Optimism bias is the tendency to expect a more favourable personal future than the evidence warrants. It has been studied almost entirely at the level of the individual. Here we ask how optimism is shaped and shared through social interaction. Across three rounds, participants imagined a shared future scenario either alone or with a partner. We measured their beliefs about related and unrelated future events, how positively they evaluated the scenario the material depicted, and their memory for it. Imagining the future increased optimism selectively for events related to the imagined scenario. This increase was comparable whether people imagined alone or together, so it was produced by imagining rather than by the interaction. Interaction instead aligned partners' beliefs. Only for discussed future, and not undiscussed, did the convergence of beliefs exceed what regression to the mean or chance grouping can account for. This alignment was not accompanied by convergence in affect and was not explained by memory for the encoded event. While interaction had a generic positive effect on emotion evaluations, participants did not become more aligned in emotion as a product of the shift. In exploratory analyses, the size of the optimistic shift covaried with the affective tone of the conversation. Our results show that optimism is shaped and aligned through social interaction. This alignment is content-specific and separable from shared affect and shared memory.
One row per participant / item / group, as described in metadata.xlsx.
| file | level | notes |
|---|---|---|
belief_items.xlsx |
participant x item x time | likelihood ratings |
subject_measures.xlsx |
participant | belief/emotion/connectedness/memory + flags |
group_conversation.csv |
group | transcript-derived tone + discussion amounts (no verbatim text) |
Transcripts and questionnaire data are not shared (participant privacy). The BDI-II, LOT-R and TIPI responses collected at screening are not included; the BDI-II was used only to apply the preregistered exclusion criterion described in the Methods, and every participant in these files scored at or below the cut-off. The conversation analyses use only the group-level aggregates in group_conversation.csv (affective tone from sentiment; discussion amounts from keyword matching, SI Note S7). The prevalence of encoded Cognitive Units within the imagination task (Results, memory section) likewise derives from the transcripts and is not reproducible from these aggregates.
group has a different meaning in each condition:
- Collaborative condition — a real group of four participants who interacted (19 groups). These are the units used in every group-level analysis.
- Individual condition — participants did not interact, so each one carries their own participant ID in this column (75 "groups" of one). They are never treated as groups; they enter the pseudo-group null only as a pool of non-interacting individuals to be partitioned at random.
The 94 distinct values are therefore 19 real groups plus 75 individual participants. Group labels are arbitrary identifiers (G01–G19 for the interacting groups, I01–I75 for the individual participants) and carry no information about when a session took place.
This work was supported by the Spanish Ministerio de Ciencia, Innovación y Universidades, part of the Agencia Estatal de Investigación (AEI), through project PID2022-140426NB-I00 (funded by MCIN/AEI/10.13039/501100011033 and by FEDER, a way to make Europe) to L.F. We thank the CERCA Programme/Generalitat de Catalunya for institutional support.