William A. Sandoval, University of California, Los Angeles

ExplanationConstructor is a software tool designed to support students' explanation of problems of natural selection and evolution, as part of technology-infused curricula developed as part of the BGuILE project . Following research showing that when students understand the purpose of experimentation as the discovery of causal relationships they are better able to design and conduct experiments , a basic premise of our approach has been to focus students' inquiry in terms of the products that inquiry should produce. Our focus has been on students' development of causal, evidence-based explanations. Our general approach is described in detail in Reiser, et al.

The discussion here is confined to a description of the design principles used to develop representations in ExplanationConstructor of important epistemic entities students work with during inquiry. This section briefly highlights how we have explored a set of trade-offs among how various knowledge representations support or hinder students' epistemic practices, and how other materials and activity can support students' interaction with these representations. This section summarizes the design history of ExplanationConstructor, and how classroom studies of its use have led to its development. The primary point we wish to make here is that knowledge representations such as those discussed in this paper, and others, support certain aspects of students' reasoning about the nature of scientific knowledge, as they create it, but that in themselves such representations are insufficient to push students to explicitly consider their own epistemological commitments or the implicit epistemological commitments underlying scientific inquiry.

Grounding epistemic forms in specific domains

ExplanationConstructor was specifically developed to structure students' efforts to construct scientific explanations. The issue was to figure out what kind of epistemic forms would scaffold students' learning to play the "scientific explanation game." There were two epistemological standards for scientific explanations that we wanted students to be able to satisfy: 1) the articulation of clear, coherent causal explanations for natural phenomena that 2) were supported by appropriate evidence. Our guiding design principle was that an epistemic form that represented these two criteria should be grounded within the domain of students' inquiry, in this case the domain of evolution and natural selection. Another principle was that students' explanations be tied to their efforts to answer specific questions, and that explanations be represented distinctly from the data used as evidence to support them (see SenseMaker and Belvedere, below).

Link explanations to specific questions

The first version of ExplanationConstructor, illustrated in Figure 1, has been described in detail elsewhere . In relation to our goals for students epistemological understanding, it was important to communicate not only what scientific explanations are like, but what they are for. Our stance was (and is) that explanations answer specific questions, so explanations should be linked to specific student questions. In ExplanationConstructor, students first recorded a question they were trying to answer, then later created one or more candidate explanations for it . Linking questions and explanations seems to have helped students monitor their progress in terms of what they felt they needed to know, their questions, and what they felt they did know, their explanations. Because these representations are distinct and persistent, students could return them over the course of their extended investigations of problems of natural selection. Moreover, group monitoring was focused in epistemic terms, concerning the ability of explanations to answer questions .

Represent theories as explanatory frameworks

One feature that distinguishes ExplanationConstructor from other collaborative inquiry tools, such as Belvedere, described below, or CSILE is that it provides domain-specific scaffolds through explanation guides . Explanation guides provide both conceptual and epistemic scaffolds. Conceptually, explanation guides focus students on the appropriate content of specific explanations. In Figure 1, for example, the "selective pressure" guide visually represents the theory of natural selection into components that prompt students for the important constituents of a natural selection explanation.

Epistemically, explanation guides encourage students' to think about theories as explanatory frameworks, super-ordinate to explanations for specific events. For each problem that students investigate, there are multiple guides to choose from. Because students have to choose explanation guides for each explanation, they are encouraged to map their emerging understanding into domain theory, to place themselves within a particular explanatory framework.

Figure 1:ExplanationConstructor 1.0. Questions, explanations, and evidence are distinctly represented, and explanations are highly structured (right).

Classroom studies of this first version of ExplanationConstructor showed that templates often provided direct guidance for students about their progress and about what data they should look for next . That is, groups could see how much of an explanation they had completed, and the specific prompts within guides suggested the kind of data that could help them complete each component. Of course, such guidance was available only prospectively, when students created new explanations before looking at all the potentially relevant data.

Linking evidence to causal claims

Data are not in themselves particular epistemic forms, but they have quite a different epistemological status from the causal claims derived from them. This distinction, as noted earlier, is often not made by students, but they seem to view explanations as being embodied in data, not interpretations given to data. ExplanationConstructor not only represents the data that students use as evidence separately from their causal claims about that data (Figure 2). Students have to actually select specific pieces of data as evidence, and then link them to specific causal claims. Thus, the theory-evidence distinction is made both in the representations students use and in their manipulations of those representations.

This first version of ExplanationConstructor seemed to provide productive supports for epistemic practices during students' small-group collaborative inquiry. Students engaged in frequent monitoring of their progress in explicitly epistemic terms, especially in evaluating the quality of their explanations as answers to their questions and whether or not they were based on appropriate data. Students' investigations were also highly planful, consistently focused on satisfying explanatory goals over the course of sustained, week-long investigations. On the other hand, the work-sheet like representation of the explanation guides sometimes worked against students' articulation of coherent explanations.. Also, the facility for linking and later viewing evidence was a bit opaque, consequently students rarely cited data for their explanations even though it was clear that they had examined relevant data and were reasoning from it.

Material and activity supports for evaluation

These findings motivated revisions to the ExplanationConstructor tool and spurred us to design other material supports that would encourage students to consider the quality of their explanations more explicitly, such as whether or not they were coherent and how specific causal claims explained specific data.

Clarifying connections between epistemic entities

A significant change to ExplanationConstructor was to make more clear the relationship between the different epistemic entities represented within the tool, namely questions, explanations, and evidence. One change was to organize questions and explanations hierarchically, to make it more clear that explanations answer particular questions. The hierarchy also allows students to explicitly record sub-questions subordinate to the over-arching questions of their investigations. For example, as shown in Figure 2, when trying to answer how the bacteria that cause tuberculosis can survive antibiotics, it might first be necessary to determine how antibiotics attack cells as a step in identifying potential differences between bacterial strains.

Another change was to make students' selected evidence for their explanations much more visible and salient (Figure 2). We also changed the way students cited data to allow them to insert a reference to data anywhere in their explanations and encouraged them to include them immediately after relevant causal claims. We found in a subsequent classroom study that not only did students cite more data in their explanations, but that they often explicitly discussed why that data was important evidence .

Another change to the software was to move the explanation guides out of the space where students wrote their explanations. Thus, they were still available for guidance, but students became more responsible for supplying connective causal language in their explanations. in fact, this is what seemed to happen. We do not yet know, although analyses are underway, whether or not these slightly removed explanation guides continued to productively direct students' investigative activities as the previous version had.

Explanation evaluation as a social practice

As far as students' epistemic practices are concerned, probably the single most important revision we made was to develop a specific rubric for students' evaluation of their explanations, and to provide them with clear opportunities to apply the rubric. This rubric highlights what we consider to be four key epistemological criteria for scientific explanations: 1) that they articulate a clear, coherent chain of cause and effect; 2) that they are supported by sufficient relevant evidence; 3) that alternative explanations of data have been explicitly considered and ruled out; and 4) that students articulate the limits of their own explanations, even those they consider their best ones. This rubric, although designed by participating teachers and the research group, becomes an object of discussion between students and teachers, and students take ownership of the criteria through their evaluation of their own and their peers' work.

ExplanationConstructor has a review facility that students can use to assess themselves or their peers. These reviews are tied to specific explanations (Figure 2), and are given using the rubric. Fundamentally, though, explanation evaluation is a social process, governed by the socioscientific norms (Tabak, 1999) of specific classrooms. An important goal of inquiry-based science learning ought to be to help students appropriate those norms, and to understand the epistemological commitments that underlie them. Consequently, a central feature of BGuILE curricula are mid-investigation peer reviews, and post-investigation consensus-building discussions. Mid-investigation reviews give students the chance to be reflective critics of each other's work, while maintaining an opportunity to respond to specific criticisms. They also give students repeated opportunities to apply the criteria of the explanation rubric. Post-investigation class discussions publicly articulate students' findings and allow students' to evaluating competing ideas to develop a consensus explanation for each problem. Throughout the unit, these consensual discussions provide teachers with opportunities to tie specific investigative experiences to broader domain principles.

Figure 2: ExplanationConstructor 2.0, showing the relations between questions, explanations, and evidence.

We found that students' critiques of each other noted a lack of specificity, especially failures to state causal mechanisms, as well as a lack of data as evidence. Students' most common self-assessments of their explanations at the end of an investigation were to acknowledge the limits of their accounts. Thus, the rubric and activities combined to encourage students to reflect on the quality of their own and each others' explanations in epistemologically important ways, and in a way that merely the construction of those explanations did not encourage.