Strategies of Knowledge Acquisition Author(s): Deanna Kuhn, Merce Garcia-Mila, Anat Zohar, Christopher Andersen, Sheldon H. White, David Klahr, Sharon M. Carver Source: Monographs of the Society for Research in Child Development, Vol. 60, No. 4, Strategies of Knowledge Acquisition (1995), pp. i+iii+v-vi+1-157 Published by: Blackwell Publishing on behalf of the Society for Research in Child Development Stable URL: http://www. jstor. org/stable/1166059 . Accessed: 16/09/2011 13:38 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . ttp://www. jstor. org/page/info/about/policies/terms. jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected] org. Blackwell Publishing and Society for Research in Child Development are collaborating with JSTOR to digitize, preserve and extend access to Monographs of the Society for Research in Child Development. ttp://www. jstor. org OF MONOGRAPHS THE IN FOR SOCIETY RESEARCH CHILD DEVELOPMENT Serial No. 245, Vol. 60, No. 4, 1995 OF STRATEGIES KNOWLEDGE ACQUISITION Deanna Kuhn Merce Garcia-Mila Anat Zohar Andersen Christopher BY WITH COMMENTARY SheldonH. White David Klahr and Sharon M. Carver BY AND A REPLY THEAUTHORS MONOGRAPHSTHE OF SOCIETY RESEARCH FOR INCHILD DEVELOPMENT SerialNo. 245, Vol. 60, No. 4, 1995 CONTENTS ABSTRACT v I. INTRODUCTION 1 II. METHOD 24 III. KNOWLEDGE IN ACQUISITION ADULTS 33 IV. KNOWLEDGE IN ACQUISITION CHILDREN 42 V. STRATEGIES STRATEGY AND CHANGE ADULTS 50 IN VI.
STRATEGIES STRATEGY AND CHANGE CHILDREN 64 IN VII. THE PROCESS CHANGE OF 75 VIII. CONCLUSIONS 98 REFERENCES121 ACKNOWLEDGMENTS 128 COMMENTARY TOWARD EVOLUTIONARY AN EPISTEMOLOGY OF SCIENTIFIC REASONING SheldonH. White 129 SCIENTIFIC THINKING ABOUT SCIENTIFIC THINKING David Klahr and Sharon M. Carver 137 REPLY SCIENTIFIC AND KNOWLEDGE THINKING ACQUISITION Deanna Kuhn 152 CONTRIBUTORS 158 STATEMENT OF EDITORIAL POLICY 160 ABSTRACT KUHN, DEANNA; GARCIA-MILA,MERCE; ZOHAR, ANAT; and ANDERSEN, CHRISTOPHER. WithCommentary Strategiesof KnowledgeAcquisition. nd H. KLAHR SHARON CARVER; and SHELDON WHITE by DAVID M. by KUHN. and a Reply by DEANNA theSociety Research in Monographs of for Child 1995, 60(4, SerialNo. 245). Development, In this Monograph, is knowledgeacquisition examinedas a processinthe coordinationof existing theorieswith new evidence. Although volving researchers studyingconceptualchange have describedchildren’sevolving theorieswithinnumerousdomains,relatively little attentionhas been given to the mechanisms meansof whichtheoriesare formed and revisedand by knowledgeis therebyacquired.
Centralto the presentworkis the claimthat strategiesof knowledge acquisitionmay vary significantlyacross (as well as within) individualsand can be conceptualizedwithin a developmental framework. To studythese strategiesand their development,we use a microgenetic method. Our application the method allowsextendedobservation the of of of acquisition knowledgewithina domain,of the strategiesused to acquire this knowledge,and of the changein these strategies overtime.
Subjectsof both ages showedprogressacrossthe 10 weeksin the level of strategiesused as well as similarity the form that this progresstook. in levelsthatdid not varygreatly,childrenshowed Despiteinitialperformance V less strategic improvement than adults and inferior knowledge acquisition. Strategic progress was maintained by both groups when new problem content was introduced midway through the sessions. The results thus indicate significant generality of strategies and strategy change across content, as well as populations.
A further indication of generality was the emergence of new strategies at about the same time in the social and physical domains, even though performance in the social domain overall lagged behind that in the physical domain. At the individual level, mixed usage of valid and invalid strategies was the norm. This finding in an adult population suggests that this variability is a more general characteristic of human performance, rather than one unique to states of developmental transition.
Another broad implication of this variability is that single-occasion assessment may provide an at best incomplete, and often misleading, characterization of an individual’s approach. Still another implication is that at least part of variability in performance across content resides in the subject, rather than exclusively in the task. That superior strategies present in an individual’s repertory are not always applied highlights the fact that more is involved in competent performance than the ability to execute effective strategies.
Metastrategic competence-the ability to reflect on and manage strategic knowledge-and metacognitive competence-the ability to reflect on the content of one’s knowledge-are emphasized as critical components of cognitive development. These competencies determine the strategies that are actually used, among those potentially available, and therefore the effectiveness of an individual’s performance. Finally, the presence of multiple strategies and multiple forms of competence greatly complicates the portrayal of developmental change. Rather than a nidimensional transition from a to b, the change process must be conceptualized in terms of multiple components following individual (although not independent) paths. VI I. INTRODUCTION Knowledge acquisition is a process fundamental to survival that begins early and continues throughout the life span. What do we know of the process? Research within the last decade has made it clear that from an early age knowledge is organized into theories that are elaborated and revised over time and that serve as vehicles for understanding the world.
In other words, knowledge acquisition to a large degree occurs through a process of theory formation and revision. Among researchers adopting a knowledge- or theory-based approach to cognitive development, the focus has been on describing the content of these evolving theories in a wide range of domains, and we now know a good deal about the progressively more elaborated knowledge that children of various ages are likely to have within numerous content domains (Gelman & Wellman, in press; Wellman & Gelman, 1992).
In contrast, relatively little attention has been given to the process of knowledge acquisition itself, that is, the mechanisms by means of which theories are formed and revised and knowledge is thereby acquired. It is this topic that is the focus of the present work. Within the knowledge-based approach, the assumption that has been at least implicit, and is occasionally voiced explicitly (Brewer & Samarapungavan, 1991; Carey, 1985a, 1986), is that these mechanisms remain more or less constant across development.
The present work rests on a contrasting claim that strategies of knowledge acquisition vary significantly across (as well as within) individuals and can be conceptualized in developmental terms. KNOWLEDGE AS ACQUISITION THEORY-EVIDENCE COORDINATION The general form of knowledge and knowledge acquisition studied here is that of the relation between one category of event and another. Most commonly, such relations are construed causally (Cheng & Nisbett, 1993), with an antecedent category of event interpreted as influencing an outcome I
KUHNETAL. category (e. g. , ingestion of food and a child’s bodily growth). Underpinning this form of knowledge is a more basic one having to do with how events or objects fit together into categories (e. g. , foods, nonfoods, and permanent vs. temporary bodily changes). Although the latter is not examined here, both forms of knowledge involve theories as organizing devices (Barrett, Abdi, Murphy, & Gallagher, 1993; Keil, 1991; Medin, 1989; Wisniewski & Medin, 1994).
Children’s and adults’ theories about causal relations undergo revision as new evidence is encountered. Hence, knowledge acquisition strategies involve the evaluation of evidence and inductive causal inference. Recent theories of inductive causal inference in adults (Cheng & Novick, 1990, 1992) are consistent with earlier accounts (Alloy & Tabachnik, 1984; Holland, Holyoak, Nisbett, & Thagard, 1986) in attributing prominent roles both to prior expectation (or theory) and to evidence of covariation (of the relevant factors) in fostering inferences of causality.
It is difficult to explain not only simple concept formation (Keil, 1991) but even basic conditioning phenomena in animals without invoking a construct that involves expectation (Holyoak, Koh, & Nisbett, 1989). A conception of inductive inference as involving a coordination of theory and evidence (Kuhn, 1989) contrasts with earlier approaches to the development of inductive inference strategies-for example, the Piagetian research on formal operations-in which such strategies were regarded as largely domain independent and therefore equally applicable to any content irrespective of prior knowledge or expectation.
In empirical studies of adults’ multivariable inductive causal inference, subjects typically are provided with a set of multiple instances in which one or more potential causes does or does not occur and an outcome is present or absent (Cheng & Novick, 1990, 1991; Downing, Sternberg, & Ross, 1985; Schustack & Sternberg, 1981). The subject is asked to evaluate the evidence and draw inferences regarding the causal status of one or more of the factors.
Although this approach can reveal much about how varying patterns of evidence affect inference, it does not lend much insight into the minimumconditions for an inference of causality, which may be as little as a single co-occurrence (of antecedent and outcome), even in the clear presence of additional covariates (Kuhn & Phelps, 1982). Moreover, in natural settings, even when multiple instances are readily available, there is no reason to believe that individuals will seek out and attend to all of them.
For both these reasons, we were interested in studying situations in which subjects are free to select the evidence on which they base their inferences, a condition that links the present work to research on scientific reasoning (Klahr, Fay, & Dunbar, 1993; Kuhn, Amsel, & O’Loughlin, 1988; Kuhn, Schauble, & Garcia-Mila, 1992), as we discuss further later in this chapter. Yet the cognitive skills examined in this Monograph are, we believe, 2 STRATEGIES KNOWLEDGE OF ACQUISITION epresentative of processes of knowledge acquisition and inductive inference more broadly (Kuhn, 1993). We therefore situate the present work in this broader context. Methodologically, this means that we examine knowledge acquisition across a broad range of domains involving both physical and social phenomena, rather than restricting the investigation to traditional scientific domains. THE MICROGENETIC METHOD To study knowledge acquisition strategies and their development, we use a microgenetic method.
The virtues of the microgenetic method as a tool for examining change have been elaborated in our own earlier work (Kuhn & Phelps, 1982) and more recently by Siegler and Crowley (1991). The evolution of behaviors that one observes over time in microgenetic study can serve to corroborate cross-sectional differences in performance. Most important, however, the method offers the opportunity for detailed analysis of the process of change. Later in this chapter, we summarize findings from previous research utilizing a microgenetic method.
An important feature of the method is that changes over time are initiated by subjects themselves, in interaction with the problems materials, rather than by the investigator, who provides no instruction or feedback with respect to a subject’s approaches to a problem. The rationale is that increased density of exercise of existing strategies may lead to change that, except for occurring comparatively rapidly, otherwise resembles a naturally occurring change process.
The researcher is thereby afforded close observation of the process. In addition, a third potential benefit of the method is its capacity to provide a fuller, more accurate picture of competence than can be attained using a single-occasion method. If a subject’s performance improves after a few sessions of engagement, it tells us that this level of performance was within the subject’s capabilities and accordingly should be recognized as part of his or her competence, or “zone of proximal development” in Vygotsky’s (1978) terms.
In several respects, the method used in the work reported in this Monograph is an elaborated form of the microgenetic method, one that has not been used in other microgenetic research. First, we simultaneously track two kinds of change over time within a domain. One is the subject’s evolving knowledge within that domain (specifically, knowledge of the causal and noncausal relations among variables that reflect the structure of the domain). The second kind of change is in strategies of knowledge acquisition, which may also evolve as knowledge is being acquired.
In other uses of the microgenetic method, typically only one form of change has been observed, 3 KUHNETAL. for example, in strategies for solving addition problems (Siegler & Jenkins, 1989). A second respect in which the basic microgenetic method is elaborated is that we observe change within multiple domains in which the subject is engaged at the same time. Doing so allows us to compare both knowledge acquisition and evolving strategy usage across domains (as well as relating the two to one another within domains).
We wished to examine a broad range of domains, involving both physical and social content, to establish the generality of the knowledge acquisition processes being examined. The research design thus stipulated that each subject undergo parallel engagement with one problem in the physical content domain and one problem in the social content domain. A number of considerations lead to the prediction of greater challenge (and hence inferior performance) in the social domain. Among these are the possibly more extensive initial knowledge (whether or not it is correct) in the social domain and possibly greater ffective investment in this knowledge (Kunda, 1990), either of which would make the task of theory-evidence coordination more difficult. A third elaboration of the microgenetic method is reflected in a research design that incorporates a traditional transfer design within a microgenetic framework. The purpose, again, is to establish generality of the knowledge acquisition strategies that we examine. The traditional transfer design used to assess generality of a skill across content domains is problematic for a number of reasons that we need not review here.
A further problem arises if (as we show here to be the case) a subject at a given point in time does not possess just a single strategy but instead selects strategies from a repertory of multiple strategies. If so, single-occasion assessment within a single content domain may produce an inaccurate and misleading characterization (since the subject could have selected a different strategy on this particular occasion and might do so on another occasion); in this case, accurate single-occasion assessment of generality acrossdomains is precluded.
The multiple-task, multiple-occasion assessment employed here allows us to assess generality in a more dynamic way than is afforded by a traditional transfer design. Each subject worked on a problem in the physical domain at one weekly session and a problem in the social domain at a second weekly session, for each of the first 5 weeks of a 10-week period of observation. At the beginning of the sixth week, new problems within each of the domains were substituted, and the sessions continued for the remaining five weeks.
The question we ask is whether the substitution of new content affects the strategies that the subject uses. To the extent that the same set of strategies that a subject uses in the final encounters with the initial problem carries over to the new content, some degree of domain generality (of both strategies and strategy change) is indicated. A final elaboration of the microgenetic method is to replicate the design 4 STRATEGIES KNOWLEDGE OF ACQUISITION with multiple age groups, enabling us to compare the knowledge acquisition process across age levels.
In addition to providing further evidence regarding the generality of knowledge acquisition processes (across populations in this case, rather than content), this comparison is important in addressing a more specific question. The pattern observed in our own as well as others’ microgenetic work has been one of mixed, or variable, strategy usage, as we describe in the next section. In other words, instead of a single, consistent approach, the subject shows variable usage of a variety of more and less competent approaches, even though the problem environment remains constant.
An ambiguity arises, however, owing to the fact that the subjects observed in microgenetic work have been either assumed or assessed to be in a state of transition with respect to the competencies in question. It is possible, therefore, that the variable strategy usage that has been observed is a particular characteristic of a developmental transition state, as dynamic systems theories of development predict (Van der Maas & Molenaar, 1992). It thus becomes important to ask whether the same variability over repeated occasions would be observed among populations at other than a characteristic age of transition.
If it is, it suggests that this variability is a more general characteristic of human performance, rather than one unique to states of developmental transition. To address this fundamental question, we chose preadolescents and community college adults as the two populations on which to base such a comparison. Previous work (Klahr et al. , 1993; Kuhn et al. , 1988) establishes the preadolescent age level as one at which the strategies in question are just beginning to emerge.
However, some young adult populations show initial levels of performance little more advanced than those characteristic of preadolescents (Kuhn et al. , 1988), enabling us to compare subjects of these two ages in a microgenetic design. In addition to establishing whether strategy change occurs at periods other than the typical period of developmental transition, the design allows cross-age comparison of the process of knowledge acquisition as well as of the interaction of knowledge acquisition and strategy change. Another set of questions centers on the effects of the exercise provided by the microgenetic method.
Despite similar starting points, does one age group show more rapid evolution of strategies than another group, both having been provided comparable exercise? Does such change differ only in degree or also in form? These questions are central to establishing the generality of knowledge acquisition strategies across populations. A final purpose of this Monographis to present a method of analysis that combines qualitative analysis of individuals with quantitative analysis of groups of individuals. Observers of the field’s progress, such as White (1994a, 1994b), have lamented the limited range of methods to which devel5
KUHNETAL. opmental researchers have restricted themselves. Especially in undertaking to study the difficult topic of processes of change, innovative methods are called for. In particular, the study of individual subjects is receiving increasing attention as an important and neglected method. As a research method, however, single-subject analysis most often is treated skeptically, and even dismissed, on the assumption that it is severely limited by its inability to provide evidence of the generality of the phenomena observed.
Here, we undertake to illustrate how individual and group, as well as qualitative and quantitative, modes of analysis can be used in conjunction to provide an enriched understanding of developmental phenomena. In the next section, we discuss previous research in more detail, in order to situate the present research effort in the context of various lines of work to which it connects. The reader wishing to focus exclusively on the present work can proceed directly to the final section of this chapter, which introduces the inference forms that figure prominently in later chapters.
THE PRESENT STUDY THE IN CONTEXT PASTRESEARCH OF FromLearning ConceptualChange to It was only a few decades ago that knowledge acquisitionand learning were treated as synonymous terms, both referring to a process of strengthening of associative bonds between stimuli and responses. In developmental psychology, Kendler and Kendler (1975) deserve the major credit for moving the field beyond a conceptualization of the developing child as a “cluster of interrelated responses” (Bijou & Baer, 1961, p. 4) and delving into the black box that represented mental phenomena. Although the Kendlers’ modeling of such phenomena in terms of covert stimuli and responses was highly restrictive, they demonstrated convincingly that the learning process cannot be studied without considering the developmental status of the organism. That insight remains a central one today. What individuals already know and how that knowledge is organized constrains what and how new knowledge will be acquired.
The burgeoning area of research that has come to be known as the study of conceptual change documents the development of knowledge in numerous domains, with physics (Vosniadou & Brewer, 1987, 1992) and biology (Carey, 1985b) the domains that have been the object of greatest study. Extensive literature reviews are provided by Gelman and Wellman (in press) and Wellman and Gelman (1992). The main tenet underlying and connecting these individual lines of 6 STRATEGIES KNOWLEDGE OF ACQUISITION research is that cognitive development can be adequately accounted for in terms of developing knowledge within content domains.
As a consequence, findings are largely specific to the domain studied. The major insight that extends across domains is the theory-like organization of knowledge. Even the properties that define simple concepts cluster and mutually support one another. Conceptions of such homeostatic causal clusters, and the mechanisms underlying them, are the “glue” that makes features cohere (Keil, 1991). At a less elementary level, evidence exists suggesting that young children’s theories have properties such as consistency, coherence, comprehensiveness, and explanatory power (Brewer & Samarapungavan, 1991; Vosniadou & Brewer, 1992).
As noted earlier, relatively little attention has been given to the mechanisms that effect theory change. When and how does new evidence lead to modification of existing theories? Despite theoretical claims that these mechanisms are developmentally invariant (Brewer & Samarapungavan, 1991; Carey, 1985a, 1986), little empirical work has been devoted to investigating them. Some research has been done to support claims that theory change will be more difficult to accomplish if it crosses ontological categories (Chi, 1992), involves radical (vs. eak) restructuring (Carey, 1990), or violates entrenched beliefs (Vosniadou & Brewer, 1992). But how should the mechanisms of change be conceptualized? Keil (1988, 1989, 1991) has addressed this question with respect to the formation of elementary concepts, contrasting accounts maintaining (a) that such concepts arise out of networks of associations observed in the environment, (b) that the process is theory guided, or (c) that at some point a developmental shift occurs from the first to the second process.
Keil (1991) rejects the possibilities of an exclusive associative network process and a developmental shift from such a process to a theory-guided one, asking how coherent theories could arise out of networks of associations. Instead, he proposes, all concepts represent a blend of an associative matrix overlaid with causal beliefs. Humans have evolved adaptations for building knowledge representations about sets of regularities in the world, but these processes are never completely data driven or completely theory driven.
In the present work, we address a similar question regarding the mechanisms of conceptual change but in this case with respect to the secondorder concepts of relations (particularly causal relations) between elementary conceptual categories. We adopt a perspective resembling Keil’s that the mechanism entails the coordination of new evidence with an existing network of theories. What are the strategies that an individual uses to achieve this coordination, and do they change with age and practice? Addressing this question leads to the topics of inductive causal inference and scientific reasoning.
First, however, we examine issues involved in the study of change. 7 KUHNETAL. Learning,Transfer,and the Study of Change The process of knowledge acquisition is likely to figure prominently in any comprehensive theory of human cognitive functioning. One prominent example is Sternberg’s (1984, 1985) triarchic theory, in which knowledge acquisition mechanisms are one of several core components of the intellect. But how is knowledge acquisition studied empirically? Psychologists studying very simple, short-term learning processes may be able to observe these processes directly in the laboratory.
The study of more comprehensive kinds of cognitive change, however, especially those involving change in knowledge acquisition strategies themselves, poses serious methodological challenges. Developmental psychologists have been in the particularly difficult position of seeking to understand developmental change without observing it directly. As has now been widely noted, the cross-sectional and even longitudinal designs that are the staples of developmental psychology may provide suggestive data regarding change, but they do not afford direct observation of the process Wohlwill, 1973). The microgenetic method has been advocated as a way out of this impasse. As described by Kuhn and Phelps (1982), the goal of the method is to accelerate the change process by providing a subject with frequent opportunities over a period of weeks or months to engage the particular cognitive strategies that are the object of investigation. This increased density of exercise of existing strategies may lead to change, allowing the researcher close observation of the process.
In the initial work by Kuhn and Phelps (1982), we chose strategies of wide applicability as a basis for exploring the utility of the methodstrategies of inductive causal inference that are fundamental to knowledge acquisition and can be identified in both scientific and informal reasoning (Kuhn, 1991, 1993). In weekly sessions, preadolescent subjects were asked to identify causal and noncausal effects as they freely investigated a domain in which multiple variables played potential causal roles in influencing an outcome.
Strategies of investigation and inference did improve in a majority of subjects over the period of observation. In a comparison condition (Kuhn & Ho, 1980), subjects each week were presented with a set of antecedentoutcome instances identical to that which the subject’s yoked control in the free investigation condition had selected for examination; these subjects also showed some, but less, change. Subsequent research (Kuhn et al. , 1992; Schauble, 1990, in press), including the present study, has followed this same paradigm of microgenetic examination of inductive inference strategies in multivariable contexts.
Meanwhile, other developmental researchers, notably Siegler and his colleagues (Siegler & Jenkins, 1989), began to use the microgenetic method, in Siegler’s case in the very different domain of elementary addition strategies. 8 STRATEGIES KNOWLEDGE OF ACQUISITION Among other researchers who have used a microgenetic method in various domains are Bidell and Fischer (1994), Granott (1993), Karmiloff-Smith (1984), Lawler (1985), and Metz (1985, 1993). In addition, a line of Genevan work beginning with a study by Karmiloff-Smith and Inhelder (1974) falls under the heading of microgenetic research.
In certain respects, modern microgenetic research connects to work in the genetic tradition of Werner (1948), although the latter was limited to observation within a single session. Enough microgenetic work has accrued by now to make comparison and generalization possible (Siegler & Crowley, 1991). Studies conducted within very different domains show convergence in several important respects. Most important, they provide a clear indication of what the change process is not-simple replacement of a less adequate approach with a more adequate one.
Instead, subjects commonly exhibit intraindividual variability in the strategies that they apply to identical problems, with less adequate strategies coexisting in a subject’s repertory together with more adequate ones. The initial appearance of a new strategy, then, does not mark its consistent application. Instead, less adequate strategies continue to compete with it, and, indeed, the more formidable challenge appears to be abandoning the old, rather than acquiring the new-a reversal in the way that development is traditionally conceived.
Change does occur, but it appears as a gradual shift in the distribution of use of a set of strategies of varying adequacy. The most recent microgenetic work (Granott, 1993; Metz, 1993) offers a number of additional insights regarding the nature of the change process. We return to them in the final chapter in discussing insights from the present work. As described earlier in this chapter, a main purpose of the present work is to extend the microgenetic method in ways that address several critical questions.
One is whether the variability and change observed in microgenetic studies is particular to subjects in a period of developmental transition or is a more general phenomenon. A second is the extent to which such change is general as opposed to domain specific. Domain specificity versus domain generality of cognitive strategies is a topic at the heart of much current debate in the field of cognitive development (KarmiloffSmith, 1994). In a previous study (Kuhn et al. , 1992), we addressed this question by having subjects work simultaneously in two domains, with separate sessions each week devoted to each.
This study provided some evidence of generality in that improvements in strategy tended to co-occur in rough synchrony across the two domains. These findings, however, do not provide an answer to the more traditional question of whether the newly developed competencies would transfer to new content to which the subject had not been previously exposed. This question is addressed in the present work. Studies of transfer have served as the traditional means for assessing generality: Does a newly acquired competency transfer to a new context? 9 KUHN ET AL.
Whether the subjects are preschool children or college adults, in a majority of cases the answer has been no. Such findings have led to critical scrutiny of the transfer construct (Detterman & Sternberg, 1993) as well as increasingly domain-specific conceptions of cognitive development (Karmiloff-Smith, 1992). Why should transfer to new contexts be expected? Two prevailing conceptualizations of transfer offer somewhat different answers. In the more common conceptualization, transfer is seen as mediated by a symbolic representation of the problem domain (Brown, 1989, 1990; Gentner, 1983, 1989; Holyoak, 1984; Singley & Anderson, 1989).
To the extent that there is overlap between the representations of two problem domains (i. e. , the extent to which the elements of one map onto the elements of the other), transfer between the two should occur. In a study by Brown and Kane (1988), for example, subjects had to recognize a connection between pulling a boat ashore with a fishing rod and pulling someone out of a hole with a spade. A somewhat different conception of transfer (Greeno, Smith, & Moore, 1993) emphasizes the activity that the problem solver engages in.
To the extent the activity is common to two settings, transfer will occur. In the words of Greeno et al. (1993, p. 146), “The structure that enables transfer is in the interactive activity of the person in the situation. … When transfer occurs it is because of general properties and relations of the person’s interaction with features of a situation. ” It is this latter conception of transfer that fits our paradigm better than the first one (which is sometimes referred to as analogical transfer).
The strategies that subjects develop are very broadly applicable across a wide range of content, but subjects learn to apply these strategies only within the context of particular, relatively narrow content. Will these strategies generalize to new and diverse kinds of content? This classic transfer question is complicated by the findings from microgenetic research. As noted earlier, microgenetic data indicate that, at a given point in time, a subject does not possess just a single strategy but instead selects trategies from a repertory that includes multiple strategies of varying adequacy. Given this situation, assessment on a single occasion within a single content domain may produce an inaccurate characterization of the subject’scompetence (since the subject might have selected a different strategy). As a consequence, studies that assess competence across domains are even more error prone. To overcome these limitations, in the present work we situate the transfer design in a microgenetic context, substituting new content midway through the observation period.
Through this technique we hope to answer a critical question about the generality of the change induced in microgenetic studies as well as to assess transfer in a more dynamic way than it has been approached in the past. 10 STRATEGIES OFKNOWLEDGE ACQUISITION Metacognition,FormalOperations,and ScientificReasoning Piaget (1950; Inhelder & Piaget, 1958, 1969) offered an explicit account of a developmental progression in strategies of knowledge acquisition. Young children construct rudimentary concepts of the type examined by Keil (1989, 1991) that we referred to earlier.
With the advent of concrete operations at the age of 6 or 7, concepts acquire the properties of systematic hierarchical classes. A further major development occurs with the appearance of formal operations at adolescence, when second-order relations between categories begin to be examined-the skill on which the present Monographfocuses. Piaget’s theoretical model of formal operations has been criticized (for a review, see Keating, 1980), and in his later work (Piaget & Garcia, 1991) there is evidence that even he came to regard the model as insufficiently concerned with the meaning of the propositions that subjects contemplated.
Empirical research relating to formal operations has been largely focused on subjects’ability to conduct scientific investigation of the relations between variables in a multivariable context, and here, in contrast, Inhelder and Piaget’s (1958) pioneering work has been substantially replicated (Keating, 1980; Moshman, in press). Both the methods and the conclusions of scientific investigation are likely to be faulty among subjects younger than midadolescence; moreover, as research subsequent to Inhelder and Piaget’s has shown, even older adolescents and many adults often perform poorly as scientists (Dunbar & Klahr, 1989; Klahr et al. 1993; Kuhn et al. , 1988; Schauble & Glaser, 1990). Although they did not use the term, Inhelder and Piaget (1958) in effect attributed poor performance in scientific reasoning tasks to metacognitive weakness, defined as the inability to contemplate one’s own thought as an object of cognition or, in their (1958) terms, to engage in second-order operations on operations. To the extent that such an ability is truly lacking, the ramifications no doubt extend well beyond the realm of scientific reasoning (Kuhn, 1992a, 1993).
Subsequent to Inhelder and Piaget’s (1958) work, metacognition has become a topic of widespread interest (Flavell, 1979, 1993; Flavell, Green, & Flavell, 1995; Flavell & Wellman, 1977; ForrestPressley, MacKinnon, & Waller, 1985; Metcalfe & Shimamura, 1994; Moshman, 1979, 1990, 1995; Schneider, 1985), but the term has been variably and often loosely defined, with the majority of investigators employing it in its initial and more restricted sense of knowledge and management of one’s cognitive strategies, particularly memory strategies.
In the present work, we make a distinction between metacognitive knowledge and metastrategic knowledge, a distinction that parallels in many respects the lower-order distinction between declarative and procedural 11 KUHNETAL. knowledge. Metacognitive knowledge involves awareness of and reflection on the content of one’s thought, ranging from simple awareness of the content of one’s present or immediately prior thought (Flavell et al. , 1995) to reflection on a set of propositions that one believes to be true or chooses to take under consideration (Moshman, 1990).
Metastrategic knowledge involves awareness and management of the strategies that are applied in the course of thinking and problem solving (Sternberg, 1984). Both metacognitive and metastrategic knowledge entail treating one’s own cognition as itself an object of cognition-a form of cognitive “distancing”(Sigel, 1993). Both metacognitive and metastrategic knowledge, we will claim, figure importantly in the development of the cognitive skills examined in this Monograph.
If knowledge acquisition is a process of theory revision, as we have claimed, to accomplish the process in a skilled way the individual needs to be aware of and reflect on a theory (metacognitive competence), coordinating it with new evidence by means of strategies that are inferentially sound and applied in a consistent manner (metastrategic competence). In the total absence of such competence, evidence and theory are not represented as distinct entities.
In this case, new evidence may lead to modification of a theory (as it does even among very young children), but the process takes place outside the individual’s conscious control (Kuhn, 1989). There is a problem, however, with attributing proficiency in knowledge acquisition or scientific reasoning to the development of metacognitive or metastrategic competencies emerging at adolescence. Competent scientific reasoning entails a number of component skills, and data exist suggesting that at least rudimentary forms of all these skills are in place well before adolescence.
In addition to the metacognitive and metastrategic abilitiesjust discussed, included among these skills are the ability to entertain alternative possibilities, to detect and interpret covariation, and to isolate and control variables. One study (Richardson, 1992) in particular stands out for its strong claim of early competence. Even young children, the author maintains, readily interpret both additive and interactive effects of three or more variables-a claim that stands in striking contradiction to data to be presented in this Monographdemonstrating the difficulty that even adults have with such coordinations.
The data from Richardson’s study, however, cannot be clearly interpreted for a number of methodological reasons, foremost among them being the failure to examine individual patterns of performance and distinguish them from group data. The remaining studies of early competence make more modest claims that certain abilities traditionally associated with scientific reasoning are present in rudimentary forms in young children. Sodian, Zaitchik, and Carey (1991), for example, undertook a study to show that young children 12 STRATEGIES KNOWLEDGE OF ACQUISITION an distinguish between an assertion and evidence that bears on the assertion if the context is simple enough. They posed first- and second-grade in their house was a large or small one, which they did by placing food in a box overnight. Two boxes were available, one with a large opening (able to accommodate a large or a small mouse) and one with a small opening (big enough for only the small mouse to pass through). The subject was asked which of the two boxes the children should put food in. Sodian et al. (1991) report that 11 of 20 first graders and 12 of 14 second graders preferred the determinate solution (i. . , chose the small-opening box), indicating both considerable competence and considerable development in this age range. Sodian et al. (1991) note that their subjects’ performance reflects a differentiation of hypothesis and evidence since the hypothesis (large or small mouse) is distinguished from the evidence that will test it (the food disappears or does not). Note, however, that the potential confusion in this case is not between theories and evidence (mice and food) but rather lies in the selection of the form of evidence appropriate to test a theory.
In a subsequent set of more detailed studies, Ruffman, Perner, Olson, and Doherty (1993) report similar evidence in comparably simple contexts even among some 5-year-olds (as well as 6- and 7-year-olds). In fact, everyday observation confirms that implicit forms of theory-evidence coordination occur at even earlier ages-illustrated, for example, by a 2-year-old who calls her parents into her bedroom with the claim that it is a ghost in her closet that is the cause of a soft “whooshing” noise that is keeping her awake.
This child understands as well as her parents that opening the closet door will provide the evidence capable of disconfirming this causal hypothesis, even though she lacks any metacognitive awareness of her own belief states as hypotheses to be coordinated with evidence. The valuable function served by Ruffman et al. ‘s (1993) study is to make clear the connection that exists between early theory-of-mind competencies (Feldman, 1992; Perner, 1991; Wellman, 1990) and competencies that figure importantly in scientific reasoning.
Both have strong metacognitive aspects. The 4-year-old child who comes to recognize that an assertion is not necessarily correct-that the candy can be believed to be in the cupboard and in truth be elsewhere (Perner, 1991)-has achieved an essential milestone in the development of scientific reasoning ability. This child has made at least a primitive differentiation between what a mind theorizes to be true and information from the external world that bears on this theory. False beliefs, by definition, are subject to disconfirmation by evidence. Although it has ometimes been treated this way in the literature, metacognition, like cognition, is not a zero-one, present-absent phenomenon that emerges in full bloom at a particular point in development. The position subjectsa problemin which some childrenwanted to find out if a mouse 13 KUHNETAL. taken in this Monographis that the development of metacognitive competence, like that of metastrategic competence, takes place very gradually over many years and involves a process of increasing “explicitation”(KarmiloffSmith, 1992) of skills present in implicit form.
Metacognitive competence develops from its most rudimentary forms (examined by Flavell and Gopnik and their colleagues in studies to be described shortly) to the more highly developed, explicit forms demanded by the activities in which subjects in the present research engage. Ruffman et al. (1993) illustrate the evolution of early emerging metacognitive capability relevant to scientific reasoning by asking subjects to reason about propositions as belief states (a requirement not present in Sodian et al. ‘s, 1991, study).
They ensure that subjects do so by explicitly characterizing these belief states as false. Many (although not all) of the 5-7-year-olds in their research judged that a story character who observes a set of dolls who usually choose red over green food will conclude that the dolls like red food, even though the subjects themselves have been told that this is not the true state of affairs (the dolls really like green food, the subject is told). In this respect, the child comprehends the relation between a pattern of evidence and a theory (the contrary-to-fact hypothesis held by the story character).
Put in different terms, the child can draw appropriate inferences from contrary-to-fact propositions (an ability that Piaget tied to the emergence of formal operations). In a follow-up experiment, Ruffman et al. showed that this comprehension extends to predictive judgments (e. g. , that the dolls will choose red food again). In theory-of-mind terms, these children are drawing appropriate inferences regarding others’ belief states (or theories, as long as we agree to use this term in its broad sense), even when they have been told that these theories are not correct. The material is deliberately designed so that the child’s own theoretical preferences are likely to be neutral. ) The portrayal of early proficiency in metacognitive competencies important to scientific reasoning that Ruffman et al. (1993) offer needs to be qualified, however, by other research demonstrating that the period between 4 and 8 years of age is one of significant development of the basic metacognitive competencies that serve as underpinnings of more complex forms of reasoning about propositions. A series of studies by Flavell et al. 1995) showed 3-5-year-old children to have considerable difficulty accurately reporting either their own immediately preceding mental activity or that of another individual, in contexts in which that mental activity had been particularly clear and salient. In contrast, 7-8-year-olds were largely (although not entirely) successful in such tasks. Distinctions between (second-order) representations (and consequent verbal reports) of thinking about an object and (first-order) representations of the object itself appeared fragile in the younger children.
The older ones, like children of a 14 STRATEGIES KNOWLEDGE OF ACQUISITION similar age in Ruffman et al. ‘s (1993) research, were better able to make inferences that depended on representations of mental states. In related work, Gopnik and her colleagues (Gopnik & Graf, 1988; Gopnik & Slaughter, 1991) showed that preschoolers have a limited awareness of the source of their beliefs-a metacognitive ability that figures prominently in the work presented in this Monograph.
Gopnik and Graf (1988) found that, even in very simple situations, 3- and 4-year-olds could not identify where knowledge they had just acquired had come from-for example, whether they had learned the contents of a drawer from seeing them or being told about them. Performance was significantly improved, however, among 5-year-olds. Some of Gopnik and Graf’s 3- and 4-year-olds might even have been successful in Sodian et al. ‘s (1991) task of differentiating and coordinating a theory (about a mouse’s size) and evidence (of food eaten or not) bearing on it, but they showed remarkablylittle differentiation of theory and evidence at he metacognitive level of distinguishing the representation of what they knew (the contents of the drawer) from a representation of the evidence that had provided this knowledge. Once the knowledge was acquired, the two evidently became fused into a single representation that encompassed only the knowledge itself. Supporting this interpretation are other findings showing that preschool children report that they have “always known” knowledge that was just acquired in the experimental situation (Gopnik & Astington, 1988; Taylor, Esbensen, & Bennett, 1994).
Evidence regarding early strategic (as opposed to metastrategic or metacognitive) competence related to scientific reasoning is largely positive. Ruffman et al. ‘s (1993) study substantiates that one of several simple strategic competencies entailed in scientific reasoning-inferring causality from covariation evidence-poses no great difficulty among young children, as earlier research had shown (Mendelson & Shultz, 1976; Shultz & Mendelson, 1975). Indeed, this ability is evident at the sensorimotor level in human infants (Piaget, 1952) as well as in nonhuman organisms.
By the end of the first year of life, infants have begun to make causal inferences based on the juxtaposition of an antecedent and an outcome. As data in the present illustrate, it is the fact that this inference strategy is overlearned Monograph that causes problems. Precursors to the critical control-of-variables strategy most closely associated with scientific reasoning are also evident. Most elementary among these are judgments of comparison, first in terms of an individual (Can I run faster than my brother? , later in terms of groups of individuals (Can the girls in the class run faster than the boys? ). Once the concept of a fair comparison emerges (What if the boys wore running shoes and the girls didn’t? ), it remains only to formalize the comparison into the framework of a controlled test of relations between variables (gender and running speed). 15 KUHNETAL. Case (1974) has shown that, although they do not do so spontaneously, children as young as age 8 can readily be taught to carry out controlled comparisons.
Early developing forms of metastrategic competence are also evident. A skill important to scientific reasoning is recognition of the indeterminacy associated with entertaining alternative possibilities. This skill is explored in a line of research beginning with studies by Pieraut-Le Bonniec (1980). During the early childhood years, children develop the ability to discriminate between situations that have determinate solutions and those that do not or, in other words, to know whether they have an answer-a competency having clear metastrategic aspects. For a review of research, see Acredolo & O’Connor, 1991, or Byrnes & Beilin, 1991. ) The study by Sodian et al. (1991) can also be interpreted in these terms. In the face of evidence of all this early competence, a perplexing problem is to explain the persistent poor performance of children, adolescents, and many adults in full-fledged scientific reasoning tasks, that is, ones in which they are asked to examine a database and draw conclusions (Dunbar & Klahr, 1989; Klahr et al. , 1993; Kuhn et al. 1988; Schauble & Glaser, 1990). Addressing this critical question is an important objective of the present Monograph. With repeated exercise, we find, knowledge acquisition strategies improve among most subjects, but these strategies remain error prone and inadequate among many adults as well as children. Microgenetic data will, we hope, provide insight into the obstacles that impede success in these fundamental forms of reasoning and knowledge acquisition. We therefore return to this question after the data have been presented.
Inductive Causal Inferencein Multivariable Contexts It is a curiosity that research on scientific reasoning (originating and remaining largely in the developmental literature) has proceeded independently of and remains largely unintegrated with research on multivariable inductive causal inference (centered in the adult cognition literature). The central difference between the two is a simple one. Whereas studies of scientific reasoning typically involve selecting instances to create a database, studies of inductive causal inference involve presenting instances from a database for examination.
In both, however, the subject must interpret the evidence and draw conclusions, these conclusions being the end product of the process in both cases. Kuhn and Brannock (1977) argued that the “natural experiment” situation involved in studies of inductive inference elicits forms of reasoning paralleling those identified in earlier studies of isolation of variables within the framework of formal operations and scientific reasoning. 16 STRATEGIES KNOWLEDGE OF ACQUISITION
Although there exists a large literature on the development of causal inference (for a review, see Bullock, 1985; Bullock, Gelman, & Baillargeon, 1982; Sedlak & Kurtz, 1981), with the exception of our own developmental studies (Kuhn & Brannock, 1977; Kuhn & Phelps, 1982; Kuhn et al. , 1988) theoretical and empirical work on multivariable causal inference has largely been located in the adult cognition literature. Like much of the literature on scientific reasoning development, the developmental literature on causal inference highlights the child’s early competence.
As noted earlier, from an early age children draw on covariation information, as well as other cues, as a basis for inferences of causality (Mendelson & Shultz, 1976; Shultz & Mendelson, 1975). Equally important, from an early age they have theories of causal mechanism that influence their causal judgments (Shultz, 1982), a finding consonant with the more recent conceptual change literature. Within the adult literature, theoretical analysis has focused largely on covariation as the most important source of information about causality.
Mill’s (1843/1973) ‘joint method of agreement and difference” identifies covariation as the appropriate basis for inferences of causality, and Kelley’s (1967) extensively researched attribution model similarly rests on covariation between antecedent and outcome. More recent investigators have followed in this tradition but have sought to identify more precisely the inductive strategies that mediate between a covariational database and an inference of causality.
In empirical studies, typically a set of multivariable instances is presented in written form and the subject asked to judge what inferences can be drawn (Briggs, 1991; Cheng & Novick, 1990; Downing et al. , 1985; Schustack & Sternberg, 1981). On the basis of such data, Schustack and Sternberg (1981) developed a linear regression model to assign weights to five types of covariation information. The first four are frequencies of the joint presence of antecedent and outcome, the joint absence of antecedent and outcome, the presence of antecedent and the absence of outcome, and the bsence of antecedent and the presence of outcome. A fifth factor is the strength of competing causes. Although adult subjects show consistency, leading to positive regression weights for the first two frequencies and negative weights for the second two, Cheng and Novick (1992) identify several theoretical anomalies in the linear regression model, for example, the role of base-level frequencies of antecedent and outcome in predicting the likelihood of a causal inference, factors that intuitively should not affect the causal status of the antecedent.
An even more critical problem, however, for such models of induction purely from an empirical database is the sheer computational weight of the task. The four frequencies in the Schustack and Sternberg model pertain to a single potential cause and outcome. Once the causal field is opened to a host of causal candidates (as it is in natural settings), the computational 17 KUHNETAL. burden quickly becomes enormous. Some means of narrowing the causal field to a set of manageable factors is needed.
Different approaches have been taken to accomplishing this objective, but they have in common restriction of the set of potential causes to the “set of events considered relevant by the attributor” (Cheng & Novick, 1990, p. 562). In other words, theoretical expectation on the part of the subject, arising from a preexisting knowledge base, is invoked as a factor in the attribution of causality. Cheng and Novick (1990, 1992) propose that, within this focal set, inferences of causality are based on estimated differences in the probabilities of the effect in the presence versus the absence of the potential cause.
Hilton and Slugoski (1986) specify “abnormal conditions”-those absent in a comparison condition-as the ones likely to be attributed as causes. Both models invoke the distinction emphasized by Mackie (1974) and others (Einhorn & Hogarth, 1986) between causes and enabling conditions. In Cheng and Novick’s (1992) model, factors yielding substantial differences across instances will be attributed as causes, whereas factors that are constant across instances will be either regarded as enabling conditions, if they are perceived as relevant, or dismissed as causally irrelevant (and hence excluded from the focal set).
Note that the latter distinction rests entirely on the subject’s theoretical belief. Covariation within a focal set of instances may well provide the basis for a judgment of causality, but, when this covariation is absent, theoretical belief offers the only basis for judging whether constant factors are causally relevant (as enabling conditions) or noncausal. Studies in the adult causal inference literature have tended to focus only on inferences of causality, treating inferences of noncausality almost as noninferences.
They have not addressed the converse of the covariation principle-evidence of noncovariation over a set of instances as a basis for an inference of noncausality-or in general examined how empirical evidence might play a role in inferences of noncausality. As discussed in the next section, we see noncausal inference as occupying a prominent place in inductive inference, scientific reasoning, and knowledge acquisition, and these inferences are a central object of attention in the present work.
We also pay a good deal of attention to another problem that Cheng and Novick (1992) acknowledge is not addressed by their model-inferences of causality based on spurious covariation of a noncausal factor with an outcome. The fact that we examine inductive inference over a period of time as a database of instances accumulates enables us to observe how a subject may gradually overcome the temptation of this invalid inference strategy as well as more generally how the subject coordinates accumulating new evidence with theoretical expectation.
Most studies of causal inference have confined subjects to the presentation of a single set of instances 18 STRATEGIES KNOWLEDGE OF ACQUISITION on a single occasion (with data analysis typically confined to the group level). In contrast, we ask subjects to seek out the evidence that they believe adequate to support their causal and noncausal inferences, and we follow them individually in their efforts to interpret this evidence and integrate it with existing knowledge.
We turn now to an examination of the inference strategies that individuals might employ as they engage in this task. AND NONCAUSAL CAUSAL STRATEGIES INDUCTIVE OF INFERENCE Causal Inference(Inclusion) On what evidence might someone base the inference that antecedent a has a causal influence on outcome o? In the framework adopted here, we assume a multivariable context, and we assume that the individual is able to select instances to attend to. The question facing the individual is whether a particular factor a makes a difference to the outcome.
For simplicity of exposition, we consider the case in which the identified factors-a, b, c, d, and e-are dichotomous (two-level) variables. (Certain differences arise if the two levels of these variables are treated as presence and absence, but, again for simplicity of exposition, they need not be taken into consideration here, and the two levels of each variable will be designated by the subscripts 1 and 2. ) A further assumption that we make is that selection of instances is at least partially theory motivated.
In other words, the individual’s prior beliefs about the causal and noncausal status of the identified factors influence the selection of instances to attend to. This selectivity takes a variety of forms that need not be identified in detail at this point; some examples are the tendencies to select instances believed to produce the most positive level of an outcome (a success rather than an explanation orientation) and to fail to investigate factors that are believed noncausal.
A minimal (but, as we shall document, frequent) basis for the inference that an antecedent a and an outcome o are causally related-an inference to which we refer henceforth as the inclusionof a-is their co-occurrence within a multivariable context: al blcdl el — ol. (1) We refer to such an inference as a co-occurrence false inclusion inference (because a and o merely co-occur on one occasion). Such inferences are based on only a single instance and are of course invalid since the cooccurrence does not establish that a played a causal role in producing o. 19
KUHNETAL. In the case in which an individual selects at least two instances for examination, an informative second instance would be (2a) a2b c1d1el – 02. Such an instance, with the outcome shown, allows the valid inclusion inference that a is causally implicated in o. This inference, based on two instances, is the product of a controlled comparison. In most natural settings, however, people do not have the luxury of selecting for observation exactly those instances that would be most informative with respect to the inferences they al