DATASET - Can Bots Teach Us How to Disagree Constructively?
Description
Anonymized dataset from a deliberate-practice study on dialogic-competence training with LLM-based conversational agents (CAs). Forty adult participants were assigned to an experimental or a control group and completed four dialogue sessions, each with a distinct CA differing in persona and argumentative style, across four contested discussion topics. The file contains one row per participant and includes background variables (gender, age band, education, field of study), self-reported frequency of and attitudes toward AI use, pre-training opinion items on the discussion topics with accompanying confidence ratings, intellectual-humility items, a 17-item self-report scale of dialogic-participation behaviors, post-conversation emotion ratings collected after the first two sessions, session timestamps, three open-ended reflection questions completed at the end of training, and PRODIGY-coded dialogue scores. The coded scores are reported per session for four dimensions (Reasoning and Argumentation, Perspective, Interpersonal Rapport, Integrity and Humility), together with per-dimension sums, an overall dialogic-performance score for each of the four sessions, baseline and post-test performance scores, and word counts. All direct and indirect identifiers were removed. The dataset contains no names; the native-language item was dropped; and the CA persona names and discussion-topic labels were de-identified to match the wording used in the manuscript. The data were translated from the original language into English.
Files
Institutions
- Hebrew University of JerusalemJerusalem, Jerusalem