When AI agents need botsitting: supporting data and code
Description
This dataset supports the article “When AI agents need botsitting: Autonomy, orchestration, and process visibility in agentic information systems.” It contains de-identified analytic data and reproducibility materials for three studies examining how autonomous execution and inter-system orchestration in agentic information systems shape sense of agency and supervisory (botsitting) burden, and how process visibility affects requests for additional process review. Study 1 is a randomized 2 × 2 scenario experiment (N = 300); Study 2 is a scenario survey with latent-variable structural equation modeling (N = 896); Study 3 is a randomized process-visibility scenario (N = 759). The archive includes analytic CSV files, an Excel workbook, R scripts, verified primary outputs, measurement and data-integrity checks, robustness analyses, R environment information, a SHA-256 manifest, and license files. The final analytic samples comprise adults aged 18 years or older and contain no direct identifiers. Data, documentation, and analysis outputs are licensed under CC BY 4.0; R code is licensed under the MIT License.
Files
Steps to reproduce
Extract AI_Botsitting_supporting_data_code_v1.0.zip and run the scripts from the extracted directory using R. The supplied scripts use fixed seed 20260618. Commands: Rscript "study 1_Analysis.R" BOOTSTRAP_REPS=2000 BOOTSTRAP_CORES=8 Rscript "study 2_Analysis.R" BOOTSTRAP_REPS=2000 BOOTSTRAP_CORES=8 Rscript "study 3_Analysis.R" Rscript measurement_and_integrity_audit.R BOOTSTRAP_REPS=2000 BOOTSTRAP_CORES=8 Rscript robustness_audit.R Study 2, Study 3, and the logistic-outcome robustness analysis use 2,000 bootstrap draws. Bootstrap confidence intervals can differ in the last displayed digit across operating systems or parallel backends. Compare outputs with the supplied verified .txt files and use SHA256SUMS.txt to verify package integrity. session_info.txt records the verified R environment.