31.2 Applied Research

Studies investigating the viability of the generative model of learning have tested the effects of simple coding strategies such as underlining, notetaking, and adjunct or inserted questions; more complex codingtorganizational strategies such as creating hierarchies, headings, summaries, and concept maps, or the manipulation of objects; elaborative integration strategies such as imaging, and creating examples, interpretations, or analogies; and, finally, metacognitive generative-leaming training. Table 31-2 organizes and summarizes some of the most significant work testing Vtrittrock's theory. The table divides the types of research into those that represent a coding generative activity and those that represent an integrative generative activity. Those that exemplify coding interrelate concepts from the instruction together to create one level of understanding through various levels and types of organizational activities. Those exemplifying integration interrelate the concepts from the instruction with prior knowledge to create a higher level of understanding by reconceptualization and elaboration. A discussion and summary of the results from each of these areas is provided. This discussion begins with the most controversial-underlining-so that it neither gets lost among the other significant, noncontroversial studies nor is given the same importance of many of the other studies reported here.

31.2.1 Simple Coding

31.2.1.1. Underlining. As previously discussed, an argument can be made for the activation of generation processes in the learner by having her or him consciously and interactively relate information in the passage with prior beliefs and conceptions. This is, in essence, what Rickards and August (1975) did in their study. They investigated subject-generated versus experimenter-provided underlining strategies under six treatment conditions. Their results indicated that when college students had an opportunity to underline text that they considered most relevant~ they performed much better on the posttests on both objective-specific and incidental learning (total recall). In fact, a very interesting result was that in the learner-generated condition, in which the subjects were asked to underline the least important items, they did poorest of all. Rickards (1979) explained that since learners were asked to underline those sentences that were more relevant to them, a mental interaction between sentences and between what they read and their own preconceptions had to occur, thereby establishing plausible evidence that learner-constructed generative learning occurred.

31.2.1.2. Notetaking. Notetaking is considered an organizational, coding strategy, which has some controversy connected with it as well. No generation of understanding occurs when a learner simply copies sentences from a page. As with the Rickards' argument, however, a learner that rewords sentences to combine ideas from the passage or relate them to prior knowledge is engaging in generative activity. This is an important distinction for teachers as they teach learners how to take notes and give them feedback in the process. To illustrate, three studies have been selected that include both high school and college students in both vocational and liberal arts areas.

Peper and Mayer (1986) found from two experiments, one with high school and one with college-level students, that notetakers performed better than non-notetakers on far-transfer tasks of problem solving but were worse on near-transfer tasks of fact retention and verbatim recognition (p. 34). Shrager and Mayer's (1989) study of college students instructed to take notes or not to take notes from a videotaped lesson confirmed these findings and found the effect on recall and transfer highest for learners with low Prior knowledge. Peper and Mayer also tested the effects of other generative strategies, such as taking summary notes and answering conceptual questions during breaks in lectures, which produced similar results. This study points Out the importance of examining the effects of generative strategies on the quality vs. quantity of learning.

Notetaking in two studies by Barnett, DiVesta, and Rogozenski (198 1) was hypothesized to aid college students ill the Processing of information. Notetaking produced better results in learning from text than no notetaking, but

elaboration of the notes during a review period, a generative activity, produced similar results in terms of amount of learning as those who simply reviewed their notes. An interesting dimension to this study was the inclusion of instructor-prepared notes vs. leamer-generated notes. For immediate recall tests during which learners had an opportunity to review or elaborate on instructor-prepared notes, they performed better than the learner-generated notes group. A second study tested whether the effects were the same for different types of questions: those common to the group, those from their own notes, those from others' notes, and those from others' elaborations. Scores for the delaved test using questions from their own notes were dramatically higher than the other groups. This provides a strong case for notetaking causing generative effects. In other words, these results showed that learners remembered what they originally perceived and encoded versus what others had intended them to remember.

To summarize, at least with these studies, notetaking has shown positive effects, but there were mixed findings when compared with type of learning. Notetaking may be a highly generative activity; however, quality of notes, type of elaborations, and opportunity for review can affect what, how much, and for how long information is learned.

31.2.1.3. Adjunct or Inserted Questions. Adjunct questions have been classified by Wittrock (1990) as a generative activity. They ftinction as a coding and organizational guide. While questions can be generated by the learner, Wittrock also believes that they serve a teaching function to induce generative thinking by causing the learner to organize the information presented by relating ideas from a passage together, thereby creating personally meaningful understanding.

Over the past 25 years, the effects of inserted or adjunct questions have been studied extensively across content areas and age levels. Two important reviews of this research have summarized those findings. Anderson and Biddle (1975) and Rickards (1979) have concluded that inserted postquestions have been shown to increase recall of incidental learning (where criterion questions are unrelated to the inserted questions) as well as increasing recall on intentional learning (i.e., where criterion questions are the same as the inserted questions). Prequestions have been shown to increase intentional learning only.

According to Anderson and Biddle (1975), adjunct questions have also been examined in terms of frequency, the need for feedback, the nature of the question, the need for overt responding, and motivation. They summarized that: The more frequent the questions the better; feedback increased learning, but so did inserted questions without feedback; while most of the research focused on fact-level questions, there was also a positive effect from higher-level questions; free recall was generally better than multiple choice; a need for overt responding was dependent on how the questions were embedded; and the questions did motivate learners in some cases. They also found that these effects held across age level, content, length of text, and medium used.

Sutliff (1986) investigated the effect of inserted questions on reducing passivity in a self-instructional slide-tape presentation as evidenced by increased learning of facts (direct learning) and inference (indirect learning). His findings were opposite those of Anderson and Biddle in that there were no significant differences between groups. He interpreted the nonsignificance to be a result of not requiring overt responses to the questions, again contrary to previous research. Because of this "veto power over learning," described by Rothkopf (1976, p. 94), results such as this need to be examined further to determine just where overt manifestations may be necessary to ensure that processing occurs.

Burton, Niles, Lalik, and Reed (1986) investigated the effect of superordinate and subordinate questions'cin the amount of mental effort (level of cognitive capacity engagement) by using a secondary task probe technique and a passage about a mythical country. They found that superordinate questions have a greater learning effect, and that the effect carries over into subsequent text. The overall result also indicated that more main ideas were recalled than details. The explanation of the effect offered was that superordinate information is pulled into short-term memory more frequently, so it gets more practice. In other words, they found that general questions are more mentally engaging than detailed ones.

Woods and Bernard (1987) also found effects contrary to those of the reviews of Anderson and Biddle and Rickards. They investigated the effects of adjunct conceptual poslquestions for encouraging greater depth of processing of verbal information of adults 60 and older. From results on intentional and incidental free-recall tests, they found that adjunct questions helped older learners process only intentional text at a greater depth.

In an interesting twist of the research question, Brody and Legenza (1980) studied the effect on learning of inserted pictures as opposed to inserted questions and hypothesized that the effect would be the same as the results on adjunct questions. Their findings supported their hypothesis that postpictures were more beneficial to reading comprehension than prepictures.

To summarize the numerous studies: Postquestions and postpictures have been shown to be most effective for increasing both intentional and incidental learning; superordinate questions have been more effective than subordinate detail questions; and overt responses have been more effective than allowing covert responses.

31.2.2.1. Organizational Strategies. This topic deals with a variety of coding/organizational activities including creating hierarchies, headings, and sentence meanings, and mapping techniques across all age levels from elementary school children to professionals in a variety of topics from

science to language arts. These organizational tasks require learners to relate ideas from a passage together by using a variety of symbolic representations. Each addresses at least one of three key questions regarding the generative model of learning: the effect of learner-generated learning vs. the effect of learner-reproductive learning; learner-generated vs. instructor-provided constructions of meaning-including organization as a variable; or the general effects of generated elaborations.

Wittrock and Carter (1975) studied free-recall responses of undergraduates in generative vs. reproductive treatments using hierarchies with varying degrees of order. The generative group was directed to organize the hierarchies, while the reproductive group was directed to simply copy them. The results showed better performance for the generative treatment groups than for the reproductive groups for the disorganized and randomly organized hierarchies. However, the organized reproductive group performed better than the unrelated generative group. This means that organization in the stimuli can compensate somewhat for a lack of learnergenerated strategies, but providing organization in the instruction and opportunities for generative activity will be the best.

In two experiments with elementary school children, Doctorow, Wittrock, and Marks (1978) studied the effect of learner-generated vs. experimenter-provided paragraph headings and sentence meanings on comprehension. Again, the combination of text organized through the use of headings plus leamer-generated sentences about the paragraphs produced dramatic gains in comprehension and recall. Generative instructions without experimenter-provided headings followed as the next most effective, and paragraph headings alone were more effective than the control group. This strategy also increased comprehension more for high-ability students than for low-ability students, perhaps because high-ability students have better organizational cognitive abilities to make sense out of disorganized information.

Beissner, Jonassen, and Grabowski (1993) tested the effects of two organizational strategies against learner differences at four levels of learning. Their findings showed an interaction between learner-generated concept vs. semantic maps and serialist learners on the problem-solving questions only, with serialists performing better with semantic maps, and holists performing better with concept maps. While this study did not compare their results with instructorprovided maps, it does contribute evidence to considering the importance of learner cognitive strengths and patterns of thinking when selecting organizational learning activities.

Also studying the effects of concept maps, Smith and Dwyer (1995) found a significant difference only on lowerlevel terminology tasks in favor of instructor-provided maps. This result ' is consistent with that of Wittrock and Carter (1975). For lower-level tasks, organization helps, especially when a learner is tested with questions that show similarity to the organization that an instructor may have possessed when creating the test.

To summarize the findings of these studies: The results show that learner-generated activities are more effective in improving achievement than instruction-provided organizational schemes. Performance is increased even more when the text is organized. The selection of activities should be tempered by cognitive ability.

31.2.2.2. Manipulation of Objects. The next organizational activity deals with manipulating objects. While this activity extends beyond the printed page as designated by Wittrock's work, it qualifies as a generative activity because a relationship is being drawn and extended between parts of the environment.

Sayeki, Ueno, and Nagasaka (1991), in a very interesting study, investigated the effects of transforming mediational objects in the learning of mathematical principles. Their results supported the hypothesis that manipulatives would increase comprehension. While they do not specifically call this a generative activity, the act of creating understanding by generating both mental and physical relationships from different shapes of a manipulable rectangle manifests the same required attributes defined by Wittrock. Their results from mathematics should be tested for conceptual learning and problem solving in other content areas.

Haag and Grabowski (1994) extended this work to computer-manipulated graphics. Most applications of moving or manipulated graphics are done through generated animation. In this study, they found that learners who manipulated the graphics on the screen using a preorganized organizational framework increased problem solving over those using no organizational framework or having the computer create the graphic statically. These results are consistent with those of other organizational strategies reported in the previous section.

Both studies lend support to the use of manipulatives for generating understanding for both children and undergraduates in math and science.

The next series of studies examine the effects of activities that require a student to relate information to prior knowledge. In these activities, learners are integrating that information through imaging, elaborations, and analogies.

31-2.3-1. Imaging. The effects of imaging have also been investigated extensively with four of those studies summarized here. They include fictitious descriptions, language arts, and economics topics studied by elementary or high school students.

Anderson and Kulhavey (1972) studied high school seniors to determine the effect of imaging on prose learning. In this study, half of the subjects were told to image, while the other half were not. Results indicated no difference in prose learning between the groups. On further probing, the researchers discovered that not all of the students in the Maging group actually created images (only 50% did), and many in the control group did create images (about one-third)! Comparing those subjects from both groups who actually used imaging with those who did not showed significant differences in favor of the imaging strategy. This illustrates the fact that mental activity cannot be strictly controlled by instruction and, again, raises the issue that requiring an overt response may be more effective in encouraging the desired result than just simply providing direction to image, as Sutliff (1986) found with adjunct questions.

Bull and Wittrock (1973) compared the effect of experimenter-provided vs. learner-generated imagery with elementary school children. Groups were directed to either draw, trace, or copy verbal information on definitions they were to learn. As predicted, results showed that the group that generated images performed significantly better than those who copied definitions; however, there was no significant difference between the imagery provided (tracing) and the copied definitions groups.

Kourilsky and Wittrock (1987) investigated what effect the sequence of the use of verbal or imaging generative activities would have on economic understanding by high school students. They found that using verbal elaborations first, followed by imaging, significantly increased economic understanding. They also found significantly greater gains by using both generative activities (verbal and imaginal) over just verbal elaboration only.

Laney (1990) found a slightly different result. Examining economic reasoning in the decision making of third-graders, he found that the verbal-only and integrated strategies were more effective than the imaging-only strategy. While using both symbol systems increased learning in both studies, the verbal-only elaboration was more effective than both the imagery-only and the use of dual-symbol systems. He felt his results were consistent with Wittrock's notion that the effective use of imagery is developmental. Laney's thirdgrade subjects had not yet developed this ability and were more familiar with verbal instruction. These are important results given the confusion that could result from the use of a generative imagery strategy too early in a learner's developmental cycle.

. These studies have shown that overt imaging is more effective than covert; learner-generated imaging is more effective than instruction-provided imaging; and visual images may be more effective than verbal ones, only in cases in which students have progressed developmentally to the point where they can understand diem.

31.2.3.2. Elaborations. Stein and Bransford (1979) conducted two studies to determine the effects of learner generated or experimenter provided by type of sentence elaborations. They hypothesized that congruence of the elaboration with the topic Would be the determining variable and, in fact, did find differences in two experiments with undergraduates. In those cases where elaborations were incongruent, students did worse than those in the treatments with no elaborations at all. Two important findings indicated that "elaborations facilitated performance only when they clarify the precise significance of target concepts . . . and that prompting subjects to ask relevant questions facilitated both the precision of elaboration and subsequent retention" (p. 769).

DiVesta and Peverley (1984), in a very complex study, tested learner-organized vs. preorganized examples on near and far transfer in a concept attainment lesson. Additional variables included variability of examples, and sequence. Their results on the active vs. passive element of their study indicated that students who generated their own examples did significantly better on both transfer tests than the preorganized group.

Johnsey, Morrison, and Ross (1992) investigated the effects of embedded vs. detached and learner-generated vs. experimenter-provided elaboration on recall, recognition, and application learning. The type of elaborations tested in this study in the area of adult professional development included two types of statements relating the content of the lesson to their job, and stating implications of the information presented to their job environment. When these elaborations were embedded in the CAI training, significant gains were found; however, there were no differences between the leamer-generated or experimenter-provided elaborations. Teaching students how to generate elaborations at the time they will need them appears to be consistent with "just-in-time" training, especially when the technique may be new or more mentally difficult to implement.

31.2.4 Combination and Comparison of Coding and Integration Strategies

Camine and Kinder (1985) expanded on the Anderson and Kulhavey study on imaging. In their investigation, elementary school subjects were asked to form an image, then verbalize it, and were then given corrective feedback. They were also asked to create a summary at the end. This strategy was compared to a "schema-based strategy" in which learners were asked structurally related questions about the passage composition. They found significant gains in reading comprehension from pre- to posttests for both narrative and expository text for both treatments. One cannot be sure whether the positive results were due to the additional instructional effects of the feedback. Nevertheless, the question of the need for feedback on learner-generated activities is an important one since significant differences favoring adjunct questioning over the imaging strategy were observed for learning of expository materials.

Linden and Wittrock (198 1) conducted a study with elementary children that found that students who were asked to generate text-related summaries, analogies, metaphors, and pictures had better comprehension than those who were not. When instructed to generate images before verbal explanations, students produced more generations.

Wittrock and Alesandrini (1990.) also investigated the effects of leamer-generated summaries and analogies by analytic and holist undergraduates. The results followed the

predicted rank ordering, with the most positive effects found for generating summaries, followed by generating analogies, both of which were significantly better than the control group, which contained no generative activities. They also found that individual differences of analytic and holist ability correlated with learning differently in the three treatments: analytic ability with learning in the generate analogies group, the holist ability with the text-only control group, and both analytic and holist abilities in the generate summaries treatment.

Finally, Hooper, Sales, and Rysavy (1994) tested undergraduates on achievement efficiency and generations when given summaries and analogies while working alone and in pairs. They found that those who generated summaries performed better than those who generated analogies. Contrary to expected predictions, students working alone did better than those working in pairs.

When using a combination of strategies, the difficulty of the task must be taken into consideration, and, where possible, the effects of cognitive strengths must be factored in. Imaging is a more difficult task than adjunct questions, and analogies more difficult than summaries. If learners are not developmentally ready for such a task, it may cause more frustration than positive effects.

Kourilsky and Wittrock (1992), in a very powerful study, investigated the effect of teaching the overall generative model of teachi ng, including its four processes and activities, to senior high school students. The seniors were taught economics in cooperative learning groups. Those students who were taught this way of thinking were found to be more confident, had significantly fewer misconceptions, andhad greater comprehension than those without this training. A fascinating result consistent with the Hooper, Sales, and Rysavy (1994) study was thatjust using cooperative learning groups alone did norproduce as great an effect.

Kourilsky (1993) taught professional teachers generative teaching strategies and economic misconceptions. She found that pre- to posttest gain on exams of comprehension and misunderstanding were significant when misconceptions were clarified.

As can be seen, a variety of studies reporting on results of generative strategies have been summarized here. This section is not intended to be exhaustive; rather the studies have been selected as representative of the kind of research that has been conducted across content areas, learning types, and age levels. However, all articles that could be found that specify generative learning as the theory being tested are included. In general, results have shown increased gains in learning when the learner is an active vs. a passive participant in the learning process and when instruction includes activities that relate new information together and new information to prior knowledge. These studies on generative learning have shown that in most cases, active learner involvement produced increased leaming-i.e., learner-generated activities have resulted in significant gains in learning, although issues of organization of lesson content and quality of response may affect the degree of the effect.