*2.1.2. Strategy development*

Strategies are sequences of procedural actions used to achieve a goal (Siegler, 1996). In the context of scientific reasoning, strategies are the steps that guide children from their initial state (e.g., a question about the effects of weight and distance in balancing a scale) to a goal state (e.g., understanding the nature of the relationship between variables). We will briefly examine two components of strategy development: *strategy acquisition* and *strategy selection*. Strategies are particularly important in the development of scientific reasoning. Children often actively explore objects in a manner that is like hypothesis testing; however, these exploration strategies are not systematic investigations in which variables are manipulated and controlled as in formal hypothesis-testing strategies (Klahr, 2000). The acquisition of increasingly optimal strategies for hypothesis testing, inference, and evidence evaluation leads to more effective scientific reasoning that allows children to construct more veridical knowledge.

New strategies are added to the repertoire of possible strategies through discovery, instruction, or other social interactions (Chen, 2007; Gauvain, 2001; Siegler, 1996). There is evidence that children can discover strategies on their own (Chen, 2007). Children often discover new strategies when they experience an insight into a new way of solving a familiar problem. For example, 10- and 11-year-olds discovered new strategies for evaluating causal relations between variables in a computerized task only after creating different cars (e.g., comparing the effects of engine size) and testing them (Schauble, 1990). Similarly, when asked to determine the cause of a chemical reaction, children discovered new experimentation strategies only after several weeks (Kuhn & Phelps, 1982). Over time, existing strategies may be modified to reduce time and complexity of implementation (e.g., eliminating redundant steps in a problem solving sequence; Klahr, 1984). For example, determining causal relations among variables requires more time when experimentation is unsystematic. In order to identify which variables resulted in the fastest car, children often constructed up to 25 cars, whereas an adult scientist identified the fastest car after constructing only seven cars (Schauble, 1990).

Children also gain new strategies through social interaction, by being explicitly taught a strategy, imitating a strategy, or by collaborating in problem solving (Gauvain, 2001). For example, when a parent asks a child questions about events in a photograph, the parent evokes memories of the event and helps to structure the child's understanding of the depicted event, a process called conversational remembering (Middleton, 1997). Conversational remembering improves children's recall of events and often leads to children spontaneously using this strategy. Parent conversations about event structures improved children's memory for these structures; for example, questions about a child's day at school help to structure this event and improved recall (Nelson, 1996). Children also learn new strategies by solving problems cooperatively with adults. In a sorting task, preschool children were more likely to improve their classification strategies after working with their mothers (Freund, 1990). Further, children who worked with their parents on a hypothesistesting task were more likely to identify causal variables than children who worked alone because parents helped children construct valid experiments, keep data records, and repeat experiments (Gleason & Schauble, 2000).

64 Current Topics in Children's Learning and Cognition

*2.1.2. Strategy development* 

constructing only seven cars (Schauble, 1990).

greater attentional resources than passive encoding (Chi, 2009).

features (Kloos & Van Orden, 2005). Although domain knowledge is helpful in directing attention to critical features, it may sometimes limit novel reasoning in a domain and limit the extent to which attention is paid to disconfirming evidence (Li & Klahr, 2006). Finally, *self-generated activity* improves encoding. Self-generation of information from memory, rather than passive attention, is associated with more effective encoding because it recruits

Strategies are sequences of procedural actions used to achieve a goal (Siegler, 1996). In the context of scientific reasoning, strategies are the steps that guide children from their initial state (e.g., a question about the effects of weight and distance in balancing a scale) to a goal state (e.g., understanding the nature of the relationship between variables). We will briefly examine two components of strategy development: *strategy acquisition* and *strategy selection*. Strategies are particularly important in the development of scientific reasoning. Children often actively explore objects in a manner that is like hypothesis testing; however, these exploration strategies are not systematic investigations in which variables are manipulated and controlled as in formal hypothesis-testing strategies (Klahr, 2000). The acquisition of increasingly optimal strategies for hypothesis testing, inference, and evidence evaluation leads to more effective

New strategies are added to the repertoire of possible strategies through discovery, instruction, or other social interactions (Chen, 2007; Gauvain, 2001; Siegler, 1996). There is evidence that children can discover strategies on their own (Chen, 2007). Children often discover new strategies when they experience an insight into a new way of solving a familiar problem. For example, 10- and 11-year-olds discovered new strategies for evaluating causal relations between variables in a computerized task only after creating different cars (e.g., comparing the effects of engine size) and testing them (Schauble, 1990). Similarly, when asked to determine the cause of a chemical reaction, children discovered new experimentation strategies only after several weeks (Kuhn & Phelps, 1982). Over time, existing strategies may be modified to reduce time and complexity of implementation (e.g., eliminating redundant steps in a problem solving sequence; Klahr, 1984). For example, determining causal relations among variables requires more time when experimentation is unsystematic. In order to identify which variables resulted in the fastest car, children often constructed up to 25 cars, whereas an adult scientist identified the fastest car after

Children also gain new strategies through social interaction, by being explicitly taught a strategy, imitating a strategy, or by collaborating in problem solving (Gauvain, 2001). For example, when a parent asks a child questions about events in a photograph, the parent evokes memories of the event and helps to structure the child's understanding of the depicted event, a process called conversational remembering (Middleton, 1997). Conversational remembering improves children's recall of events and often leads to children spontaneously using this strategy. Parent conversations about event structures

scientific reasoning that allows children to construct more veridical knowledge.

Children also acquire strategies by interacting with an adult modeling a novel strategy. Middle-school children acquired a reading comprehension strategy (e.g., anticipating the ending of a story) after seeing it modeled by their teacher (Palinscar, Brown, & Campione, 1993). Additionally, children can acquire new strategies from interactions with other children. Monitoring other children during problem solving improves a child's understanding of the task and appears to improve how they evaluate their own performance (Brownell & Carriger, 1991). Elementary school children who collaborated with other students to solve the balance-scale task outperformed students who worked alone (Pine & Messer, 1998). Ten-year-olds working in dyads were more likely to discuss their strategies than children working alone and these discussions were associated with generating better hypotheses than children working alone (Teasley, 1995).

More than one strategy may be useful for solving a problem, which requires a means to select among candidate strategies. One suggestion is that this process occurs by adaptive selection. In adaptive selection, strategies that match features of the problem are candidates for selection. One component of selection is that newer strategies tend to have a slightly higher priority for use when compared to older strategies (Siegler, 1996). Successful selection is made on the basis of the effectiveness of the strategy and its cost (e.g., speed), and children tend to choose the fastest, most accurate strategy available (i.e., the most adaptive strategy).

Cognitive mechanisms provide the basic investigation and inferential tools used in scientific reasoning. The ability to reason about knowledge and the means for obtaining and evaluating knowledge provide powerful tools that augment children's reasoning. *Metacognitive abilities* such as these may help explain some of the discrepancies between early scientific reasoning abilities and limitations in older children, as well as some of the developmental changes in encoding and strategy use.

#### **2.2. Metacognitive and metastrategic processes**

Sodian, Zaitchik, and Carey (1991) argue that two basic skills related to early metacognitive acquisitions are needed for scientific reasoning. First, children need to understand that inferences can be drawn from evidence. The theory of mind literature (e.g., Wellman, Cross, & Watson, 2001) suggests that it is not until the age of 4 that children understand that beliefs and knowledge are based on perceptual experience (i.e., evidence). As noted earlier, experimental work demonstrates that preschoolers can use evidence to make judgments

about simple causal relationships (Gopnik, Sobel, Schulz, & Glymour, 2001; Schulz & Bonawitz, 2007; Schulz & Gopnik, 2004; Schulz, Gopnik,& Glymour, 2007). Similarly, several classic studies show that children as young as 6 can succeed in simple scientific reasoning tasks. Children between 6 and 9 can discriminate between a conclusive and an inclusive test of a simple hypothesis (Sodian et al., 1991). Children as young as 5 can form a causal hypothesis based on a pattern of evidence, and even 4-year-olds seem to understand some of the principles of causal reasoning (Ruffman, Perner, Olson, & Doherty, 1993).

Emergence of Scientific Reasoning 67

of evidence evaluation and inference strategies. With respect to encoding, increases in task complexity require attending to more information and making judgments about which features are relevant. This encoding happens in the context of prior knowledge and, in many cases, it is also necessary to inhibit prior knowledge (Zimmerman & Croker, in press).

Sodian and Bullock (2008) also argue that mature scientific reasoning involves the metastrategic process of being able to think explicitly about hypotheses and evidence, and that this skill is not fully mastered until adolescence at the very earliest. According to Amsel et al. (2008), metacognitive competence is important for hypothetical reasoning. These conclusions are consistent with Kuhn's (1989, 2005, 2011a) argument that the defining feature of scientific thinking is the set of cognitive and metacognitive skills involved in differentiating and coordinating theory and evidence. Kuhn argues that the effective coordination of theory and evidence depends on three metacognitive abilities: (a) The ability to encode and represent evidence and theory separately, so that relations between them can be recognized; (b) the ability to treat theories as independent objects of thought (i.e., rather than a representation of "the way things are"); and (c) the ability to recognize that theories can be false, setting aside the acceptance of a theory so evidence can be assessed to determine the veridicality of a theory. When we consider these cognitive and metacognitive abilities in the larger social context, it is clear that skills that are highly valued by the scientific community may be at odds with the cultural and intuitive views of the individual reasoner (Lemke, 2001). Thus, it often takes time for conceptual change to occur; evidence is not just evaluated in the context of the science investigation and science classroom, but within personal and community values. Conceptual change also takes place in the context of an individual's personal epistemology, which can undergo developmental transitions (e.g.,

Returning to the encoding and retrieval of information relevant to scientific reasoning tasks, many studies demonstrate that both children and adults are not always aware of their memory limitations while engaged in investigation tasks (e.g., Carey, Evans, Honda, Jay, & Unger, 1989; Dunbar & Klahr, 1989; Garcia-Mila & Andersen, 2007; Gleason & Schauble, 2000; Siegler & Liebert, 1975; Trafton & Trickett, 2001). Kanari and Millar (2004) found that children differentially recorded the results of experiments, depending on familiarity or strength of prior beliefs. For example, 10- to 14-year-olds recorded more data points when experimenting with unfamiliar items (e.g., using a force-meter to determine the factors affecting the force produced by the weight and surface area of boxes) than with familiar items (e.g., using a stopwatch to experiment with pendulums). Overall, children are less likely than adults to record experimental designs and outcomes, or to review notes they do keep, despite task demands that clearly necessitate a reliance on external memory aids.

Children are often asked to judge their memory abilities, and memory plays an important role in scientific reasoning. Children's understanding of memory as a fallible process

Sandoval, 2005).

*2.2.1. Encoding and strategy use* 

Second, according to Sodian et al. (1991), children need to understand that inference is itself a mechanism with which further knowledge can be acquired. Four-year-olds base their knowledge on perceptual experiences, whereas 6-year-olds understand that the testimony of others can also be used in making inferences (Sodian & Wimmer, 1987). Other research suggests that children younger than 6 can make inferences based on testimony, but in very limited circumstances (Koenig, Clément, & Harris, 2004). These findings may explain why, by the age of 6, children are able to succeed on *simple* causal reasoning, hypothesis testing, and evidence evaluation tasks.

Research with older children, however, has revealed that 8- to 12-year-olds have limitations in their abilities to (a) generate unconfounded experiments, (b) disconfirm hypotheses, (c) keep accurate and systematic records, and (d) evaluate evidence (Klahr, Fay, & Dunbar, 1993; Kuhn, Garcia-Mila, Zohar, & Andersen, 1995; Schauble, 1990, 1996; Zimmerman, Raghavan, & Sartoris, 2003). For example, Schauble (1990) presented children aged 9-11 with a computerized task in which they had to determine which of five factors affect the speed of racing cars. Children often varied several factors at once (only 22% of the experiments were classified as valid) and they often drew conclusions consistent with belief rather than the evidence generated. They used a positive test strategy, testing variables believed to influence speed (e.g., engine size) and not testing those believed to be non-causal (e.g., color). Some children recorded features without outcomes, or outcomes without features, but most wrote down nothing at all, relying on memory for details of experiments carried out over an eight-week period.

Although the performance differences between younger and older children may be interpreted as potentially contradictory, the differing cognitive and metacognitive demands of tasks used to study scientific reasoning at different ages may account for some of the disconnect in conclusions. Even though the simple tasks given to preschoolers and young children require them to understand evidence as a source of knowledge, such tasks require the cognitive abilities of induction and pattern recognition, but only limited metacognitive abilities. In contrast, the tasks used to study the development of scientific reasoning in older children (and adults) are more demanding and focused on hypothetico-deductive reasoning; they include more variables, involve more complex causal structures, require varying levels of domain knowledge, and are negotiated across much longer time scales. Moreover, the tasks given to older children and adults involve the acquisition, selection, and coordination of investigation strategies, combining background knowledge with empirical evidence. The results of investigation activities are then used in the acquisition, selection, and coordination of evidence evaluation and inference strategies. With respect to encoding, increases in task complexity require attending to more information and making judgments about which features are relevant. This encoding happens in the context of prior knowledge and, in many cases, it is also necessary to inhibit prior knowledge (Zimmerman & Croker, in press).

Sodian and Bullock (2008) also argue that mature scientific reasoning involves the metastrategic process of being able to think explicitly about hypotheses and evidence, and that this skill is not fully mastered until adolescence at the very earliest. According to Amsel et al. (2008), metacognitive competence is important for hypothetical reasoning. These conclusions are consistent with Kuhn's (1989, 2005, 2011a) argument that the defining feature of scientific thinking is the set of cognitive and metacognitive skills involved in differentiating and coordinating theory and evidence. Kuhn argues that the effective coordination of theory and evidence depends on three metacognitive abilities: (a) The ability to encode and represent evidence and theory separately, so that relations between them can be recognized; (b) the ability to treat theories as independent objects of thought (i.e., rather than a representation of "the way things are"); and (c) the ability to recognize that theories can be false, setting aside the acceptance of a theory so evidence can be assessed to determine the veridicality of a theory. When we consider these cognitive and metacognitive abilities in the larger social context, it is clear that skills that are highly valued by the scientific community may be at odds with the cultural and intuitive views of the individual reasoner (Lemke, 2001). Thus, it often takes time for conceptual change to occur; evidence is not just evaluated in the context of the science investigation and science classroom, but within personal and community values. Conceptual change also takes place in the context of an individual's personal epistemology, which can undergo developmental transitions (e.g., Sandoval, 2005).

#### *2.2.1. Encoding and strategy use*

66 Current Topics in Children's Learning and Cognition

and evidence evaluation tasks.

out over an eight-week period.

about simple causal relationships (Gopnik, Sobel, Schulz, & Glymour, 2001; Schulz & Bonawitz, 2007; Schulz & Gopnik, 2004; Schulz, Gopnik,& Glymour, 2007). Similarly, several classic studies show that children as young as 6 can succeed in simple scientific reasoning tasks. Children between 6 and 9 can discriminate between a conclusive and an inclusive test of a simple hypothesis (Sodian et al., 1991). Children as young as 5 can form a causal hypothesis based on a pattern of evidence, and even 4-year-olds seem to understand some

Second, according to Sodian et al. (1991), children need to understand that inference is itself a mechanism with which further knowledge can be acquired. Four-year-olds base their knowledge on perceptual experiences, whereas 6-year-olds understand that the testimony of others can also be used in making inferences (Sodian & Wimmer, 1987). Other research suggests that children younger than 6 can make inferences based on testimony, but in very limited circumstances (Koenig, Clément, & Harris, 2004). These findings may explain why, by the age of 6, children are able to succeed on *simple* causal reasoning, hypothesis testing,

Research with older children, however, has revealed that 8- to 12-year-olds have limitations in their abilities to (a) generate unconfounded experiments, (b) disconfirm hypotheses, (c) keep accurate and systematic records, and (d) evaluate evidence (Klahr, Fay, & Dunbar, 1993; Kuhn, Garcia-Mila, Zohar, & Andersen, 1995; Schauble, 1990, 1996; Zimmerman, Raghavan, & Sartoris, 2003). For example, Schauble (1990) presented children aged 9-11 with a computerized task in which they had to determine which of five factors affect the speed of racing cars. Children often varied several factors at once (only 22% of the experiments were classified as valid) and they often drew conclusions consistent with belief rather than the evidence generated. They used a positive test strategy, testing variables believed to influence speed (e.g., engine size) and not testing those believed to be non-causal (e.g., color). Some children recorded features without outcomes, or outcomes without features, but most wrote down nothing at all, relying on memory for details of experiments carried

Although the performance differences between younger and older children may be interpreted as potentially contradictory, the differing cognitive and metacognitive demands of tasks used to study scientific reasoning at different ages may account for some of the disconnect in conclusions. Even though the simple tasks given to preschoolers and young children require them to understand evidence as a source of knowledge, such tasks require the cognitive abilities of induction and pattern recognition, but only limited metacognitive abilities. In contrast, the tasks used to study the development of scientific reasoning in older children (and adults) are more demanding and focused on hypothetico-deductive reasoning; they include more variables, involve more complex causal structures, require varying levels of domain knowledge, and are negotiated across much longer time scales. Moreover, the tasks given to older children and adults involve the acquisition, selection, and coordination of investigation strategies, combining background knowledge with empirical evidence. The results of investigation activities are then used in the acquisition, selection, and coordination

of the principles of causal reasoning (Ruffman, Perner, Olson, & Doherty, 1993).

Returning to the encoding and retrieval of information relevant to scientific reasoning tasks, many studies demonstrate that both children and adults are not always aware of their memory limitations while engaged in investigation tasks (e.g., Carey, Evans, Honda, Jay, & Unger, 1989; Dunbar & Klahr, 1989; Garcia-Mila & Andersen, 2007; Gleason & Schauble, 2000; Siegler & Liebert, 1975; Trafton & Trickett, 2001). Kanari and Millar (2004) found that children differentially recorded the results of experiments, depending on familiarity or strength of prior beliefs. For example, 10- to 14-year-olds recorded more data points when experimenting with unfamiliar items (e.g., using a force-meter to determine the factors affecting the force produced by the weight and surface area of boxes) than with familiar items (e.g., using a stopwatch to experiment with pendulums). Overall, children are less likely than adults to record experimental designs and outcomes, or to review notes they do keep, despite task demands that clearly necessitate a reliance on external memory aids.

Children are often asked to judge their memory abilities, and memory plays an important role in scientific reasoning. Children's understanding of memory as a fallible process

develops over middle childhood (Jaswal & Dodson, 2009; Kreuzer, Leonard, & Flavell, 1975). Young children view all strategies on memory tasks as equally effective, whereas 8- to 10-year-olds start to discriminate between strategies, and 12-year-olds know which strategies work best (Justice, 1986; Schneider, 1986). The development of metamemory continues through adolescence (Schneider, 2008), so there may not be a particular age that memory and metamemory limitations are no longer a consideration for children and adolescents engaged in complex scientific reasoning tasks. However, it seems likely that metamemory limitations are more profound for children under 10-12 years.

Emergence of Scientific Reasoning 69

Metacognitive abilities are necessary precursors to sophisticated scientific thinking, and represent one of the ways in which children, adults, and professional scientists differ. In order for children's behavior to go beyond demonstrating the correctness of one's existing beliefs (e.g., Dunbar & Klahr, 1989) it is necessary for meta-level competencies to be developed and practiced (Kuhn, 2005). With metacognitive control over the processes involved, children (and adults) can change what they believe based on evidence and, in doing so, are aware not only that they are changing a belief, but also know *why* they are changing a belief. Thus, sophisticated reasoning involves both the use of various strategies involved in hypothesis testing, induction, inference, and evidence evaluation, and a meta-

Much of the existing laboratory work on the development of scientific thinking has not *overtly* acknowledged the role of contextual factors. Although internal cognitive and metacognitive processes have been a primary focus of past work, and have helped us learn tremendously about the processes of scientific thinking, we argue that many of these studies focused on individual cognition have, in fact, included both social factors (in the form of, for example, collaborations with other students, or scaffolds by parents or teachers) and cultural

**3.1. Instructional and peer support: The role of others in supporting cognitive** 

control of them are facilitated by both the social and physical environment.

Our goal in this section is to re-examine our two focal mechanisms (i.e., encoding and strategy) and show how the development of these cognitive acquisitions and metastrategic

Children must learn to encode effectively, by knowing what information is critical to pay attention to. They do so in part with the aid of their teachers, parents, and peers. Once school begins, teachers play a clear role in children's cognitive development. An ongoing debate in the field of science education concerns the relative value of having children learn and discover how the world works on their own (often called "discovery learning") and having an instructor guide the learning more directly (often called "direct instruction"). Different researchers interpret these labels in divergent ways, which adds fuel to the debate (see e.g., Bonawitz et al., 2011; Hmelo-Silver, Duncan, & Chinn, 2007; Kirshner, Sweller, & Clark, 2006; Klahr, 2010; Mayer, 2004; Schmidt, Loyens, van Gog, & Paas, 2007). Regardless of definitions, though, this issue illustrates the core idea that learning takes place in a social context, with guidance that varies from minimal to

level awareness of when, how, and why one should engage in these strategies.

**3. Scientific reasoning in context** 

tools that support scientific reasoning.

**development** 

*3.1.1. Encoding* 

didactic.

Likewise, the acquisition of other metacognitive and metastrategic skills is a gradual process. Early strategies for coordinating theory and evidence are replaced with better ones, but there is not a stage-like change from using an older strategy to a newer one. Multiple strategies are concurrently available so the process of change is very much like Siegler's (1996) overlapping waves model (Kuhn et al., 1995). However, *metastrategic competence* does not appear to routinely develop in the absence of instruction. Kuhn and her colleagues have incorporated the use of specific practice opportunities and prompts to help children develop these types of competencies. For example, Kuhn, Black, Keselman, and Kaplan (2000) incorporated performance-level practice and metastrategic-level practice for sixth- to eighthgrade students. Performance-level exercise consisted of standard exploration of the task environment, whereas metalevel practice consisted of scenarios in which two individuals disagreed about the effect of a particular feature in a multivariable situation. Students then evaluated different strategies that could be used to resolve the disagreement. Such scenarios were provided twice a week during the course of ten weeks. Although no performance differences were found between the two types of practice with respect to the number of valid inferences, there were more sizeable differences in measures of understanding of task objectives and strategies (i.e., metastrategic understanding).

Similarly, Zohar and Peled (2008) focused instruction in the control-of-variables strategy (CVS) on metastrategic competence. Fifth-graders were given a computerized task in which they had to determine the effects of five variables on seed germination. Students in the control group were taught about seed germination, and students in the experimental group were given a metastrategic knowledge intervention over several sessions. The intervention consisted of describing CVS, discussing when it should be used, and discussing what features of a task indicate that CVS should be used. A second computerized task on potato growth was used to assess near transfer. A physical task in which participants had to determine which factors affect the distance a ball will roll was used to assess far transfer. The experimental group showed gains on both the strategic and the metastrategic level. The latter was measured by asking participants to explain what they had done. These gains were still apparent on the near and far transfer tasks when they were administered three months later. Moreover, low-academic achievers showed the largest gains. It is clear from these studies that although meta-level competencies may not develop routinely, they can certainly be learned via explicit instruction.

Metacognitive abilities are necessary precursors to sophisticated scientific thinking, and represent one of the ways in which children, adults, and professional scientists differ. In order for children's behavior to go beyond demonstrating the correctness of one's existing beliefs (e.g., Dunbar & Klahr, 1989) it is necessary for meta-level competencies to be developed and practiced (Kuhn, 2005). With metacognitive control over the processes involved, children (and adults) can change what they believe based on evidence and, in doing so, are aware not only that they are changing a belief, but also know *why* they are changing a belief. Thus, sophisticated reasoning involves both the use of various strategies involved in hypothesis testing, induction, inference, and evidence evaluation, and a metalevel awareness of when, how, and why one should engage in these strategies.
