A rarity heuristic for hypothesis testing

摘要: A Rarity Heuristic for Hypothesis Testing Aidan Feeney Jonathan St.B.T. Evans & Simon Venn Department of Psychology University Durham Science Laboratories South Road DH1 3LE United Kingdom aidan.feeney@durham.ac.uk Centre Thinking and Language Plymouth Drake’s Circus PL4 8AA {jevans, svenn}@plymouth.ac.uk Abstract This paper presents the results two experiments designed to investigate processes underlying effects beliefs about probabilities on an hypothesis testing task. Both demonstrate that although such exist, they are inflexible in face explicit statistical implicit contextual manipulations likely information be gained from selecting evidence concerning rare features. It is argued these suggest operation a rarity heuristic whilst possible adaptive functions discussed. Probabilities Over last ten or more years, human testing, which previously had been viewed as being prone bias (Wason, 1960; Doherty et al, 1979), has rehabilitated. One central claims made during this process rehabilitation some way adapted probabilistic structure our environment. For example, Klayman Ha (1987) have confirmation, verification matching biases, amongst others, may result generalised positive test strategy testing. demonstrated how could good environments with realistic structure. Their claim tasks Wason’s 2 4 6 task lead non-normative behaviour because encourage participants adopt generally sensible experimental situation whose does not match strategy. More recently, Oaksford Chater (1994), spirit Anderson’s (1990) general ‘rational analysis’ cognition, proposed decision-theoretic account selection 1966). based both itself assumptions people’s understanding abstract conditional hypotheses. Our aim extend study another gain insight into mechanisms effects. Pseudodiagnosticity Feeney, Clibbens (1997) considered role background determining performance pseudodiagnosticity (PD) (Doherty 1979) tendency select relevant just one pair hypotheses when trying decide between them. An example standard paradigm used taken Mynatt, Dragan (1995): Your sister car she bought couple years ago. It’s either X Y but you can’t remember which. You do her over 25 miles per gallon any major mechanical problems she’s owned it. following information: A. 65% X’s gallon. Three additional pieces also available: B. The percentage Y’s C. no first ownership. D. Assuming find out only three (B, C D) would want order help your owns? Please circle answer. In PD task, above, anchor provided (item A) provides potentially supportive presented scenario. we term focal hypothesis. According his colleagues, normatively correct answer problem choose item B - Bayesian terms completed likelihood ratio allows diagnosticity assessed.