20.2 Improving traditional instruction: Cognitive Load Theory

For a number of years, John Sweller, an Australian psychologist from the University of New South Wales, has examined instructional implications of a model of memory called cognitive load theory, which is based on a straightforward reading of information-processing concepts of memory, schema development, and automaticity of procedural knowledge:

• Human working memory is limited; we can only keep in mind a few things at a time. This poses a fundamental constraint on human performance and learning capacity.
• Two mechanisms to circumvent the limits of working memory are: (1) schema acquisition, which allows us to chunk information into meaningful units, and (2) automation of procedural knowledge.

The first mechanism deals primarily with processing and understanding information; the second deals with the acquisition of skills. Each mechanism helps us overcome the limits of working memory by drawing on our long-term memories, which are very detailed and powerful.

Sweller's model of instructional design is based on these concepts:

1. Our limited working memories make it difficult to assimilate multiple elements of information simultaneously.
2. When multiple information elements interact, they must be presented simultaneously. This imposes a heavy cognitive load on the learner of the information and threatens successful learning.
3. High levels of element "interactivity" and their resulting cognitive load can be inherent in the content; e.g., learning language grammar inherently involves more element interactivity than simple vocabulary learning. However, weak methods of presentation and instruction may result in unnecessarily high overhead. An example would be to present a student a figure whose understanding requires repeated consultation of the text. The extra work required in decoding and translating the figure competes with the content for precious working-memory resources as the learner attempts to comprehend the material.

Cognitive load theory leads to some specific predictions for student learning:

• Simple content-i.e., content with relatively few intrinsic interactive elements-is not threatened by weak instructional methods. Learners are generally able to fit the demands of content and instruction within their working memories in such cases.
• Content containing high levels of interactivity among its elements cannot be learned effectively through weak instructional methods, that is, methods that require extra processing by learners. The demands of content and/or the method exceed the limits of the learner's working memory, and learning does not occur.

Sweller's cognitive load theory has led to a number of instructional prescriptions, including:

• Carefully analyze the attention demands of instruction. Sweller's method defines elements and then counts the number of elements in instructional messages. Processing troubles arise when the learner must attend to too many different elements at the same time.
• Use single, coherent representations. These should allow the learner to focus attention rather than split attention between two places, e.g., between a diagram and the text or even between a diagram with labels not located close to their referents (Chandler & Sweller, 1991; see discussion in Sweller & Chandler, 1994, pp. 192-193).
• Eliminate redundancy. Redundant information between text and diagram have been shown to decrease learning. (See Saunders & Solman, 1984; Reder & Anderson, 1982; Lesh, Landau & Hamilton, 1983; and Schooler & Engstler-Schooler, 1990, for research on redundant information on other tasks.)
• Provide for systematic problem-space exploration instead of conventional repeated practice (Pierce, Duncan, Gholson, Ray & Kamhi, 1993).
• In multimedia instruction, present animation and audio narration (and/or text descriptions) simultaneously rather than sequentially (Mayer & Anderson, 1991, 1992; Mayer & Sims, 1994).
• Provide worked examples as alternatives to conventional problem-based instruction (Paas & Van Merrienboer, 1994; Carroll, 1994; in the area of analogical reasoning tasks, see Robins & Mayer, 1993, and Pierce et al., 1993).

In the section below, we present an overview of research on worked examples to illustrate the implications of cognitive load theory for instruction. For more discussion of the other instructional strategies briefly listed above, refer to Sweller (1989) and Sweller and Chandler (1994).

20.2.1 Worked Examples

Conventional models of instruction in many domains involve the presentation of a principle, concept, or rule, followed by extensive practice on problems applying the rule. This approach at first glance seems like common sense-providing ample skills practice is "teaming by doing." However, cognitive load theory suggests that such instructional approaches may actually be hurting learners' understanding of the subject matter.

Swelter and Cooper (1985) examined the cognitive-load effects of methods for teaching algebra to high school students. They hypothesized that when learners confront a conventional end-of-chapter practice exercise, they devote too much attention to the problem goal and to relatively weak search strategies such as means-end analysis- Students already know how to use general search strategies to solve problems; what they lack is the specific understanding of how cases relate to the general rule.

Swelter and Cooper hypothesized that learners might benefit from studying worked examples until they have "mastered" them, rather than working on conventional practice problems as soon as they have "obtained a basic familiarity with new material" (p. 87). The authors developed an alternative teaching model that emphasized the study of worked examples. After learners acquire a basic understanding of the algebraic principle, they study a series of examples; then the teacher answers any questions the learners have. When the learners indicate they understand the problems, they are required to explain the goal of each sample problem and to identify the mathematical operation used in each step of the problem. The teacher provides assistance to any learners who have difficulty with the questions. Then the learners complete similar problems, repeating them until they are solved with no errors; if too much time elapses, the teacher provides the answer.

Sweller and Cooper found that in the worked-examples model, acquisition of knowledge was significantly less time consuming than in the conventional practice-based model. Furthermore, learners required significantly less time to solve similar problems (i.e., problems identical in structure) and made significantly fewer errors than did their counterparts. There were no significant group differences in solving novel problems. Thus teaming was more efficient with no discerned loss in effectiveness. The authors concluded that "the use of worked examples may redirect attention away from the problem goal and toward problem-state configurations and their associated moves" (p. 86).

Sweller (1989) summarizes his position toward problem solving and teaming by arguing that:

a. Both schema acquisition and rule automation are the building blocks of skilled problem-solving performance.

b. Paradoxically, a heavy emphasis on conveying problem solving is not the best way to acquire schemas or facilitate rule automation, because the means-end strategy commonly used focuses attention inappropriately and imposes a heavy cognitive load.

Alternatives to conventional problem solving such as ... worked examples must be carefully analyzed and, if necessary, modified to ensure that they, too, do not inappropriately direct attention and impose a heavy cognitive load.

d. For the same reasons as for point c, the format of instructional materials should be organized to minimize the need for students to attend to and men integrate disparate sources of information (Sweller, 1989, p. 465, reformatted).

Sweller's critics might claim that students under worked-example treatment were indeed actively engaging in problem-solving and practice activities, but that the nature of the practice shifted from traditional word problems to the study of worked examples. Instead of engaging in a multi activity (e.g., translating the word problem into one or more formulas, and performing calculations), the task narrowed to articulating the goal of the worked example and the appropriate mathematical operation. Swelter would likely agree with the critic. The point of the research is to suggest that not all "problem-solving" activities are equally effective. Some problem-solving activities actually leave learners at a loss, forcing them to resort to "weak" problem-solving methods - which they already know-rather than "strong" or domain- specific methods-which they are trying to learn. Ben and Scardamalia (1992) discuss this issue:

In novel situations, where no strong methods have been devised, weak methods are all anyone has. We use then the time, whenever we are stumped. But just because, everyone uses them, could hardly survive without doing so, and therefore practices them extensively, there is reason to question the value of teaching them. Teaching problem-solving skills may be an illusion, like teaching babies to talk (Bereiter & Scardamalia, 1992, p. 528).

If our goal is to teach students certain well-defined domains such as algebra or physics, then giving them problems requiring extensive use of "weak" methods may be counterproductive and may even interfere with learning the domain.

20.2.1.1. Worked Examples and Self-Explanations. One limitation of the Sweller and Cooper study was that only indirect inferences could be made concerning learners' cognitive processes. Chi and her colleagues (e.g., Chi, Bassok, Lewis, Reimann & Glaser, 1989; Chi, de Leeuw, Chiu & LaVancher, 1991; Chi & VanLehn, 1991) addressed this issue in the area of college-level physics. They analyzed the think-aloud protocols of good and poor problem solvers to identify the cognitive processes learners used in studying worked examples in physics. Learners in the study did not differ significantly in their prior knowledge of physics; instead, good and poor problem solvers were identified by their performance on objective tests.

Chi's worked examples differed from those used by Sweller and Cooper (1985) in significant ways. Sweller and Cooper presented worked-example sheets that were not p of a. text and that included no verbal commentary. By contrast, the physics examples were part of the text and included step-by-step verbal commentary, although the learn, had to infer the "why's and wherefore's" of each step.

Self-explanations were one kind of student response the worked. examples. "A self-explanation is a comment about an example statement that contains domain-relevant information over and above what was stated in the example line itself" (p. 69). Chi et al. (1989) found that good problem solvers generated more self-explanations than poor problem solvers, and that poor problem solvers used the examples rotely as prompts for solving subsequent problems. Chi and VanLehn (1991) conjectured that "the act of self-explaining may make the tacit knowledge ... more explicit and available for use" (p. 101). They identified two general sources for self-explanations: "deduction from knowledge acquired earlier while reading the text part of the chapter... [and] generalization and extension of the example statements" (p. 69).

In an intervention study, Chi et al. (199 1) found that high-ability and average-ability students benefited equally from being prompted to generate self-explanations. This finding counters other research on strategy training, which has found that such training generally benefits low-ability students while it doesn't benefit and may even interfere with the performance of high-ability students (e.g., Brown & Campione, 1981). These discrepant findings might be partially explained by the fact that the earlier studies tended to teach skills rather than strategies (see Duffy & Roehler, 1989, for a discussion of the confusion about skills and strategies).

20.2.1.2. Summary. Cognitive load theory bears a strong resemblance to traditional instructional-design theories (Reigeluth, 1983, 1987). The prescriptions for instruction require a careful task analysis that especially considers the memory load implications of different content combinations and instructional methods. The emphasis on well-defined content, worked examples, and careful doses of presented information is reminiscent of Merrill's (1983; Merrill & Tennyson, 1977) Rule-Example-Practice prescriptions for teaching concepts and procedures. The emphasis on careful control over presentation and pacing, and the strongly positive gains attributable to managing cognitive load, serve as prudent reminders of the importance of task and memory variables.