Do we advise as one likes? The alignment bias in social advice giving
Fig 4
The conceptual schematic of the winning model M7.
This conceptual schematic illustrates the psychological computation of advice-giving behavior within a single trial of interaction. Two parallel components (non-social value , and social value
) jointly determined the likelihood of choosing either judgement option (‘higher’ vs. ‘lower’).The non-social value
reflected advisors’ tendency to adhere to their personal opinion, as measured by their judgement option in the non-social session (
), with the strength of this tendency indexed by their confidence (
), as scaled by post-judgement wagering magnitude. The social value
captured advisors’ tendency to align or misaligned with advisees’ opinions. The value of the chosen type (aligned vs. misalign) each trial was updated by the feedback from advisees (
) with acceptance serving as reward and rejection serving as punishment, with the updating pace scaled by the learning rate (
). Notably, we modeled a confirmation bias in feedback learning (highlighted in yellow). Learning rates were parameterized separately for each combination of advice type and feedback type, allowing the model to capture asymmetries that reinforce the link between aligned advice and acceptance, and between misaligned advice and rejection.