The Belief Bias
Eight syllogisms. For each, you decide if the conclusion follows logically from the premises. Assume the premises are true. Ignore whether the conclusion is true in the real world. WIZ measures how much your real-world belief overrode the logic check.
“The form of the argument is one thing; the truth of its content is another. To judge the form independently of the content is the first move in deductive reasoning.”paraphrased from Aristotle, Prior Analytics
Wilkins (1928) gave subjects matched sets of categorical syllogisms — the classical “all A are B, C is A, therefore C is B” structure that has been studied since Aristotle. One set used symbolic content (letters and nonsense words). A second set used neutral familiar content. A third set used emotionally loaded content. Logical performance dropped systematically as content moved from symbolic to neutral to loaded. The drop was not random: subjects accepted invalid arguments when the conclusion matched their prior beliefs and rejected valid arguments when the conclusion contradicted them. The logic of the syllogism was being read through a belief filter.
Evans, Barston & Pollard (1983) Memory & Cognition vol 11 picked up the thread with a clean 2x2 design. Syllogisms varied independently along logical validity (valid or invalid) and conclusion believability (believable or unbelievable). The instructions were explicit: assume the premises are true, judge whether the conclusion follows logically, ignore the real-world truth of the conclusion. Acceptance rates across three experiments: V+B 89%, V+U 56%, I+B 71%, I+U 10%. The pattern is the signature of belief bias. Subjects accepted 89% of valid arguments when the conclusion was believable but only 56% when it was unbelievable — a 33-point penalty for valid arguments whose conclusion they did not already accept. They accepted 71% of invalid arguments when the conclusion was believable, against 10% when it was unbelievable — a 61-point bonus for invalid arguments whose conclusion they already believed.
The bias has been replicated several hundred times. It operates at the conclusion-evaluation stage rather than at premise-encoding (Newstead Pollard Evans & Allen 1992). It produces a robust 35-45 point gap across studies (Klauer Musch & Naumer 2000). It survives explicit warning. It correlates only weakly with cognitive ability (Stanovich & West 2008). It even amplifies under time pressure (De Neys 2006). Goel & Dolan (2003) ran fMRI on the Evans paradigm and found different neural signatures for the two response modes: belief-based judgments activated ventral medial prefrontal cortex (affective evaluation), while logic-based judgments activated left lateral parietal cortex (rule-based deduction). The same syllogism produces different brain activity depending on whether belief or logic is dominating.
The exercise. Eight syllogisms in the four hidden cells of the Evans 2x2. Two valid + believable conclusions. Two valid + unbelievable. Two invalid + believable. Two invalid + unbelievable. For each, judge VALID or INVALID, assuming the premises are true. A subject reading the logic correctly gets all eight right. A subject reading the belief correctly but the logic incorrectly gets exactly four right (the V+B and I+U baselines). At the end I compute your Belief Bias Gap: the difference between your accuracy when belief and logic agree and your accuracy when they conflict. The Evans 1983 modal subject has a gap of 33 to 47 points. A pure logician has a gap of zero. A pure belief-follower has a gap of 100.