Replication data and code: Invitation Messages for Business Surveys: A Multi-Armed Bandit Experiment

This replication package contains the experimental data and code from a study investigating how elements of a survey invitation message targeted to businesses influence their participation in a self-administered web survey. The experiment was conducted in collaboration with the German Business Panel (GBP) during its fifth survey wave, spanning from August 16, 2022, to November 25, 2022. A full factorial design was implemented, varying five key components of the email invitation. Unlike conventional experimental setups with static group assignments, the study employed adaptive randomization, wherein a Bayesian learning algorithm sequentially allocated more observations to invitation messages exhibiting higher survey starting rates. Over the 15-week experimental period, 738,598 invitation messages were distributed to business contacts, of which 176,000 were opened within one week. A total of 7,833 recipients initiated the survey, and 3,733 completed it. The dataset includes detailed records of message distribution, survey engagement metrics, and adaptive randomization adjustments, providing a comprehensive basis for analyzing the effectiveness of invitation design in business survey participation. Keywords: Adaptive Randomization, Reinforcement Learning, Nonresponse, Email Invitation, Web Survey, Firm Survey, Organizational Survey This replication package contains the experimental data and code from a study investigating how elements of a survey invitation message targeted to businesses influence their participation in a self-administered web survey. The experiment was conducted in collaboration with the German Business Panel (GBP) during its fifth survey wave, spanning from August 16, 2022, to November 25, 2022. A full factorial design was implemented, varying five key components of the email invitation. Unlike conventional experimental setups with static group assignments, the study employed adaptive randomization, wherein a Bayesian learning algorithm sequentially allocated more observations to invitation messages exhibiting higher survey starting rates. Over the 15-week experimental period, 738,598 invitation messages were distributed to business contacts, of which 176,000 were opened within one week. A total of 7,833 recipients initiated the survey, and 3,733 completed it. The dataset includes detailed records of message distribution, survey engagement metrics, and adaptive randomization adjustments, providing a comprehensive basis for analyzing the effectiveness of invitation design in business survey participation. Keywords: Adaptive Randomization, Reinforcement Learning, Nonresponse, Email Invitation, Web Survey, Firm Survey, Organizational Survey