Further Exploration of the Belief Bias in Transitive Inference
Belief bias in reasoning is the tendency to draw or accept conclusions on the basis of the believability of the problem content, rather than whether the conclusions follow logically from the premises. For example, given the premises, The horse is larger than the mouse, and The mouse is larger than the elephant, accepting the conclusion, The elephant is larger than the horse would indicate belief bias. The logically correct transitive inference is The horse is larger than the elephant. Dual-process models of reasoning propose that heuristic (belief-based) and analytic processes are used in reasoning and that belief bias occurs if heuristic processing dominates when the two processes produce conflicting responses. The current study extended previous research on belief bias in transitive inference. The 230 participants evaluated the conclusions of six types of transitive inference problems. These were valid and invalid problems with believable, neutral, and unbelievable content. Participants also rated their confidence in each decision. A subset of 73 participants completed a test of fluid intelligence. Conclusion acceptance rates on invalid problems were lower for unbelievable than neutral problems, indicating a strong negative belief bias. Acceptance rates on valid problems were higher for believable than neutral problems, indicating a weak positive belief bias. These findings correspond to de-biasing effects of beliefs on transitive reasoning. That is, unbelievable content made it easier to reject invalid conclusions whereas believable content made it easier to accept valid conclusions. Confidence was higher for problems with belief-laden than neutral content. Accuracy and confidence ratings were well-calibrated overall, but there was some evidence that participants who performed more poorly on neutral problems were over-confident when evaluating the conclusions of invalid problems with believable conclusions. Participants who performed better on neutral problems and those with higher fluid intelligence used more analytic processing on problems with belief-laden content. The findings are consistent with the dual-process models of reasoning.
Advances in Psychology Research. Volume 102
Psychology not elsewhere classified