When participants are not misled they are not so bad after all: A pragmatic analysis of a rule discovery task

When Participants Are Not Misled They Are Not so Bad After All: A Pragmatic Analysis of a Rule Discovery Task Jean-Baptiste Van der Henst (jvanderhenst@hotmail.com) Laboratory of Experimental Psychology, K.U. Leuven, Tiensestraat 102 Leuven, 3000 Belgium Institut Jean Nicod, 1 bis avenue de Lowendal Paris, 75007 France Sandrine Rossi (rossi@scvie.unicaen.fr) Laboratoire de Psychologie Cognitive et Pathologique, Universite de Caen, Esplanade la Paix Caen, 14032 France Walter Schroyens (Walter.Schroyens@psy.kuleuven.ac.be) Laboratory of Experimental Psychology, K.U. Leuven, Tiensestraat 102 Leuven, 3000 Belgium Abstract In this paper we present a pragmatic analysis of a widely used task in the field of hypothesis testing: the 2-4-6 problem (Wason, 1960). In this task participants have to discover the rule “three increasing numbers” by testing triples of numbers and are given the “2-4-6” as an example of triples compatible with the rule. We argue that most people fail because the givens of the task are conversationally misleading: first because the 2-4-6 is communicated and is thus presumed to be relevant (Sperber & Wilson, 1995) and second because the rule to be discovered is too simple in the context of the task. In a first experiment we showed that providing the triple without communicating it improved performance in the task. In a second experiment we contextually increased the relevance of the rule and observed that people were thus more inclined to discover it. Introduction Imagine that you have to discover a rule that generates triples of numbers. Some triples are consistent with the rule and some are not. Now, somebody who knows the rule – a trustworthy person like an experimental psychologist – is telling you that ‘2-4-6’ is a triple that is consistent with the rule. Will you consider this as helpful information or not? Surely, you will and you will also probably think that the experimental psychologist would expect you to regard this triple as helpful in order to succeed in the task. Hence, you will consider that the salient properties conveyed by ‘2-4-6’ (like “evenness” or “increase by 2”) must be taken into account in order to discover the rule. However, considering that these properties are important is in fact deceptive since the rule to be discovered does not relate to them: the rule is simply “three increasing numbers”. Focusing on 2-4-6’s most salient properties is thus not the good way to solve such a task! This task is the well-known ‘2-4-6’ problem designed by Peter Wason more than forty years ago (Wason, 1960) in order to investigate hypothesis testing ability (see Gorman, 1995 and Poletiek, 2001 for reviews). It has become the most commonly used task by researchers in the field of hypothesis testing. In its standard version, it consists in proposing sequences of triples to discover a rule the experimenter has in mind. For each triple, the experimenter indicates whether or not it is consistent with the rule. Participants have to test triples until they are sure of having discovered the rule. As for the other famous Wason’s task, namely the selection task (Wason, 1968), one stimulating aspect of the ‘2-4-6’ problem is that few people succeed in it despite of its apparent simplicity. In the initial study (Wason, 1960), only 21% of participants succeeded in discovering the rule in their first announcement. Typically, the rules proposed by participants inherit the salient properties of ‘2-4-6’ and are more specific than the rule to be discovered. For instance, they propose rules such as “three even numbers”, “numbers increasing by 2”, “even numbers increasing by 2”. The failure in the 2-4-6 task has often been viewed as a sign of irrationality. Wason argued that participants exhibited a “confirmation” bias, and Evans (1983; 1989) argued that people exhibited a “positivity” bias. The protocols indeed reveal that people tend to propose instances of triples compatible with their hypothesis whereas the most efficient strategy consists of proposing instances inconsistent with the held hypothesis. It is commonly accepted that people overly rely on a positive testing strategy and focus on too narrow hypotheses (Poletiek, 2001). What is the reason for this? Researchers assume that positive testing is