Welcome to the fourth-grade science fair, with its baking-soda volcanoes, bread mold grown in drawers, proud parents and thoughtful judges. The teachers can't help but wonder if the young would-be scientists can tell good science from bad. In science, how is critical thinking best taught?

This question may be answered by David Klahr, PhD, a psychology professor at Carnegie Mellon University, and Milena Nigam, a research associate at the University of Pittsburgh's Center for Biomedical Informatics. They have new evidence that "direct instruction"--explicit teaching about how to design unconfounded experiments--most effectively helps elementary school students transfer their mastery of this important aspect of the scientific method from one experiment to another.

Their assertion is based on years of research, including a controlled study presented in March at the U.S. Department of Education's first-ever Secretary's Summit on Science, attended by several hundred policy-makers, education, corporate and foundation leaders and teachers.

For decades, early science education has emphasized "discovery learning," in which children, given experimental materials such as springs and pulleys, marbles and ramps, are expected to "discover" scientific principles on their own. The approach is a legacy from two intellectual giants: developmental psychologist Jean Piaget and educational philosopher John Dewey. Piaget believed children locked in learning better when they learned on their own; Dewey sought to motivate students with hands-on, real-world problems.

Still, science and similarly complex subjects may well require a distinct teaching methodology, says Klahr. His controlled studies continue to demonstrate that, at least for many of the multistep procedures used in science, direct instruction works and generalizes better.

The findings come at a critical juncture, notes Klahr, because, in his words, "The United States produces proportionately fewer scientists than many other 'competitor' countries, so better science teaching is certainly a national priority if we are to maintain our scientific leadership. Early mastery of the basics of the scientist's toolkit can help kids to understand and appreciate science. More generally, a critical understanding of the difference between good and bad science is essential to informed adult decisions in the marketplace and in the voting booth."

Direct versus discovery

Klahr saw three main reasons to challenge discovery learning. First, most of what students, teachers and scientists know about science was taught, not discovered, he says. Second, teacher-centered methods (in which teachers actively teach, as opposed to observe or facilitate) for direct instruction have been very effective for procedures that are typically harder for students to discover on their own, such as algebra and computer programming. Third, he adds, only vague theory backed the predicted superiority of discovery methods--and what there is clashes with data on learning and memory. For example, discovery learning can include mixed or missing feedback, encoding errors, causal misattributions and more, which could actually cause frustration and set a learner back, says Klahr.

Yet discovery learning has persisted, he says, partly because of a lingering notion that direct instruction would not only be ineffective in the short run, but also damaging in the long run. Piaget thought interfering with discovery blocked complete understanding. More recent cognitive research, says Klahr, shows that "this is just plain wrong."

Studying the study of science

In the late 1990s, Klahr, Zhe Chen, PhD, and Anne Fay, PhD, first studied teaching methods and the Control of Variables Strategy (CVS), which allows scientists to design unconfounded experiments and draw valid conclusions from experimental outcomes.

"This 'process skill' must be taught," says Klahr. In short, it's fun to collect bugs in the backyard--but to learn how to design experiments to test specific hypotheses about how bugs behave when hot or cold (i.e., under different conditions), children need explicit guidance.

That earlier research showed that most elementary students could learn the principles of CVS in less than 30 minutes of direct instruction and retain the skills seven months later. Later this year, Klahr and Nigam will share even more promising data. Beyond comparing instructional types, they have tested a critical claim about the advantage of discovery learning over direct instruction--that it transfers to other tasks, a cornerstone of real learning.

In the study, which will appear this fall in Psychological Science, the researchers studied 58 third-graders and 54 fourth-graders in four Pittsburgh-area schools. They randomly assigned children from both grades to a direct instruction or a discovery learning condition. In direct instruction, teachers controlled the goals, materials, examples, explanations and pace of instruction. In discovery learning, teachers did not intervene beyond suggesting a learning objective.

On day one, the researchers determined the children's baseline competence in CVS; they then gave them little wooden ramps and had the children design experiments to study how factors such as steepness and ramp length affected how far a ball rolls after it comes down the ramp.

In direct instruction, the children watched as the instructor designed several experiments. Some controlled all but one variable, directly comparing, for example, the effects of rubber ball versus golf ball, short ramp versus long ramp, rough ramp versus smooth ramp while holding everything else the same. Others had confounds, such as golf balls down rough ramps versus rubber balls down smooth ramps. Each time, researchers asked the children if the design would let them "tell for sure" if the studied variable affected the outcome. The instructor explained why each of the unconfounded experiments singled out the critical factor (and vice-versa for the confounds). Meanwhile, in discovery learning, children were asked to design the same number and type of experiments, but without any instruction in CVS or feedback.

Experimenters then rated student designs of two new experiments, one to measure the effects of an earlier factor (run length) and one to measure the effect of a new factor (surface). The latter design revealed whether the children could transfer their experimental strategy to something new. After direct instruction, 77 percent of the children were able to design at least three out of four experiments without confounds. After discovery learning, 23 percent--significantly fewer students--were able to do the same. About a week later, a different experimenter asked the children to evaluate two science-fair posters by suggesting how to make them "good enough to enter in a state-level science fair." Both posters described deeply flawed experiments. Again, significantly more children exposed to direct instruction were able to critically evaluate experiments. Discovery learning's purported advantage was not supported.

The debate continues

Still, "No single study ever settles a debate once and for all," says James Stigler, PhD, a professor of psychology at the University of California, Los Angeles and director of video studies for the Third International Mathematics and Science Study. He notes that the study's two teaching approaches exaggerated their real-world counterparts, limiting generalizability, but thinks the study does underscore that labs work best when integrated with explicit instruction in critical science concepts and methods.

Psychologist Rich Shavelson, PhD, professor of education and (by courtesy) psychology at Stanford University, notes that totally unguided discovery of the type used in the study is rarely used in the classroom. Still, he says, "This study uses a strong research design. I'd like to see a replication with [the more typical] guided discovery. Plus, the extent to which results would travel to classrooms with varying teacher quality, opportunity to learn, et cetera, has yet to be found out."

Leona Schauble, PhD, a cognitive development psychologist at Vanderbilt University, agrees. "Educators do not believe that children should stumble around and reinvent modern science," she says. She views the teaching of controlled experiments as one small piece of a science education that includes many other conceptual challenges.

Rachel Adelson is a writer in Raleigh, N.C.