Learning styles are one type of unproductive instructional mythology pervasive in the training profession. At best, most learning style programs are a waste of resources, and at worst, they lead to instructional methods that actually retard learning. A good example is the popular idea that there are visual learning styles and verbal learning styles. According to this myth, some people learn more from their visual senses, while others learn more from their auditory senses. In a misguided attempt to accommodate visual and auditory styles, many e-learning courses present content using visuals, text to describe the visuals, and audio narration of that text. However, as we have seen in Chapter 5, presenting words simultaneously in text and in audio results in a psychological redundancy that overloads working memory and depresses learning.

The concept of learning styles is based on an assumption of individual differences among learners that reliably interact with different types of instruction. For example, according to the visual-auditory learning style myth, an individual with a visual learning style will learn more from a visual presentation than from an auditory presentation, whereas an individual with an auditory learning style shows the reverse pattern.

Any consistent response to a specific instructional method by a group of learners that share a common attribute results in what instructional scientists call an interaction. We've touched on interactions in previous chapters, but since this entire chapter is about interactions, it's an opportune time to take a closer look.

What Are Interactions?

An interaction means that individuals of Type A will have a better or worse learning outcome with a given instructional Method Z than will individuals of Type B. When two instructional methods have opposite effects on two different types of learners, the result is called a disordinal interaction. A disordinal interaction exists, for example, when Type A individuals learn more from Method Y than from Method Z and, at the same time, Type B individuals learn more from Method Z than from Method Y. Figure 10.1 shows what a disordinal interaction looks like when the outcome data are plotted. As you can see, for learners of Type A, instructional Method Y resulted in much better learning than instructional Method Z. In contrast, Type B individuals show the reverse pattern. To maximize learning when you have this type of interaction, you would first determine whether any given learner was a Type A or a Type B and then tailor the instruction to match.

Figure 10.1. A Disordinal Interaction Between Method Y and Z for Type A and Type B Learners.

In order to adopt learning-style-based programs, we need replicated valid evidence (based on random assignment of learners to different instructional programs) showing that learners of Style A do better with one type of instructional approach, while learners of Style B do better with a different approach. We have not seen valid replicated evidence behind many of the popular learning styles, such as verbal and visual. However, prior knowledge is the one individual difference that has consistently been shown to interact with different instructional methods. Luckily, prior knowledge is an individual difference that is easier to assess than many other differences. In this chapter we will present guidelines, evidence, and psychology that show how you should adapt your instruction based on the amount of relevant prior knowledge possessed by your target learners.

Recall the experiments we summarized in Chapter 2 that compared expert and novice chess player memory for chess boards. Not surprisingly novices needed to refer back to a mid-play chess board more times than did experts in order to reconstruct it. Most interesting, however, was the reversal in this pattern when the chess board was a random board—one on which pieces did not reflect any real-world play patterns. When the board was random, expert memory was actually poorer than that of novices. This result is analogous to a pattern that Kalyuga, Ayres, Chandler, and Sweller (2003) term the expertise reversal effect.

As a novice chess player, your understanding of a mid-play and a random chess board are the same. Neither has any special meaning beyond a mix of game pieces placed on squares. However, as you gain expertise, your understanding of play patterns makes viewing a random board disruptive. Applied to training, expertise reversal predicts that a given instructional method that works well for novice learners is not only not useful for individuals with more expertise, but also results in depressed learning outcomes! In other words, in expertise reversal you will see an interaction between instructional method A that is effective for novices and instructional method B that is effective for experts.

The reason that expert chess players don't do well remembering random chess boards is that they view chess boards through eyes that are trained to look for meaningful patterns. These patterns are based on schemas that experts have built in their long-term memory over many chess games. When faced with a real mid-play chess board, these schemas allow experts to represent several pieces as one large item of information in working memory. A consequence of interpreting the world through schemas is a larger virtual working memory capacity for incoming information. However, when those familiar patterns disappear, as in a random chess board, expert performance is degraded due to the resources diverted to resolving the conflict between their schemas and the environment.

As a consequence of expertise reversal, efficient instruction for more advanced learners will require different methods than training designed for entry-level learners. If your training is extensive and supports a transition from low to high proficiency, you will need to adjust your more advanced lessons to accommodate the greater experience of your learners. A typical example is new hire training. Many new hire training programs extend over a number of weeks and combine formal training with on-the-job experience. During this process, trainees who started as complete novices build skills reaching relatively high levels of proficiency by the end of the training. In an ideal situation, the initial lessons that the new trainees study apply most of the load management guidelines we have suggested throughout this book. However, as learning progresses, the instructional materials transition to different methods that we summarize in this chapter.

If you are delivering instruction by self-paced e-learning, you can take advantage of a newly discovered efficient testing method to quickly assess the expertise of each learner as she progresses in the training. Applying this method results in an ongoing dynamic evaluation of learning accompanied by the assignment of the best instructional methods for a given level of expertise. We will provide more details of this method in the next chapter.

Throughout this chapter we will summarize evidence for the expertise reversal effect. The experiments we describe use one of two approaches that we call comparison group and staged.

Staged Experiments

The other experimental design frequently used to study differences in expertise is called a staged design and is illustrated in Figure 10.2. In these studies, learning outcomes are compared in the same group over time as learning progresses. The experiment starts with all novice learners, who are randomly assigned to study either the text alone version or the text plus diagram version. After the initial study period, all learners are tested. Then the entire group completes a well-designed lesson to ensure that everyone has an equally effective opportunity to build expertise. After completing the well-designed training, Stage 2 begins. Stage 2 may immediately follow Stage 1 or may be scheduled at a later time, such as the following week. During Stage 2, the learners are again assigned to either the text alone version or the text plus diagram version. After a study period, learners are tested again. By Stage 2, learners have gained more expertise, and the experimental lesson versions (text alone or text with diagrams) might have a different effect on learning. This plan may be followed for several stages. Each stage would include a period of study of the experimental lesson materials plus a training period with well-designed materials to ensure that everyone gains expertise.

Figure 10.2. A Generalized Plan of a Staged Experiment.

When the expertise reversal effect applies, the version that worked well for the learners in the early stages will not work as well for them as they gain expertise and eventually will result in less learning than the comparison versions. When the data is plotted on a graph, the pattern will look similar to that shown in Figure 10.1, with Type A learners replaced with Stage 1 and Type B learners replaced with Stage 2. This experimental design is frequently called a "longitudinal" design because we use the same learners whose expertise alters over time, rather than a cross section of learners with different levels of expertise.

In either the comparison group or the staged approach, if the learners rate the difficulty of the experimental instructional materials at the end of their study period, an efficiency metric can be calculated. These metrics can be plotted on the efficiency graph, as we discussed in Chapter 1. The main difference you will see on these efficiency graphs is two data points for each test lesson: one set of data points representing efficiencies for low knowledge learners and one set representing efficiencies for high knowledge learners.

In this chapter we provide evidence-based guidelines for instructional methods that work best with low knowledge learners and with high knowledge learners. We summarize what is known about individual differences regarding learning from text alone, text with and without diagrams, worked examples and problem assignments, as well as directive and guided discovery course designs.

This first guideline might sound odd, since this is a chapter about instructional methods for learners with high knowledge. However, we have evidence that you will get a return on investing resources in making texts very clear for novices—but not for readers with relevant experience.

WHAT THE RESEARCH SAYS ABOUT TEXT COHERENCE

McNamara, Kintsch, Songer, and Kintsch (1996) compared learning among high and low prior knowledge students after reading high and low coherent texts on heart disease. You can compare a segment from the low and high coherent versions in Figure 10.3. The underlined text in the revised version indicates edits made to add clarity to the less coherent version. For example, note that section headers have been added. Section headers are useful to signal paragraph topics to readers (Lorch, 1989). Second, pronouns such as "it" in the low coherence text were replaced with the appropriate reference, such as "heart" when there was a possibility of ambiguity in interpretation. Third, new sentences were added to explain unfamiliar concepts. For example, note the sentence added to the bottom of the coherent version that explains what heart valves are and how they function. Finally, new preview sentences were added. For example, at the end of the first paragraph the authors added the following preview sentence: "There are many kinds of heart disease, some of which are present at birth and some of which are acquired later." This preview sentence orients readers to the topics to follow. The low coherent text is understandable; however, the ideas have not been explicated to the same extent as the coherent version. For novices, comprehension of the low coherent text is likely to require greater mental effort.

Figure 10.3. Excerpts from Low and High Coherent Texts (edits are underlined in the coherent version).

Prevalent wisdom suggests that the more coherent text version would be the best resource for everyone. After all, isn't high coherence in writing always better? Surprisingly, the research team found that high prior knowledge readers who studied the low coherent version performed better on a problem-solving test than high prior knowledge learners who studied the high coherent version. As you can see in Figure 10.4, low prior knowledge readers showed the opposite pattern.

Figure 10.4. Opposite Learning Outcomes from High and Low Coherent Text by High and Low Prior Knowledge Readers.

Here we see an example of the expertise reversal effect. A high coherence text that most effectively helped low prior knowledge learners understand its meaning exerted the opposite effect on high prior knowledge readers. High prior knowledge readers learned more from a low coherent text. An explanation for this result based on cognitive load theory is that high prior knowledge readers already have adequate schemas about heart functions. Adding schema support in the form of extra explanations and preview sentences was redundant for high prior knowledge readers. The load added to their working memory had a negative overall impact on their processing of the text. In contrast, because low prior knowledge readers lacked appropriate schemas, they experienced additional extraneous cognitive load when reading the less coherent text.

We need more research on text coherence before we recommend writing low coherent texts for more experienced learners. We can say that adding explanations that are unnecessary for a given level of readership should, according to cognitive load theory, have negative effects, as found by McNamara, Kintsch, Songer, and Kintsch (1996). We do recommend that extra investment be made to prepare highly coherent instructional texts for novice learners. From a cognitive load perspective, reading difficulty was not rated, so we do not know whether the outcome observed in this study reflects a redundancy effect. We look forward to more research for evidence-based writing guidelines to guide our writing practices for high and low knowledge readers.

While we are on the topic of learning from reading, we briefly mention research showing that it's not a good idea to ask poor readers to answer questions about what they read while they are reading. That's because individuals with weak reading skills need to invest all of their cognitive effort in processing the text. To ask them to respond to questions at the same time leads to cognitive overload and less learning.

WHAT THE RESEARCH SAYS ABOUT ASKING QUESTIONS DURING READING

Van den Broek, Tzeng, Risden, Trabasso, and Basche (2001) compared recall of story facts among readers of different skill levels, including fourth graders, tenth graders, and college students. As you can see in Figure 10.5, the test scores show a significant interaction. The readers with lowest skills (fourth graders), learned less when having to answer questions during reading than fourth graders who did not answer any questions. The reverse pattern is seen for the college students, who are skilled readers.

Figure 10.5. Answering Questions During Reading Had Opposite Learning Effects Among Readers of Different Expertise.

While most organizational training professionals are not creating materials for fourth graders, in a global economy many learners may not be native speakers of English. In addition many trainees may have low reading proficiencies. Based on this evidence, we recommend that, when designing written materials for lower skilled readers, you should avoid inserting questions into the materials. Instead you can promote understanding by providing opportunities for questions and discussion after completing the reading assignment.

In Chapter 3, we saw that relevant diagrams help learners to perform procedural tasks more effectively than text, as well as promote a deeper understanding of content. Further, we recommended that you explain visuals with audio narration. However, when a diagram is self-explanatory, it becomes redundant to add words in any form, as we discussed in Chapter 5. In this section we review guidelines showing that more experienced learners will profit as well as or better from either text or diagrams alone. Experienced learners will not need the redundant information contained in the combination of text and diagram. In fact, it is likely that such redundant information will depress their learning.

WHAT THE RESEARCH SAYS ABOUT PROVIDING EXPERIENCED LEARNERS TEXT ALONE

Mayer and Gallini (1990) used a comparison group experimental design to evaluate the effectiveness of diagrams added to text on high and on low knowledge learners. They developed different lesson versions that described three different mechanical processes, including the operations of brakes, pumps, and generators. Among their lesson versions one included text alone and the other included the same text plus relevant diagrams. Figure 10.6 shows two sections from the text plus a diagram lesson on how brakes work. Prior to the experiment, participants answered a survey that assessed their prior knowledge of mechanical systems. On the basis of the survey, the researchers divided the participants into a low and a high knowledge group. These participants were then randomly assigned to study the two lesson versions. After the study period, the learners were tested with questions that asked them to apply what they learned. For example, they were asked, "Why do brakes get hot? What could be done to make brakes more reliable?" As you can see in Figure 10.7, the lesson versions with diagrams added to the text were very helpful for the low prior knowledge learners, but had no beneficial effect on the learning of high prior knowledge learners. The high prior knowledge learners had a sufficient schema to make sense of the text alone. In contrast, the diagram provided a schema substitute for low prior knowledge learners.

Figure 10.6. Text Plus Diagram Lesson Version of How a Brake Works.

Figure 10.7. Lesson Versions with Diagrams Aid Understanding of Low but Not High Prior Knowledge Learners.

In a summary of several experiments comparing effects of diagrams on high and low knowledge learners, Mayer (2001) reports that low prior knowledge learners studying texts with relevant diagrams will on average show performance gains 61 percent greater than performance gains of high prior knowledge readers. As with the reading studies summarized under Guideline 25, the data suggest that we invest more resources in design of effective materials for low prior knowledge than for high prior knowledge learners. First, we should pay attention to the coherence of the written text and, second, we should add relevant diagrams to illustrate the ideas in the text.

Note, however, that the expertise reversal effect was not seen in the Mayer and Gallini (1990) study. Learning of high prior knowledge learners was the same from versions with text alone as from versions with text and graphics. Mayer and Gallini did not specifically seek out high and low knowledge individuals. Rather, they took a group of individuals and divided them into higher and lower knowledge groups. By increasing the knowledge base of the high knowledge group, the probability of a full expertise reversal effect is increased.

WHAT THE RESEARCH SAYS ABOUT PROVIDING EXPERIENCED LEARNERS DIAGRAMS ALONE

Kalyuga, Chandler, and Sweller (1998) used a staged experiment to evaluate the effects of two lesson versions designed to teach how to interpret circuit diagrams of a motor on learning of novice and more experienced participants. One version included diagrams plus text, and the other included just the diagrams. The effects of the text alone versus text plus diagram were tracked among the same group of learners as they gained expertise over three learning stages similar to the design shown in Figure 10.2.

As you can see in Figure 10.8, at Stage 1 the novice participants benefited from lessons that included a combination of diagrams and explanatory text. The novice learners lacked relevant schemas to make sense of a circuit diagram alone. Notice, however, that by Stage 2, either lesson version was equally effective. By this point, the learners had sufficient expertise to profit from the diagram alone but insufficient expertise to be disrupted by the combination of text and diagram. However, as their expertise developed even more over the month between Stages 2 and 3, the text became redundant. By Stage 3, learning was better from a diagram alone. Here we see the expertise reversal effect. As novices, learners needed the schema support provided by text to explain the diagram. As they gradually developed expertise, the benefits of the text disappeared and, once they had gained relatively high levels of expertise, the text became redundant and resulted in diminished learning.

Figure 10.8. Diagrams Alone Resulted in Better Learning with More Expert Learners.

These results were replicated in a second experiment in which the diagrams were explained by words presented in audio rather than by text (Kayluga, Chandler, & Sweller, 2000). In Figure 10.9 we show a complex diagram for determining cutting speeds for drills that also includes a table of cutting speeds for various materials as well as the steps to follow to determine a cutting speed. Two lesson versions incorporated diagrams only or diagrams explained by words presented in audio narration. Similar to the previous experiment, the same group of learners who started out as novices were trained and tested over three stages during which they gained expertise. The results were similar to those shown in Figure 10.8. The diagram explained by audio was most effective for novice learners. However, as their expertise increased, there was a point at which there was no significant learning difference between the two versions, and eventually the diagram alone was best.

Figure 10.9. A Worked Example Using Audio to Explain How to Use the Diagram to Determine Cutting Speeds for Drills.

As in the previous experiment (Kalyuga, Chandler, & Sweller, 1998) that used circuit diagrams, the task of determining a cutting speed could not be performed without the diagram. Therefore, once learners had developed a schema adequate to interpret the diagram, the explanatory words became redundant. The efficiency diagram shown in Figure 10.10 shows the reversal of efficiency of different lesson versions as the learners gained expertise.

In both of the previous experiments, words presented as text or as audio narration became redundant to a diagram that was essential to accomplish the task. However, in the Mayer and Gallini experiment, we saw that more experienced learners profited as much from a text-only version as from a version with text and diagram. So you might wonder whether you should drop diagrams or drop text for more experienced learners.

Figure 10.10. Diagrams Plus Words Are More Efficient for Novices; Diagrams Alone Are More Efficient for Experts.

In Chapter 8, we presented a number of guidelines on how to best sequence and structure faded worked examples in order to gradually impose mental work on learners. Therefore here we will only briefly review how worked examples interact with learners of low and high prior knowledge.

A Review of Research on Worked Examples as Learning Progresses

Kalyuga, Chandler, Tuovinen, and Sweller (2001) compared the learning benefits of lessons with worked example-problem pairs to lessons with all-problem assignments for training how to write programs for relay circuits of various levels of complexity. Experiment 1 used a staged learning approach. They found that worked example-problem pairs led to greater learning than all-problem lessons for the first two stages. However, by Stage 3, the learning effects of the two lesson versions were equivalent. An expertise reversal effect had not yet occurred. The research team concluded that an expertise reversal would take place if the learners had even more experience with this type of programming. Because they had limited time to work with the participants, they conducted a second experiment using an easier task that could be learned faster. This experiment involved two stages. As you can see in Figure 10.11, sufficient expertise was gained by the learners so that by Stage 2 worked examples became redundant and learning was better with all practice problems.

Figure 10.11. Lessons with All Problems Led to Better Learning of Experienced Learners.

Recall from Chapter 7 that a directive lesson presents guidelines or steps, shows worked examples of how to apply the guidelines or steps, and then asks the learner to practice through faded worked examples and/or practice problems. For example, in our sample spreadsheet e-lessons on the CD, we initially explain rules for formatting Excel formulas followed by faded worked examples and then practice assignments.

In contrast, lessons with a discovery or guided discovery design use a more inductive approach. Inductive lessons provide learners with stories, experiences, or exploratory opportunities from which learners are expected to derive the relevant guidelines or steps. For example, at one point in our Excel virtual classroom lesson, the instructor showed three sample formulas and asked learners to derive the major formatting rules.

Various types of guided discovery course architectures are frequently touted as better because the learner is constantly engaged. However, engagement not directed toward schema acquisition and automation can impose an extraneous working memory load. Next, we review two experiments that compared learning from a directive lesson in which worked examples were provided with learning from guided discovery lessons in which learners were asked to provide their own examples.

WHAT THE RESEARCH SAYS ABOUT DIRECTIVE AND GUIDED DISCOVERY LESSONS

Kalyuga, Chandler, and Sweller (2001) prepared two lesson versions, both designed to teach how to write Boolean switching equations to design relay circuits of different levels of complexity. An initial lesson provided everyone with basic knowledge needed for both simple and complex circuits. Next, in Stage 1, participants were randomly assigned to study two experimental lesson versions. In the directive version, the lessons provided worked-example problem pairs for either simple or complex circuits. In the guided discovery version, learners were directed to design high and low complexity relay circuits. If their circuit design was appropriate, they were directed to type in the Boolean equation for their circuit. Therefore, in the guided discovery lesson, learners were basically constructing their own examples. Upon completion of their lessons, all learners rated lesson difficulty and were tested. At this point, all study participants were given a training session designed to build expertise in constructing circuits. Stage 2 started with a training session similar to the one that ended Stage 1 and proceeded to experimental sessions using directive or guided discovery approaches.

As you can see in Figure 10.12, there was no difference in learning between the directive and guided discovery lessons on simple tasks. However, the worked example lessons were completed much faster and thus offer a more efficient approach. However, for complex tasks the results are different. As shown in Figure 10.13, the directive lesson using worked examples resulted in better learning among novice learners than the guided discovery lesson. However, by the end of Stage 2 when the learners had more expertise, both versions resulted in equivalent learning.

In summary, in this experiment learning of high and low complexity content was compared from a directive lesson that used worked-example problem pairs to a guided discovery lesson over time as learning increased. Working memory would be most vulnerable to overload when learners were novices and were working with more complex circuits. As expected, the directive lessons were most effective for novice learners working with high complexity content. In contrast, for low complexity circuits there was little difference in learning between the two lesson versions, although instructional time was faster with directive instruction.

Figure 10.12. A Comparison of Directive with Guided Discovery Lesson Design on Learning and Training Time of Novice Learners on Simple Tasks.

Figure 10.13. A Comparison of Directive and Guided Discovery Lesson on Learning of Complex Tasks by Novice and Experienced Participants.

A similar result was reported by Tuovinen and Sweller (1999) using instruction on how to work with database programs. In this study a directive lesson that used worked example-problem pairs was more effective for learners unfamiliar with database programs than a guided discovery lesson in which learners were asked to create their own examples. However, for participants who had prior familiarity with databases, either version was equally effective.

There are solid research and psychological reasons for recommending that you:

In this chapter we have recommended that you adapt your instruction according to the expertise of your learner. In e-learning, individual measures of learner knowledge can be used as the basis for adapting instructional methods dynamically as each learner progresses through the instruction. Adaptive testing has not been used extensively, however, because it is time-consuming for learners to take tests throughout the training. However, Kalyuga and Sweller (2004) have reported a brand new method of rapid testing that makes tailored instruction in e-learning practical. We summarize their research in Chapter 11.