Anonymized Data for “When Should AI Recommendations Be Assertive? Matching Recommendation Content with Message Assertiveness in Online Product Comparisons”
Description
This dataset contains the analysis datasets for three online experiments examining how analytical versus empathetic AI recommendation content interacts with assertive versus non-assertive messages to influence consumers’ recommendation acceptance and purchase intention in online product comparisons. Study 1 examined the interaction using wireless earbuds (N = 361). Study 2 used running shoes and examined perceived recommendation diagnosticity and psychological comfort as mediators, controlling for product category involvement (N = 445). Study 3 used an interactive carry-on suitcase comparison prototype and additionally manipulated user- versus system-initiated recommendations (N = 655). It also included a forced-choice outcome and measures of perceived recommendation certainty. The repository contains 11 SPSS (.sav) files covering the three main studies, three manipulation pretests, three no-recommendation baseline studies, and two additional message assertiveness validation studies. The files include experimental condition codes, item-level responses, composite scores, manipulation checks, and participant background measures. Variable and value labels are embedded in the SPSS files.
Files
Steps to reproduce
1. Open the 11 SPSS (.sav) files in IBM SPSS Statistics and use the embedded variable and value labels to identify the study variables and experimental conditions. 2. Reproduce the manipulation checks, baseline comparisons, and validation analyses using the corresponding Pretest, Control, and Validation check files. 3. For Study 1, conduct a 2 × 2 MANOVA. For Study 2, conduct a 2 × 2 MANCOVA and parallel mediation analyses using PROCESS Model 8, with product category involvement as a covariate. For Study 3, conduct a 2 × 2 × 2 MANCOVA, follow-up interaction tests, logistic regression for the forced-choice outcome, and parallel mediation analyses. 4. Use 5,000 bootstrap resamples and 95% confidence intervals for indirect effects, and compare the results with the manuscript and Appendix D.
Institutions
- Beijing Institute of TechnologyBeijing, Beijing