This study explores consumers' inference strategies in a mixed choice task involving memory, external information, and missing information on attribute values for some brands. Accessibility of relevant information was manipulated, and both instructed and uninstructed or natural inferences were studied. Instructed inferences by low accessibility subjects conformed more with prior overall evaluations of the brands, displaying evaluative consistency. Instructed inferences by high accessibility subjects tended to follow a correlational rule linking missing information to other attribute information in memory, displaying probabilistic consistency. Choices conformed to inferences, and both were more variable when inferences were uninstructed.S ometimes only partial product information is available during choice, and consumers may have to infer how brands perform on specific features. For example, on an initial shopping trip, a consumer examines and evaluates some cameras based upon a then salient set of attributes. Later, at another store, the consumer examines more cameras. However, this store does not carry some of the previously examined brands, and a salesman now tells the consumer about an important feature not previously considered. The consumer has complete, externally available information on all cameras in the second store, but for brands examined earlier, some information (prior evaluations and brand features) must be retrieved and may not be readily accessible in memory. Moreover, for these brands, the new attribute information was never collected and, therefore, is missing.This scenario represents a "mixed" choice task with memory and external information Chakravarti 1983, 1986;Lynch, Marmorstein, and Weigold 1988; Lynch and Srull 1982). Moreover, missing information for some brands complicates choice of the best brand. The consumer may handle this situation in different ways. Sometimes, the missing information can be collected with minimal cost or effort, and the risks of a bad decision may make
