The dynamics model, which is based on L. Talmy’s (1988) theory of force dynamics, characterizes causation as a pattern of forces and a position vector. In contrast to counterfactual and probabilistic models, the dynamics model naturally distinguishes between different cause-related concepts and explains the induction of causal relationships from single observations. Support for the model is provided in experiments in which participants categorized 3-D animations of realistically rendered objects with trajectories that were wholly determined by the force vectors entered into a physics simulator. Experiments 1–3 showed that causal judgments are based on several forces, not just one. Experiment 4 demonstrated that people compute the resultant of forces using a qualitative decision rule. Experiments 5 and 6 showed that a dynamics approach extends to the representation of social causation. Implications for the relationship between causation and time are discussed.

This article summarizes research on the roles of prior knowledge in learning. Educators often focus on the ideas that they want their audience to have. But research has shown that a learnerÕs prior knowledge often confounds an educatorÕs best efforts to deliver ideas accurately. A large body of findings shows that learning proceeds primarily from prior knowledge, and only secondarily from the presented materials. Prior knowledge can be at odds with the presented material, and consequently, learners will distort presented material. Neglect of prior knowledge can result in the audience learning something opposed to the educatorÕs intentions, no matter how well those intentions are executed in an exhibit, book, or lecture. Consider a hypothetical book on wool production in Australia. Australian ranchers raise sheep in an extremely hot desert climate. The sheep are raised to have wool so thick that without yearly trimmings the sheep would be unable to walk. To many children, these facts together are absurd. Children think wool is hot; if you put a thermometer inside a wool sweater, the mercury would go up (Lewis, 1991). WouldnÕt sheep grow more wool in cold places where they need to stay warm? Is wool hot because the sheep absorb the desert warmth? Alternatively, consider a hypothetical exhibit on fish schooling. Fish follow each other in close formation that looks highly organized. But no single fish is the leader, and none of the fish know how to command

Abstract: We stand poised to marry the fruits of qualitative research on children’s conceptions with the machinery of psychometrics. This merger allows us to build upon studies of limited groups of subjects to generalize to the larger population of learners. This is accomplished by reformulating multiple choice tests to reflect gains in understanding cognitive development. This study uses psychometric modeling to rank the appeal of a variety of children’s astronomical ideas on a single uniform scale. Alternative conceptions are captured in test items with highly attractive multiple choice distractors administered twice to 1250 8th through 12th-grade students at the start and end of their introductory astronomy courses. For such items, an unusual psychometric profile is observed—instruction appears to strengthen support for alternative conceptions before this preference eventually declines. This lends support to the view that such ideas may actually be markers of progress toward scientific understanding and are not impediments to learning. This method of analysis reveals the ages at which certain conceptions are more prevalent than others, aiding developers and practitioners in matching curriculum to student grade levels. This kind of instrument, in which distractors match common student ideas, has a profoundly different psychometric profile from conventional tests and exposes the weakness evident in conventional standardized tests. Distractor-driven multiple choice tests combine the richness of qualitative research with the power of quantitative assessment,

The purpose of this article is to provide an introduction to the growing body of research on the development of scientific reasoning skills. The focus is on the reasoning and problem-solving strategies involved in experimentation and evidence evaluation. Research on strategy use in science has undergone considerable development in the last decade. Early research focused on knowledge-lean tasks or on tasks in which subjects were instructed to disregard prior knowledge. The purpose of this article is to provide a general introduction to the body of research in cognitive and developmental psychology conducted under the labels ''scientific reasoning,'' ''scientific discovery,'' and ''scientific thinking.'' There are three main reasons for reviewing this literature. First, within the last decade, there was a call to develop a distinct ''psychology of science' Correspondence and reprint requests should be addressed to Corinne Zimmerman, Learning Research and Development Center, University of Pittsburgh, Pittsburgh, PA 15260. E-mail: czimmϩ@pitt.edu. 99 0273-2297/00 $35.00 Copyright © 2000 by Academic Press All rights of reproduction in any form reserved. 100 CORINNE ZIMMERMAN science educators also have been interested in children's understanding of both scientific concepts and the scientific enterprise (e.g., This intersection with science education provides the third rationale for this review. In a recent Handbook of child psychology chapter, DEVELOPMENT OF SCIENTIFIC REASONING 101 in science, but reviews only a handful of studies that exist on the development of domain-general scientific reasoning skills. To bridge this gap, my third goal is to provide a review of this literature for both audiences and to demonstrate how developmental research on scientific reasoning can be used to inform science educators. The plan of the article is as follows. In the first section, I will provide a general overview of the two main approaches to the study of scientific thinking: one focused on the development of conceptual knowledge in particular scientific domains and a second focused on the reasoning and problemsolving strategies involved in hypothesis generation, experimental design, and evidence evaluation. Both approaches will be introduced to distinguish two different connotations of ''scientific reasoning,'' but it is the second line of research that is the primary focus of this review. Next, I will introduce In the second section, I will review major empirical findings using the SDDS model as a framework. The first subsection will include a brief review of research that has been focused on experimentation skills. In the second subsection, I will review the research on evidence evaluation skills. The last subsection will include a discussion of self-directed experimentation studies. In these integrative investigations of scientific reasoning, participants actively engage in all aspects of the scientific discovery process so that researchers can track the development of conceptual knowledge and reasoning strategies. I will conclude this subsection with some generalizations about development of skills and individual differences. In the third and final section of the paper, I will provide a general summary around the points outlined above. Based on this review of the skills and strategies of scientific reasoning, I suggest the idea that science may best be taught as both an academic skill and a content domain. APPROACHES AND FRAMEWORK The most general goal of scientific investigation is to extend our knowledge of the world. ''Science'' is a term that has been used to describe both a body of knowledge and the activities that gave rise to that knowledge. In parallel, psychologists have been interested in both the product, or individuals' knowledge about scientific concepts, and the processes or activities that foster that knowledge acquisition. Science also involves both the discovery of regularities, laws, or generalizations (in the form of hypotheses or theories) and the confirmation of those hypotheses (also referred to as justification or verification). That is, there has been interest in both the inductive processes involved in the generation of hypotheses and the deductive processes 102 CORINNE ZIMMERMAN used in the testing of hypotheses. 2 Scientific investigation broadly defined includes numerous procedural and conceptual activities such as asking questions, hypothesizing, designing experiments, using apparatus, observing, measuring, predicting, recording and interpreting data, evaluating evidence, performing statistical calculations, making inferences, and formulating theories or models 3 The Domain-Specific Approach: Conceptual Knowledge about Science One approach to studying the development of scientific reasoning has involved investigating the concepts that children and adults hold about phenomena in various content domains in science, such as biology (e.g., Carey, 1985; DEVELOPMENT OF SCIENTIFIC REASONING 103 or instruction. Historically, this approach has roots in the pioneering work of Piaget, who was interested in the development of various concepts such as time, number, space, movement, and velocity (e.g., In the domain-specific approach, a typical scientific reasoning task consists of questions or problems that require participants to use their conceptual knowledge of a particular scientific phenomenon. For example, children were asked to answer questions about the earth such as ''Is there an edge to the earth?'' and ''Can you fall off the edge? '' (Vosniadou & Brewer, 1992, p. 553). In the domain of genetic inheritance, participants were asked to reason about issues such as the origins of anatomical and behavioral differences in species by answering questions such as ''How does it come about that people have different color of eyes/hair?'' and ''How did the differences between horses and cows originate? '' (Samarapungavan & Wiers, 1997, p.174). In the domain of physics, individuals were instructed to draw the path of a ball as it exits a curved tube In the previous examples, participants were using their current conceptual understanding (or ''misconceptions'') to generate a solution to the task. They were not required to evaluate evidence, make observations, or conduct experiments to verify their solutions or answers. 4 As there are several different domains of science, and numerous concepts of interest within each (e.g., within the domain of physics alone, different researchers have studied the concepts of gravity, motion, velocity, balance, heat/temperature, electricity, and force, to name a few), a thorough discussion of these literatures is not appropriate given the scope of this review. Reviews and collections of work on domain-specific concepts can be found in Carey (1985), CORINNE ZIMMERMAN The Domain-General Approach: Introduction and Background A second approach that has been taken to understand the development of scientific thinking has involved a focus on domain-general reasoning and problem-solving strategies that are involved in the discovery and modification of theories about categorical or causal relationships. These strategies include the general skills implicated in experimental design and evidence evaluation, where the focus is on the cognitive skills and strategies that transcend the particular content domain to which they are being applied. In their review of research on the acquisition of intellectual skills, Voss, Riley, and Carretero (1995) classified scientific reasoning as a ''general intellectual skill.'' Scientific thinking as defined in this approach involves the application of the methods or principles of scientific inquiry to reasoning or problemsolving situations This approach has historical roots in experimental psychology, in the body of research on reasoning and problem solving (e.g., In addition to examining the deductive components of scientific inference, Initial investigations with adults focused on a what seemed to be a perva-DEVELOPMENT OF SCIENTIFIC REASONING 105 sive ''confirmation bias'' that existed, even among scientists (e.g., Given Studies using the 2-4-6 task with children are rare, however. This dearth of developmental studies could be due, in part, to Integration of Concepts and Strategies: A Framework for the Scientific Discovery Process The two contrasting approaches outlined represent different conceptualizations about what the development of scientific reasoning involves. In some respects, the different approaches reflect a lack of agreement concerning which type of acquisition (i.e., concepts or strategies) is more important for accounting for developmental differences in scientific reasoning Background The SDDS model was influenced by the work and assumptions of Simon and his colleagues (e.g., Scientific ''reasoning'' is the most common label for the research approaches outlined thus far. A careful examination of what is involved in scientific inquiry should reveal that it involves aspects of both problem solving and reasoning (Copi, 1986, Chap. 14; One of the main generalizations about problem-solving processes is the use of heuristic searches (e.g., The SDDS Model Klahr and With respect to searching hypothesis space, Klahr and Dunbar (1988) noted the difference between ''evoking'' and ''inducing'' a hypothesis. The key difference is that in some situations, one can use prior knowledge in order to constrain the search of hypothesis space, while in other situations, one must make some observations (via experimentation) before constructing an initial hypothesis. The latter scenario relies more heavily on inductive reasoning, while the former may rely on memory retrieval. One implication of this distinction is that the search through experiment space may or may not be constrained by a hypothesis. Initial search through the space of experiments may be done in the service of generating observations. In order to test a hypothesis, once induced, the search process involves finding an experiment that can discriminate among rival hypotheses. The search through these 108 CORINNE ZIMMERMAN two spaces requires different representations of the problem and may require different heuristics for moving about the problem spaces. The first two cognitive processes of scientific discovery involve a coordinated, heuristic search. The third process of the SDDS model involves evidence evaluation. This process was initially described as the decision made on the basis of the cumulative evidence, that is, the decision to accept, reject, or modify the current hypothesis. Initially, Klahr and Dunbar emphasized the ''dual search'' nature of scientific discovery, while the evidence evaluation process was somewhat neglected in the overall discovery process. In more recent descriptions, Klahr has elaborated upon the evidence evaluation process, indicating that it involves a comparison of results obtained through experimentation with the predictions derived from the current hypothesis Klahr and Dunbar's original description of the model highlighted the dual search coordination, but updated descriptions acknowledge that scientific discovery tasks depend upon the coordination and integration of all three components The SDDS framework captures the complexity and the cyclical nature of the process of scientific discovery. The framework incorporates many component processes that previously had been studied in isolation. Summary Scientific discovery is a complex activity that requires the coordination of several higher level cognitive skills, including heuristic search through problem spaces, inductive reasoning, and deductive logic. The main goal of scientific investigation is the acquisition of knowledge in the form of hypotheses that can serve as generalizations or explanations (i.e., theories). Psychologists have investigated the development of scientific concepts and the development of strategies involved in the discovery and verification of hypotheses. In initial studies of scientific thinking, researchers examined these component processes in isolation or in the absence of meaningful content (e.g., the 2-4-6 task). MAJOR EMPIRICAL FINDINGS Thus far I have described only in very general terms the main approaches to studying scientific reasoning and an attempt to integrate the major components of scientific activity into a single framework. In this section I will 110 CORINNE ZIMMERMAN describe the main findings or generalizations that can be made about human performance and development on simulated discovery tasks. 5 Initial attempts to study scientific reasoning began with investigations that followed a ''divide-and-conquer'' approach by focusing on particular cognitive components, as represented by the cells in Research on Experimentation Skills Experimentation is an ill-defined problem for most children and adults DEVELOPMENT OF SCIENTIFIC REASONING 111 teristics common to experimentation across content domains. At a minimum, one must recognize that the process of experimentation involves generating observations that will serve as evidence that will be related to hypotheses. Ideally, experimentation should produce evidence or observations that are interpretable in order to make the process of evidence evaluation uncomplicated. One aspect of experimentation skill is to isolate variables in such a way as to rule out competing hypotheses. An alternative hypothesis can take the form of a specific competing hypothesis or the complement of the hypothesis under consideration. In either case, the control of variables and the systematic combinations of variables are particular skills that have been investigated. Early approaches to examining experimentation skills involved minimizing the role of prior knowledge in order to focus on the strategies that participants used. That is, the goal was to examine the domain-general strategies that apply regardless of the content to which they are applied (i.e., cell E in An analogous task is the colorless liquid task used by Tschirgi (1980) looked at one aspect of hypothesis testing in ''natural'' problem situations. Story problems were used in which two or three variables were involved in producing either a good or a bad outcome (e.g., baking a good cake, making a paper airplane) and therefore involved some domain knowledge (i.e., cells B and E of Tschirgi (1980) found that in familiar, everyday problem situations, the type of outcome influenced the strategy for generating an experiment to produce evidence. In all age groups, participants looked for confirmatory evidence when there was a ''positive'' outcome. That is, for positive outcomes, they used a ''Hold One Thing At a Time'' (HOTAT) strategy for manipulating variables (choice a above). They selected disconfirmatory evidence when there was a ''negative'' outcome, using the more valid ''Vary One Thing At a Time'' (VOTAT) strategy (choice b above). The only developmental difference was that the sixth graders and adults (but not second and fourth graders) were aware of the appropriateness of the VOTAT strategy. Tschirgi suggested that the results supported a model of natural inductive logic that develops through everyday problem-solving experience with multivariable situations. That is, individuals base their choice of strategy on empirical foundations (e.g., reproducing positive effects and eliminating negative effects), not logical ones. Sodian, Zaitchik, and Carey (1991) investigated whether children in the early school years understand the difference between testing a hypothesis and reproducing an effect. The tasks used by Over half of the first graders answered the series of questions correctly (with justifications) and 86% of the second graders correctly differentiated between conclusive and inconclusive tests. It is important to point out, however, that the children were provided with the two mutually exclusive and exhaustive hypotheses and, moreover, were provided with two mutually exclusive and exhaustive experiments from which to select In summary, researchers interested in experimentation skills have focused on the production of factorial combinations and the isolation of variables on tasks in which the role of prior knowledge was minimized. An important precursor for success in producing a combinatorial array in the absence of domain-specific knowledge is systematic or rule-governed behavior, which appears to emerge around age 5. An awareness of memory limitations and of the importance of record keeping appears to emerge between the ages of 10 and 13. With respect to the isolation of variables, there is evidence that the goal of the experiment can affect the strategies selected. When the hypothesis to be tested can be construed as involving a positive or negative outcome, second-, fourth-, and sixth-grade children and adults select valid experimental tests when the outcome is negative, but use a less valid strategy when the outcome is positive. What develops is an awareness of the appropriateness of the VOTAT strategy selected in the case of negative outcomes. When domain knowledge can be used to view the outcome as positive, even adults do not appear to have developed an awareness of the inappropriateness 114 CORINNE ZIMMERMAN of the HOTAT strategy. The research reviewed in this section provides evidence that, under conditions in which producing an effect is not at issue, even children in the first grade understand what it means to test a hypothesis by conducting an experiment and, furthermore, that children as young as 6 can differentiate between a conclusive and an inconclusive experiment. Research on Evidence Evaluation Skills The evaluation of evidence as bearing on the tenability of a hypothesis has been of central interest in the work of Kuhn and her colleagues In the evidence evaluation studies to be reviewed, the evidence provided for participants to evaluate is covariation evidence. In the first section, I will provide a general description of what is meant by covariation evidence and generalizations about tasks used in the studies to be reviewed. Then, early studies of rule use in the evaluation of covariation evidence will be described. In the second section I will summarize the landmark work of Covariation Evidence: General Description and Early Research With respect to determining causal relationships, Hume (1758/1988) identified the covariation of perceptually salient events as one potential cue that DEVELOPMENT OF SCIENTIFIC REASONING 115 two events are causally related. Even young children have a tendency to use the covariation of events (antecedent and outcome) as one indicator of causality (e.g., In covariation, there are four possible combinations of the presence and the absence of antecedent (or potential cause) and outcome (see CORINNE ZIMMERMAN In evidence evaluation tasks involving covariation of events, participants are provided with data corresponding to the frequencies in the cells of a 2 ϫ 2 contingency table (i.e., represented in either tabular or pictorial form). The pattern could represent perfect covariation, partial (or imperfect) covariation, or no correlation between the two events. The task may require participants to evaluate a given hypothesis in light of the evidence or to determine which hypothesis the pattern of data support. In either case, the focus is on the inferences that can be made on the basis of the evidence (i.e., in most cases, participants were instructed to disregard prior domain knowledge). Experimental design skills are not of interest. Early work on covariation detection was conducted by Shaklee and her colleagues (e.g., Shaklee and her colleagues found a general trend in the rules used to weigh the evidence in the contingency tables. It was predicted that younger children (Grades 2, 3, and 4) would use the frequency reported in cell A to make a judgment and proceed developmentally to a rule in which they compare frequencies in cells A vs. B. However, the cell A strategy was not common. The most sophisticated strategy that participants seemed to use, even as adults, was to compare the sums-of-diagonals. The conditional probability rule was used only used by a minority of participants, even at the college level. Adults could readily learn this rule, if they were shown how to compare the relevant ratios (see footnote 6). Children in Grades 4 through 8 could be taught to use the sums-of-diagonals rule The Work of Kuhn, Amsel, and O'Loughlin (1988) Kuhn et al. Kuhn et al. 's (1988) general method involved the presentation of covariation data sequentially and cumulatively. Participants were asked a series of questions about what the evidence shows for each of the four variables. Responses were coded as either evidence-based or theory-based. To be coded as evidence-based, a participant's response to the probe questions had to make reference to the patterns of covariation or instances of data presented (i.e., the findings of the scientists). For example, if shown a pattern in which type of cake covaried with getting colds, a participant who noted that the sick children ate chocolate cake and the healthy kids ate carrot cake would be coded as having made an evidence-based response. In contrast, theorybased responses made reference to the participant's prior beliefs or theories about why the scientists might have found that particular relationship. In the previous example, a response that chocolate cake has ''sugar and a lot of bad stuff in it'' or that ''less sugar means your blood pressure doesn't go up '' (Kuhn, 1989, p. 676) would be coded as theory-based. Kuhn et al. were also interested in both inclusion inferences (an inference that two variables are causally related) and exclusion inferences (an inference of no relationship between variables). Participants' inferences and justification types could be examined for covariation evidence versus noncovariation evidence and in situations where the prior theory was causal or noncausal. Other variations in the other studies included (a) examining the effects of explicit instruction; (b) use of real objects for evidence (e.g., tennis balls with various features) versus pictorial representations of data; (c) task instructions to relate the evidence to multiple theories instead of a single theory; and (d) a reciprocal version of the task in which the participant generates the pattern of evidence that would support and refute a theory. Through the series of studies, Kuhn et al. found certain patterns of responding. First, the skills involved in differentiating and coordinating theory 118 CORINNE ZIMMERMAN and evidence, and bracketing prior belief while evaluating evidence, show a monotonic developmental trend from middle childhood (Grades 3 and 6) to adolescence (Grade 9) to adulthood. These skills, however, do not develop to an optimum level even among adults. Even adults have a tendency to meld theory and evidence into a single representation of ''the way things are.'' Second, participants have a variety of strategies for keeping theory and evidence in alignment with one another when they are in fact discrepant. One tendency is to ignore, distort, or selectively attend to evidence that is inconsistent with a favored theory. For example, the protocol from one ninth grader demonstrated that upon repeated instances of covariation between type of breakfast roll and catching colds, he would not acknowledge this relationship: ''They just taste different . . . the breakfast roll to me don't cause so much colds because they have pretty much the same thing inside [i.e., dough]' ' (Kuhn et al., p. 73, elaboration added). A second tendency was to adjust a theory to fit the evidence. This practice seems perfectly reasonable, but the bothersome part is that this ''strategy'' was often outside a participant's conscious control. Participants were often unaware of the fact that they were modifying a theory. When asked to recall their original beliefs, participants would often report a theory consistent with the evidence that was presented and not the theory as originally stated. An example of this is one ninth grader who did not believe type of condiment (mustard versus ketchup) was causally related to catching colds. With each presentation of an instance of covariation evidence, he acknowledged the evidence and elaborated a theory based on the amounts of ingredients or vitamins and the temperature of the food the condiment was served with (Kuhn et al., p. 83). Kuhn argued that this tendency suggests that the subject's theory does not exist as an object of cognition. That is, a theory and the evidence for that theory are undifferentiated-they do not exist as separate entities. Third, there were a variety of errors involved in understanding covariation evidence and its connection to causality. There were also problems in understanding noncovariation. For example, when asked to generate a pattern of evidence that would show that a factor makes no difference in an outcome, participants often produced covariation evidence in the opposite direction of that predicted by their own causal theory. Criticisms of Ruffman, Perner, Olson, and Doherty (1993) examined children's abilities (aged 4 to 7) to form hypotheses on the basis of covariation evidence. They also used less complex tasks with fewer factors to consider than In order to rule out the possibility that children were simply describing a state of affairs, Ruffman et al. tested if 4-to 7-year-olds understood the predictive properties of the hypothesis formed on the basis of covariation evidence. Children were asked to evaluate evidence and then form a hypothesis about which characteristics of tennis rackets were responsible for better serves (e.g., racket size, head shape). They were then asked which tennis racket they would buy and how good the next serve would be. The results were consistent with the idea that by age 7, children understood that the newly formed hypothesis could be used to make predictions. Ruffman et al. deliberately chose factors that were all equally plausible. Correct performance in the Kuhn et al. tasks was defined by considering covariation evidence as more important than the implausible hypothesis it was intended to support. For example, in Studies 3 and 4 of Kuhn et al., adults and third, sixth, and ninth graders were to evaluate evidence to determine the features of tennis balls that resulted in good or poor serves (i.e., color, texture, ridges, and size). Most children and adults do not believe that color is causally related to the quality of a tennis serve. Ruffman et al. argued that revising prior beliefs (e.g., about the causal power of color) is more difficult than forming new theories when prior beliefs do not exist or are not held with conviction. Literature on inductive inference supports this claim (e.g., Unlike Ruffman et al.'s (1993) criticism about strong prior beliefs, the participants in Amsel and Brock's study were selected only if they did hold strong prior beliefs concerning the variables. That is, participants believed that a relationship exists between the health of plants and the presence/absence of sunshine and that no relationship exists between health of plants and the presence/absence of a charm (represented as a four-leaf clover). Children in second/third grade, sixth/seventh grade, college students, and noncollege adults were presented with four data sets to evaluate given by the factorial combination of prior belief (causal or noncausal) by type of contingency data (perfect positive correlation vs. zero correlation). Participants were asked whether giving the plants (sun/no sun) or (charm/no charm) was causally related to whether the plants were healthy or sick and to respond only based on the information given and not what they know about plants. Standard covariation evidence served as the control (four instances in a 2 ϫ 2 contingency table), while three conditions involved ''missing data.'' Participants in the control group were presented with four data instances that represented covariation (or noncovariation) between the putative causal factor and the outcome. Participants in the three missing data conditions were shown two additional instances in which either the antecedent was unknown, the outcome was unknown, or both were unknown. Amsel and Brock reasoned that if participants were evaluating the evidence independent of their strongly held prior beliefs, then the judgments in the control and missing data conditions should be the same. That is, participants would simply ignore the evidentially irrelevant missing data. If they were using prior beliefs, however, they might try to explain the missing data by judging the variables as consistent with their prior beliefs. If they were using newly formed beliefs, then judgments would be consistent with the new belief and pattern of evidence (causal with covariation evidence; noncausal with noncovariation). College adults were most like the ''ideal reasoner'' (i.e., defined as someone whose causal certainty scores were based solely on the four instances of contingency data). Both groups of children (second/third grade and sixth/ seventh grade) judged the causal status of variables consistent with their prior beliefs even when the evidence was disconfirming. The noncollege adults' judgments tended to be in between, leading the authors to suggest that there are differences associated with age and education in making causal judgments independently of prior beliefs. In the missing data conditions, participants did not read into or try to explain the missing data. Rather, the effect was to cause the children of both grade levels, but not the adults, to be less certain about the causal status of the variables on the outcome. There DEVELOPMENT OF SCIENTIFIC REASONING 121 was an age and education trend for the frequency of evidence-based justifications. When presented with evidence that disconfirmed prior beliefs, children from both grade levels tended to make causal judgments consistent with their prior beliefs. When confronted with confirming evidence, however, both groups of children and adults made similar judgments. In this section I have outlined the research that was conducted based on criticisms of Kuhn et al.'s (1988) methodology, including issues about task complexity, plausibility of factors, participants' method of responding (e.g., certainty judgments versus forced choice), and data coding (e.g., causal judgments and justifications assessed jointly or separately). In the next section, I discuss Koslowski's (1996) criticisms of Kuhn et al. in particular and of studies employing covariation evidence in general. The role of causal mechanism. Koslowski questioned the assumptions about the primacy of covariation evidence. One of the main concerns in scientific research is with the discovery of causes. Likewise, in much of the research on scientific reasoning, tasks were employed in which participants reason about causal relationships. Psychologists who study scientific reasoning have been influenced by the philosophy of science, most notably the empiricist tradition which emphasizes the importance of observable events. Hume's strategy of identifying causes by determining the events that covary with an outcome has been very influential. In real scientific practice though, scientists are also concerned with causal mechanism, or the process by which a cause can bring about an effect. Koslowski noted that we live in a world full of correlations. It is through a consideration of causal mechanism that we can determine which correlations between perceptually salient events 122 CORINNE ZIMMERMAN should be taken seriously and which should be viewed as spurious. For example, it is through the identification of the Escherichia coli bacterium that we consider a causal relationship between hamburger consumption and illness or mortality. It is through the absence of a causal mechanism that we do not consider seriously the classic pedagogical example of a correlation between ice cream consumption and violent crime rate. In the studies by In subsequent studies, participants were given problem situations in which a story character is trying to determine if some target factor (e.g., a gasoline additive) is causally related to an effect (e.g., improved gas mileage). They 7 We also use this pedagogical example to illustrate the importance of considering additional variables that may be responsible for both outcomes (i.e., high temperatures for this example). DEVELOPMENT OF SCIENTIFIC REASONING 123 were then shown either perfect covariation between target factor and effect or partial covariation (four of six instances). Perfect correlation was rated as more likely to indicate causation than partial correlation. Participants were then told that a number of plausible mechanisms had been ruled out (e.g., the additive does not burn more efficiently, the additive does not burn more cleanly). When asked to rate again how likely it was that the additive is causally responsible for improved gas mileage, the ratings for both perfect and partial covariation were lower for all age groups. Koslowski also tried to determine if participants would spontaneously generate information about causal mechanisms when it was not cued by the task (Experiment 16). Participants (sixth grade, ninth grade, adults) were presented with story problems in which a character is trying to answer a question about, for example, whether parents staying in the hospital improves the recovery rate of their children. Participants were asked to describe whatever type of information might be useful for solving the problem. Half of the participants were told that experimental intervention was not possible, while the other half were not restricted in this manner. Almost all participants showed some concern for causal mechanism, including expectations about how the target mechanism would operate. Although the sixth graders were less likely to generate a variety of alternative hypotheses, all age groups proposed appropriate contrastive tests. In summary, Koslowski argues that sound scientific reasoning requires ''bootstrapping,'' that is, using covariation information and mechanism information interdependently. Scientists, she argues, rely on theory or mechanism to decide which of the many covariations in the world are likely to be causal (or merit further study). To demonstrate that people are reasoning in a scientifically legitimate way, one needs to establish that they rely on both covariation and mechanism information and they do so in a way that is judicious. As shown in the previous studies, participants did treat a covarying factor as causal when there was a possible mechanism that could account for how the factor might have brought about the effect and were less likely to do so when mechanism information was absent. Moreover, participants at all age levels showed a concern for causal mechanism even when it was not cued by the task. Considerations of plausibility. In another study (Experiment 5), participants were asked to rate the likelihood of a possible mechanism to explain covariations that were either plausible or implausible. Participants were also asked to generate their own mechanisms to explain plausible and implausible covariations. When either generating or assessing mechanisms for plausible covariations, all age groups (sixth graders, ninth graders, adults) were comparable. When the covariation was implausible, sixth graders were more likely to generate dubious mechanisms to account for the correlation. In some situations, scientific progress occurs by taking seemingly implausible correlations seriously

As researchers studying student reasoning in introductory physics, and as instructors teaching courses, we often focus on whether and how students apply what they know in one context to their reasoning in another. But we do not speak in terms of “transfer.” The term connotes to us a unitary view of knowledge as a thing that is acquired in one context and carried (or not) to another. We speak, rather, in terms of activating resources, a language with an explicitly manifold view of cognitive structure. In this chapter, we describe this view and argue that it provides a more firm and generative basis for research. In particular, our resources-based perspective accounts for why it is difficult, and perhaps unnecessary, to draw a boundary around the notion of “transfer”; provides an analytical framework for exploring the differences between active transfer involving metacognition and passive transfer that “just happens”; helps to explain many results in the transfer literature, such as the rarity of certain kinds of transfer and the ubiquity of others; and provides an ontological underpinning for new views of transfer such as Bransford, Schwartz, and Sears ’ (this issue) “preparation for future learning.”

Despite the accumulation of substantial cognitive science research relevant to education, there remains confusion and controversy in the application of research to educational practice. In support of a more systematic approach, we describe the Knowledge-Learning-Instruction (KLI) framework. KLI promotes the emergence of instructional principles of high potential for generality, while explicitly identifying constraints of and opportunities for detailed analysis of the knowledge students may acquire in courses. Drawing on research across domains of science, math, and language learning, we illustrate the analyses of knowledge, learning, and instructional events that the KLI framework affords. We present a set of three coordinated taxonomies of knowledge, learning, and instruction. For example, we identify three broad classes of learning events (LEs): (a) memory and fluency processes, (b) induction and refinement processes, and (c) understanding and sense-making processes, and we show

Conceptual change researchers have made significant progress on two prominent but competing theoretical perspectives regarding knowledge structure coherence. These perspectives can be broadly characterized as (1) knowledge-as-theory perspectives and (2) knowledge-as-elements perspectives. These perspectives can be briefly summarized in terms of the following questions. Is a student’s knowledge most accurately represented as a coherent unified framework of theory-like character (e.g., Carey, 1999; Chi, 2005; Ioannides & Vosniadou, 2002; Wellman & Gelman, 1992)? Or is a student’s knowledge more aptly considered as an ecology of quasi-independent elements (e.g., Clark, 2006;

Abstract not found

To evaluate how top-down and bottom-up processes contribute to learning from animated displays, we conducted four experiments that varied either in the design of animations or the prior knowledge of the learners. Experiments 1–3 examined whether adding interactivity and signaling to an animation benefits learners in developing a mental model of a mechanical system. Although learners utilized interactive controls and signaling devices, their comprehension of the system was no better than that of learners who saw animations without these design features. Furthermore, the majority of participants developed a mental model of the system that was incorrect and inconsistent with information displayed in the animation. Experiment 4 tested effects of domain knowledge and found, surprisingly, that even some learners with high domain knowledge initially constructed the incorrect mental model. After multiple exposures to the materials, the high knowledge learners revised their mental models to the correct one, while the low-knowledge learners maintained their erroneous models. These results suggest that learning from animations involves a complex interplay between top-down and bottom-up processes and that more emphasis should be placed on how prior knowledge is applied to interpreting animations.

short standard long