|
Ahn, W. & Bailenson, J. (1996). Causal attribution as a search for underlying mechanisms: An explanation of the conjunction fallacy and the discounting principle. Cognitive Psychology, 31, 82-123. Purpose The aim of the article was to show how two seemingly paradoxical phenomena (the conjunction fallacy and the discounting principle) can be explained by one underlying process. The conjunction fallacy (or conjunction effect) is the tendency to assign a greater likelihood to a pair of causes than to either cause in and of itself. For example, when told that "Karen was in a car wreck", people assert that it is more likely that the cause was poor weather and Karen's lousy vision than just the poor weather. The supposedly fallacious nature of the conjunction fallacy is based on probability theory. The probability of any given event is always greater than the probability of the given event and something else (i.e. P(A) > P(A&B)). The discounting principle is the tendency to reject possible explanations for an event subsequent to being presented with a plausible explanation. For example, if people are told that Karen was in a car wreck because of her lousy vision, they will later discount the alternative that Karen wrecked because of poor weather. These two phenomena appear paradoxical. The conjunction fallacy seems to show people (irrationally) prefer multiple causes for an event, while the discounting principle shows that people prefer single causes. The authors aim was to resolve this paradox, and also to redescribe the conjunction fallacy as a rational component of human cognition (hence the new term, conjunction effect). To this end, the authors' main proposal is that people seek to discover underlying processes and mechanisms when reasoning about a causes and effects (i.e. people seek to answer the question "How did this effect come about?"). This Mechanism Approach was contrasted with the Covariation Approach. In its simplest form, the covariation approach proposes that people attribute a cause to an effect based on correlational information alone (i.e. people seek to answer the question "With what is this effect correlated?"). According to the mechanism approach, both the conjunction and discounting effects arise from people's construction of mechanism-based explanations for events, as opposed to covariation between causes and effects. The apparent paradox arises because the task used to elicit the conjunction effect is quite different from the task used to elicit the discounting effect. The conjunction effect happens when people are presented with two causes that can be easily combined into a coherent story suggesting a single mechanism. Thus, referring to the above example, both Karen's lousy vision and the poor weather suggest a common mechanism for the car wreck, i.e. sub optimal visibility. The mechanism approach also predicts situations for which the conjunction effect will not occur. If two causes cannot be combined into a common mechanism, as should be the case when given only covariation information, the conjunction effect should not occur. The discounting effect may be observed when people infer a sufficient mechanism based on one cause, and then have to judge the likelihood of another explanation given the first. If the second explanation conflicts with the original mechanism, then it will be discounted. Again, referring to the above example, if people are told that Karen has lousy vision and was in a car wreck, they might expect Karen's vision to be sufficiently bad to cause her to wreck even in nice weather. Thus, people may subsequently deny that poor weather played a role in the wreck. Again, the mechanism based approach predicts certain circumstances for which discounting effects will not be observed. If the second explanation is inconsistent with the mechanistic explanation constructed from the first explanation (as might be the case if given only correlational information), the discounting principle should not be observed. In short, the authors propose that the conjunction effect arises because people often infer a single mechanism from two causes, and the discounting effect occurs because people infer a mechanism based on one cause that does not require a second cause. The purpose of this paper is to provide evidence for this claim. Experimental Work Experiment 1 For Experiment 1, one group of subjects performed a conjunction task, and another group of subjects performed a discounting task. For the conjunction task, participants were asked to rate the likelihood that an explanation was the cause of some event. On different trials, participants judged the likelihood of conjunctive explanations and the constituents that were either mechanism or covariation based. For example, on mechanism based trials, participants would judge the likelihood that Karen wrecked because of lousy vision and poor weather, and on two different trials, the same participant would judge the likelihood that Karen wrecked because of poor vision, and the likelihood that she wrecked because of lousy weather. On covariation based trials, the mechanism-based explanations were replaced with covariation based explanations that matched the subjective likelihood of the mechanism based explanations as determined by a previous experiment. For example, the explanation that Karen wrecked because of the weather was replaced by the explanation that Karen wrecked because it was more likely for wrecks to occur on that given evening. Note that, for any given participant, different events were used with covariation and mechanism based explanations. Participants who performed the discounting task had to judge the strength of one cause with or without another cause already present. Again, the causes were either mechanism or covariation based. For a given event, participants made four judgments: two judgments concerning the strength of a each factor alone and two judgments concerning the strength of each factor given that the other factor was true. Results were consistent with the mechanism approach. For the conjunctive task, participants judged conjunctive explanations as more likely only when explanations were mechanism based. For the discounting task, participants discounted alternate explanations when another explanation was given only for mechanism based explanations. Experiment 2 In Experiment 2, for each event, pairs of explanations were constructed such that it was difficult for a person to come up with a single mechanism without a cue. The context cues, when given, were intended to unify two explanations or to bias the participants for or against one of the explanations. Again, participants performed either the conjunction task or the discounting task. For both tasks, after participants viewed a target event, and were then presented with a context cue (or the letters XXX in the no cue condition). After viewing the cue, the conjunction or discounting task was performed. Again, the results were consistent with the mechanism approach. For the conjunction task, conjunctive explanations were rated as stronger than single explanations when a unifying cue was given and weaker when a cue that biased subjects for only one of the explanations was given. Conversely, the discounting effect was increased when subjects were cued to construct a story supporting only one of the explanations. When no cue was given, there was no difference between single and double explanations for either of the tasks. Experiment 3 The methods of Experiment 3 were similar to Experiment 2, except that all explanations were covariation based. In demonstrating conjunction and discounting effects using covariation based explanations, the authors refute the criticism that the null result in Experiment 1, where covariation information alone did not have any effects, was not due to an unfair questioning procedure, as the questioning procedure in Experiment 3 was nearly identical. Therefore, explanations conveying information about covariation are acceptable so long as an underlying mechanism is available.
Conclusions The authors conclude that both the conjunction effect and the discounting effect occur only when underlying mechanisms mediating the cause and the effect are known, consistent with the mechanism approach to causal attribution. The more traditional covariation based approaches cannot account for the results. Covariation approaches assume that the conjunction and discounting effects are mutually exclusive, but all three experiments demonstrated both effects for the same stimuli. Further, conjunction effects were observed when participants performed a discounting task, and vice versa. Thus, when reasoning about causal relationships, the authors believe that people do not simply rely on information about covariation. Rather, the crucial information concerns how well explanations cohere to underlying mechanisms. Points for Discussion 1) At certain points in the paper, the authors seem to waiver (albeit ever so slightly) in their assault on covariation based explanations for causal attributions. For example, at one point the authors write "… covariation based models, in their current form, cannot explain why there are different levels of discounting and conjunction effects…." (p. 114, italics added). How might an ardent supporter of the covariation approach attempt to explain the current results? 2) How might processes related to making causal attributions be related to processes involved in concept formation? Might it be the case that concept formation is not based on covariation information, but rather mechanism based information? |
|
This page was last updated: |