Developmental Psychology 1997, Vol. 33, No. 6, 1074-1090 Copyright 1997 by the American Psychological Association, Inc. 0012-1649/97/$3.00 Setters and Samoyeds: The Emergence of Subordinate Level Categories as a Basis for Inductive Inference in Preschool-Age Children Sandra R. Waxman, Elizabeth B. Lynch, K. Lyman Casey, and Leslie Baer Northwestern University Basic level categories are a rich source of inductive inference for children and adults. These 3 experiments examine how preschool-age children partition their inductively rich basic level categories to form subordinate level categories and whether these have inductive potential. Children were taught a novel property about an individual member of a familiar basic level category (e.g., a collie). Then, children's extensions of that property to other objects from the same subordinate (e.g., other collies), basic (e.g., other dogs), and superordinate (e.g., other animals) level categories were examined. The results suggest (a) that contrastive information promotes the emergence of subordinate categories as a basis of inductive inference and (b) that newly established subordinate categories can retain their inductive potential in subsequent reasoning over a week's time. One of the most remarkable aspects of early cognitive devel- opment is children's capacity to form categories of objects and to use these categories as a basis for inductive inference. Al- though initial attempts at object categorization may rely heavily on perceptual commonalities among objects (Gentner & Wax- man, 1994; Waxman & Markow, 1995), children and adults hold strong expectations that members of object categories also share deep, underlying commonalities that go beyond those available from perceptual inspection (Atran, 1990; Gelman, Coley, & Gottfried, 1994; Gelman & Medin, 1993; Gelman & Wellman, 1991; Medin, 1989; Medin & Ortony, 1989; Mur- phy & Medin, 1985; Shipley, 1993). These expectations are directly related to inductive power. Once a deep, nonperceptual property is projected onto a given individual, this property is typically extended to other members of that object category. For example, if we discover that one paramecium has a digestive tract, we infer that this property is true of all paramecia. This tendency to project a newly discovered property of an individual object onto other members of its category is essential in ex- tending the limits of knowledge beyond direct observation. It also promotes cognitive stability and coherence in the face of the diversity among objects (Rips, 1975; Smith & Medin, 1981). There is now considerable evidence that preschool-age chil- Sandra R. Waxman, Elizabeth B. Lynch, K. Lyman Casey, and Leslie Baer, Department of Psychology, Northwestern University. This research was supported by National Institutes of Health Grant HD28730. We are grateful to the children, parents, and teachers from the Baby-Toddler Nursery, Central Evanston Childcare, Child Care Center, Discovery Preschool, and the School for Little Children for participating in these studies and to Katherine Maggs for her assistance in data collec- tion. We are indebted to Douglas Medin, Lance Rips, John Coley, Laura Namy, and Evan Heit for comments and discussion. Correspondence concerning this article should be addressed to Sandra R. Waxman, Department of Psychology, Northwestern University, 2029 Sheridan Road, Evanston, Illinois 60208-2710. Electronic mail may be sent via Internet to s-waxman@nwu.edu. dren share with adults the expectation that category members share deep resemblances and that categories serve as a basis for inductive inference. In an elegant series of studies, Gelman and her colleagues (Gelman & Coley, 1990; Gelman & Mark- man, 1986, 1987; Gelman & O'Reilly, 1988; Gelman, 1988) have demonstrated that preschool-age children use basic level categories as a guide in induction. For example, children as young as 2 years 6 months of age will project an enduring property from one object to another on the basis of category membership, even when category membership conflicts with perceptual appearance. However, one limitation of virtually all research on the devel- opment of inductive inference is that it has focused on categories at the basic level. To the best of our knowledge, there are only two exceptions. In one, Shipley (1992) reported that preschool children are more likely to use basic level than subordinate level categories as the range of inductive inference. In the other, Gelman and O'Reilly (1988) reported that preschool children are more likely to use basic level than superordinate level cate- gories as the range of inductive inference. These findings are consistent with claims regarding the developmental primacy of basic level categories (Mervis, 1987; Rosch, Mervis, Gray, John- son, & Boyes-Braem, 1976), and, with the position that al- though basic level categories may initially be perceptually based, they rapidly acquire conceptual power (see Gentner and Waxman, 1994; Shipley, 1993; Waxman, in press, for evidence that the conceptual power of the basic level derives, at least in part, from infants' and children's experience with objects and their names). However, the conceptual power and inductive strength of fa- miliar basic level categories present children with a potential challenge: How do children come to partition these rich and inductively powerful basic level categories into subordinate level categories that have inductive potential? This question serves as the focus of the three experiments reported here. A review of recent literature suggested that two factors—the type of information and the structure of the information—are 1074