Running head: CONSTRAINTS ON SOCIAL CATEGORIZATION Constraints on the Acquisition of Social Category Concepts

Determining which dimensions of social classification are culturally significant is a developmental challenge. Some suggest this is accomplished by differentially privileging intrinsic visual cues over non-intrinsic cues (Atran, 1990; Gil-White, 2001) whereas others point to the role of noun labels as more general promoters of kind-based reasoning (Bigler & Liben, 2007; Gelman, 2003). A Novel Groups procedure was employed to examine the independent effects of noun labels and visual cues on social-categorization. Experiment 1 demonstrated that in the absence of a visual cue, a noun label supported social categorization among 4-year-olds and 7-year-olds. Experiments 2 and 3 demonstrated that children and adults fail to differentiate between intrinsic and nonintrinsic visual cues to category membership, suggesting that this distinction is not central to the acquisition of social category concepts, and showed that in the absence of a shared noun label, visual cues were not sufficient for younger children to form social categories. Experiment 4 ruled out a potential demand characteristic in the previous experiments. Together, these results reveal the primacy of verbal labels over visual cues for social categorization in young children and suggest a developmental change between ages 4 and 7 in the ability to construct new representations of social category concepts. DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 4 Constraints on the Acquisition of Social Category Concepts To successfully navigate the social world, children must reason about people’s social group memberships. In doing so, they face two learning problems: They must determine which properties pick out social groups in their local milieu and they must determine what properties are implied by social group membership. Each of these problems is far from trivial. Dimensions of social classification differ remarkably in their outward appearance, causal history, and the permeability of group boundaries. For example, membership in the category “blue-eyed people” is determined by a polygenetic trait affecting the amount of eumelanin in the iris. In contrast, membership in the category “Red Sox fan” is highly fluid, determined by expressions of preference more directly under volitional control such as how much an individual cares about the Red Sox winning, the clothes worn to a Red Sox game, and the amount of money paid to watch the Red Sox in action. Similarly, the properties projected on the basis of social group membership (e.g., behaviors, dispositions, beliefs, preferences, language, friendship choices, rights, etc.) differ as widely as the dimensions of social classification (Hirschfeld, 1993, 1994, 1995b; Mervis & Rosch, 1981; Markman, 1989; Bloom, 1998; Ahn et al., 2000; Murphy, 2002; Rosch & Levitin, 2002). Despite these complexities, early in life children reveal an exquisite ability to hone in on culturally meaningful groupings such as gender, race, and language spoken, and demonstrate an understanding that not all social categories project the same properties (Gelman & Taylor, 2000; BarHaim, Ziv, Lamy, & Hodes, 2006; Kinzler, Dupoux, & Spelke, 2007; Heyman & Gelman, 2000a; Hirschfeld, 1994, 1995a,b, 2001; Rhodes & Gelman, 2009; Solomon, DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 5 2002; Astuti et al., 2004; Shutts, Kinzler, McKee, & Spelke, 2009; Shutts, Banaji, & Spelke, 2010; Fawcett & Markson, 2010a,b; Diesendruck & Eldror, 2011). Several theories attempt to account for how children create and deploy representations of new social categories. In a series of studies, Gelman and Heyman (Gelman & Heyman, 1999; Heyman & Gelman, 1999, 2000b) demonstrated that socialcategories marked with a noun label support significantly stronger inductive inferences than when labeled either with an adjective or a verbal predicate, suggesting that noun labeling promotes the deployment of representations of social categories and likely contributes to their initial acquisition. This claim is not specific to social category reasoning, as the propensity to treat nouns as picking out kinds is well documented across a variety of developmental domains (Markman, 1989; Waxman, 2010), and is even visible in 12-month-old infants’ expectations about object identity (Xu, Cote, & Baker, 2005). Of course, even if labeling a category with a noun promotes inductive inference within both the social and non-social domains, it is still possible that other aspects of social categorization might diverge in interesting ways from categorization in other domains. On this note, some have argued that young children assimilate social categories to biological ones, essentializing them as they do species kinds. For example, Medin & Atran (2004), Atran (1990), and Gil-White (2001) suggest that children essentialize some social categories by misapplying species-kind reasoning drawn from the biological domain (e.g., plants, animals) to the social domain (e.g., race, gender). This motivates the prediction that children might privilege properties that appear to have a biological basis (e.g. apparently intrinsic visual cues such as skin color) since these cues are suggestive of DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 6 species differences. Although this specific claim is still untested, recent work has explored children’s inferences about essentialized and non-essentialized categories while beginning to challenge the assumption that representations of social categories are derived from biological concepts (Diesendruck & Eldror, 2011). For example, Diesendruck & Eldror (2011) examined 4-6-year-olds expectations about social groups and the property inductions licensed by shared group membership. They observed that children may privilege psychological properties and causal properties when making judgments about category members. In contrast, Aboud (1988) has suggested that all salient perceptual cues to similarity (e.g., distinctive clothes, skin color, hair length, distinctive behaviors) that cooccur with group membership and thereby distinguish among groups are likely to form a basis for social categorization. Drawing largely on the work of Piaget (1929), Aboud argues that domain-general cognitive faculties of logical reasoning underlie the construction of all categories, and that qualitative changes in these reasoning abilities drive stage-like developmental changes in children’s categorization skills, social and nonsocial alike. For younger children (e.g., 6 and below), this entails a reliance on salient perceptual cues, suggesting that the earliest categories children acquire will likely be grounded in those that are marked by shared visual cues in particular (for a similar analysis of the role of perceptual salience, see Bigler & Liben, 2007). This view stands in contrast to the claim that intrinsic visual cues (and their implication of a difference in biological kind) are more salient when establishing representations of social categories. Further, this view stands in contrast to the noun-labeling hypothesis, which bestows little importance on visual properties to category membership because the inductive power of DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 7 social categorization is derived from deeper representations of kinds supported by the presence of a common count noun label. In one of the few studies to directly weigh in on these accounts, Diesendruck & haLevi (2006) explored the role of noun labeling as well as several dimensions of social categorization (e.g., shared personality, shared religion, shared physical appearance) in determining the properties project by social categorization (e.g., preferences). Five-yearolds were introduced to pictures of two test individuals and were taught several unique properties about each individual (e.g., “This girl is religious. She is shy. This shy religious girl likes to flep; This girl is secular. She is friendly. This friendly secular girl likes to dax.”). Next, participants were introduced to a target individual and were told that this individual shares one property in common with each of the test characters (e.g., “This girl is secular like her. She is she shy like her.”) and then participants were asked which preference the target individual shared (e.g., “Do you think she likes to flep like her or dax like her?”). The dependent measure was the proportion of inferences participants drew based on shared social category membership (e.g., religion, gender) as compared with shared physical property (e.g., height), or psychological trait (e.g., shy). In two experiments shared properties were marked visually and verbally. In two further experiments verbal labels were contrasted with physical similarity. Across these experiments, children drew more inductive inferences for labeled social categories than for groups that shared a common physical property or psychological trait (Diesendruck & haLevi, 2006; see also Hirschfeld, 1995b, for a similar observation). In other words, children were more likely to think that the target individual shared the same novel preference as the person who matched him/her in social DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 8 category membership (e.g., religion, gender) than the person who shared a common physical or psychological trait. This pattern of results held constant even when visual cues to group membership were eliminated, dovetailing with findings from the broader categorization literature in which visual cues to similarity are not necessary for categorization, and supporting a privileged role for noun labels as compared with other bases of similarity (in this case, shared personality or physical traits; Markman, 1989; Gelman & Heyman, 1999; Heyman & Gelman, 1999, 2000b). Most developmental studies of social categories focus on actual social groups well-represented in children’s experience, and to which they know themselves to either belong or not (e.g. race in North America: Aboud, 1988; Bigler & Liben, 2007; Hirschfeld, 1994; Jews and Arabs in Israel: Diesendruck & haLevie, 2006). While of course valuable, such research may obscure generalization about the general process of social category acquisition for at least two reasons. First, merely belonging to social groups—even previously unfamiliar ones—is enough to induce children, at least from the preschool years, to prefer “ingroups” on both explicit and implicit measures (Dunham, Baron & Carey, 2011; Bigler, Jones, & Lobliner, 1997; Patterson & Bigler, 2006; Nesdale & Flesser, 2001). Thus, merely belonging to a group can shift children’s pattern of attention and inference, implying that results with familiar groups could depend on this form of identification. Second, the focus on familiar, highly salient social groups raises the possibility that children paid attention to the social categories they did because they had learned that these social groups are supremely important in their cultural setting. More broadly, by early elementary school children have knowledge of at least some of the stereotypes and consensus cultural evaluations of the groups that surround them DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 9 (Dunham, Baron & Banaji, 2008; Cvencek, Greenwald & Meltzoff, 2011; Cvencek, Meltzoff & Greenwald, 2011). This rich learning history makes it difficult to identify more general expectations about how groups function. Investigation of the psychological processes involved in the acquisition, representation and deployment of social category concepts necessarily requires a focus on both the naturally occurring categories in the real world and novel categories created in the laboratory that can control for such variables as familiarity and group membership. One powerful approach to studying children’s initial intuitive expectations about groups is to focus on the acquisition of novel social categories. Because they have not previously been encountered, we can be sure that children’s subsequent judgments are based on the specific information provided in the experimental setting. And because we can present children with scenarios in which they do not belong to the groups in question, we can also avoid the biasing effects of membership. This approach allows us to examine the abstract principles by which children create social categories in the first place. We here build on previous research using a Novel Groups procedure (e.g. Gregg, Seibt, & Banaji, 2006; Ford & Stangor, 1992; Levy, Stroessner, & Dweck, 1998; Diesendruck & Eldror, 2011), in which participants are introduced to previously unfamiliar social groups, taught some information about a few exemplars (e.g., evidence that they performed antisocial behavior) and then asked to make judgments (e.g., inferences, evaluations) about other members of the group. In the present study, we present stories about individuals who might be seen as members of the same social group, and explore what cues to group membership promote social categorization. We illustrate our stories with pictures. To even further reduce the possibility that children would draw on pre-existing DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 10 knowledge, or identify preferentially with one or the other group, the drawings depicted the individuals as non-human cartoon characters. Within this framework, the present inquiry focuses on five questions. First, we examined whether the following cues were sufficient, in the absence of any shared perceptual features among category members, to promote the formation of a new social category among 4and 7-year-olds: a noun label uniting two individuals who were always depicted together, engaging in coordinated anti-social acts. Second, we examined whether in the absence of a noun label, but in the presence of shared perceptual features, social categorization would emerge in the same vignettes. Third, we explored the hypothesis that children will differentially privilege intrinsic cues like skin color over non-intrinsic cues like clothing when forming representations of social groups. Fourth, we examined the additive effect of a noun label and a visual cue, asking whether noun labels lead to a closer focus on the type of visual cues that support social categorization. Finally, we examined whether evidence of social categorization in the first 3 experiments can be explained in terms of a simple correlation among coordinated behavior between individuals who share a noun label and visual properties, when that shared behavior had no socially relevant content. We attempted to ensure that our experiments test the formation of categories that are social. They involve categories of people, or at the very least, agentive individuals with human-like properties. The information we present involves a series of social interactions amongst individual group members where individuals collectively and collaboratively engage in antisocial behavior. Third, the dependent measures we employ involve expectations regarding dispositions to act anti-socially or pro-socially, as well as DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 11 judgments about the friendship potential of members of the different groups. In sum, these studies investigate whether variation in the features defining groups (noun labels, various visible cues, nature of coordinated action) affects children’s tendency to form inductively and affectively rich social categories. Experiment 1 Experiment 1 directly examined whether, in the absence of a shared visual cue to group membership, children could use a noun label correlated with individuals engaging in antisocial behavior to establish a representation of a social category. Children heard a story about two individuals labeled as Lups who engaged in antisocial behaviors while several individuals labeled as Nifs engaged in neutral to mildly positive behaviors. To ensure that children would not identify with the target individuals and could bring no prior knowledge of social groups to bear on the task, the Lups and Nifs were not pictured as actual people but rather as cartoon figures. Despite this, the basis of categorization picked out by the label was clearly social: each Lup repeatedly engaged in antisocial acts in coordination with another individual bearing the same noun label, Nifs engaged in neutral acts, and many intergroup social interactions were depicted. Our measures of social categorization were predictions about whether new Lups and Nifs not depicted in the story would engage in novel positive and negative social behaviors, and relative friendship choices among Lups and Nifs also not presented earlier in the story. Children are provided with two bases for establishing a representation of a social category: shared noun label and evidence of socially relevant behavioral consistency among individuals. Although individual characters differed from one another visually (e.g., hair style, shape of facial features, presence of freckles, etc.), this individual DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 12 variation did not vary systematically with group membership. Specifically, individuals only differed in the noun label used to refer to the them and by the tendency to perform antisocial acts. Thus, no physical differences between individuals could serve as a cue to group membership. In addition to supporting inductive inferences about social behavior, social categories also support intergroup preferences. These preferences often serve as crucial mediators of intergroup behavior including predicting friendship choices among children and predicting hiring decisions among adults. As such, we also measured children’s preference for the two novel groups. If categorization of novel social groups requires clear visual markers to differentiate categories, we would not expect children to generalize novel antisocial behaviors to new members of the group portrayed negatively in the story (the Lups) and they should neither generalize more prosocial behaviors toward nor prefer members of the other group (the Nifs) as potential friends. Method Participants Thirty-five 4-year olds (mean age 4 years, 5 months; females = 18) and 30 7-year olds (mean age 7 years, 7 months; females = 15) were recruited. Child participants were recruited from the community and were tested either in our laboratory or at local daycares, pre-schools, after-school programs and museums. A legal guardian provided informed consent for all children. Procedure All participants were tested individually and worked one-on-one with an experimenter who read aloud the instructions and who recorded their responses on a DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 13 laptop computer. Using a repeated measures design to control for individual baseline differences, participants were first asked to make attributions of positive and negative social behaviors to novel agents. Subsequently, participants were asked to report a preference among further of these individuals. During this pretest the characters were labeled as “this person”, however, these novel agents would eventually be labeled as Lups and Nifs during the post-test assessments. Next, participants heard a story in which individuals were either labeled as a Lup or as a Nif and learned that individual Lups behaved in an antisocial manner. The story involved multiple vignettes involving the same 2 people from each group. Next, participants were presented with the same attribution and preference questions from the pre-test, however, this time the targets were labeled as either a Lup or as a Nif. The specific individuals that appeared in the story were not included in the set of generalization and preference questions, thereby ensuring that any systematic judgments reflected categorical judgments about the Lups and Nifs and not about specific individual Lups and Nifs. Further, each individual in the pre-test and post-measures were never consistently linked with either a positive or negative behavior. While the story provides information about the behavioral tendencies of specific individuals who share a common noun label, the dependent measures require participants to use the noun label to facilitate categorization of new individuals and subsequent category dependent inferences (e.g., that Lups, more so than Nifs, are likely to engage in novel antisocial acts, are less likely to engage in prosocial behavior and are less desirable as friends). Pre-test Measures DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 14 Attribution Task. Participants first began by responding to 8 ambiguous situations each involving two novel actors. Four of these situations involved a negative social behavior (e.g., someone knocked someone else over, someone caused a car crash) and four involved a positive social behavior (e.g., helping a friend at school, cleaning up spilled milk). See Appendix 1 for a full list of these questions. For each question, participants saw two cartoon drawings of unique people-like agents side-by-side (had eyes, ears, mouth, nose, arms, feet and were said to have engaged in people-typical activities such as skateboarding, drinking milk, writing, etc.). Participants were then instructed to, “Point to the person who did the thing I tell you about”. The experimenter entered each response into the computer. See Figure 1 for example illustrations of these characters. Group Preference Task. Following the Attribution Task, participants viewed two novel individuals side-by-side and were asked to indicate which person they liked more. There were a total of 8 such trials involving different individual exemplars not seen previously. Story Manipulation Following the Group Preference Task, the experimenter read a short story about two novel social groups. Individuals in the story differed systematically in the noun label used to refer them, either as a Lup or as a Nif, and in the type of behavior performed (either antisocial acts or neutral to mildly positive acts). During the story participants observed two individuals sharing a common label (the Lups) engage together in antisocial behaviors (e.g., not sharing, stealing a pie, tearing up someone’s artwork). Importantly, these two individuals were always the transgressors and when there was a known victim, DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 15 it was a member of the other group (the Nifs). See Appendix 2 for the full text of the story and see Figure 2 for example illustrations that accompanied the story text. Whereas the text of the story refers to the two main characters as both “these Lups” and “the Lups”, the full context of the story strongly implies that these phrases refer to the two particular individuals displayed on the screen before them as opposed to a generic interpretation referring to all Lups; indeed, each time the experimenter referenced either “these Lups” or “the Lups”, she pointed to both of the Lups on the screen. Post-test Measures Next, participants responded to the same Attribution Task questions followed by the Group Preference Task questions as above. The questions within each task appeared in a random order, but task order was fixed. This time, each individual was labeled with a noun label (either as a Lup or as a Nif). For example, on a particular trial of the Attribution Task a participant would be asked, “Who stole some money? This Lup or this Nif?” and participants responded by pointing to one of the two individuals on the monitor. Importantly, children’s judgments were elicited concerning the likelihood that new members of both groups would engage in new behaviors not previously depicted in the story, and they were asked their friendship preferences among a new Lup and a new Nif not seen before. In this way, the inductive potential of the groups was measured. In each case, participants had to decide to select one Lup or one Nif. Results All participants completed the experiment. Preliminary analyses revealed no effect of participants’ gender at any age and therefore subsequent analyses collapsed across this factor. For the Attribution task, two percentages were calculated: One for the DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 16 frequency with which bad behaviors were attributed to the group portrayed negatively (out of four possible trials) and one for the frequency with which good behaviors were attributed to the group portrayed negatively (out of four possible trials). For the preference task, one percentage was calculated for the frequency with which a member from the group portrayed negatively was chosen over a member of the group not portrayed negatively (out of 8 possible trials). For each task separate percentages were calculated for pre-test and post-test responses. During the pre-test measures no visual cues or labels could be used to identify the Lups and Nifs. Thus, pretest responses were linked to specific individuals who would eventually be paired with a particular group label at post-test, allowing us to calculate the effect of being part of either group. Across all age groups and tasks, the pre-test percentages did not differ from chance (50%). Difference scores are reported below to convey the magnitude and direction of the change in attributions and preference following the story manipulation; a summary of results is presented in Figure 3. For ease of graphical presentation only post-test percentages are displayed. Finally, for the analyses reported below we perform separate ANOVAs for the generalization of negative behaviors, generalization of positive behaviors and reported group preference. Research has suggested that positive and negative intergroup evaluations are conceptually independent and follow different developmental trajectories (Aboud, 2003), leading us to make planned comparisons between these individual measures of generalization and preference. Negative Attributions DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 17 Observing two individuals engaging in negative behavior and labeled with the same noun was sufficient to lead children of both ages to project the tendency to perform antisocial acts to new individuals bearing the same category label. Participants in both age groups generalized more antisocial behaviors to the “bad” group following the story manipulation (4-year-olds, Mdiff = 30% and for 7-year-olds, Mdiff = 50%). A 2x2 (Test Time: pretest and post-test generalization of negative behaviors; Age Group: 4-year-olds and 7-year-olds) repeated measures analysis of variance (ANOVA) confirmed this, revealing a significant main effect of test time (F1, 63 = 35.02, p < .05, η = 0.36) such that the rate of negative generalizations to new members of the “bad” group (the Lups) was higher after the story manipulation. Neither a main effect of age nor a Test Time X Age Group interaction was observed (all ps > .1). Positive Attributions Following the story manipulation, participants were less likely to generalize positive behaviors to the group portrayed negatively (the Lups: 4-year-olds, Mdiff = 11.5% and for 7-year-olds, Mdiff = -26.5%). As with the attribution of negative behaviors, a 2x2 (Test Time: pretest and post-test generalization of positive behaviors; Age Group: 4-year-olds and 7-year-olds) repeated measures ANOVA revealed a main effect of test time (F1, 63 = 8.39, p < .01, η = 0.12). Similarly, no main effect of age group or a Test Time X Age Group interaction was detected (all ps > .1). However, it is worth noting that as a group 4-year-olds did not reveal a significant tendency to generalize fewer positive behaviors to the Lups, compared to pretest; the -11.5% change did not differ significantly from 0 (t34 = -1.02, p >.3). In contrast, 7-year-olds were significantly more likely to generalize fewer positive behaviors to the Lups on the post-test (t29 = -2.98, p < .01). DO NOT CITE WITHOUT PERMISSION Constraints on Social Categorization 18