Effects of Direct Instruction and Discovery Learning

In a study with 112 thirdand fourth-grade children, we measured the relative effectiveness of discovery learning and direct instruction at two points in the learning process: (a) during the initial acquisition of the basic cognitive objective (a procedure for designing and interpreting simple, unconfounded experiments) and (b) during the subsequent transfer and application of this basic skill to more diffuse and authentic reasoning associated with the evaluation of sciencefair posters. We found not only that many more children learned from direct instruction than from discovery learning, but also that when asked to make broader, richer scientific judgments, the many children who learned about experimental design from direct instruction performed as well as those few children who discovered the method on their own. These results challenge predictions derived from the presumed superiority of discovery approaches in teaching young children basic procedures for early scientific investigations. A widely accepted claim in the scienceand mathematics-education community is the constructivist idea that discovery learning, as opposed to direct instruction, is the best way to get deep and lasting understanding of scientific phenomena and procedures, particularly for young children. ‘‘The premise of constructivism implies that the knowledge students construct on their own, for example, is more valuable than the knowledge modeled for them; told to them; or shown, demonstrated, or explained to them by a teacher’’ (Loveless, 1998, p. 285). Advocates of discovery learning concur with Piaget’s assertion that ‘‘each time one prematurely teaches a child something he could have discovered for himself, that child is kept from inventing it and consequently from understanding it completely’’ (Piaget, 1970, p. 715). Moreover, they argue that children who acquire knowledge on their own are more likely to apply and extend that knowledge than those who receive direct instruction (Bredderman, 1983; McDaniel & Schlager, 1990; Schauble, 1996; Stohr-Hunt, 1996). There are pragmatic, empirical, and theoretical grounds for questioning this position. Pragmatically, it is clear that most of what students (and teachers and scientists) know about science was taught to them, rather than discovered by them. Empirical challenges come from studies demonstrating that teacher-centered methods using direct instruction are highly effective (cf. Brophy & Good, 1986; Rosenshine & Stevens, 1986), particularly for teaching multistep procedures that students are unlikely to discover on their own, such as those involved in geometry, algebra, and computer programming (Anderson, Corbett, Koedinger, & Pelletier, 1995; Klahr & Carver, 1988). Finally, most developmental and cognitive theories predict that many of the phenomena associated with discovery learning would make it a relatively ineffective instructional method (Mayer, 2004). For example, children in discovery situations are more likely than those receiving direct instruction to encounter inconsistent or misleading feedback, to make encoding errors and causal misattributions, and to experience inadequate practice and elaboration. These impediments to learning may overwhelm benefits commonly attributed to discovery learning—such as ‘‘ownership’’ and ‘‘authenticity.’’ However, our aim in this study was to go beyond a comparison of the immediate effectiveness of two radically different types of instruction. In addition, we tested the prediction that if children achieve mastery of a new procedure, then the way that they reached that mastery will not affect their ability to transfer what they have learned. This pathindependent transfer, if supported, has implications for discovery learning, because one of its purported advantages is that it has longterm benefits on how children ultimately transfer what they have learned—benefits that justify its admittedly lower efficiency. In order to evaluate this hypothesis, we had to create exemplars of both the discovery-learning and the direct-instruction approaches, expose children to them, and then challenge learners with an approAddress correspondence to David Klahr, Department of Psychology, Carnegie Mellon University, 5000 Forbes Ave., Pittsburgh, PA 15213-3890; e-mail: [email protected]. PSYCHOLOGICAL SCIENCE Volume 15—Number 10 661 Copyright r 2004 American Psychological Society priate transfer task. However, at the outset, we faced a difficult definitional problem because nearly 100 years of research (cf. Winch, 1913) had yet to produce a consistent definition of discovery learning. Therefore, we intentionally magnified the difference between the two instructional treatments in order to provide a strong test of the pathindependent transfer hypothesis. In our discovery-learning condition, there was no teacher intervention beyond the suggestion of a learning objective; there were no guiding questions and no feedback about the quality of the child’s selection of materials, explorations, or self-assessments. Correspondingly, we used an extreme type of direct instruction in which the goals, the materials, the examples, the explanations, and the pace of instruction were all teacher controlled. The specific context in which we contrasted these two instructional approaches was an important elementary-school science objective known as the control-of-variables strategy (CVS). Procedurally, CVS is a method for creating experiments in which a single contrast is made between experimental conditions. The logical aspects of CVS include an understanding of the inherent indeterminacy of confounded experiments. In short, CVS is the basic procedure that enables children to design unconfounded experiments from which valid causal inferences can be made. Its acquisition is an important step in the development of scientific reasoning skills because it provides a strong constraint on search in the space of experiments (Klahr, 2000; Klahr & Simon, 1999). CVS mastery is considered a central instructional objective from a wide variety of educational perspectives (DeBoer, 1991; Duschl, 1990; Murnane & Raizen, 1988; National Research Council, 1995). Chen and Klahr (1999) demonstrated that direct instruction on CVS led to statistically and educationally significant improvement in children’s ability to design simple, unconfounded experiments. For children receiving direct instruction, mean performance on CVS increased in the psychology lab from 40% correct prior to instruction to 80% correct following instruction, whereas children in the discoverylearning condition showed no significant improvement. Moreover, students receiving direct instruction were superior to control students on a far-transfer test of experimental design administered 7 months later. When the direct-instruction procedure was adapted from an experimental script to a lesson plan implemented in a classroom setting, mean CVS performance increased from 30% prior to instruction to 96% following instruction (Toth, Klahr, & Chen, 2000). However, the type of direct instruction used in these CVS studies has been criticized with respect to both its content and its epistemology (Chinn & Malhotra, 2001). The content critique is that CVS— as taught in these studies, as well as similar investigations of children’s experimental skill (e.g., Germann, Aram, & Burke, 1996; Metz, 1985; Palincsar, Anderson, & David, 1993; Schauble, 1990; Zohar, 1995)—is a relatively circumscribed and inflexible procedure. The epistemological critique is that direct instruction in CVS, although apparently effective in improving students’ CVS scores, does not provide them with a basis for exploring broader issues surrounding ‘‘authentic’’ scientific inquiry, such as detecting potential confounds and other validity challenges in complex experimental situations. According to these critiques, when children taught via direct instruction are faced with such complexities, they will show little transfer beyond the specific context in which they were taught. The aim of the present study was to evaluate these critiques by investigating the relative effectiveness of direct instruction and discovery learning not only with respect to the acquisition of CVS, but also with respect to children’s ability to reason in more authentic contexts. Our operational definition of authentic scientific reasoning is based on measures of children’s ability to evaluate science-fair posters created by other children. We chose this transfer task because one important aim of science education—from elementary school through graduate training—is to increase students’ ability to evaluate the soundness of other people’s scientific endeavors (National Research Council, 1995) and to assess the validity of their scientific claims (Zimmerman, Bisanz, & Bisanz, 1998). The poster-evaluation task used in this study challenged children to reason about the quality of other children’s research—to evaluate the design, measurements, data analysis, alternative hypotheses, and conclusions depicted in their posters. We tested three hypotheses: Direct instruction is more effective than discovery learning in teaching children CVS. That is, we expected to replicate earlier comparisons of different instructional approaches to CVS training (Chen & Klahr, 1999; Klahr, Chen, & Toth, 2001; Toth et al., 2000). When evaluating science-fair posters, children who have mastered CVS outperform those who have not. This hypothesis is based on the expectation that mastery of the procedural and conceptual rationale for controlling variables in experimental settings provides children with a basis for generating a rich array of critiques about not only experimental design, but also other aspects of ‘‘good science,’’ such as the adequacy of the design, manipulations, measurements, inferences, and conclusions associated with an experimental study. What is learned is more important than how it is taught. This hypothesis led to the prediction that the association between CVS mastery and good performance on the poster-evaluation task would be independent of the learning path, that is, independent of whether that mastery was achieved under the direct-instruction condition or the discovery-learning condition. Conversely, we predicted that the association between poor CVS scores and poor poster-evaluation scores would be independent of the instructional condition under which children failed to achieve CVS mastery. The importance of this hypothesis is that it runs counter to one of the primary purported benefits of discovery learning: that it has a positive influence on long-term transfer.