Examining Employee Reactions to AI Moral Voice

This dataset originates from a series of studies investigating employees' reactions to morally oriented advice generated by artificial intelligence (AI). It comprises data from three sub-studies. All data were collected from a sample of full-time employees in China via the specialized online survey platform Credamo.Data Generation Process and Handling Methods: Study 1 employed a cross-sectional survey methodology. Using the platform's screening tools, 300 participants meeting the predefined criteria (full-time employment, team-based work, bachelor's degree or higher) were recruited. Participants were initially asked whether they had received AI-generated moral advice in the past year. Those who responded "yes" were prompted to describe the specific incident in detail. Only responses where the advice involved explicit moral content (e.g., care, harm, justice) and provided a concrete description of the event were included in the final analysis. Ultimately, 174 valid questionnaires passed the attention checks and were used for analysis. Studies 2 and 3 utilized scenario-based experimental designs. In Study 2, participants were randomly assigned to one of three conditions: "AI Promotive Moral Voice," "AI Prohibitive Moral Voice," or a "Control" condition. Study 3 employed a 2 (advice type: promotive vs. prohibitive) × 2 (AI explainability: high vs. low) between-subjects design. Both experimental studies incorporated attention check items during data collection and excluded participants who failed these checks or exhibited clear patterned responses. The final valid sample sizes were 358 for Study 2 and 722 for Study 3.Data Content and Structure: The dataset primarily contains anonymized individual-level data collected during the research period. Data are stored in tabular format (.csv files). The total number of records is 174 for Study 1, 358 for Study 2, and 722 for Study 3. In each data table, rows represent individual participants, and columns represent variables,主要包括:Demographic Variables: Such as participant gender, age, education level, and work experience.Experimental Manipulation and Grouping Variables: Such as the "moral voice type" and "AI explainability" level in Studies 2 and 3.Psychological Mechanism Variables: Measured using established psychological scales, such as "human nature threat" and "moral reflection," typically using Likert 5-point or 7-point scales.Outcome Variables: Such as the degree of "voice endorsement."Control Variables: Such as negative attitudes toward AI.All psychological scale variables are dimensionless scores. Demographic variables like age and work experience are measured in "years."Data Quality Statement: Data quality was ensured through rigorous participant screening, random assignment, attention checks, and data cleaning procedures. Data missingness was minimal, primarily arising from occasional skipped items by participants, and was handled during analysis using methods like listwise deletion or maximum likelihood estimation. Potential measurement error primarily stems from the self-reported nature of the data, potentially introducing common method bias. We implemented measures in both the research design (e.g., temporal separation) and statistical controls to mitigate this concern. All data files can be opened and processed using common data analysis software such as SPSS, R, Python, or Excel.