ity of the tasks Kuhn used as problematic. They chose to employ simpler tasks that involved story problems about phenomena for which children did not hold strong beliefs. Children’s performance on these tasks demonstrated that even first and second graders could differentiate a hypothesis from the evidence. Likewise, Ruffman et al. (1993) used a simplified task and showed that 6-year-olds were able to form a causal hypothesis based on a pattern of covariation evidence. A study of children and adults (Amsel and Brock, 1996) indicated an important role of prior beliefs, especially for children. 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. Looking across these studies provides insight into the conditions under which children are more or less proficient at coordinating theory and evidence. In some situations, children are better at distinguishing prior beliefs from evidence than the results of Kuhn et al. suggest.
Koslowksi (1996) criticized Kuhn et al.’s work on more theoretical grounds. She argued that reliance on knowledge-lean tasks in which participants are asked to suppress their prior knowledge may lead to an incomplete or distorted picture of the reasoning abilities of children and adults. Instead, Koslowski suggested that using prior knowledge when gathering and evaluating evidence is a valid strategy. She developed a series of experiments to support her thesis and to explore the ways in which prior knowledge might play a role in evaluating evidence. The results of these investigations are described in detail in the later section of this chapter on the role of prior knowledge.
Researchers have also looked at reasoning about cause in the context of full investigations of causal systems. Two main types of multivariable systems are used in these studies. In the first type of system, participants are involved in a hands-on manipulation of a physical system, such as a ramp (e.g., Chen and Klahr, 1999; Masnick and Klahr, 2003) or a canal (e.g., Gleason and Schauble, 2000; Kuhn, Schauble, and Garcia-Mila, 1992). The second type of system is a computer simulation, such as the Daytona microworld in which participants discover the factors affecting the speed of race cars (Schauble, 1990). A variety of virtual environments have been created in domains such as electric circuits (Schauble et al., 1992), genetics (Echevarria, 2003), earthquake risk, and flooding risk (e.g., Keselman, 2003).
The inferences that are made based on self-generated experimental evidence are typically classified as either causal (or inclusion), noncausal (or exclusion), indeterminate, or false inclusion. All inference types can be fur-