Generics designate kinds but not always essences

Significance

How people reason about “kinds” (richly structured categories with high inductive potential) is at the heart of theories of cognition. One theory characterizes people as biased to essentialize, concluding that people are biased to infer that kinds are naturally determined by internal properties (essences). Generic statements (“men like sports”) are purported to evoke this bias. We advance methods by contrasting beliefs that a category is a kind with beliefs that a category has essential structure. We reveal flexibility in the presence of generics: Across varied domains and property content, generic statements increased participants’ belief that a category is richly structured with high inductive potential. Generics caused essentialism in interaction with other cues (e.g., biological properties) that reasonably prompt essentialist explanation.

Abstract

People believe that some categories are kinds with reliable causal structure and high inductive potential (e.g., tigers). Widely endorsed theories propose that people are biased to assume kinds are essential, and so naturally determined by internal causal properties. Generic language (e.g., “men like sports”) is 1 mechanism thought to evoke this bias. We propose instead that generics principally designate that categories are kinds. Participants can entertain diverse causal structures in the presence of generics: Hearing that biological properties generalize to a category (e.g., “men grow beards”) prompts participants to infer essential structure, but hearing neutral or social properties (“women are underpaid”) generalized prompts other causal beliefs. Thus, generics induce essentialism only in interaction with cues that reasonably prompt essentialist explanation. We tested our model with adult participants (n = 739 total), using measures that disentangle essentialist beliefs from kind beliefs. In study 1, we replicate prior methods with our new measures, and find that generics influence kind beliefs more than essentialism. In study 2, we vary property content (biological vs. cultural properties), and show that generics only increase essentialism when paired with biological properties. In study 3, we show that generics designate kinds but not essentialism when neutral properties are used across animals, tools, and people. In study 4, we show that believing a category is a kind increases the spontaneous production of generic statements, regardless of whether the kind is essential or socially constructed. Generics do not necessitate essentialist beliefs. Participants were flexible in their reasoning about kinds.

Some categories are shallow (white things) and others are richly structured (tigers). Richly structured categories support higher-order thought such as causal and inductive reasoning by enabling inferences based on the information contained in those structures (1, 2). We will call the assumption of a category as richly structured the “kind assumption” (3). Here, we investigate how people determine that a category is a kind, and how kinds are mentally represented. We pay special attention to social categories. The mental representation of kinds is a critical topic in most theories of high-level cognition (3–5).

A mundane but potent cue to kinds is generic language (6). Generic statements convey nonquantified generalizations: “Boys like blue.” Children and adults reason intuitively using the logic of generics rather than quantified statements (7–9). When participants determine whether “likes blue” generalizes to boys, they reason whether “boys like blue” is true rather than whether “all boys like blue” or “most boys like blue” is true. Generics often imply the property has a principled relationship to the category (10, 11). Thus, “boys like blue” often implies boys like blue in virtue of the kind of person they are. Children and adults also intuitively assume that generic properties are more central to the causal structure of the category (12, 13). Thus, hearing generics about a category communicates that the category has many generalizable properties, many principled relationships between category and properties, and many central causal features important to generating the category’s properties, all heavily implying that the category is richly structured.

By virtue of their links to high inductive potential and rich causal structure, generics might also seem to induce essentialism (10, 14, 15). Psychological essentialism is a disposition to view certain categories as having internal properties that cause their observable characteristics (3, 16). More broadly, essentialism entails a naturalistic perspective about the category (3, 16), wherein the category is treated as a natural kind (as opposed to a social construct or artifact). For example, believing that boys like blue because of their sex characteristics rather than their cultural upbringing is an essentialist belief. We will call the belief that a category is a natural kind with internal causal properties the “essence assumption” (17).

The theory of psychological essentialism envisions the kind and essence assumptions as components of the same disposition (17, 18). Though sometimes acknowledged as conceptually separable (17), even the kind and essence assumptions are said to be psychologically closely interrelated. If people assume one, they should tend to assume the other. Nowhere is this perspective clearer than the pervasive methodological assumption that inductive potential is a measure of essentialism (17–19). Additionally, participants may have a tendency to explain properties using inherent features [the so-called inherence heuristic (20)]. Attempts to interpret generics should favor inherent explanations; thus, generics should tend to favor essentialism (21). Thus, the view that generics induce essentialism is more than a specific proposal about generic language and its cognitive effects. It is also an important piece of evidence supporting the view that people are systematically biased to infer that kinds have essences (3). Proponents of the view often acknowledge that generics do not create essentialist thought (see ref. 14). Instead, inherent and essentialist thinking are preexisting cognitive biases that generics evoke. Generics presumably induce essentialism in part because our conceptual systems are fundamentally biased toward essentialism. Showing that generics fail to invariably induce essentialism would not only call for a dramatic shift in theories of generics but would raise the possibility that essentialism is not as pervasive as previously thought.

We propose an alternative model (Fig. 1) in which generics principally induce the kind assumption. That is, they primarily highlight the rich, inductive potential of categories. They only induce the essence assumption in interaction with other cues that suggest the rich structure is essential. In our model, the kind assumption is a superordinate belief, and the essence assumption is merely 1 particular interpretation of a kind’s causal structure. There are diverse kinds in the world, including tigers, oak trees, chairs, money, artists, and Christians. They vary considerably in their underlying causal structure, including whether they are natural or social (or both) and whether they have or lack an essence. A systematic account of the full range of common concepts and their interrelations must capture the heterogeneous nature of concepts and the corresponding causal patterns associated with their referents. Generics communicate that properties are generalizable and principled. Neither of these features necessitates any particular causal interpretation; thus, if participants are able to entertain different sorts of causal structures, then generics should selectively induce the kind assumption. Causal interpretations about the nature of the kind (e.g., social vs. natural) should therefore vary in the presence of generic statements. “Jews believe eating pork is wrong” expresses a possible cultural property, and so should prompt social explanations. Hearing “Jews have a high risk of breast cancer” generically expresses a possible biological property, and so should prompt natural explanations and attribution of essences. Generics designate kinds, and only trigger essentialism when the generalized properties reasonably prompt essentialist explanations.

Fig. 1.

Overview of our model. Generic language principally communicates the kind assumption. In interaction with other cues, like property content, it can communicate a variety of causal structures. For example, biological properties can communicate the essence assumption when expressed generically.

Prior accounts assume that accidental, empty, or otherwise nonprincipled properties should not induce essentialism, because such properties say nothing about the structure of the category (22). Accidental and empty properties undermine the very mechanism that props up the cognitive effects of generics (i.e., that generics usually communicate that something is a kind). Our account proposes that even if the properties are meaningful, rich, and principled, generics do not induce essentialism on their own. Rich and principled properties can support diverse causal representations, including causal representations that are not essential.

Precursors to our account include evidence that believing a social category is inductively informative is independent of believing it is natural (23, 24). These findings can be construed as showing that essentialism is multifaceted and represents a cluster of related biases (25). Essentialism may well be multifaceted, but our argument goes further: Kindhood does not imply essentialism, and essentialism is not necessary to kindhood. Essentialism usually implies that something is a kind because essentialism usually assumes the presence of rich causal structure. But kindhood and rich structure do not imply that something is essential because kinds are heterogeneous in their causal structures.

There are coherent and meaningful categories like artifacts (26), social roles (27), and relational categories (28) that lack internal essences (3). But, others argue that these categories are less richly structured and support fewer generalizations than categories like animals, plants, and minerals (29); indeed, they may be less richly structured in proportion to their reduced essential structure (3, 5). Thus, the existence of nonessential categories is potentially consistent with the conclusion that the kind assumption is strongly interrelated with the essence assumption. Generics provide a stronger test case. Generics are a vehicle for asserting rich causal structure and generalizations. Showing that generics increase the kind assumption independent of the essence assumption, and that both essential and nonessential categories can support robust generic production, provides stronger evidence for the possibility that representations of kinds are flexible. A single set of studies is insufficient to fully adjudicate whether essentialist interpretations of the world are weighted more heavily than other (i.e., social constructionist) interpretations. However, if the essence assumption can vary in the presence of generics as a function of property content, such a finding would reveal that participants can recognize that kinds can be generated by diverse causal systems, and it would highlight a psychological mechanism that affords flexible interpretations of the world.

Social categories are a critical test case for debates about generics and kind representation. Social categories can encompass a variety of causal structures. For example, senators, artists, black people, women, and people with Down syndrome span both social and natural forms of explanation. If participants are biased to assume generics imply essentialism about social categories despite the heterogeneity of their causal structures, such assumptions would provide powerful evidence for the model that generics induce the essence assumption, and for the broader proposal that people are biased to assume kinds are essential. But if generics about social categories selectively induce the kind assumption, and only contingently induce the essence assumption under certain conditions, the cognitive effects of generics would need to be reappraised, and would support a more flexible picture of kind concepts.

In several studies with children and adults, generics induced essentialism (10, 14, 15). However, these studies did not distinguish between the model that generics induce the essence assumption and our model that generics induce the kind assumption and only selectively the essence assumption. As stated earlier, inductive potential and related measures are employed as measures of essentialism, and the kind assumption is taken as evidence of the essence assumption. Additionally, these past studies provided at least some properties that support natural explanations.

To better distinguish between these 2 accounts, we employed separate measures of the kind and essence assumptions to retest prior research (study 1). One substantive departure from prior work is that we relied exclusively on adult participants. Prior work found that generics had the most powerful and reliable effect on adult cognition (10, 14); therefore, we reasoned that assessing adults’ reasoning was an important test of our account of generics and kind representation. We then distinguished between properties that prompt biological and cultural explanations (study 2). We also examined whether generics specifically designate the kind assumption across artifacts, animals, and human categories (study 3). Finally, we examined the reverse relationship by investigating whether diverse causal beliefs support the spontaneous production of generic statements (study 4).

Results

Study 1.

We first retested the generic statements used in prior work (14) using new measures that disentangle the kind assumption and the essence assumption. Participants read 16 statements about a novel social category called Zarpies in study 1 and Vawns in studies 2 to 4. The Zarpies/Vawns were diverse in race and gender but wore distinctive, shared clothing. The 16 statements were either generic (“Zarpies are…”) or specific (“This Zarpie is…”).

There were 2 measures, each containing 4 questions. These items were derived from a factor analysis that found these items (among others) factored into 2 separate dimensions (30) (see also ref. 23 for a similar factor analysis that supported the validity of these measures). The guiding assumption was that the kind assumption corresponds to a belief in the inductive potential of the category and the strength of the category–property links, and the essence assumption corresponds to the belief that the category and its properties derive from natural rather than social processes (17). Participants rated their agreement with the following statements: Kind assumption: 1) Underneath superficial similarities and differences, all Vawns are basically the same. 2) Individual Vawns have very little in common. (Reverse-coded) 3) If someone tells you a fact about an individual Vawn, that fact is very likely true of other Vawns as well. 4) For some properties that Vawns have, it makes sense to say: “This person has that property because it is a Vawn.” Essence assumption: 1) Vawns have internal or microscopic properties that cause their characteristic appearance and behavior. 2) The category Vawn was invented by people. (Reverse-coded) 3) The boundary between the category Vawn and non-Vawn is something decided by people. (Reverse-coded) 4) A Vawn can never change into a non-Vawn.

These measures allowed us to see whether generic statements induce both the kind assumption and essence assumption, or whether they selectively induce the kind assumption. We expected that the apparent large increase in essentialism uncovered in prior work was primarily a substantial increase in the kind assumption, and that the essence assumption was less affected. However, given that the stimuli included some biological properties, we did not expect an absence of an effect. In study 2, we isolate the effects of property content.

There was a large effect of generics (vs. specifics) on the kind assumption, t(136.46) = 6.46, P < 0.001, d = 1.09. This replicates the large effect found in prior work (14). In contrast, there was a small effect of generics on the essence assumption, t(133.37) = 2.16, P = 0.032, d = 0.37. Using a multilevel model, we examined whether this difference in effect was significant. We used a multilevel model with condition (generic vs. specific), measure (kind vs. essence), and their interaction, accounting for participant as a random effect. There was a significant interaction between measure and condition, b = −0.78, SE = 0.21, P < 0.001. See Table 1 and SI Appendix, Fig. S1. The large effect found in prior studies is thus attributable to the increase in the kind assumption rather than a large increase in the essence assumption.

Table 1.

Results for study 1 to 3

	Specific	Generic	P	d
Study 1—Human category; mixed properties
Kind	3.11*	4.26†	<0.001	1.09
Essence	2.87*	3.23*	0.032	0.37
Study 2—Human category; cultural properties
Kind	3.24‡	4.32†	<0.001	1.18
Essence	2.82*	2.51*	0.140	−0.30
Study 2—Human category; biological properties
Kind	3.14*	4.16†	<0.001	1.16
Essence	2.97*	3.72‡	<0.001	0.70
Study 3—Animal category; neutral properties
Kind	4.40†	4.80†	0.011	0.52
Essence	3.84†	4.07†	0.225	0.25
Study 3—Artifact category; neutral properties
Kind	4.55†	4.92†	0.049	0.42
Essence	3.08*	3.25*	0.320	0.21
Study 3—Human category; neutral properties
Kind	3.63‡	4.58†	0.001	1.14
Essence	3.07*	3.46‡	0.099	0.37

Generic vs. specific contrasts: P value and Cohen’s d.

^*Comparison to chance: below midpoint.

^†Comparison to chance: above midpoint.

^‡Comparison to chance: midpoint.

Study 2.

Study 1 found a much larger effect on the kind assumption than on the essence assumption, but both were influenced by generic statements. To better distinguish our view from prior accounts, we manipulated the property content of those generic statements in study 2. Specifically, we varied whether the properties were plausibly biological (e.g., “Vawns have freckles on their feet”) or plausibly cultural (e.g., “Vawns believe the sun is their god”). Our model predicts a similar effect on the kind assumption in both cases, since a similar number of generic utterances were provided, but we predicted a divergence in the essence assumption, since the properties support different causal inferences. Participants should infer that the category is essential when the properties are biological, and infer the category is socially constructed when the properties are cultural. This would support our hypothesis that generics induce the essence assumption only in interaction with property content that prompts essentialism (Fig. 1).

Generics (vs. specifics) greatly increased the kind assumption when either biological, t(90.24) = 5.61, P < 0.001, d = 1.15, or cultural properties, t(80.46) = 5.86, P < 0.001, d = 1.19, were used. In contrast, generics nonsignificantly reduced the essence assumption when cultural properties were used, t(83.06) = −1.49, P = 0.140, d = −0.30, but greatly increased the essence assumption when biological properties were used, t(87.63) = 3.37, P = 0.001, d = 0.70. The 3-way interaction between property content, measure (kind vs. essence), and condition (generic vs. specific) was significant, b = 1.12, SE = 0.36, P = 0.002. See Table 1 and SI Appendix, Fig. S2. Cultural generics had opposite effects on the kind assumption and essence assumption, b = −1.39, SE = 0.25, P < 0.001, whereas biological generics had the same effect on both the kind assumption and essence assumption, b = −0.27, SE = 0.25, P = 0.295. The effect of property content on the essence assumption was comparable to the effect of generics (vs. specifics) on the kind assumption, t(88.11) = 6.72, P < 0.001, d = 1.40. Therefore, generics robustly and consistently increased the kind assumption, but only contingently influenced the essence assumption; the essence assumption was influenced by biological properties stated generically rather than generics per se.

Study 3.

Study 3 examined the generalizability of the conclusion that generics principally designate the kind assumption. Social categories were a critical test case because a priori novel social categories could be natural or social. One may worry, though, that social categories are a rare case where the kind and essence assumptions come apart (see ref. 3 in response to ref. 23). Animal categories are at the heart of essentialist accounts, and the crux of the work on generics and essentialism (3, 10).

To test the generalizability of our model, we examined 3 domains: social categories, tools, and animals. We kept properties constant. Thus, we chose properties that an animal, person, or tool could plausibly possess: For example, “Vawns can pick apples quickly.” By design then, the properties did not demand any particular causal explanation. They were not empty, however, and indeed could be readily interpreted as principled relationships.

Generics increased the kind assumption in artifacts, animals, and human categories, collapsing across domain: t(249.75) = 5.23, P < 0.001, d = 0.64. Generics negligibly increased the essence assumption, collapsing across domain: t = 1.71, P = 0.089, d = 0.21. Indeed, the interaction between measure (kind vs. essence) and condition (generic vs. specific) was significant, b = −0.35, SE = 0.15, p = 0.019. See Table 1 for individual t tests; all t tests patterned the same for all 3 domains.

Notably, animals, M = 4.40, 95% CI: [4.15: 4.64], and artifacts, M = 4.55, 95% CI: [4.26: 4.84], were assumed to be kinds at default (using the specific condition as default and comparing with the midpoint, 3.5). Yet, at default, artifacts were assumed to be nonessential, M = 3.08, 95% CI: [2.80: 3.35], whereas animals were assumed to be essential, M = 3.84, 95% CI: [3.61: 4.06]. Generics were especially important for human categories: The effect size for the human category was noticeably larger than the effect sizes for animals and artifacts. Unlike animals and artifacts, novel social categories were not considered kinds prior to the use of generics: before, M = 3.63, 95% CI: [3.35: 3.90]; after, M = 4.58, 95% CI: [4.32: 4.84]. Thus, generics not only amplify the conceptual structure of social categories (as they did with animals and artifacts) but seemed to alter it more substantively. This might suggest that generics are particularly important parts of how members of a cultural group communicate to each other which social categories they deem worth attending to. Overall, then, we find strong support for the conclusion that generics induce the kind assumption but not the essence assumption across diverse domains.

Study 4.

Prior models of generics also examined the effect of essentialism on the production of generics, suggesting that essentialism preferentially supports the spontaneous production of generic statements (14, 29). Further support for our model over prior models would be provided by demonstrating that the production of generics is driven by the kind assumption rather than the essence assumption. We therefore described the novel social category Vawns as a natural kind or a social kind, and compared that with a condition where the category was not described as a kind at all. The descriptions were closely derived from the statements used in the measures of the kind assumption and essence assumption in studies 1 to 3.

The manipulation checks confirmed that the prompts inculcated the correct concepts. Prompts for the biological kind, M = 5.05, 95% CI: [4.79: 5.32], and social kind, M = 4.85, 95% CI: [4.60: 5.11], condition led participants to say Vawns were kinds and at similar rates, t(92.82) = 1.08, P = 0.285. In contrast, participants said Vawns were not a kind in the baseline (not kind) condition, M = 2.09, 95% CI: [1.74: 2.44], which was lower than the other conditions, P < 0.001. Likewise, the biological kind prompt led participants to say Vawns were essential, M = 3.87, 95% CI: [3.58: 4.17], which was greater than baseline, M = 3.11, P < 0.001. The social kind prompt led participants to say Vawns were nonessential, and so socially constructed, M = 2.29, 95% CI: [2.06: 2.53]. Indeed, this was even lower than baseline, P < 0.001.

Despite drastically different concepts of Vawns, participants spontaneously produced comparable levels of generics, χ² = 1.16, P = 0.281, OR (odds ratio) = 1.24, in the social (38%) and natural (31%) conditions (Table 2). Both of these were greater than the baseline condition (14%) (collapsing across social and biological kinds), χ² = 19.61, P < 0.001, 2.52. Thus, generics are linked to kinds, not exclusively natural kinds; social kinds encourage participants to produce generic utterances too.

Table 2.

Production of generics by condition

	Nonkind	Kind	Social kind	Natural kind
Generic	32	160	91	69
Category statements	233	462	238	224
Total statements	380	767	398	369
Generic/category, %	0.14	0.35	0.38	0.31
Generic/total, %	0.08	0.21	0.23	0.19
Category/total, %	0.61	0.60	0.60	0.61

Note: There were 3 conditions: natural kind, social kind, and nonkind. “Kind” indicates either social kind or natural kind.

Discussion

It is possible to distinguish between beliefs that a category is a kind (supports generalizations) and beliefs that a category has essential structure. Across 4 studies relying on this distinction, we found support for the model of generics as designating kinds. Generics consistently increased the kind assumption, and the kind assumption increased the production of generics. In contrast, generics had a contingent relationship to the essence assumption. Biological properties expressed generically induced essentialism; neutral or social properties expressed generically did not. Moreover, describing a category as socially constructed led participants to spontaneously produce just as many generic utterances as telling them a category was essential—so long as the category was described as a kind. Thus, our participants did not appear to be systematically biased in the presence of generics, but were versatile reasoners: As there should be, there are diverse causal structures in the world, and diverse causal structures can generate coherent and inductively significant kinds. Across 4 studies, we have demonstrated not only 1 substantial way in which people are flexible in their causal interpretations of the world but we also have uncovered a mechanism by which this flexibility is introduced into ordinary concepts. That is, participants can ascertain heterogeneous causal structures by attending to the types of properties that generalize to a category, and by identifying the causal structure that plausibly produced those properties.

It is premature to conclude there is no overarching bias toward essential or inherent explanations; a broader set of evidence is needed to identify whether alternative forms of causal reasoning (like social construction) are equally likely in ordinary cognition as essentialist reasoning. Indeed, some theorizing suggests that these biases should be particularly pronounced in young children (i.e., preschoolers 4 to 5 y old) and diminish across development (e.g., ref. 31). This developmental trajectory is supported by several developmental studies finding greater reliance on essential and inherent explanations in children than adults (31–33). In addition, social constructionist reasoning has primarily been observed in middle childhood [i.e., around the seventh year (33–35)]. Thus, a key question for future research is whether young children also interpret generics flexibly, or whether they are biased to assume essences in the presence of generics. It is also important to evaluate the relationship between generics and essentialism in adults from other populations (see ref. 36 for integration of cultural psychology and essentialism). Although essentialist biases have been documented cross-culturally (37), because researchers tend not to distinguish between the kind and essence assumptions (17) adults from other populations might differ in the flexibility of kind concepts as explored here. Our results qualify the pervasiveness of essentialist biases in adult participants in 1 English-speaking population. We cannot yet conclude that biases toward essentialism (or for that matter, biases toward the kind assumption) are not widespread in other populations.

We demonstrate that biological properties—when conveyed generically—are an important cue for evoking essentialism. Other cues could also indicate essentialism, such as distinctive physical traits and shared visual appearance; indeed, these cues seem prevalent (whether real or imagined) among many real-world social categories. Our guess is that these cues induce essentialism independent of generics. Property content is embedded in statements; whether those statements are generic or specific changes the evidentiary value of the property content (whether the properties generalize or not). But, observing first-hand that members of a category share distinctive bodily features independently substantiates the generalizability of a biological property; thus, they could independently induce essentialism. Generic statements provide a mechanism for efficiently communicating the generalizability of properties, like distinctive bodily features, without needing first-hand evidence. Nevertheless, it remains an open question how untested cues like distinctive visual appearance contribute to essentialism, and whether these cues interact with generic language.

One of the appealing features of prior research on generics was their role as a mechanism of cultural transmission. Our model retains this feature. The kind assumption induced generics, and generics induced the kind assumption. Thus, generics are a subtle and ordinary linguistic cue by which the kind assumption can be transmitted within a culture. Our model also allows generic statements to transmit diverse kind representations because we show that participants are able to infer different causal structures from the properties contained within generic statements. Thus, though our model speaks against the proposal that generics invariably induce essentialism, our model does support the interest in generics as a mechanism of cultural transmission (14). Indeed, it strengthens the potential importance of generics by greatly extending their influence.

Talking about social categories with generics, and even talking about social categories explicitly as kinds, does not induce essentialism at default, and so it is possible that generics may not induce the harms associated with essentialism in adults. Yet, the kind assumption by itself may support stereotypes (23). And our findings strongly warn against generic statements about physical and medical properties (e.g., “black people are prone to sickle cell anemia”). A negative view of generic language may be counterproductive in some cases, though. Structural and cultural properties are generic, and indeed are even principled: Black people are oppressed by virtue of being black, and Jews avoid pork by virtue of being Jewish. Like “colorblindness” (38), a blanket negative view of generics could reduce sensitivity to discrimination and may stifle benign and productive attention to cultural diversity. That being said, the interpretation of any given generic statement will involve factors not fully explored here. Consequential real-world social categories like race and gender involve a number of additional considerations. Participants reason about gender and race using a mix of social and biological explanations (39), and may not always clearly distinguish between them. Likewise, gender and race involve ideological and motivated processes that could color the interpretation of generics. Future work should consider whether generics highlighting discrimination (“women are underpaid”) or cultural diversity (“Muslims pray 5 times a day”) in real-world cases can have positive effects, or whether they are routinely distorted to produce negative effects.

The questions considered here fit into broader debates relevant to scientific practice and our speech and thought about kinds. As evidence mounted that human races are not biologically real (40–42), many philosophers and researchers argued that we should eliminate race from our scientific inquiries as well as our everyday language and thought [“racial eliminativism” (43–46)]. The idea that generics induce essentialist thought, and the idea that we are biased to essentialize kinds with or without generics, accords with such a view, since it suggests that we should worry about everyday language and thought that treats race as a kind. A more recent philosophical perspective is that race is socially real; race is a socially constructed kind (47, 48). Such a view suggests that race is a valid and necessary part of scientific inquiry, so long as race is understood as a social-scientific term rather than a natural-scientific one. Under such a view, thinking and talking about race as a real kind is not only consistent with progressive political goals but very likely necessary for achieving them.

This philosophical shift goes beyond race, and constitutes a broader recognition that real kinds include not only the natural and essential ones (e.g., tigers and gold) but include the many terms familiar to the social sciences like money and corporations (48, 49). We make a related psychological claim that our conceptual systems are not biased to infer that some categories are essential and others are mere labels without corresponding reality; instead, we are prepared to reason about the diversity of real causal structures in the world, which includes appreciating the real but socially constructed entities. The evidence provided here, in combination with a growing literature on concepts of distinctly social entities (33–35), raises the possibility that the claim “people are essentialists” might be better replaced with statements like “people are realists.” Participants generally believe most categories refer to entities in the world that really exist, be they chairs, teams, or dogs, and participants attempt to decipher how they work (50–55). But far from being a bias, realism is a widely accepted philosophical doctrine assumed in most scientific inquiry (56), and could prompt attempts to understand the world (57). Thus, though participants often have sparse and incomplete theories of the world (58), they may at least be flexible in the theories they are willing to entertain.

Materials and Methods

Study 1.

Participants.

Participants in all studies were provided consent information prior to the survey; clicking the link was taken as consent. All studies and consent procedures were approved by the Yale Institutional Review Board. We apportioned 100 “Hits” from Amazon Mechanical Turk. In the allotted time, 139 participants managed to submit a completed survey. All participants were retained. We assumed a Cohen’s d of 0.75 based on prior work (14); we recruited ∼50 participants per condition to have 95% power for t test between conditions. Demographics were provided by Turk Prime: gender, 64% male, 36% female; race, 75% white, 8% black, 8% Asian, 6% Hispanic/Latino, 3% other.

Design and procedure.

The design was to measure the influence of generic language (Vawns…X) versus specific language (This Vawn…X) on the kind and essence assumptions. Participants read 16 properties about a novel social category that were either in generic or specific format. Participants then answered 2 measures in counterbalanced order, testing the kind and essence assumptions.

Stimuli.

We presented a novel social category diverse in age, race, and gender. Members wore dark blue clothing decorated with a 5-pointed yellow star. They were depicted in various island settings and said to be residents of a remote island. The Vawns were posed in 8 pictures featuring 2 Vawns each. Each picture was accompanied by text that reported 2 properties. The properties were taken from prior work (14); see SI Appendix, Table S1. In the generic condition, the properties were reported in generic format (Vawns…); in the specific condition, the properties were reported in the specific format (This Vawn…).

Study 2.

Participants.

We again recruited ∼50 participants per condition for a total of 200 participants. In the time allotted, 188 participants submitted the survey. The sample was 59% male and 41% female. The sample was 77% white, 11% black, 7% Asian, 3% Hispanic/Latino, and 2% other.

Design and procedure.

The overall design of the study was to measure the influence of generic language (Vawns…X) versus specific language (This Vawn…X) on the kind and essence assumptions as a function of the property content (biological vs. cultural). Participants read 16 statements in either generic or specific format; the properties were either plausibly biological or plausibly cultural. Participants then answered the essence assumption and kind assumption measures. See SI Appendix, Table S1 for properties.

Study 3.

Participants.

We again recruited ∼50 participants per condition for a total of 300 participants. In the time allotted, 271 participants submitted the survey. The sample was 52% male and 48% female. The sample was 75% white, 8% black, 7% Asian, 5% Hispanic/Latino, and 5% other.

Design and procedure.

The overall design of the study was to measure the influence of generic language (Vawns…X) versus specific language (This Vawn…X) on the kind and essence assumptions as a function of category domain (animal, artifact, and human). Participants were randomly assigned to see pictures of an animal (a novel bird), artifact (a novel fruit-picking device), or human category. Participants read 16 statements in either generic or specific format; the properties were causally neutral and appropriate for animals, artifacts, and humans (SI Appendix, Table S1). Participants then answered the essence assumption and kind assumption measures.

Study 4.

Participants.

We again recruited ∼50 participants per condition for a total of 150 participants. In the time allotted, 141 participants submitted the survey. The sample was 52% male and 48% female. The sample was 75% white, 8% black, 7% Asian, 5% Hispanic/Latino, and 5% other.

Design.

The overall design of the study was to investigate the influence of the kind and essence assumptions on the spontaneous production of generic statements. We varied whether a human category was described as being a natural kind, a social kind, or not a kind (SI Appendix, SI Text). We then measured the spontaneous production of generic utterances.

Procedure.

Participants read prompts about the Vawns. Participants were then presented with 4 scenes, depicting Vawns engaged in varied activities. Participants were instructed to describe the people and events depicted. No other information was provided. After participants described the Vawns, they answered the same kind and essence assumption questions as studies 1 to 3 as a manipulation check.

Coding.

To code for generics, we looked for statements that made a claim about the Vawn category in general. The coding parameters are adapted from prior work (19); see SI Appendix. There was 94.45% agreement between 2 blind coders; disagreements were resolved prior to analyses.