H4a). Work locus of control also moderated the indirect effect of algorithmic scheduling on voice through felt responsibility for constructive change, such that the indirect negative effect of algorithmic scheduling on voice was stronger for employees with an external work locus of control than for those with an internal work locus of control (H4b). Similarly, work locus of control moderated the indirect effect of algorithmic monitoring on voice through felt responsibility for constructive change, such that the indirect negative effect of algorithmic monitoring on voice was stronger for employees with an external work locus of control than for those with an internal work locus of control (H4c). Finally, work locus of control moderated the indirect effect of algorithmic feedback on voice through felt responsibility for constructive change, such that the indirect positive effect of algorithmic feedback on voice was stronger for employees with an internal work locus of control than for those with an external work locus of control (H4d). All hypotheses (H1–H4) were supported (

Table 5. The results of hypothesis testing.

Hypothesis	Results
Hypothesis 1a: Algorithmic directing is negatively related to felt responsibility for constructive change.	Supported
Hypothesis 1b: Algorithmic scheduling is negatively related to felt responsibility for constructive change.	Supported
Hypothesis 1c: Algorithmic monitoring is negatively related to felt responsibility for constructive change.	Supported
Hypothesis 1d: Algorithmic feedback is positively related to felt responsibility for constructive change.	Supported
Hypothesis 2: Felt responsibility for constructive change is positively related to voice.	Supported
H3a: Work locus of control moderates the relationship between algorithmic directing and felt responsibility for constructive change.	Supported
H3b: Work locus of control moderates the relationship between algorithmic scheduling and felt responsibility for constructive change.	Supported
H3c: Work locus of control moderates the relationship between algorithmic monitoring and felt responsibility for constructive change.	Supported
H3d: Work locus of control moderates the relationship between algorithmic feedback and felt responsibility for constructive change.	Supported
H4a: The indirect negative effect of algorithmic directing on voice through felt responsibility for constructive change is stronger for employees with an external work locus of control than for those with an internal work locus of control.	Supported
H4b: The indirect negative effect of algorithmic scheduling on voice through felt responsibility for constructive change is stronger for employees with an external work locus of control than for those with an internal work locus of control.	Supported
H4c: The indirect negative effect of algorithmic monitoring on voice through felt responsibility for constructive change is stronger for employees with an external work locus of control than for those with an internal work locus of control.	Supported
H4d: The indirect positive effect of algorithmic feedback on voice through felt responsibility for constructive change is stronger for employees with an internal work locus of control than for those with an external work locus of control.	Supported

).

5.1. Theoretical Implications

This study offers important theoretical contributions to the existing literature on algorithmic management and voice. First, it extends research on the antecedents of voice by incorporating algorithmic management as an important explanatory factor. Recent studies on voice have continued to show how contextual cues and managerial practices shape voice behavior by focusing on such issues as the interactive process of voice, voice motivation, voice efficacy, managerial responses, and the conditions under which employees speak up in specific contexts [56,57,58]. By comparison, although algorithmic management, as a digital managerial mechanism increasingly embedded in organizational operations, has already come to play an important role in task allocation, process coordination, behavioral supervision, and performance feedback, its effects on voice have yet to be fully understood [20,22]. This study introduces algorithmic management into the voice literature and further distinguishes among four dimensions—algorithmic directing, algorithmic scheduling, algorithmic monitoring, and algorithmic feedback. In doing so, it shows that different algorithmic functions convey different role-expectation signals to employees and therefore exert differentiated effects on voice. Accordingly, this study not only responds to the need to extend voice research into digital management contexts, but also enriches the perspective that views algorithmic management as an antecedent of voice.

Second, drawing on signaling theory, this study extends understanding of the mechanism through which algorithmic management influences voice. Existing research suggests that felt responsibility for constructive change is not merely a general positive attitude, but rather a form of responsibility cognition through which employees internalize improving and optimizing work as part of their own role obligations; this sense of responsibility can, in turn, stimulate voice [5,59,60]. On this basis, the present study introduces felt responsibility for constructive change as a mediating variable and argues that, in the context of algorithmic management, whether employees choose to speak up depends not only on their external perceptions of managerial arrangements, but also on whether they interpret the implementation of algorithmic management as signaling that the organization encourages participation in improvement, continuous optimization of work, and shared responsibility for constructive change. This perspective complements prior algorithmic management research that has mainly emphasized fairness, trust, opacity, and algorithmic evaluation as important bases of employee reactions [30]. The findings show that the implementation of algorithmic management shapes employees’ understanding of how work should be carried out and where role boundaries lie, thereby activating their felt responsibility for constructive change and, in turn, promoting voice. In this way, the study advances the explanation of the relationship between algorithmic management and voice to the level of how employees understand their own responsibility role in organizational improvement.

Third, this study further enriches research on the boundary conditions under which algorithmic management influences voice. Existing research has already begun to examine the contextual boundary conditions of this relationship. For example, Liang found that voice endorsement strengthens the effect of perceived algorithmic fairness on voice [19]. The present study shows that employees’ individual control beliefs shape how they interpret the signals conveyed by algorithmic management, and in turn influence the process through which algorithmic management affects voice via felt responsibility for constructive change. This finding is consistent with research suggesting that work locus of control reflects employees’ generalized beliefs about whether work outcomes are shaped by personal agency or external forces, and that such beliefs are important for understanding employees’ motivation, behavioral orientation, and responses to work contexts [19,37]. Specifically, the findings indicate that, compared with employees with an internal work locus of control, those with an external work locus of control are more likely to experience a reduction in felt responsibility for constructive change when facing algorithmic directing, algorithmic scheduling, and algorithmic monitoring. In contrast, compared with employees with an external work locus of control, those with an internal work locus of control are more likely to experience an increase in felt responsibility for constructive change when exposed to algorithmic feedback. In this way, the present study advances existing research on boundary conditions from the level of organizational response contexts to the level of employee individual differences, thereby offering a more fine-grained explanation for why employees may respond differently to the same algorithmic management arrangements.

5.2. Practical Implications

First, organizations should recognize algorithmic management as an important managerial antecedent of voice and pay close attention to the differentiated design of its various functions. As organizations advance the use of algorithmic management, they should avoid treating it as a uniform control tool. Instead, they should differentiate its design across functional dimensions, so that while maintaining efficiency, they can appropriately limit the intensity of algorithmic directing, algorithmic scheduling, and algorithmic monitoring, and fully leverage the positive role of algorithmic feedback. Doing so helps preserve the necessary space for employees to participate in improvement and express their ideas.

Second, when seeking to foster voice in the context of algorithmic management, organizations should place greater emphasis on cultivating employees’ felt responsibility for constructive change. In implementing algorithmic management, organizations should make full use of the informational support and improvement-oriented nature of algorithmic feedback by providing clear, timely, and actionable performance information and optimization cues, thereby strengthening employees’ felt responsibility for constructive change. At the same time, they should avoid allowing overly strong algorithmic directing, algorithmic scheduling, and algorithmic monitoring to continually reinforce employees’ self-perception as passive executors. Only when employees perceive from algorithmic management that the organization expects them to participate in improvement and assume responsibility for constructive change is their felt responsibility for constructive change likely to be activated.

Third, employees’ responses to algorithmic management vary substantially across individuals. Therefore, organizations should pay attention to differentiated management when implementing algorithmic management. For employees with a more external work locus of control, organizations should be cautious about the inhibiting effects of overly strong algorithmic control on their sense of responsibility and willingness to engage in voice. For employees with a more internal work locus of control, by contrast, clearer and more timely feedback can be used to stimulate their proactive orientation toward improvement. Only by aligning algorithmic management with employees’ individual characteristics can organizations more effectively realize the positive value of digital management and genuinely promote voice and continuous organizational improvement.

5.3. Limitations and Future Directions

Although this study has generated meaningful insights at both the theoretical and practical levels, several limitations remain and warrant further attention in future research. First, the study still has certain limitations with respect to causal inference. Although we employed a three-wave matched survey design and reduced common method bias to some extent through temporal separation and multisource data collection, a time-lagged design cannot fundamentally resolve issues of causal identification, and survey research still cannot fully rule out reverse causality or omitted-variable bias among the focal variables [61,62]. Future research could adopt longitudinal designs over a longer time span or combine methods such as scenario experiments, field experiments, and cross-lagged panel models to test more rigorously the causal mechanisms linking algorithmic management, felt responsibility for constructive change, and voice.

Second, this study focused on felt responsibility for constructive change as the key mediating mechanism. This focus is consistent with our signaling theory perspective, because algorithmic management functions convey role-expectation signals that shape whether employees perceive themselves as responsible for promoting constructive change. However, although this study identifies felt responsibility for constructive change as the core mediating mechanism based on signaling theory, it does not rule out the possibility of other mediating pathways from different theoretical perspectives. For example, employees’ perceptions of fairness and sense of gain may influence their voice and other proactive behaviors [6,10,63]. Therefore, future research could further examine these alternative mechanisms to provide a more comprehensive understanding of how algorithmic management influences voice.

Third, the empirical context of this study may limit the generalizability of the findings. On the one hand, this study was conducted at Meituan service stations in China, and the sample included delivery workers, service station employees, station operation personnel, and administrative staff involved in platform-based station operations. Although this highly algorithmized context is suitable for examining algorithmic management, employees in such settings may experience algorithmic directing, scheduling, monitoring, and feedback more intensively than employees in traditional organizations, manufacturing firms, or knowledge-intensive work settings [20]. We therefore encourage future research to further examine our model in traditional firms that have adopted algorithmic management practices, so as to assess whether the proposed relationships hold across different organizational contexts. On the other hand, because this study was conducted in the Chinese context, cultural features such as collectivism, high-context communication, and Confucian relational norms may shape how employees interpret algorithmic management signals and decide whether to speak up [63,64,65]. Therefore, future research could conduct cross-industry, cross-organization, and cross-national comparative studies to further assess the external validity of the present findings.