AI Authorship and Investor Allocation Dataset
Description
This dataset supports an experimental study examining how investors respond to AI-generated versus CEO-prepared earnings communication and to the disclosure of formal governance review. The study uses a 2×2 between-subjects randomized design with four conditions: CEO-prepared/no formal review, CEO-prepared/formal review, AI-generated/no formal review, and AI-generated/formal review. The experiment was designed to separate the effects of AI authorship from the effects of governance oversight, addressing whether investor reactions are driven by skepticism toward AI-generated communication, by the presence of formal review, or by the interaction between both factors. The analytic dataset includes anonymized respondent-level observations from investment-experienced participants. Variables include experimental condition indicators, portfolio allocation outcomes across common stock, corporate bond, and risk-free asset alternatives, and behavioral perception measures including trust/credibility, perceived accountability, governance oversight, expected return, perceived risk, investment attractiveness, and decision confidence. The dataset also includes basic investor background controls such as investment experience, financial knowledge, risk tolerance, familiarity with AI, and general trust in AI. The workbook contains a concise README/codebook, the cleaned analysis dataset, and a compact results summary. It is intended to support replication, transparency, and robustness checks for the associated manuscript on AI authorship, governance review, and investor allocation behavior in earnings communication.
Files
Steps to reproduce
Open the workbook and begin with the README_Codebook sheet to review the study design, variable definitions, coding rules, and sample construction. Use the Analysis_Data sheet as the primary respondent-level dataset. Each row represents one anonymized participant in the final analytic sample, and each participant is assigned to one of four experimental conditions: CEO_NoReview, CEO_Review, AI_NoReview, or AI_Review. To reproduce the descriptive results, group observations by Condition and calculate means and standard deviations for the main allocation outcomes: StockAllocation_USD, BondAllocation_USD, RiskFreeAllocation_USD, and TotalAureviaAllocation_USD. Repeat the same procedure for the behavioral perception measures, including TrustCredibilityIndex, AccountabilityIndex, GovernanceOversightIndex, ExpectedReturn_Pct, PerceivedRisk, InvestmentAttractiveness, and DecisionConfidence. To reproduce the main factorial tests, estimate ordinary least squares models using StockAllocation_USD as the primary dependent variable. The key predictors are Authorship_AI, GovernanceReview, and their interaction term AI_x_Review. The coefficient on Authorship_AI captures the AI-authorship effect when no formal review is disclosed. The coefficient on GovernanceReview captures the review effect for CEO-prepared communication. The interaction term tests whether formal governance review moderates the AI-authorship effect. Robust standard errors may be used. To reproduce the planned contrasts, compare condition means directly: AI_NoReview minus CEO_NoReview, CEO_Review minus CEO_NoReview, AI_Review minus AI_NoReview, AI_Review minus CEO_Review, and the difference-in-differences contrast comparing the AI penalty with and without formal review. To reproduce the mechanism evidence, estimate models linking experimental condition indicators to AccountabilityIndex, GovernanceOversightIndex, and TrustCredibilityIndex, followed by models predicting StockAllocation_USD from these perception measures and investor controls. The Results_Summary sheet provides compact benchmark outputs that can be used to verify the reproduced descriptive statistics, planned contrasts, main model coefficients, and mechanism diagnostics.
Institutions
- Ca' Foscari University of VeniceVeneto, Venice