Characterizing Children’s Expectations About Expertise and Incompetence: Halo or Pitchfork Effects?

Human culture is crucially dependent on the transmission of information from one person to another. Just as a single individual cannot possibly perform all the everyday tasks required for survival in the modern world, necessitating a division of physical labor, a single individual cannot possibly know everything required for survival in the modern world, necessitating a comparable division of cognitive labor (e.g., Keil, 2006). Depending on interest, opportunity, and ability, different people will have different pockets of expertise, and they can share that expertise when necessary. An epistemological problem, long studied by philosophers (e.g., Goldman, 1999; Hume, 1975; Plato, 2005; Putnam, 1975; Reid, 1975) concerns how a novice in a given domain determines who is qualified to provide information about that domain. As Plato (2005) asked in his dialogue Charmides, how can a layperson determine whether someone who purports to be an expert in an area really is one?

One prerequisite would seem to be recognizing that a division of cognitive labor exists in the first place, that no single individual can know everything. Although young children do sometimes act as if they believe adults are omniscient (e.g., Wimmer, Hogrefe, & Perner, 1988), Keil and his colleagues have argued that they have a nascent understanding that knowledge is not distributed equally among individuals (e.g., Danovitch & Keil, 2004; Keil, 2006; Lutz & Keil, 2002). For example, Lutz and Keil (2002) found that in a forced-choice situation, 3-year-olds could indicate that a doctor would be more likely than a car mechanic to know why people get runny noses, and that a car mechanic would be more likely than a doctor to know how to fix a flat tire. Slightly older children in that study additionally recognized that differences in expertise can be rooted at a deeper level—that a doctor would be more likely than a car mechanic to know why plants need sunlight to grow, for example, and that a car mechanic would be more likely than a doctor to know how a yo-yo works.

One important question raised by Lutz and Keil’s (2002) study—and the focus of our first study—is the extent to which children impose epistemic limits or boundaries on what they expect a given expert to know. Although children systematically selected one expert over the other when asked to choose between them in Lutz and Keil’s work, there is reason to believe that they may impute more knowledge than appropriate to a given expert when there is no obvious alternative expert available.

For example, in Taylor, Esbensen, and Bennett (1994, Study 1), preschoolers were introduced to a child who was said to know more about animals than anyone in his classroom; everyone called him an “animal expert.” When children were asked whether he would know something outside the domain of animals—how a carburetor works—4-year-olds tended to respond affirmatively, even though they denied this knowledge to themselves or to nonexpert peers. In a second, similar study, 4-year-olds also claimed that a child described as a chemistry expert would know how bird bones are different from other animals’ bones, again, even though they also said that neither they nor a nonexpert peer would know. The primary focus of these studies was on children’s understanding of knowledge acquisition rather than expertise, but Taylor et al. suggested that “our preliminary results suggest [4-year-olds] do not realize that expertise can be domain-specific” (p. 1602).

This expectation that an expert in one domain would also be knowledgeable in other domains suggests that children’s reasoning about expertise may be subject to a kind of “halo effect.” A number of studies have shown that children (and adults) often assume that an individual who has one desirable characteristic will have other desirable characteristics. For example, Cain, Heyman, and Walker (1997) found that 4- and 5-year-olds predicted that a nice child would not just be more likely than a mean child to engage in prosocial behavior; they also predicted that he or she would be smarter and more athletic (see also Benenson & Dweck, 1986; Heyman, Gee, & Giles, 2003; Stipek & Daniels, 1990; Yuill, 1997).

In our first study, we set out to investigate specifically whether children’s expectations about expertise are subject to a halo effect. Rather than pitting experts from two different domains against each other (as in Lutz & Keil, 2002), or asking yes–no questions about the knowledge of a single expert (as in Taylor et al., 1994), we presented children with an expert in one domain, along with an individual who was said to know relatively less in that domain. Children were simply asked which informant they would seek out and endorse information from—a situation that approximates the dilemma posed by Plato about how a novice can judge the knowledge of a putative expert. In one condition, the information concerned a topic squarely within the expert’s domain of expertise; in a second condition, the information concerned a topic that was unrelated. Our interest was whether children would prefer the expert informant regardless of domain (consistent with the results of Taylor et al., 1994, and the halo effect), or whether they would prefer the expert only for matters germane to his or her expertise (consistent with Lutz & Keil’s, 2002, work on children’s understanding of the division of cognitive labor).

Our procedure was derived from a recent line of work that has addressed preschoolers’ ability to track the accuracy of two potential informants (e.g., Birch, Vauthier, & Bloom, 2008; Clément, Koenig, & Harris, 2004; Jaswal, McKercher, & VanderBorght, 2008; Jaswal & Neely, 2006; Koenig, Clément, & Harris, 2004; Koenig & Harris, 2005; Pasquini, Corriveau, Koenig, & Harris, 2007; Scofield & Behrend, 2008). In these studies, 3- to 5-year-olds have preferred to learn new words from someone who has a history of having been correct about what familiar objects are called rather than someone who has been incorrect or ignorant. A subset of these studies has touched upon the scope of this preference, but this has always been in closely related domains.

For example, Koenig and Harris (2005) found that preschoolers preferred a formerly accurate labeler over a formerly ignorant one as a source for the functions of new objects. Birch et al. (2008) asked children to make inferences in the reverse direction and found that 4-year-olds preferred an informant who had been accurate about the functions of familiar objects over one who had been inaccurate when choosing which of the two would be the best source for novel object labels. Of course, knowledge about object names and object functions often do go hand-in-hand (e.g., Bloom, 2000; Kemler Nelson, Russell, Duke, & Jones, 2000), so this is perhaps not surprising. Similarly, Jaswal et al. (2008) found that preschoolers preferred a formerly accurate morphologist over an inaccurate one as the source for novel labels, but again, this may not be surprising as both concern conventional aspects of language use.

Finally, Sobel and Corriveau (2010) recently demonstrated to 4-year-olds that one informant was knowledgeable about a nonobvious internal property of a novel object (e.g., what color pin was hidden inside it) and ignorant about a nonobvious external property of the same object (e.g., what color sticker was hidden on its back), while another was knowledgeable about the external property and ignorant about the internal one. Later, children preferred the “insides” expert over the “outsides” expert as an informant about the name of the object. As Sobel and Corriveau argue, this preference is consistent with the finding that children expect an object’s insides to be relevant to its category membership, but do not have the same expectation about its outsides (e.g., Gelman, 2003). This represents another example of a situation in which the domains in which expertise generalized were closely related.

In Study 1, children learned that one individual was an expert about dogs and one was less expert about dogs. Crucially, and different from the work described above, the individual who was less expert about dogs did not commit errors, nor was she ignorant about them; she was simply less knowledgeable compared to the expert. Children in one condition chose which of the two individuals would be a better informant about the names of dogs, and children in another condition chose which would be a better informant about the names of artifacts. We chose dogs and artifacts because they represent distinct domains, and from an adult perspective, a dog expert has a very circumscribed type of knowledge.

We were not interested in the means by which children came to see the two informants as holding differential expertise about dogs, but it was important that they recognize that one was more knowledgeable about dogs than the other. Thus, we provided them with two overlapping types of evidence regarding the informants’ contrasting levels of expertise. First, the experimenter explained that one of the informants was a dog expert and knew more about dogs than anyone else (a la Taylor et al., 1994), including the other potential informant. Second, children themselves heard the dog expert refer to a series of dogs by breed names (e.g., “That’s a Sharpei dog”) whereas the other informant referred to perceptually obvious characteristics of the same dogs (e.g., “That’s a furry dog”). We thought it unlikely that children would recognize the breed names but expected that their unfamiliarity might add to the cover story that the informant who used them was especially knowledgeable about dogs—particularly when contrasted against the informant who referred to the same dogs in perceptually obvious ways.

In Study 2, we used a similar procedure to investigate the domain-specificity or generality of children’s expectations about incompetence. Just as children are more likely to confront individuals who possess pockets of expertise than general omniscience, they are also more likely to encounter people who display concise pockets of incompetence rather than an all-encompassing incompetence. To explore whether children bring domain-specific expectations to an incompetent source, we presented them with a speaker who was clearly incompetent in the limited domain of dogs. Children in Study 2 learned that one of the informants was inexperienced with dogs and knew less about dogs than anyone else, including the other informant. Next, children saw this less experienced informant refer to a series of dogs incorrectly (e.g., “That’s a Sharpei cat”) paired with an informant who made only neutral remarks (e.g., “That’s a nice one”). Our question in Study 2 was parallel to that of Study 1: Would children mistrust the incompetent source across domains or just for information regarding dogs?

Study 1

Method

Participants

Participants were sixteen 3-year-olds (mean age = 3 years 5 months [3;5], range = 3;1–3;10; 7 girls) and sixteen 4-year-olds (mean age = 4;4, range = 4;0–4;10; 8 girls). The sample was randomly selected from a database consisting of children from a Midwestern university city. Children from this pool are predominantly Caucasian, native English speakers from middle- to high-socioeconomic-status (SES) homes. An additional three children were dropped from the final sample because of parental interference or not completing the study.

Design

Children were randomly assigned to condition. Half of the children at each age participated in a dog condition, where they made judgments about which of two informants to trust about the names of dogs, and half participated in an artifact condition, where they made judgments about which of two informants to trust about the names of artifacts.

Materials

Children were presented with video clips featuring two female adult actors (one in a pink shirt and another in a blue shirt), sitting at a table. The actors were shown 8 × 10 in. color pictures of four dogs during familiarization and, depending on condition, either four additional dogs or four novel artifacts during the test phase. As will be described next, all children experienced exactly the same scaffolding, familiarization and explicit judgment trials. Depending on condition, half of the children saw the informants label dogs (dog condition) and half of the children saw the informants label artifacts (artifact condition). Hard copies of the same pictures shown to the actors were also shown to children.

Procedure

To introduce the task, the experimenter pointed to still images of the two informants and said, “I’ve got these two friends. See? One has a blue shirt and one has a pink shirt. Do you want to meet them?” The remainder of the procedure consisted of the following phases in this order: scaffolding, first explicit judgment question, familiarization, second explicit judgment question, and test trials.

Scaffolding. Children were shown two still images of the two informants on the computer screen. The experimenter pointed to each one, and explained,

Here’s my friend, Jenny, and here’s my friend, Kate. Kate is a grown up just like your mom. Jenny is a grown up just like your mom too but she knows a lot about dogs. She is an animal doctor and works with dogs all the time at the animal hospital. She knows more about dogs than anyone else I know and she knows more about dogs than Kate. All of our friends call her a dog expert because she knows so much about dogs.

For half of the participants, the pink-shirted actor was the dog expert, and for the other half, the blue-shirted actor was the dog expert.

First explicit judgment trial. Immediately following the scaffolding, the experimenter asked the first explicit judgment question, “Who knows more about dogs?” Children did not receive any feedback in response to their answers.

Familiarization. The purpose of the familiarization period was to provide children with the opportunity to observe for themselves the informants’ differential expertise about dogs. The experimenter explained,

Jenny and Kate are going to tell us about some dogs. I want you to listen very carefully to what they say and then I’m going to ask you what you think.

She began the familiarization clip, in which color photographs of four dogs of different breeds appeared, one at a time, on an easel next to the actors. One actor commented on all four pictures, and the other actor did the same. The dog expert labeled each one using subordinate breed names (e.g., “See? That’s a Basenji dog. Yep, that’s a Basenji dog”), and the novice described a perceptually obvious characteristic of the dog (e.g., “See? That’s a brown dog. Yep, that’s a brown dog”). To ensure both that the dogs were distinct and that breed names were simple and unfamiliar, the breed names did not always actually match the breed of the depicted dog. After each utterance, the experimenter paused the video and repeated what the speaker had said. Half of the participants heard the dog expert first and half heard the novice first. Table 1 shows a list of the dogs and the actors’ comments about each one.

Table 1.
Dog Pictures and Statements Made by Informants During the Familiarization Period of Studies 1 and 2

Second explicit judgment trial. After familiarization, the children were shown still images of the two informants and asked the explicit judgment question a second time, “Who knows more about dogs?” Again, children received no explicit feedback from the experimenter on their responses.

Test trials. In the test phase, children were shown four video clips in which the two informants made conflicting claims about four new dogs or artifacts, depending on condition. Prior to each clip being shown, the experimenter presented children with a picture of a new dog or artifact (depending on condition) and asked them to indicate which informant they would like to ask about it (hereafter, “ask” trials): “Do you know what kind of dog this is/what this is called? I bet one of these people can help. Which person would you like to ask?” (Children who claimed to know the name of an object or dog breed were told, “Actually, I don’t think that’s what it’s called” before the experimenter asked them whom they’d like to ask.) Both verbal (e.g., “Kate,”“girl in the pink shirt”) and nonverbal (i.e., pointing) responses were recorded.

After each ask trial, the experimenter started the videoclip corresponding to the novel dog or artifact the child had just been shown. In each clip, a previously unseen female actor presented one informant with the picture, and asked, “Can you tell me what this is called?” The first informant responded by saying, for example, “That’s a blicket dog” in the dog condition, or “That’s a blicket” in the artifact condition. The same question was then posed to the second informant who responded by saying, for example, “That’s a jeter [dog].” Whether the novice or expert was asked to name an item first alternated across the four clips.

After the two informants had labeled a picture, the experimenter paused the video, repeated the informants’ labels, and asked children to endorse one of the two labels (hereafter, “endorse” trials). For example, “She said it’s a blicket [dog] and she said it’s a jeter [dog]. What do you think it’s called, a blicket [dog] or a jeter [dog]?”Table 2 shows a list of the test stimuli in both conditions as well as the actors’ conflicting comments about each item.

Table 2.
Test Stimuli for Both Dog and Artifact Conditions of Studies 1 and 2

The entire session lasted approximately 10 min and was videotaped. Children received a point for indicating the expert for every explicit judgment and ask and endorse question.

Study 2

In Study 2, we used the same procedure to investigate the scope of children’s expectations about an informant who was incompetent about a particular domain. One informant was identified by the experimenter as knowing very little about dogs, and subsequently referred to each of several dogs as a “cat”; the other speaker was introduced neutrally and made neutral comments about the same dogs. If children treat a person’s incompetence as domain specific, we expect them to avoid the incompetent informant when gathering information about dogs but not when gathering information about artifacts. If they treat incompetence as domain general, then they should avoid the incompetent informant in both cases.

Study 1 Versus Study 2 Comparison

We compared children’s preference for the dog expert in Study 1 to their preference for the neutral informant in Study 2. Although cross-study comparisons must be interpreted cautiously, given that the two studies made use of the same stimulus materials, counterbalancing, general procedure, and test questions, we felt such a comparison was warranted and could be informative.

Using only those children in the studies who passed both explicit judgment questions, we conducted a 2 × 2 ANOVA (Study 1/Study 2 × Dog/Artifact) on the likelihood that children preferred the dog expert (for Study 1 participants) or the neutral informant (for Study 2 participants). The reason for comparing children’s preference of these two informants is that they were the ones preferred in the dog condition of each study, and our interest was in comparing whether that preference was maintained in the artifact condition of each study. Children in Study 2 were more likely to have a preference for one of the two informants overall than children in Study 1, F(1, 49) = 4.96, p < .04, d = .59. However, this main effect must be considered in light of a significant interaction between study and condition, F(1, 49) = 3.91, p = .05. The interaction reflects the fact that children in Study 1 preferred the dog expert for information about dogs on an average of 64% of test trials, and children in Study 2 preferred the neutral informant for information about dogs 65% of the time. In contrast, children in Study 1 preferred the dog expert for information about artifacts on an average of just 44% of test trials, whereas children in Study 2 continued to have a preference for the neutral informant for information about artifacts—on an average of 67% of test trials. There was no main effect of condition, F(1, 49) = 2.55, p = .117.

General Discussion

Our goals were, first, to investigate whether children expect an expert in one domain to be an expert in an unrelated domain and, second, whether children expect an incompetent source to be incompetent in an unrelated domain. Study 1 demonstrated that children do not expect expertise to generalize to an unrelated domain. They preferred a dog expert over a dog novice when it came to the names of dogs but showed no preference when it came to the names of artifacts. In contrast, results of Study 2 suggest that children do expect incompetence to generalize across unrelated domains. Regardless of whether the task involved the names of dogs or artifacts, they preferred a neutral informant over one who was said to be inexperienced with and had demonstrated incompetence about dogs. In other words, children gave a domain-specific treatment to expertise and a domain-general treatment to incompetence. We next consider three important implications of these findings.

Study 1’s results suggesting that expertise may be domain specific are consistent with the results from Lutz and Keil’s (2002) work on the division of cognitive labor, but they stand in contrast to many demonstrations of a “halo effect” in the developmental literature. In these studies, children have been shown to make global judgments of others’ abilities: If a person is known to be highly ranked in one domain, they expect that he or she is also highly ranked in other domains (Benenson & Dweck, 1986; Cain et al., 1997; Miller, 2000). In fact, at the youngest ages, the domains need not be closely related: being smart may be associated with being nice and perhaps being a fast runner as well (Droege & Stipek, 1993; Heyman et al., 2003; Stipek & Daniels, 1990; Yuill, 1997).

Our failure to find a halo effect seems most at odds with the findings of Taylor et al. (1994). In that study, 4-year-olds often credited an acknowledged expert in one domain (e.g., a “chemistry expert”) with special knowledge in another domain (e.g., how bird bones differ from other animals’ bones). There are at least two methodological differences between the studies that could help to explain these conflicting results. First, the primary emphasis of Taylor et al.’s study was on children’s understanding of knowledge acquisition; the single yes–no question about expertise was a minor part of their project, resulting in what they acknowledged were “preliminary results” (p. 1602). In contrast, we assessed children’s expectations about expertise on eight trials (four ask and four endorse; statistically, there was no difference between the two types of test trials). This allowed us to compare children’s performance to chance, across conditions, and to conduct analyses by individual. In each case, we found differences between the dog condition and the artifact condition in Study 1, which suggested that children’s expectations about expertise are domain specific.

A second methodological difference between the two studies is that the domain of acknowledged expertise in our Study 1 was lower in the taxonomic hierarchy than the domains used by Taylor et al. (1994). Specifically, Study 1 used an expert about dogs, whereas Taylor et al. used an “animal expert” or “chemistry expert.” Perhaps 3- and 4-year-olds expect a subordinate or basic level expert to have less knowledge than experts at a higher level. If our expert had been an animal expert, perhaps children would have had broader expectations about the scope of her expertise. This remains an intriguing question for future work.

Although we found no evidence in Study 1 of a halo effect for expertise, Study 2 suggested a “pitchfork effect” for incompetence. Someone who was clearly incompetent about dogs was avoided not just when it came to dogs, but also when it came to artifacts. We suggest that the difference in the domain generality of expertise and incompetence may be related to a general psychological phenomenon widely studied among adults, known as a “negativity bias” or a “positive–negative asymmetry” (Baumeister, Bratslavsky, Finkenauer & Vohs, 2001; Vaish, Grossman, & Woodward, 2008). In the adult literature on impression formation, one common finding is that negative information about a person or a new acquaintance generally carries more weight than good information about a person (e.g., Peeters & Czapinski, 1990).

For example, when college students were asked to judge the likability of a target person who was said to perform a series of both negative and positive behaviors (he or she ordered the most expensive drink when someone else bought a round but also purchased groceries for an ill neighbor), negative behaviors were weighted more heavily (Vonk, 1996; see also Fiske, 1980; Vonk, 1999). Similarly, children require fewer examples of negative behaviors to infer negative traits about other people than they do positive behaviors to infer positive traits (e.g., Aloise, 1993), and their memory for individuals who have done negative things (e.g., stolen the class’s cookies) is better than their memory for individuals who have done positive things (e.g., brought in cookies for the class; Kinzler & Shutts, 2008).

Our studies seem consistent with this negativity bias in that information, suggesting that a particular individual was incompetent in one domain (i.e., “bad”) readily led to a global evaluation of that individual as incompetent, but information suggesting that a particular source was knowledgeable (i.e., “good”) in one domain did not lead to a global evaluation of that source as especially knowledgeable across domains. Although the “negativity bias” has been recently emphasized, specifically in early emotional development (Vaish et al., 2008), the possibility of a cognitive or epistemic negativity bias seems reasonable given the nonemotional cases of the bias in adult memory, information processing, social learning, stereotyping, and neurological processing (for a review of the adult literature, see Baumeister, Bratlavsky, Finkenauer, & Vohs, 2001).

As noted in the Introduction, a few previous studies using the reliability paradigm have found that children’s preference for one informant over another generalized from one domain to another closely related one (e.g., names to functions, internal properties to labels; Birch et al., 2008; Koenig & Harris, 2005; Sobel & Corriveau, 2010). In each of these studies, one informant demonstrated reliability in a particular domain, whereas the other informant was either inaccurate or ignorant in that domain. It is not clear from these studies whether children were preferring the more reliable source, or avoiding the inaccurate or ignorant person. Our findings may help to shed some light on this question.

Specifically, in our studies, one informant demonstrated expertise or incompetence in a particular domain, whereas the other informant was merely neutral. (Note that this is different from the work just described, where the other informant was always inaccurate or ignorant.) Children preferred the expert over the neutral informant in the domain of the expert’s expertise, but not in a different domain (Study 1), and they avoided the incompetent individual in favor of the neutral one both in the domain in which the incompetent one demonstrated incompetence and in a different domain (Study 2). Thus, the generalization from names to functions (or functions to names) or insides to names in the previous work may reflect the fact that children have a relatively undifferentiated mistrust of speakers who have been inaccurate or ignorant, not that they have an undifferentiated trust of those who have been competent. The implication of this (along with the negativity bias described earlier) is that when reasoning about how trustworthy someone is, examples of when they have been wrong may be weighted more heavily than examples of when they have been right.

Indeed, in Corriveau, Meints, and Harris (2009), 3-year-olds treated a previously accurate informant and a previously neutral one as equally good informants, suggesting that being accurate did not confer any special status on a potential informant when the alternative was someone who had been neutral. Furthermore, in Pasquini et al. (2007), 3-year-olds consistently avoided an informant who had been wrong, even if that individual had erred just once in the past. We suggest that young children may not form impressions of someone’s trustworthiness (and perhaps a host of other traits as well) by adding up and averaging all of the information they have about that person, good and bad alike. Instead, like adults, negative information may have stronger effects on their final impressions. Given children’s early sensitivity to negative information and to inaccuracy in particular (Koenig & Woodward, 2010), it seems possible that children younger than the 3- and 4-year-olds tested here might also show an asymmetry in which incompetence generalizes to other domains, but competence does not.

As noted in the Introduction, we were not concerned with the means by which children came to recognize the differential dog knowledge of the two informants in the present work—the scaffolding, the observations, or some combination of the two. Our goal in this initial investigation was to understand whether they expected an individual’s expertise or incompetence in a particular domain to be limited to that domain, or to generalize across domains. The possibility that children treated the subordinate breed names as a piece of evidence in favor of a speaker’s expertise deserves further attention. Thus, an important question for future work will be to investigate precisely the circumstances that lead children to recognize differences in expertise.

Other questions that emerge from these experiments concern delineating the scope of children’s attributions and the way they may change with development. For example, children’s deference to the dog expert in Study 1, though systematic and unique to dog information, was not at ceiling levels. If children’s appeal to the expert becomes more marked with age, it is possible that halo effects could emerge with development. Furthermore, children in Study 1 might have preferred the dog expert if the test questions had been about a more scientific domain or a distinct animal species. Children may appreciate that dog expertise does not reach far enough to encompass everyday artifacts, but they may believe that it encompasses other, less distally related things (e.g., sharks, insects, people, etc.; Keil, Stein, Webb, Billings, & Rozenblit, 2008; Lutz & Keil, 2002).

In conclusion, results from Study 1 suggest that there is no halo effect for children’s expectations about expertise: An expert in one narrow domain is not necessarily especially knowledgeable in another, unrelated domain. However, results from Study 2 point toward a pitchfork effect for children’s expectations about incompetence: Someone who is clearly incompetent in one narrow domain is avoided not just as a source in that domain but also in another, unrelated one.