Two Faces of Representativeness: The Effects of Response Format on Beliefs about Random Sampling

Two beliefs that act in concert have been proposed as the basis for the representativeness heuristic in general, and judgments about random sampling in particular: samples resemble their parent populations (resemblance), and random sampling is a self-correcting process (balancing). Based on the results of a preliminary experiment, we proposed the ‘rule-cuing’ hypothesis, which is that different aspects of sampling problems can invoke these two beliefs separately. We found that when response formats required subjects to estimate themean of a sample, subjects’ responses reflected resemblance beliefs, whereas when subjects estimated the total score in a sample, balancing beliefs were elicited. In additional experiments we eliminated two rival hypotheses: theproblem difficulty hypothesis, and thearithmetic inconsistency hypothesis. Results suggest that beliefs, as well as preferences, may be constructed on-line in response to task characteristics.