What Young Children Know about Contamination and Contagion and

What That Tells Us About Their Concepts of Illness.

Charles Kalish

University of Wisconsin-Madison

To appear in Siegal & Peterson, Children’s Understanding of Biology and Health, Cambridge University Press.

Acknowledgments: Preparation of this chapter was supported by a grant from the Graduate School of the University of Wisconsin-Madison and by a Spencer Foundation-National Academy of Education Postdoctoral Fellowship.

12/11/97

Conceptions of illness provide researchers an excellent opportunity to study children’s understandings of different domains of experience. To what extent, for example, do young children distinguish biological from social or psychological aspects of illness experience? This question has taken on particular importance in light of ongoing debates about the development of biological knowledge. Susan Carey and colleagues (Carey, 1985a, 1995; Solomon, Johnson, Zaitchik, & Carey, 1996) have argued that a differentiated notion of biology does not emerge until middle-childhood. Other researchers suggest that some forms of biological knowledge are present much earlier (e.g., Simons & Keil, 1995; Wellman & Gelman, 1992). In this chapter I will focus on children’s reasoning about one central aspect of illness experience: contagion and contamination. How children think about contamination and contagion is an interesting and important question in its own right. Such an investigation also promises to shed light on larger issues within cognitive development.

Beliefs about contagion and contamination represent reasoning about causal relationships surrounding illness. At least for modern adults in western cultures, contagion and contamination, though separate concepts, are understood as components or aspects of a single model of disease transmission: an infection model of illness. Infection involves the idea of some vector or agent of disease (a contaminant) being transmitted from an object or medium to a host who becomes ill as a result of this transmission (contagion). Because they may be parts of a model or theory of illness causation, conceptions of contagion and contamination are of great interest to researchers studying cognitive development. To what extent are children's ideas organized into a coherent model of infection? The coherence or organization of children’s causal reasoning is a central issue within accounts of cognitive development based on the acquisition and elaboration of intuitive theories (see Wellman & Gelman, 1992 for review). Particularly in the domain of biology, there is significant debate regarding the coherence of children's reasoning (e.g., Solomon & Cassimatis, 1997). Thus a central focus of this chapter will be a consideration of the degree to which children's ideas about contagion and contamination are organized into a coherent model of infection. Given this focus, and for ease of exposition, I will often use the term "infection" to refer to the process (and conceptual model) involving contamination and contagion. By adopting this usage I do not mean to prejudge the issue of coherence.

Biological Models and Conceptions of Illness

Contagion and contamination are likely to be the disease processes most familiar to young children. Most of the illnesses that affect children involve infection (e.g., colds, chicken-pox, measles). Many of the rules children know surrounding causes of illness relate to these processes (e.g., "Wash your hands before eating."). Further there is something cognitively compelling about contagion and contamination. Causal reasoning of this sort is pervasive. For example, cultures that lack a scientific understanding of disease transmission often nonetheless believe that disease, or other properties, may be transmitted from one person to another and that otherwise innocuous substances may become dangerous upon contact with contaminants (cf. Frazer, 1890/1981). Analogously, children extend infection beyond illness contexts, and beyond the range of (what adults understand to be) the underlying mechanism (e.g., the game of "cooties"). Since it is a virtual given that even quite young children think about contagion and contamination, in this chapter I will explore some of the attributes and implications of these conceptions. In particular I will focus on two questions. First, how do young children understand the processes of contagion and contamination? What kind of models do they have of infection? The second focus will be on exploring how children’s beliefs about infection figure in their conceptions of illness.

Do young children conceive of uniquely biological processes and entities? Contagion and contamination are prime candidates for such biological knowledge. As adults (at least as modern, western adults) our model of infection is grounded in biological conceptions of germs and the actions of pathogens in the body. However, a biological model is not the only way to conceive of infection. For example, we also seem to think about contamination and contagion in associational terms, as operating according to the principles of sympathetic magic (Frazer, 1890/1981; Rozin & Nemeroff, 1990). In the course of this review I will outline four models or conceptions of infection: associational, physical, simple-biological, and differentiated-biological. I believe the evidence suggests that young children are not limited to associational models: they understand something of the physical nature of infection. However, we do not (yet) have compelling evidence that young children have biological conceptions of infection.

Children’s models of infection are also important as sources of evidence about their conceptions of illness. Illness is a concept with significant and intriguing connections to theories of biology (cf. Reznek, 1987; Caplan, Engelhardt, & McCartney, 1989) and so is relevant to the biology debate. Illness is also intrinsically important. We would like to know how children think about illness so we can communicate with them more effectively about prevention, treatment, consequences, etc. In addition, "illness" is part of a family of related concepts. As with conceptions of living things, conceptions of illness involve a hierarchy (for example, chicken-pox, infectious disease, illness; compare systematics with nosology). Discussion of conceptual representation often contrast kinds of living things with kinds of human-made objects (Gelman & Markman, 1986; Kalish, 1995; Keil, 1989). Kinds in the domain of illness provide a context for extending accounts of children’s concepts. In particular, I will argue that children’s reasoning and inferences about infection provide important information about their concepts of illness. Children’s, and adults’, inferences about contagion (as well as other data) suggest that they represent illness as part of a kind-hierarchy (Shipley, 1989; 1993). This system of kinds is relatively unfamiliar within discussion of natural kinds of living things and prototypes/cluster concepts of human-made objects. Thus considering the implications of reasoning about contagion and contamination may also lead us to expand our ideas about the possible forms of conceptual representations.

Contagion & Contamination

The terms "contagion" and "contamination" have a venerable history and have developed many different meanings. For purposes of this discussion it is important to distinguish the familiar, commonsense, adult model of infection involving contamination and contagion from other possible conceptions. In particular, I would like to use a characterization of contagion and contamination that is broad enough to span a number of different models. In this definition, the core of both contagion and contamination is the idea that contact with an object may produce illness. Contamination refers to the idea that an otherwise innocuous object may be negatively effected by contact or mixture with some substance. Contagion refers to the idea of transfer of contamination from one object to another. The central cases of contagion are when the objects involved are people or other living things and the transmission involves a generative (potentially infinite) chain rather than a single, limited, instance. The central cases of contamination involve foods or other inanimate objects (e.g., a food becomes contaminated). However, in my characterization it is appropriate to describe people becoming contaminated (which results in illness) and inanimate objects being contagious. For example, consider the following chain: A fly falls into some juice and contaminates it. Person A drinks the juice. By a process of contagion Person A becomes contaminated (sick). Person A sneezes on Person B and passes the contamination on via contagion. The usage is not standard in this example, the terms "contagion" and "contamination" are applied more broadly than adult commonsense would warrant. Similarly, describing each part of this chain as involving a process of infection is also an extension of everyday usage. For example, we typically would not think of Person A as being contaminated, at least in the same way as the juice is; the person gets infected, the juice does not. However, I believe these distinctions are rooted in our underlying causal models. As adults we see an important difference in what is happening in the juice and in the person. With other models and conceptual schemes this distinction may not be motivated. I will use the terms "contagion," "contamination," and "infection" in this broad sense and rely on the specifications of particular models to articulate distinctions (for example, the juice may be infected in a physical sense, the person in a biological sense).

The above definition, and the following discussion, captures only a limited perspective on contamination and contagion. The focus is restricted in limiting the effects of infection to illness, rather than, for example, including disgust as a possible outcome (Rozin, 1990). To the extent that infection is seen as involving biological relationships and is part of a coherent theory of illness, this distinction is well motivated. It remains possible that this distinction is artificial and at some points in development, infection involving illness is not differentiated from infection involving other effects (see below). It is also important to point out a second sense in which this discussion of infection is restricted. I will focus on conceptions and developments only within a single cultural context-- modern, western societies. Given that much of the input children receive in this culture reflects modern scientific findings it is clear that universal claims would be inappropriate.

There is a long tradition of claims that preschool-aged (or preoperational) children’s conceptions of infection involve only associations between two things or ideas. An associational conception of infection is very much like belief in sympathetic magic, described by Frazer (1981/1890) as a mistaking of mental associations for real relationships (Rozin & Nemeroff, 1990). For example, Bibace & Walsh (1981) describe children’s use of contagion in the following way: "The cause of illness is located in objects or people that are proximate to, but not touching, the child. The link between the cause and the illness is accounted for only in terms of mere proximity or magic.(p. 36)" Similarly, Rozin (Rozin, Fallon, & Augustoni-Ziskind, 1985) describe "associational contagion" where no physical processes are involved. For example, a piece of clothing worn by a dis-favored person may evoke a disgust reaction despite the knowledge that no physical trace of the person remains on the clothing (Rozin & Fallon, 1987). While adults show associational contagion for judgments of disgust, they tend not to ascribe illness to associational contagion. It is not clear whether children make this distinction. They may not have a different model or conception of infection for illness and for emotional reactions. This account of children’s understanding of contamination and contagion as a cause of illness fits well with Piaget’s description of preschoolers as thinking according to a transductive logic. However, more recent research has challenged claims that young children think about contagion and contamination in purely associational terms.

One piece of evidence cited in support of the contention that children have an associational understanding of contagion is that they view all illnesses as contagious (e.g., Kister & Patterson, 1980). Children predict an illness will be passed from one person to another simply because they associate the two people (e.g., the two people appear spatially close together). There are no limits on the conditions which may be associated this way. A mature understanding of contagion implies that only some conditions are contagious. Contagion involves a physical mechanism (germs, for example) present in some illnesses but not others.

More recent research suggests young children may understand something of the physical basis of contagion. Siegal (1988; Siegal, Patty, & Eiser, 1990) has argued that children limit their judgments of contagion (e.g., a scraped knee is not contagious; see also Kalish, 1996a). Similarly, Keil (1992, 1994) has argued that even young children recognize that behaviors are not contagious. Most directly, I (Kalish; 1996b) have argued that children’s predictions of contagion are not based on simple associations, but rather involve some idea of intermediate mechanism. For example, children reliably judge that getting sneezed on by a person will make you sick. However, when provided information about a mechanism (in this case, that no germs were transferred) children no longer predict illness. Though the association between getting sneezed on and getting sick still holds, children no longer judge the illness to be contagious. In another study (Kalish, 1997) three- to five-year-old children were asked to predict whether story characters would get sick and how they would react emotionally to different types of contacts with contaminated foods. Children predicted emotional reactions (e.g., sadness) on the basis of associational contagion; it was what a character knew and thought that determined emotional reaction. However, illness was seen as the result of a physical process of contagion; it was what a character touched or ate, not what they knew, that determined whether they got sick. This pattern of predictions was clearer for older children in the study. There may be some developmental changes occurring between the ages of three and five in terms of an understanding of the physical basis of contagion. These data begin to suggest that young children may conceive of contagion in a fairly adult-like way; as physical processes. More details of this physical conception are being revealed by research on young children’s understanding of contamination.

While evidence regarding associational contagion came from children’s overascriptions of contagion, evidence for an associational understanding of contamination is said to come from children’s underascriptions of contamination. Young children, unlike older children and adults, judge that foods are safe to eat (are un-contaminated) after all large, visible portions of a contaminant are removed. So, for example, while young children will not drink juice with a grasshopper in it, they will drink the juice once the grasshopper is removed (see Rozin, 1990 for a review, though see also Siegal & Share, 1990 for conflicting results). This response is consistent with an associational conception of contagion. When the contaminant is visible the associations are salient, when the contaminant is removed the association is broken. Young children have been said not to understand that invisible particles of the contaminant may remain in the juice (e.g., Fallon, Rozin, & Pliner, 1984). Invisible particles (e.g., germs, toxins) are part of the mature understanding of the physical bases of contamination. More recent research with preschool-aged children suggests that by age 3 or 4 some understanding of invisible particles is present. They recognize that properties of visible matter (such as the sweetness of sugar) may remain when the matter is dissolved and no longer visible (Au, Sidle, & Rollins, 1993; Rosen & Rozin, 1993). These children often provide or endorse explanations for the persistence of tastes and contamination phrased in terms of invisible particles (Au et al, 1993). Further understanding of the physical bases of contamination is demonstrated by children’s insistence that physical contact between a contaminant and a food is necessary to render the food contaminated (Rosen & Rozin, 1993; Springer & Belk, 1994). However, a substantial minority of preschool-aged children accept that one object may contaminate another simply by being put in close proximity, but not touching. It is not clear if children think anything is transferred in this case. These data suggest (as does the contagion work described above) that an understanding of the physical bases of infection is developing during the preschool years.

Most recent research suggests that, at least by the end of the preschool years, children have a physical rather than associational view of infection. However, there remains dispute regarding just how close or comparable preschool-aged children’s understanding is to that of older children and adults. This dispute concerns the coherence of infection as a model of illness causation. While young children may have some of the pieces, and roughly the correct ontology, have they put these elements together to form something like an integrated model of infection? For example, children may know that causes of illness are part of the physical world, and they may know many facts about how things get contaminated and how to avoid contagion, etc. These elements, though, may not be integrated into a unified explanatory structure (Solomon & Cassimatis, 1997). Part of the problem is that the (modern, western) understanding of infection is bound up with a germ theory of illness. It is our understanding of the action of germs that provides a coherent model of infection. It is not clear whether germs play this role for young children. Apriori there is no reason to believe that an understanding of germs is necessary for a coherent, integrated model of infection–other models are conceivable. However, it is also conceivable that children just know a bunch of facts about contagion and contamination without having a model. Whether children have a germ-based model of infection and whether their understanding of infection might be coherent in the absence of a clear understanding of germs, are issues of current debate.

At least for adults, infection is a coherent model of illness causation. The coherence comes in an understanding of underlying causal processes that provide a link between contamination and contagion. The way a person becomes sick (or an object becomes contaminated) determines whether or how they will be contagious. Contagion and contamination are seen as aspects of the same causal process. We often think about this process in terms of germs (which is taken to encompass viruses and bacteria, and possibly other micro-organisms). In some cases, germs are what contaminate food and make people sick. When germs are involved (and perhaps only in those cases) contamination and illness may be contagious–transferred from one object or person to another. There may be many other details to our commonsense model (e.g., reproduction and growth of germs, processes of decontamination) but an understanding of the role of germs in contamination and contagion seems central. Thus, in assessing the degree to which young children have a coherent model of infection it is important to examine their understanding of the causal basis of the phenomena.

A fundamental element of anything like our commonsense adult model of infection is that the mechanism which caused an illness or a case of contamination is crucial in determining whether that contamination is contagious. In previous research I have argued that cause is at least part of the basis for young children’s judgments of contagion (Kalish, 1996a). In one study, children heard stories describing characters displaying a set of symptoms. Among other things, children were asked to judge whether that character was contagious ("Could you catch it if you played" with that person?). Of particular interest was comparing these judgments with those in a second condition in which children were presented the same stories with the addition of information about the causes of symptoms. When the cause of a character’s symptoms were contagion (i.e., they caught the illness from someone else) characters were frequently judged to be contagious. Other sorts of causes (e.g., a headache resulting from a blow) were not predicted to result in contagious illness. In a second study, children judged all conditions said to be caused by germs to be contagious at the same rate despite variation in symptoms produced (e.g., positive vs. negative symptoms). This research demonstrates that children do focus on how a condition was caused (how someone became contaminated) when judging contagion.

This link between cause and contagion has been challenged in some recent research. Solomon & Cassimatis (1997) report that before age 10 children are relatively insensitive to the causes of conditions. For example, they judge conditions caused by germs, by poisons and by "events" (e.g., sneezing because of pepper) to be contagious all at the same rates. To the extent that young children made differentiated predictions about contagion they seemed to do so on the basis of symptoms presented. For example, coughs were judged to be more contagious than tummy aches. The relationships between symptoms and contagion may be evidence of a set of empirical generalizations rather than any organized models.

Clearly these discrepant results need to be reconciled. Importantly, young children may have a different view of the range of causes of contagious illness. Agents other than germs (e.g., "cold air," McMenamy & Wiser, 1997 or poisons, Keil, 1994) may be thought to mediate infection. Thus they question of whether children understand of germs as agents of infection must be distinguished from the more general question of whether they have a model of infection. Young children may or may not realize the unique role germs play. They may, however, understand that contagion and contamination are linked together as part of some coherent causal process. While important as a necessary component of a recognizable understanding of infection, knowing that the cause of contamination affects contagion is far from sufficient for providing a complete model. It remains to be demonstrated how young children conceive of the causes underlying infection. In particular, a central question is whether children have a biological, germ-based, model.

Young children do see germs as a mechanism of infection (Kalish, 1995). For example, they recognize that events which often contaminate food will not do so if no germs are involved (e.g., a cookie that falls on the floor but gets no germs on it is not contaminated). Similarly, events which typically result in contagion will not do so in the absence of germs (e.g., playing with a sick child doesn’t make you sick if no germs are transferred). However, children may not see germs as unique and distinctive agents of infection. McMenamy and Wiser (1997) find that preschool-aged children judge illnesses caused by "folk agents" (e.g., cold air) to be contagious. There is some suggestion that these young children saw folk agents as operating independently of germs. One way the role of germs in infection has been addressed is by asking whether young children distinguish between germs and poisons. Young children are equally likely to judge as contagious illnesses caused by germs and illnesses caused by poisons (Keil, 1994; Solomon & Cassimatis, 1997). Moreover children may not see germs as living things and as different from chemical poisons. Keil (1992) found children more likely to attribute biological properties to germs than to poisons, however in some cases children were also given the information that germs (viruses) had purposes. The information about goals may have led them to ascribe other biological attributes. Solomon and Cassimatis (1997), in contrast, found that young children did not distinguish between germs and poisons and tended to treat both as non-living. The conservative conclusion may be that children do not see germs as a special kind of mechanism. This suggests they may have a physical, but not biological conception of infection.

Poisons and other chemical/physical entities can be viewed as mechanisms of contagion and contamination. Poison is clearly a contaminant. If one contacts poison one may become ill. Poison may also be a vehicle for contagion. For example, if someone gets a particularly virulent poison on his hand and then touches someone else, that second person may come to show the effects of the poison as well. This transfer of materials (and the effects of the materials) represents a coherent model of physical infection. This model seems consistent with what we know about young children’s beliefs about infection. It is not associational, but rather based on material transfer. Further, it seems that young children’s understanding of the role of germs may also fit within this physical model. Germs function like poisons, as physical agents of contamination.

A physical model of infection based on material transfer does differ from our adult commonsense biological understanding. A key feature of the adult model is generativity. Transfer of poisons strikes us as a marginal case–the process is too limited. Contagion and contamination are thought to be potentially infinite. This attribute is accounted for by our belief that the agents of infection, germs, reproduce and multiply. Similarly, a physical model of infection seems too static. In our biological model contamination spreads within the host. Contamination originally contacted by one part of the body (e.g., the mouth) may result in contagious transfer from other sites (e.g., mucus, saliva, feces). Again it is the living nature of germs, and the way germs interact with the biological host, that seems to account for these intuitions. Thus a biological conception of infection seems to require some additional knowledge over and above a physical model. Importantly, our core adult model of infection involves living things as both agents and hosts. While transfer of inanimate particles from one inanimate thing to another may be infection in "some sense" it is not the typical sense.

While there is little direct evidence, existing data suggest that young children do not hold a biological conception of infection. Au and Romo (1997, this volume) report that children’s untutored explanations for contamination (e.g., rotting food) involve mechanical forces–movements of invisible particles. The possible physical conceptions of germs (and their conflation with poisons) discussed above would also be incompatible with a biological model of infection. Children have been held to be generally unaware of internal bodily processes (see Carey, 1985a). Yet it is what happens when a contaminant enters the body–how it reproduces, spreads, and causes reactions on the part of the host–that seem at the heart of a biological conception of the process. Most preschool-aged children do not recognize that illness is a delayed effect of contact with germs (Kalish, 1997). Children think people become sick immediately upon contact with contaminants. This suggests some ignorance of the processes occurring inside the body in response to contamination. Clearly the understanding of the action of germs within the body is a complex matter. It is likely that many adults have only sketchy beliefs about these relationships. However some understanding that more than simple material transfer is going on seems crucial to a truly biological model.

One way of characterizing a biological model of infection is that it is an elaborated, more complex, version of a physical model. In a physical model, the agents of infection are understood to be material entities. A biological model further specifies the nature of these entities–they are living things (e.g., that reproduce). A physical model requires that a contaminant must physically contact a host. A biological model further specifies what happens inside the host (bodily processes). However, an alternative criteria for identifying a body of knowledge as biological is the scope of phenomena or domain within which it is applied. Simons and Keil (1995) argue that young children may have only a skeletal or abstract notion of biological phenomena. What would identify a model of infection as biological on this conception would be how notions of infection interact with other biological knowledge. For example, infection may be seen as unique to living things, or be thought to have different properties when involving living and non-living things. One piece of evidence of this sort is Keil’s (1992) demonstration that children limit contagion to physical/biological attributes. In Keil’s work, preschool-aged children denied that behaviors (such as obsessive hand washing) may be caught from another person. Young children seemed to overextend contagion to all biological attributes: Congenital conditions were thought to be just as contagious as acquired conditions. This suggests that children may identify infection with the domain of biological properties of living things. While more research is needed on the relationships between conceptions of infection and other elements of biological knowledge, the possibility of such relationships means that children may have a biological model of infection (of some sort) without understanding specific details.

In assessing young children’s understanding of infection it is necessary to make distinctions between different types or models of infection. The first distinction drawn was between an associational and a material conception. Young children do not seem to be limited to an associational view, they can understand the material basis for contagion and contamination.. A second distinction concerns whether children have an coherent model of infection or whether their knowledge is more fragmented and piecemeal. While children have been said to see infection as mediated by some underlying causal processes (e.g., transfer of material) there is some debate on this point. Assuming that children do have a causal model of infection the question arises, what sort of model? Here it is useful to make a distinction between physical and biological models of infection. While young children may understand that infection is based on material transfer, they may not have a particularly biological understanding. A biological model seems to involve more detail and specific knowledge than a physical conception For example, one crucial element seems to be the recognition that agents of infection act in particular ways inside the body (e.g., they reproduce). The existing evidence provides some suggestion that young children may hold a physical rather than a biological model.

This discussion of infection has focused almost exclusively on children in the preschool years. In part this reflects the focus of research in this area. However it is also possible to make some conjectures about later developments based on data from early childhood. Many researchers believe that a biological understanding or way of viewing phenomena emerges from physical mechanical knowledge (Au & Romo, this volume; Kalish; 1997) or, perhaps, social knowledge (Carey, 1985a). This process requires some conceptual change and is thought to occur in middle childhood (Carey, 1985a) and/or in response to formal instruction (Au & Romo, this volume). Thus we might predict that children will be developing biological models of infection in middle childhood. Certainly as children grow older and are exposed to more information their knowledge about the specifics of germs and bodily processes increases. However, Simons & Keil (1995) argue we should not equate biological knowledge with knowledge of specifics. How much detail is required for a biological model remains a matter for further debate and research.

Conceptions of Illness

Among the reasons that children’s understanding of contagion and contamination has attracted so much attention is that children cite these processes as causes of illness. While there may be some debate surrounding the particular model children hold, infection does seem to play a central role in children’s understanding of illness. Exactly how central is the next question taken up. For example, researchers have argued that at some points in development children think that all illnesses are contagious. This suggests that the model children have of infection may also be their model of illness. What it is to be sick is to have a contagious ailment. While there are reasons to doubt this strong claim (see below), it does seem that children’s reasoning about infection is quite informative regarding their more general conceptions of illness.

One suggestion is that contagious illness is the prototypical case. A prototype is a probabilistic summary, or a best instance, and provides predictions about likely features (Rosch, 1975). The claim that infection is prototypical implies that then when children think of illness they tend to think of a process involving contagion and contamination. Infection is thought to be the most typical, most likely, default sort of illness. There is some evidence that adults’ have prototype representations of disease and specific illnesses (cf., Bishop, 1991; Bishop & Converse, 1986; Reznek, 1987). Contagious ailments, specifically acute viral infections (e.g., colds, measles, chicken-pox), may be the prototypical or "best" cases of disease (Campbell, Scadding, & Roberts, 1979). In fact, it has been argued that one of the changes that occurs in medical school is that students learn to give up this "cognitive bias" towards thinking about disease in terms of infection (Stefan & McManus, 1989).

To what extent might infection be the prototype of illness for young children? It does seem that adults have a notion of a "prototypical" contagious illness (Bishop, 1991). This prototype serves to organize thinking about many illness. However, adults also recognize non-contagious illnesses. If infection is one among many prototypes for adults, we might expect it to play a more dominant role for children. Most of the illnesses most children experience are the acute viral infections–while adults may be familiar with more non-prototypical illnesses (heart disease, cancer). In fact, many of these best examples of contagious illness are often associated with childhood (e.g., chicken-pox, mumps, measles). Keil (1989) has argued that children’s concepts develop from being organized primarily around characteristic features (prototypes) to an organization involving defining or causal features. For example, young children do not distinguish features which are merely associated with category memberhsips (e.g., that uncles are usually adults) from those features which are part of the definition or core of the category (e.g., that uncles are the brothers of a parent). Thus we might expect that children’s earliest conceptions of illness are most heavily influenced by notions of prototypicality. One piece of evidence is that children often indicate that all illnesses are contagious (e.g., Hergenrather & Rabinowitz, 1991). Even congenital illnesses may be thought to have the prototypical property of being contagious (Keil, 1992). In the absence of other information, a prototype is the default. Kalish (1996a) found that children judged 72% of the conditions they identified as illnesses to be contagious (in the absence of information about the causes of the symptoms). Eighty percent of those judged to be illnesses were predicted to involve a fever, a symptom usually associated with infections of some sort. To the extent children’s conceptions and identifications of illness are similar to adults (see Kalish [1996a] for arguments) this suggests they too see some elements of infection as prototypical of illness. However, the role of prototypes in children’s thinking about illness has not been directly addressed.

Finding that infection is prototypical would provide information about how children might typically reason about illness but, in an important way, this finding would not be informative about the nature or structure of children’s concept of illness. For example, any concept from nominal kinds (such a "even number") to natural kinds (such as "dog") may be associated with prototypes (Armstrong, Gleitman, & Gleitman, 1983; Rey, 1983). Thus many possible conceptions of illness are consistent with a claim that infection is a prototype. In particular, one possibility is that, more than just being a prototypical case, infection defines illness for young children.

Taking infection as definitive of illness is to hold a natural kind conception of illness. A natural kind is a grouping or type that seems to have an objective basis–there is some fundamental identity that we discover (Mill, 1872/1973). For a kind such as "illness" the objective basis would be some particular biological process (Reznek, 1987). Contagion and contamination could provide a unitary underlying causal model for illness. For example, it has been reported that a germ-theory of illness was popular in Europe following the discoveries of Pasteur (Reznek, 1987). All illness was thought to involve germs and infection. This model would provide a natural kind conception of illness. Illness would have the characteristics of other natural kinds (basic level animal kinds being the central cases in the psychological literature). For example, there could be anomalous cases–ailments which appeared to be illnesses but were discovered not to be because they did not involve germs (see Gelman & Coley, 1991; Markman, 1989 for further discussion). At least for modern western adults, illness is not a natural kind: there is no unifying underlying biological process with which illness may be identified (cf. Reznek, 1987). In particular adults recognize that some illnesses are contagious and caused by contamination, but not all are (e.g., inherited diseases, vitamin deficiencies, etc.). However, children may hold different views.

Discussions of concepts often posit three types of representation; concepts are natural kinds, nominal kinds, or property clusters (cf. Keil, 1989). If children’s and adults’ conception of illness is not a natural kind, and is not nominal (based on an analytic definition), then it must be a property cluster. Property cluster concepts are based on similarity and prototypes. Thus illness may have a prototype structure as well as showing prototype effects (see Lakoff, 1986 for this distinction). However, there is a fourth alternative. Within a given domain the natural kinds might exist only at a low level of generality but be collected together into higher order kinds. Those higher order kinds need not themselves be natural kinds but might have some other structure. Shipley (1989, 1993) and Goodman (1955) have described a set of principles for this kind of organization; an organization involving "kind-hierarchies" and "over-hypotheses." A kind-hierarchy is consistent with prototype effects but posits more structure or specific relationships between concepts and attributes. In the remainder of this chapter I will outline the notion of a kind-hierarchy and argue that this type of conceptual structure best matches children’s and adults thinking about illness. In particular, reasoning about infection will be a crucial piece of evidence in favor of a kind-hierarchy account of the illness concept.

As part of an account of induction, Goodman (1955) proposed the idea that people were sensitive to fairly abstract relationships between classes of categories and dimensions. These relationships he termed "over-hypotheses." An over-hypothesis is a proposition such as "Instances of the same kind of illness have the same cause." It is a hypothesis about a set of categories (the kinds of illnesses; chicken-pox, sickle-cell anemia, etc.) and a dimension (cause). It specifies that within a category the values on the dimension will be uniform (e.g., all instances of chicken-pox will have a common cause). Systems of concepts built out of over-hypotheses are called "kind-hierarchies" (Shipley, 1989). Higher order kinds (e.g., illness) involve over-hypotheses about subordinate level kinds (e.g., chicken-pox). One important point for the present discussion that the higher order kinds are characterized by over-hypotheses, rather than by features shared among instances of the higher order kind. On a kind-hierarchy view, there is no claim that illnesses share any qualities or that there are a particular set of properties unique to illnesses. That is, there are not necessarily any laws that apply to all and only illnesses. However, "illness" is not simply an arbitrary or unstructured concept. The concept captures some important similarities shared by its members, each of which is a natural kind. To identify a kind as an illness is to say what type of natural kind it is.

There is some existing data suggesting that people’s conceptions of illness are part of a kind-hierarchy. Illness is not a natural kind but it is a superordinate kind that identifies a set of natural kinds. The kinds of illnesses (which I will refer to as "diseases") are natural kinds of a particular type. One piece of evidence comes from the analysis of experts’ conceptions. In his exploration of the concept of "disease," Reznek (1987) suggests particular diseases (such as chicken-pox) are natural kinds. There are characteristic biological processes underlying these particular diseases. Reznek argues that higher-level concepts of "disease" or "illness" are social constructs which exist for only peculiar historical reasons (e.g., professional disputes, technological limitations), though he does not consider that they may provide some organization for the natural kinds of diseases. While the claims about expert usage are suggestive, more direct evidence comes from accounts of lay adults illness representations. In particular, Leventhal’s (Leventhal, Meyer, & Nerenz, 1980) theory of a dimensional representation is also consistent with a kind-hierarchy account.

An influential model of commonsense conceptions of illness (Leventhal et al., 1980; also Lau & Hartmann, 1983) states that people think about illness in terms of values on dimensions. An example of a set of dimensions used to think about illness is presented in Table 1. This model has been proposed for lay adults, and has also been said to characterize young children’s representations of illness (Goldman, Whitney-Saltiel, Granger, & Rodin, 1991). One source of data for these claims come from studies in which people are asked to describe a particular illness (e.g., their most recent illness). The finding is that statements can be classified into one of the five dimensions (e.g., statements about identity, symptoms, cause, cure, or time course). Values on these dimensions are used to characterize particular illness episodes. Similarly, knowledge about types of illness (e.g., colds) is also organized around these dimensions (Goldman, et al., 1991). What people know, or want to find out, about an illness is how it was caused, what will make someone better, etc. Importantly the claim is not that all illnesses have some features in common (e.g., all have the same cause) but rather that questions about cause are relevant in cases of illness. Thus the concept of illness involves beliefs about what kinds of information or features are relevant to instances. In this way the characterization of illness is very similar to Keil’s (1979) description of an ontological level of knowledge representation. In Keil’s terms, Leventhal’s account is a hypothesis about a characteristic set of predicates that span illnesses.

While related to an ontological characterization, the suggestion that illness is part of a kind-hierarchy is a stronger claim. The supposition is not just that "illness" is associated with a particular set of spanning predicates, but also that values of the predicate/dimension will be shared within sub-types of illnesses. Similarly, a kind-hierarchy is a stronger constraint on conceptions of illness than that directly implied by Leventhal’s account. From Leventhal’s theory one could predict that upon encountering a novel illness ("Lumenza") people would want to know (roughly) five things (its symptoms, consequences, time course, cause, and cure). A kind-hierarchy account would further predict that people would generalize ("project," Goodman, 1955) these five pieces of information to all other instances of the same illness. This is because the concept of "illness" includes over-hypotheses such as "cause is shared across instances of the same kind of illness." For example, if told that one case of Lumenza is contagious (or acquired via contamination) people will assume other cases of Lumenza are also contagious. Importantly, there are predicates that span illness (e.g., began in June) that are not part of over-hypotheses (and so not projected). Thus the kind-hierarchy hypothesis is that the concept of illness specifies the types of features (dimensions) projectable across instances (cases) of particular diseases.

Since a kind-hierarchy account involves stronger claims than a dimensional or ontological view, it is appropriate to ask for some more data in support of this account. One way to test this hypothesis is to study the projection of properties across instances of novel diseases. Given two pieces of information about a condition, first that it is an illness and second that some property is associated with the condition, over-hypotheses allow projection of the property to other cases of the same disease. If children’s and adults’ concept of illness includes over-hypotheses then they should be willing to project some properties of a novel disease based on a single instance. If these projections are based on over-hypotheses, rather than general inductive strategies, they should be limited to some properties. Not all properties of a novel disease will be projected. These predictions were borne out in a study of predictions of illness properties. This study is only a preliminary attempt to assess the kind-hierarchy hypothesis. While more work needs to be done, the results are suggestive.

Thirty-one aduts and thirty children (5-yar-olds) participated in the experiment. Participants were told stories about characters suffering from one of two novel diseases. In one condition participants heard about two (training) characters; 12-year-old Johnny–sick with "lumenza" and six-year-old Lisa–sick with something else. The characters were ascribed two opposite properties (e.g., one is contagious the other isn’t). The task was to rate which property another (test) character would have. The test character shared gender and age with one of the training characters but had the same disease as the other (six-year-old Julie–sick with lumenza). In this condition the phrasing of the question to the child was "Will Julie be X (e.g., contagious) like Johnny or will Julie be not-X (not contagious) like Lisa?" In the No-Training condition, subjects were simply asked to predict the properties of the test character with no other information ("Will Julie be contagious or not?").

People were asked about two different kinds of properties. There were five sickness properties, which were expected to be shared within disease kind. These properties were chosen to express values on some of the five dimensions identified as components of lay illness representations (Goldman, et. al, 1991; Leventhal, et. al, 1980). There were also five social/psychological dimensions, which were expected to be shared within age/gender (or be randomly distributed). A list of properties is presented in Table 2. There were two sets of stimuli (set A and set B). These sets differed in the properties ascribed to the sick characters. For example, in set A the training character who had the same illness as the test character was said to have a fever. In set B training character was said to not have a fever.

One thing this experiment does is provide information about prototypes or default reasoning about illness. Illness properties were not projected randomly in the absence of other information. Though only 10 children participated in the No-Training condition, there was a significant tendency to judge that someone suffering from an unfamiliar illness would have a fever, need medicine and be contagious (8 or more judgments out of 10, p = .05, binomial theorem). This is consistent with prototype theory of illness representation. A prototype provides default information. However, this may not be the only way the general concept of illness provides prediction. In particular, the kind-hierarchy hypothesis specified that the concept of illness should guide inductions from cases of illnesses.

The main question was how people would use information about a single case of a novel illness to make inferences about future cases. Would properties be projected to the test character on the basis of the disease she shared with the training character or on some other basis (e.g., randomly)? One way to address this question is to look at the differences between projections for set A and set B. If people judged that the test character would have the same property as the same-illness training character then their ascriptions would be reversed in sets A (where lumenza is said to be contagious) and in sets B (where lumenza is said to not be contagious). If properties were projected on some other basis (e.g., using base rate or prototypicality judgments, for example, that all illnesses are contagious) then there should be no differences in judgments across sets. Finally, if gender or age is used as the basis for projections sets A&B would differ but in the opposite direction (e.g., if lumenza is contagious then the test character is not because she is a girl and the other, training, girl was not contagious).

Figure 1 shows the patterns of children’s and adults’ property projections. For illness dimensions (those predicted to be projectable from illness) there was a large difference between property judgments for set A and for set B. For non-illness dimensions the difference was small. Similarly, we can compare inferences in the Training and No-Training conditions. For illness dimensions, predictions in these two conditions are very different. For non-illness dimensions differences are small. Thus, the information that one case of a novel disease is associated with a particular property (e.g., is contagious) affected projection of that property to a second case of the novel disease for some properties but not others. Children and adults have intuitions about which properties are projectable within categories of illness and which are not. Properties projectable within illness categories were those identified as components of lay illness representations in previous research (e.g., Leventhal, et al, 1980).

Children and adults have some important and consistent assumptions about the general properties of illnesses. In particular, knowing just that a character was suffering from an illness led people to make many other inferences. For example, a novel illness was predicted to lead to a fever and to be contagious. These judgments (in the No-Training condition) are consistent with a prototype of illness based on acute infections and processes of contagion and contamination. However, people also have more assumptions. There is more to their concept of illness than a set of default expectations. In particular there also seem to be beliefs about which properties of a novel illness are projectable. These over-hypotheses specify what can be learned about a type of illness from experience with individual instances.

People judge that two instances of the same illness will be alike in symptoms displayed (having a fever or not), the time-course, cure (requiring medicine or not), and cause (involving germs or not, and contagious or not). This suggests that the concept of illness contains over-hypotheses for these dimensions (e.g., "cause is projectable within types of illnesses"). Importantly, not all properties are thought to generalize. People judge that two characters suffering from the same illness need not have similar preferences or emotional reactions. The concept of illness does not contain over-hypotheses for these dimensions. Contagion and contamination, for example, are important components of people’s conceptions of illness in a way that preferences or emotions are not. However, it is not the properties themselves that form part of the illness concept. Rather it is beliefs about the consistency of properties that are part of the general notion. "Is contagious" is not a component of people’s representation of illness. Something like "Is contagious or not" may be.

The results of the experiment described above suggest that children’s and lay adults’ conceptions of illness do include over-hypotheses about projectable dimensions. This is certainly one piece of evidence consistent with a kind-hierarchy representation of illness. Another element of the hypothesis is that the categories at a more specific level (types of diseases) are thought to specify common features. The results lend some support this hypothesis for a novel illness. People were willing to project properties, such as contagion, within categories of diseases. However, this is only suggestive of the stronger belief that these properties will be always shared across instances of the same disease. For example, people may have been establishing a prototype of the novel illness, rather than viewing it as a natural kind. Part of the kind-hierarchy hypothesis was that illness is a general category which collects together a set of natural kinds; it says what kind of natural kind a specific disease is. Thus, one remaining question is whether or not conceptions of specific diseases are natural kinds.

One of the pieces of evidence used to support the claim that experts view diseases as natural kinds was that they see each disease as involving a characteristic underlying process (Reznek, 1987). At least for familiar, common illnesses, this seems to be the lay adult view as well. One of the indications that a category is organized around a characteristic underlying causal process is that it serves as a productive basis for induction (cf. Gelman & Coley, 1991; Keil, 1989; Markman, 1990). From the above study people did use disease categories as bases of induction (e.g., that if one case involves infection then another will). However, this is not definitive proof that people identify diseases with particular underlying process. For lay adults (again, modern western adults) this characterization does seem plausible (cf. Campbell, et al, 1979). For example, most of us probably belief that a specific disease, say chicken-pox, is not defined by its symptoms (e.g., red spot, contagious) but rather what makes a case of illness chicken-pox is that it involves a particular biological process and disease agent (a chicken-pox virus which causes spots and contagion). Whether adults believe this to be true of all types of illnesses (or whether, for example, lay adults recognize some conditions to be syndromes–defined by symptoms) is a matter for future research. What is also unclear is whether young children (ever) have this essentialistic, natural kind, view of specific diseases.

While there is not much research on children’s (or adults) conceptions of specific diseases, Keil (1994) reports some data suggesting that young children do not have natural kind representations of diseases. Keil presented subjects with cases of illnesses which shared either observable symptoms or shared origins/etiology, but not both. The question was whether children would identify types of illnesses on the basis of symptoms or on the basis of causal process (as indicated by the origins and developmental progress of the disease). Thus Keil asked children which cases of illness were the same kinds. What Keil found was kindergarten-aged children almost always judged that illnesses with the same symptoms were the same kind and ignored differences in origins. Second-graders paid some attention to origins, but not until fourth grade were children consistently using causal process to identify types of illnesses. This suggests that young children do not have a natural kind conception of specific diseases.

Why is it that children do not use causal process to identify diseases? It is not that they generally ignore causes. For example, Keil (1989) has found that quite young children recognize the importance of origins and causes in making the general distinction between biological (living things) and non-biological kinds. Similarly the results of Kalish (1996a) suggest that children attend to causes when judging whether something is an illness or not. It may be that it is level of detail of the distinction that leads children to ignore causes. Children’s causal models may not be developed in enough detail to account for or represent distinctions at the level of specific diseases (or specific species, Keil, 1989; cf. Simons & Keil, 1995). In particular, their model of infection may be too general to account for specific diseases.

Children know that infection is a causal process that underlies some cases of illness. However, to understand how cause distinguishes one illness from another (e.g., chicken-pox from measles from mumps) one needs a specific, differentiated, understanding of infection. Our adult view is that infection is a super-ordinate level concept; there are many types of infectious processes, each of which involves a unique germ (or other agent). While both chicken-pox and measles are contagious, there are fundamental differences between the causal processes involved in these two diseases. If children do not understand that there are different germs which may be involved in cases of infection they could not use this causal model to distinguish particular diseases. For example from an undifferentiated model of infection it is not clear why it is measles that is caught from contact with a measles-sufferer rather than, say, chicken-pox or any other set of symptoms. Knowing that illness is contagious does not, necessarily, imply that type of illness is conserved. Awareness of the specific nature of infection may be part of what it takes to really understand the uniqueness of particular diseases. In turn, awareness of the distinctive natures of particular diseases may prompt children to develop a more differentiated understanding of infection. If the initial results indicating that young children have a weak understanding of specific diseases are supported, this would be suggestive evidence for the possibility that these children have an undifferentiated view of infection. Thus, not only may children’s understanding of infection inform us about their concepts of illness, their conceptions of illness may provide us some ideas about the models of infection they hold.

Summary

In the course of the above review of the literature on conceptions of contagion, contamination and illness we can identify four models of infection. First is an associational model: contagion and contamination are a mistaking of associations among ideas for associations between physical entities, akin to the principles of sympathetic magic (Frazer, 1890/1981; Rozin & Nemeroff, 1990). A second model treats infection as a physical relationship based on transfer of material particles. Third is a biological model where the agents of infection are understood to be living (e.g., reproducing) entities that infect and act on other living things (hosts). Fourth is a differentiated biological model. Here agents of infection are understood to come in distinct types, species, with unique attributes. Infection becomes a higher-order description of a set of related, but distinct, processes. A tentative conclusion from this review is that preschool-aged children hold something like the second of these models. There is good reason to believe they are not associationists, but we do not (yet) have compelling evidence they view infection in uniquely biological terms.

In the grand Piagetian tradition it is possible to describe a developmental progression of infection models, with subsequent models incorporating, elaborating on, and, at times, superseding, the previous models. It would seem reasonable to predict an invariant sequence in the appearance of these models (to the extent they do appear it should be 1->2->3->4, rather than some other ordering). However, a crucial disanalogy may be the cognitive changes underlying the transitions between models. Rather than general changes in the organization of thought, it seems more likely that changes in models result from changes in specific knowledge or beliefs about germs, the body, and illness (cf. Carey, 1985a, 1985b; Simons & Keil, 1995). In particular, it does not seem appropriate to view model 2 as a product of an inferior logic or less sophisticated thinking than models 3 and 4. As discussed above, adults hold model 2 and use it to reason about interactions in the physical domain. If children lack models 3 and 4 it is likely because they lack the relevant knowledge about the biological specifics.

From this perspective of domain specific change, even the earliest form, associational infection, may not be adequately characterized as deriving from inferior logic or errors of thought. Rather, an early sensitivity to infection may represent a species-typical adaptation to the problem of food selection and illness avoidance (Rozin, 1990). This early (and perhaps universal?) conception is elaborated upon by individual experience and cultural forms of knowledge (e.g., magic, science). This interaction produces new forms of thinking about infection which are partially co-extensive with older forms. Different phenomena may come to be seen as illustrating (or being explained by) particular types of infection. For example, in western culture most adults probably see a biological model of infection involved in illness causation, rather seeing illness as mediated by associational or physical mechanisms. However, the ‘assignment’ of phenomena to models need not be universally agreed upon. Moreover, as Rozin and colleagues have demonstrated (Rozin & Nemeroff, 1990), early models of contagion and contamination may coexist with later.

As models of infection are likely dependent on knowledge of biology and biological relationships, infection may be taken as an index of that knowledge. How children think about infection is diagnostic (and constitutive) of their general understanding of biology. More directly, thinking about infection is also informative about general conceptions of illness. Infection is not definitive of illness. Young children recognize illnesses that are non-contagious and cases of contamination which are not illness. This suggests illness is not a natural kind, based on a causal model of infection, for young children. While there have been some suggestions that older children see infection as underlying all illness, more data is needed to support this claim. Infection does seem to be part of children’s and adults prototype for illness (cf. Bishop, 1991). For example, in the absence of other information (e.g., in the study reported above) cases of a novel disease are assumed to be contagious. However, inferences about infection (and other properties of illness) also provide evidence for more structure to the concept of illness. In particular, children’s and adults’ conception of illness contains over-hypotheses about dimensions projectable within sub-types of illnesses (specific diseases). If told that one case of illness is contagious people will infer another instance of same type is also. This pattern of inference is consistent with a kind-hierarchy conception of illness.

One general theme that emerges from the study of conceptions of infection and illness is the importance of identifying on the right level of generality to study children’s thinking. For the study of conceptions of living things it is the folk-generic level of kinds (e.g., dog, cat, horse, cf. Atran, 1987) at which the most important identities and differences are conceptualized. In the sub-domain of illness the fundamental level is less clear. For adults, both "illness" and "contagious disease" are too broad. The most significant distinctions, the natural kinds, are more specific; at the level of types of diseases and microorganisms. For children the level is less clear; "illness" seems to broad, but they may not see the significance of, and the basis for distinctions among, specific types of diseases. Perhaps it is at an intermediate level of generality, distinguishing between infection and other types of causal processes, that children’s knowledge is organized. Certainly investigations of contagion and contamination have been informative about many issues regarding children’s thinking. There is every reason to believe that future studies of beliefs about infection will continue to enlighten us about children’s cognitive development.

References

Footnotes

Table 1

Properties Used in Projection Study

Figure Captions

Figure 1: Property inferences for test Characters in Training and No-Training Conditions. The vertical axis represents the mean proportion of judgments that test characters would have properties from Set A. Predictions of Set B properties are the inverse of Set A predictions (0% A predictions = 100% B predictions). The midline indicates no-preference for Set A or Set B properties (chance performance). Open boxes indicate how frequently Set A properties were predicted in the No-Training Condition. Arrows from these boxes indicate the direction and magnitude of change between the No-Training and Training conditions. In the Set A condition, the training character who shared disease with the test character was said to have Set A properties. In the Set B condition that character was said to have Set B properties.