Bias in Labeling

Labeling refers to using specific word(s) to denote a social category. The labels we use reflect which aggregates of people are singled out as meaningful categories. By using conventional category labels, people continuously confirm and maintain which categories are considered as meaningful for categorization within a (sub)culture. Cultures commonly have category labels for age groups (e.g., elderly, adolescents), for ethnicity (e.g., Asian, Caucasian), nationality (e.g., Canadian, Dutch), gender (e.g., men, women), and professions (e.g., plumber, professor).

There are, however, many arbitrary characteristics based on which we could—but conventionally do not—categorize aggregates of people. We do not, for instance, have a label for people who regularly eat carrots, or people who wear colored (vs. black) shoe laces. However, such unconventional categorizations may suddenly become relevant in particular circumstances (e.g., carrots turn out to be poisonous, or shoe laces become a fashion statement). In such circumstances, the group may become a topic of discussion, and people will likely develop a category label to refer to them. In communications, they will likely initially be described (e.g., “people who regularly eat a carrot”), but soon a more definite noun label will evolve (e.g., carrot-eaters). After an initial categorization, subsequent use of pronouns (e.g., we, us vs. they, them) can distinguish individuals as members of such an identifiable out-group.

The use of category labels has important consequences for impression formation. Once a group is linguistically labeled, it is explicitly defined and distinguished from other groups, and thereby gains in apparent reality. Several studies have shown that even trivial category labels induce perceivers to accentuate perceived similarity among members within the labeled category (i.e., they are all alike) and to exaggerate the differences between categories (e.g., Corneille & Judd, 1999; McGarty & Turner, 1992). Foroni and Rothbart (2011, 2013), for instance, presented participants with silhouette drawings of female body types, ordered on a continuum from very thin to very heavy, and had participants estimate their absolute weight and similarity in “personality,” “life style,” and “body type.” The silhouettes were presented in various conditions, in visual groupings, and either with or without a label. Results showed that the presence of a label (compared to no label), even when participants self-generated these, reduced perceived differences between members of the same (labeled) category, while the perceived differences between members of different categories became larger. This shows that once a label is imposed on an aggregate of individuals it obscures our perception of diversity between individual category members.

Moreover, the formation of a stereotype becomes more likely when a category label is used. That is, the more a collection of individuals is perceived as a meaningful, coherent group, the more likely perceivers will seek stereotypic characteristics that are considered to be essential to its members (Abelson, Dasgupta, Park, & Banaji, 1998; Crawford, Sherman, & Hamilton, 2002; Hamilton & Sherman, 1996; Yzerbyt, Corneille, & Estrada, 2001). Once such a stereotype is formed, the use of the category label will be enough for recipients to bring this existing stereotype to mind (Dijksterhuis & Van Knippenberg, 1996).

Interestingly, however, different types of linguistic labels are associated with a different strength in recipient inferences. When individuals are labeled with a noun (e.g., “a carrot-eater,” “a chocolate-eater”), recipients infer that the relevant characteristic is more essential, stable, and resilient to the person, compared to when the individual’s characteristic is described by verbs (e.g., “eats carrots whenever she can,” “eats chocolate a lot”; Gelman & Heyman, 1999; Walton & Banaji, 2004).

A similar difference in recipient inferences has been shown for even more similar grammatical forms, namely nouns versus adjectives. Nouns and corresponding adjectives can be very similar in form (e.g., being a German vs. being German; a homosexual vs. homosexual) as well as in function (e.g., ascribing particular characteristics, such as nationality; Graf, Bilewicz, Finell, & Geschke, 2013). Nevertheless, Carnaghi, Maass, Gresta, Bianchi, Cadinu, and Arcuri, (2008) showed that nouns have a more powerful impact on impression formation than adjectives. Nouns, such as “an athlete,” more strongly induce stereotype-congruent inferences about the target, and simultaneously inhibit counter-stereotypical inferences compared to adjectives, such as “athletic.” Furthermore, nouns induce stronger dispositional inferences compared to adjectives. That is, the relevant characteristic (i.e., being athletic) is seen as part of a profound and unchangeable disposition, as more informative about the person, to have a higher enduringness, and higher likelihood of future repetition (Carnaghi et al., 2008). A noun category label, compared to an adjective, thus conveys more stereotype confirming inferences. Moreover, when in- and out-group members are labeled with nouns (compared to adjectives), this induces a stronger intergroup bias (i.e., ingroup-favoritism; Graf et al., 2013).

Another important difference between label types is whether they are formulated with a plural noun (e.g., Germans are . . . ) or a singular noun (e.g., This German is . . . ). Plural nouns typically have a generic reference, in that they refer to the category as a whole. By contrast, singular nouns single out an individual group member. Plural, compared to specific singular labels, induce stronger recipient inferences in both children and adults (Rhodes, Leslie, & Tworek, 2012). That is, when a new category is referred to using a plural noun (or using indefinite singular generics that have a generic reference, e.g., The German is . . . ), the associated properties are expected to be innate and inevitable and to generalize to other category members. A singular label still induces stronger inferences compared to no label, but the use of a plural label leads to the strongest category expectancies. Other research in developmental psychology with pre-school children confirms the importance of plural noun statements (e.g., dogs are friendly) in the formation of (social) category stereotypes (Bigler & Liben, 2006; Cimpian & Markman, 2009).

Importantly, the use of these different types of labels depends on the information that is communicated. The use of strong definite noun labels (e.g., a Muslim) is reserved for people showing characteristics and behavior that fits with existing stereotype expectancies for the applied category (e.g., regularly visits mosque; fasts during Ramadan). In contrast, to refer to people showing characteristics or behaviors that are inconsistent with existing social category expectancies, speakers are less likely to use a strong noun category label. Individuals showing stereotype inconsistent characteristics and behavior are more likely referred to using modified noun labels (i.e., subtypes; non-practicing Muslim), or with adjectives (e.g., Muslim person), or instead categorized in an alternative category (e.g., an Arab; Richards & Hewstone, 2001). Likewise, generic noun labels are particularly used when speakers believe the category is a distinct group rather than a non-distinct kind (Rhodes et al., 2012).

In some cases, speakers can use different category labels to refer to the same social group. Different labels, even when referring to the same group, may become associated with a different stereotype content (Maass, Suitner, & Merkel, 2014). Derogatory group labels or stigma, for instance, are associated with a different, more negative representation than neutral labels (Carnaghi & Maass, 2007; Dovidio, Major, & Crocker, 2000). When negative and derogatory (metaphorical) nouns are used in this way to describe certain social groups, this can have disastrous, negative consequences for the group described. Musolff (2014), for instance, describes how the use of a “parasite” label to refer to immigrants has legitimized negative and discriminatory language towards this specific (minority) group. The type of label used may call for and justify particular responses or even policies dealing with individuals from the category (Moncrieffe & Eyben, 2013).

In sum, any (sub)culture develops its own conventional labels for (social) categories. In some cases, speakers can use different category labels to refer to the same social group. The choice of one of these labels over the other implies different characteristics of the described group or individuals. Research on linguistic biases shows that, even within the choice of one specific label, different linguistic choices lead to different inferences about the described group. That is, using a noun label over an adjective label leads to stronger inferences in recipients. The use of strong definite noun labels, however, is reserved for people showing stereotype consistent characteristics and behaviors for the applied category. To refer to stereotype inconsistent individuals, in contrast, people tend to use labels that induce weaker stereotype inferences (e.g., adjective labels, or nouns modified by an adjective indicating subtypes). Thus, the choice of one label over the other not only follows from, but also induces recipient inferences that are in line with the social category stereotype of those using the label.

Bias in What We Communicate About Categorized Individuals

A second type of bias is related to stereotype content and deals with the question of what information is shared about people once they have been categorized. Numerous studies on serial reproduction and dyadic conversation have revealed a stereotype-consistency bias. This shows that stereotype-consistent information, compared to stereotype-inconsistent information, is more prominent in communications about categorized individuals (e.g., Kashima, 2000; Klein, Tindale, & Brauer, 2008; Ruscher, 1998; Schaller, Conway, & Tanchuk, 2002). Once target individuals are labeled as a member of a specific category (e.g., by using a label like “an alcoholic”), conversation dyads increase their focus on label-congruent information (Ruscher, 1998; Ruscher, Hammer, & Hammer, 1996). Particularly when information is already part of the common ground (i.e., shared cognition) between communication partners, communication of stereotype-congruent knowledge becomes more likely (Fast, Heath, & Wu, 2009; Lyons & Kashima, 2001, 2003).

Such a general tendency for people to systematically communicate more about certain aspects and traits of categories relative to others has the potential to determine stereotype content (Schaller et al., 2002). Members of (sub)cultures are exposed relatively more to congruent information of shared stereotypes, which leads to a continuous confirmation of existing stereotypic associations. The more often a link between a social category and a certain attribute is perceived, the stronger the attribute in question becomes associated with the social category. This, in turn, increases accessibility and likelihood of being activated and shared again in subsequent communications. By repeatedly retelling a certain stereotypic representation, the representation can thus become a widely shared, cultural belief among members of a subculture (Bratanova & Kashima, 2014).

Next to increased accessibility of existing stereotype content, another important explanation put forward for the tendency to predominantly communicate shared stereotype-consistent information, is that sharing this information is relationally beneficial to communication partners. People prefer to communicate information they believe resonates with their audiences, because it allows them to develop common ground and it facilitates positive perceptions among group members, such as feelings of similarity, liking, and agreeableness (Bratanova & Kashima, 2014; Clark & Kashima, 2007; Higgins, 1992). In addition, conveying shared stereotype-consistent information to a conversation partner likely allows for a relatively smooth interaction. That is, sharing stereotype-consistent information usually requires fewer resources (in communication time, number, and length of utterances) than stereotype-inconsistent information, because conversation partners can easily reconcile the shared information with existing stereotypic beliefs. Furthermore, sharing stereotype-consistent information less likely leads to misunderstanding or disagreement (Klein et al., 2008).

In sum, what is shared about a social category in general, and about individuals who are categorized in a given category, is an important determinant of stereotype content. The tendency to predominantly share information that is part of existing stereotypes is an important factor in the maintenance of stereotypes.

Bias in How We Formulate Stereotype-Related Information

A third type of bias is related to linguistic variations in communications about categorized people. When speakers describe the behaviors of categorized individuals, their formulations subtly refer, in various ways, to whether or not the information was stereotypically expected. Research on linguistic biases shows that behavioral information that is in line with social-category knowledge (i.e., stereotype-consistent) is formulated differently compared to incongruent information (i.e., stereotype-inconsistent). Several subtle variations in verbal formulation can be used to communicate whether a behavior was expected or unexpected for a categorized target (Beukeboom, 2014).

Any language contains an abundance of potential explicit words to indicate that a behavior is typical, or rather that it is an unexpected one-time-event. Such words include adjectives (e.g., expected, invariable, unexpected, surprising), adverbs (e.g., always, constantly, invariably, indeed, unexpectedly, once), determiners (e.g., every, few) and set phrases (e.g., at all times, as always). These words and phrases explicitly refer to existing expectations about the target’s behavior that were either confirmed or challenged.

Language also has more implicit ways to convey existing expectancies. One of the most studied linguistic means associated with stereotype expectedness is language abstraction, as defined by the Linguistic Category Model (LCM; Semin, 2011; Semin & Fiedler, 1988, 1992; Wigboldus & Douglas, 2007). The LCM distinguishes five different word categories that vary on a concrete-abstract dimension. Most concrete are descriptive action verbs (dav). These describe single, observable actions (e.g., “Jack winks at Sue”) and preserve perceptual features of the event. The second category (interpretive action verbs and state action verbs) describes specific observable events, yet are more abstract in that they refer to a general class of behaviors and do not preserve the perceptual features of an action (e.g., “Jack flirts with Sue"). The next category (state verbs) typically describes an unobservable emotional state and not a specific event (e.g., “Jack likes Sue”). Most abstract are adjectives (e.g., “Jack is flirtatious”) that describe only the subject, show no reference to context or specific acts and thus generalize across specific events and objects. The level of abstraction in descriptions of social events shapes the types of inferences that recipients draw. With increasing abstraction, the more recipients infer that the behavior was due to stable and resilient dispositions, rather than contextual and situational forces, and the more likely it is to recur in the future (Semin & Fiedler, 1988, 1992).

The LCM formed the basis for a major contribution to a linguistic mechanism underlying the communication of stereotypes; the Linguistic Intergroup Bias (LIB; Maass, Salvi, Arcuri, & Semin, 1989). The LIB refers to the hypothesis that desirable behaviors of in-group members (i.e., members of the same group the speaker belongs to) and undesirable behaviors of out-group members (i.e., members of a rival group compared to the speaker’s group) are described at a relatively high level of language abstraction (e.g., adj; the in-group member is helpful; the out-group member is aggressive). In contrast, to describe an out-group member showing desirable behavior and an in-group member showing undesirable behavior, relatively low levels of language abstraction are used (e.g., dav; the in-group member hits somebody; the out-group member opens the door for someone; Maass et al., 1989). Because the different LCM categories elicit different cognitive inferences, the implicit meaning that is communicated varies as a function of level of abstraction. By describing desirable behavior of in-group members and undesirable behavior of out-group members abstractly, these behaviors are portrayed as stable and highly diagnostic traits. Undesirable behavior of in-group members and desirable behavior of out-group members, in contrast, are portrayed as situationally determined and exceptions to the rule.

Research demonstrated that the LIB mechanism also operates outside an intergroup context and may result from general expectancies (Maass, Ceccarelli, & Rudin, 1996; Maass, Milesi, Zabbini, & Stahlberg, 1995). Given that expected behavior is considered to be more stable, diagnostic, and typical than unexpected behavior, expected behavior is more appropriately described with abstract terms. Wigboldus, Semin, and Spears (2000) demonstrated that stereotypic expectancies give rise to differences in language abstraction and termed this phenomenon the Linguistic Expectancy Bias (LEB). For example, to describe behavior that is inconsistent with the male stereotype (e.g., crying), people use relatively more concrete language (e.g., he has tears in his eyes). In contrast, when describing a woman demonstrating the same—but female stereotype-consistent—behavior, people tend to use more abstract language (e.g., she is emotional).

Given that abstract descriptions provide more information about the actor’s stable dispositional qualities and less information about the specific situation (Semin & Fiedler, 1988), abstract descriptions of stereotype-consistent behavior endorse existing stereotypic beliefs. That is, abstract words used for stereotype-consistent behaviors imply attribution to stable traits that are likely to be repeated (internal attribution), whereas concrete words used for stereotype-inconsistent behaviors imply attribution to transient situational causes, suggesting exceptions to the rule (external attribution; Wigboldus et al., 2000; Wigboldus & Douglas, 2007). As recipients infer this information from these subtle variations in behavior descriptions, this endorses the stereotypic expectancies the speaker conveys in his/her language.

Besides language abstraction, research revealed a number of other linguistic means that function to maintain stereotypes. One way by which speakers convey that a person’s behavior is stereotype inconsistent is shown in the Stereotypic Explanatory Bias (SEB; Hamilton, & Sherman, 1996; Hastie, 1984; Sekaquaptewa, Espinoza, Thompson, Vargas, & von Hippel, 2003). The SEB proposes that speakers tend to produce more explanatory comments for stereotype-inconsistent (vs. consistent) behaviors. Such explanations of stereotype-inconsistent behavior are argued to reflect a speaker’s attempt to make sense of a surprising inconsistency (e.g., the man was crying, because he had a rough day). For stereotype-consistent behavior (e.g., the woman was crying), in contrast, such an explanation is not called for. The explanations thus reflect a speaker’s attempt to resolve why the inconsistency occurred in order to maintain coherence in one’s existing stereotypic impression. Both concrete situational descriptions and explanations are considered to be linguistic reflections of such cognitive attempts to resolve inconsistencies.

When communicating about stereotype-inconsistent information, speakers can also introduce stereotype-consistent terms, for instance by using negations or irony. Research on the Negation Bias (NB; Beukeboom, Finkenauer, & Wigboldus, 2010) revealed that the use of syntactic negations (e.g., not stupid, rather than smart) is more pronounced in descriptions of stereotype-inconsistent compared to stereotype-consistent behaviors. For example, if one holds a stereotypic expectancy that girls are emotional, but a particular girl violates this expectancy by being very tough, a speaker is likely to reveal his prior expectancy by using a negation like The girl was not emotional. In contrast, for stereotype-consistent behavior (e.g., the boy was tough; the girl was emotional), the use of negations is less likely. Negations thus allow one to introduce stereotype-consistent concepts in communications about stereotype-inconsistent information, and thereby to re-affirm existing associations with a category. Because negations tend to activate rather than suppress the negated concept, negations likely reinforce the association between the target and the negated concept (e.g., girls are emotional) in a recipient (Beukeboom et al., 2010).

A similar mechanism occurs in the Irony Bias (IB; Burgers & Beukeboom, 2016). Research on the IB shows that speakers find ironic remarks particularly appropriate to comment on stereotype-inconsistent (vs. stereotype-consistent) behaviors. An ironic remark about stereotype inconsistent behavior (e.g., “He really is a tough guy” about a man who is crying) allows speakers to introduce the expectancy (i.e., guys are tough) and also indicate that the expectancy is disconfirmed in the given situation. Like negations, ironic remarks introduce opposite terms to describe a behavioral situation and allow speakers to refer to implicit stereotypes. Both negations and ironic comments about stereotype-violating behavior can thus activate and communicate the implicit stereotypic expectancy in message recipients, even though the actual situation is stereotype disconfirming.

Moreover, recipients infer that a behavior is unexpected for a described person when it is described with a negation (compared to an affirmation; Beukeboom et al., 2010) or using ironic (compared to literal) comments (Burgers & Beukeboom, 2016). Just like the LEB and SEB, these biases thus serve to maintain knowledge about what is and what is not expected for members of a given social category. Even when people are confronted with stereotype-inconsistent information, a biased description may induce recipients to draw stereotype-confirming inferences.