Digital Human CEO Conjoint Allocation Experiment
Description
This repository contains the anonymized observed dataset, codebook, stimulus documentation, derived variables, bonus-return rules, manipulation checks, scale reliability summaries, robustness checks, regional summaries, and package-contrast outputs for the study “Do Investors Price Governance in Digital Human CEO Communication? A Multi-Regional Conjoint Allocation Experiment.” The study used a multi-regional, incentivized, repeated-choice conjoint allocation experiment with 960 completed retail-investor participants from five regional groups: United States, Italy, Rest of Europe, China, and Rest of Asia. Each participant completed eight paired allocation rounds, with two fictitious listed-firm profiles per round, producing 7,680 participant-round observations and 15,360 firm-profile evaluations in the completed dataset. Participants allocated a simulated USD 10,000 portfolio across Firm A stock, Firm B stock, a broad market ETF, a bond ETF, and cash or money market. Firm profiles varied randomized attributes related to Digital Human CEO communication and financial information, including disclosure, avatar style, explainability, named executive accountability, public redress, financial context, and financial fundamentals. The main behavioral outcome is USD allocation to the focal firm stock. The workbook includes the full completed dataset, participant-level data, stimulus attributes, avatar metadata, post-task measures, manipulation checks, reliability outputs, AMCE results, robustness checks, regional results, and full-governance versus weak-governance package contrasts. The dataset is anonymized and does not include names, emails, IP addresses, addresses, or other direct personal identifiers. The data are observed experimental responses.
Files
Steps to reproduce
To reproduce the study outputs, open the workbook and begin with the README and Codebook sheets, which describe the study design, observation units, variable definitions, coding conventions, exclusion flags, and sheet structure. Use Raw_Data as the main profile-level dataset. Each row represents one firm-profile evaluation within a paired allocation round. The completed dataset contains 960 participants, eight rounds per participant, two profiles per round, and 15,360 firm-profile rows. Define the headline completed sample as all 960 completed participants. Define the completed-valid main analysis sample by excluding only rows from participants flagged for attention failure or speeding, while retaining manipulation-check failures. This produces 926 participants and 14,816 profile evaluations. Define the strict robustness sample by excluding attention failures, speeders, and manipulation-check failures, producing 813 participants and 13,008 profile evaluations. Use allocation to the focal firm stock as the primary behavioral outcome. Estimate average marginal component effects by comparing mean focal-firm allocation across randomized attribute levels for disclosure, avatar style, explainability, accountability, redress, financial context, and fundamentals. Attribute coding is documented in Stimulus_Attributes, Derived_Variables, and Codebook. Reproduce the main AMCE table using the completed-valid main sample. Then reproduce the strict-sample estimates using the workbook-coded strict inclusion flag. Compare both outputs to the AMCE_Results and Robustness_Checks sheets. Reproduce package contrasts by comparing focal-firm allocation for profiles combining all five governance elements with profiles combining none of them. Use Package_Contrasts to verify the full-sample, routine-context, and adverse-context contrasts. Use Participant_Data to reproduce sample descriptives, regional counts, demographics, financial-literacy summaries, investment-experience indicators, and sample-inclusion counts. Use Manipulation_Checks to reproduce manipulation-check pass rates, and use Scale_Reliability and Post_Task_Items to reproduce post-task scale summaries and reliability results. Use Bonus_Return_Rules to verify that the performance bonus depends on financial fundamentals, asset class, and simulated market state, and not on governance attributes. Use Regional_Results only for exploratory regional summaries because the study was powered primarily for pooled inference. All reported manuscript values should be checked against the final workbook sheets before submission. The anonymized workbook is the source of truth for sample definitions, AMCEs, robustness checks, manipulation checks, scale reliabilities, package contrasts, and bonus-return rules.
Institutions
- Ca' Foscari University of VeniceVeneto, Venice