How social and non-social information influence classification decisions: A computational modelling approach

ABSTRACT

Social information such as observing others can improve performance in decision making. In particular, social information has been shown to be useful when finding the best solution on one’s own is difficult, costly, or dangerous. However, past research suggests that when making decisions people do not always consider other people’s behaviour when it is at odds with their own experiences. Furthermore, the cognitive processes guiding the integration of social information with individual experiences are still under debate. Here, we conducted two experiments to test whether information about other persons’ behaviour influenced people’s decisions in a classification task. Furthermore, we examined how social information is integrated with individual learning experiences by testing different computational models. Our results show that social information had a small but reliable influence on people’s classifications. The best computational model suggests that in categorization people first make up their own mind based on the non-social information, which is then updated by the social information.

KEYWORDS:

• Classification
• Computational modelling
• Decision making
• Social learning

Acknowledgements

We would like to thank Laura Wiles for her help with editing the manuscript.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. In the individual condition the test objects did not include any social information and were each presented 8 times, whereas in the social conditions each new test object was presented 8 times with no, 4 times with incorrect, and 4 times with correct social information. For the ANOVA we calculated performance for the objects with no social information based on the 8 repetitions in the individual and social conditions. To calculate performance for the objects with correct and incorrect social information we used the 4 presentations with the respective social information in the social conditions. In the individual condition we calculated performance by randomly drawing for each test object 4 repetitions from the 8 repetitions (once for the correct and once for the incorrect comparison) to keep the number of repetitions comparable to the social conditions.

2. To compare performance in the individual and social conditions, we used for each object in the social condition (separate for type of social information) the respective repetition of the object in the individual condition. As in Experiment 1, the pattern of results stays the same if all objects and not only the new test objects are considered.

Additional information

Funding

This work was supported by SNF research [grant number 146169] to the second and third authors.