18.5 AN EXAMINATION OF THE PROPOSITIONS OF A CONDITIONS-BASED THEORY

As noted in introduction to this chapter, the primary propositions of conditions-based theory can be summarized to four main assertions: (a) Learning goals can be categorized as to learning outcome or knowledge type; (b) related to (a), different outcome categories require different internal conditions (or, one can view the proposition as "different internal conditions leading to different cognitive outcomes"); (c) outcomes can be represented in a prerequisite relationship; (d) different learning outcomes require different external conditions for learning. In this section, issues relating to each of the primary propositions will be discussed.

18.5.1 Learning Outcomes Can Be Categorized

What is meant by a learning outcome? The meaning we attribute to "outcomes" differs depending on whether we perceive these outcomes as external (as a category of task or goal) or internal (as an acquired capability, perhaps supported by a unique memory system). Gagné (1985) clearly described his classification system of outcomes as "acquired capabilities," an internal definition. Merrill (1983) has described his outcome categories as "perfon-nances," "categories of objectives," and "learned capabilities," rather a mix of internal and external connotations. Reigeluth's categorization is of "types of content," which somewhat implies the categorized of an external referent. Landa describes his kinds of knowledge as "psychological phenomena," suggesting an internal orientation. Clearly, there is no consensus even within the models described in this chapter as to what the term learning outcomes actually implies. Indeed, the evidence to support the validity of each category system would vary in its type and complexity, depending on whether the phenomena are viewed as entities "out there" that can be pinned down and observed, or "within," where we only see circumstantial evidence of their presence.

The statement "learning outcomes can be categorized" is both a philosophical and psychological assertion. Indeed, both philosophers, such as Ryle (1949), and psychologists, such as Anderson (1990), have posited ways to categorize knowledge. Interestingly, Ryle and Anderson agreed on a similar declarative/procedural classification system. Certainly, instructional theorists have suggested a variety of category systems. (However, most are compatible with the declarative/procedural classification. Gagné certainly adds additional categories to these: attitude, motor skill, and, perhaps, cognitive strategies. Tennyson and Rascb add a third class of learning, contextual knowledge.) For each group, the philosopher, the psychologist, and the instructional theorist, the evidence for the "truth" of the proposition would vary. For philosophers, this is an epistemological question, and the manner for determining its truth would depend on the philosophic school to which a particular philosopher ascribes. We will not pursue this approach for determining the validity of our assertion directly.

Reigeluth (1983) suggests a utility criterion for determining whether.a categorization system is appropriate:

When we say concepts are human-made and arbitrary, we mean phenomena can be conceptualized (i.e., grouped or categorized) in many alternative ways.... Practically all classification schemes will improve our understanding of instructional phenomena, but concepts are not the kind of knowledge for which instructional scientists are looking, except as a stepping stone. Instructional scientists want to determine when different methods should be used-they want to discover principles of instruction-so that they can prescribe optimal methods. But not all classification schemes are equally useful for forming highly reliable and broadly applicable principles.... The same is true of classes of instructional phenomena: Some will have high predictive usefulness and some will not. The challenge to our discipline is to find out which ones are the most useful (pp. 12-13).

The psychologist would want empirical evidence that the categories are distinct, which leads to our second proposition.

18.5.2 Different Outcome Categories Require Different Internal Conditions

Most of the models within the conditions-based theory propose that learning categories are different in terms of cognitive-processing demands and activities. All of the major seven design models described in this chapter appear to make this assumption, to a greater or lesser degree. Although all models in this chapter suggest that a general information-processing procedure occurs in learning, they also suggest that this processing significantly and predictably differs for each of the categories of learning that they have identified. For example, R. Gagné suggested that in particular the cognitive processes of retrieval of prior knowledge, encoding, and retrieval and transfer of new learning would differ significantly in nature, depending on the type of learning goal. Indeed, several of the model developers, including Tennyson and Rash (1988), R. Gagné (1985), Merrill (1983), and Smith and Ragan (1993), postulated different memory structures for different types of learning outcomes.

A slightly different statement of the proposition allows for a closer relationship to the first proposition (outcomes can be categorized): Different internal conditions lead to different cognitive outcomes. This more descriptive (and less prescriptive) assertion seems to be supported by additional educational theorists. For example, both Anderson (1990) and E. Gagné (1993) propose that different cognitive processes lead to declarative and procedural learning. They also propose that these two types of learning have different memory systems, schemata for declarative knowledge and productions for procedural learning. They both provide some empirical evidence that these cognitive processes and storage systems are indeed unique to the two types of learning.

We must point out that even if connectionists (Bereiter, 1991) are correct, that there is only one memory system (neural networks) and only one basic cognitive process (pattern recognition), this does not necessarily preclude the possibility of different types of learning capabilities. For example, there may be generalized activation patterns that represent certain types of learning.

18.5.3 Outcomes Can Be Represented in a Prerequisite Relationship

Gagné's work on learning hierarchies would appear to be sufficient to confirm this assumption rather resoundingly, as reported previously in this chapter. In addition to work by Gagné and others working directly in his tradition, research by individuals working from entirely different frames of reference appears also to solidly confirm this assumption.

Although early learning hierarchy research appeared highly confirmatory, R. T. White developed an important review of learning hierarchy research in the early 1970s (White, 1973). In this review, studies validating the idea of learning hierarchies were sought. Due to methodological weaknesses, White found no studies that were able to validate a complete and precise fit between a proposed learning hierarchy and optimal learning: "All of the studies suffered from one or more of the following weaknesses: small sample size, imprecise specification of component elements, use of only one question per element, and placing of tests at the end of the learning program or even the omission of instruction altogether" (White, 1973, p. 371).

In research following White's review, research that applied his recommendations to correct methodological weaknesses, a series of studies providing confirmation of the learning hierarchy fon-nulation were published. (White 1974a, b, c; Linke, 1973) These results led Gagné to conclude: "The basic hypothesis of learning hierarchies is now well established, and sound practical methods for testing newly designed hierarchies exist" (White & Gagné, 1974, p. 363). Other research from what may be considered within the Gagné tradition which appears to confirm the learning hierarchy hypothesis includes Resnick, 1967; Resnick and Wang, 1969; Merrill, Barton, and Wood, 1970; and Linke, 1973.

Work on learning hierarchies from outside the Gagnd tradition or a conditions theory perspective includes studies by Winkles, Bergan, and associates, and Kallison. Winkles (1986) investigated the learning of trigonometry skills with a learning hierarchy validation study identifying both lateral and vertical transfer. Two experiments with eighth- and ninthgrade students involved instructional treatments described as "achievement with understanding" and "achievement only." Results reported "achievement with understanding treatment is better for the development of lateral transfer for most students, and of vertical transfer for the more mathematically able students, whereas the differences between the treatment groups on tests of achievement and retention of taught skills are not significant. A small amount of additional instruction on vertical transfer items produces much better performance under both treatments" (p. 275).

Bergan, Towstopiat, Cancelli, and Karp (1982), also not working from the conditions tradition, reported a study that provided what appears to be a particularly interesting form of confirmation of the learning hierarchy concept and some insights into rule learning:

This investigation examined ordered and equivalence relations among hierarchically arranged fraction identification tasks. The study investigated whether hierarchical ordering among fraction identification problems reflects the replacement of simple rules by complex rules. A total of 456 middleclass second-, third-, and fourth-grade children were asked to identify fractional parts of sets of objects. Latent class techniques reveal that children applied rules that were adequate for simple problems but had to be replaced to solve more complex problems (Bergan, Towstopiat, Cancelli & Karp, 1982, p. 39).

In a follow-up study to the 1982 work, Bergan, Stone, and Feld (1984) employed a large sample of elementaryage children in their learning of basic numerical skills. Students were presented with tasks that required rules of increasing complexity. The researchers were again studying the replacement of relatively simple rules with more complex extensions of them:

Hypotheses were generated to reflect the assumption of hierarchical ordering associated with rule replacement. In addition, restrictive knowledge and variable knowledge perspectives were evaluated. Latent-class models were used to test equivalence and ordered relations among the tasks. The results provided evidence that the development of counting skills is an evolving process in which parts of a relatively simple rule are replaced by features that enable the child to perform an increasingly broad range of counting tasks. The results also suggested that rule replacement in counting plays an important role in the development of other math skills. The results also give support for the restrictive knowledge perspective, lending credence to the stair-step learning theory (Bergan, Stone & Feld, 1984, p. 289).

An unusual and indirect, but interesting and suggestive, view of the importance of hierarchies in learning intellectual skills is found in a study by Kallison (1986), who varied sequence (proper vs. manipulated, i.e., reasonable vs. modified to disrupt clarity) and explicitness of lesson organization (organization of lesson explained/organization hidden). In the disrupted sequence treatment, even though care was taken to make an unclear presentation, the hierarchical nature of content relationships was preserved. Four treatments resulted and were used with three ability levels (2 X 2 X 3). In the study, 67 college students were taught intellectual skills: numeration systems, base 10 and base 5, and how to convert from one system to the other. Although sequence modification did not affect achievement substantially, the explicitness of lesson organization explicit did significantly impact achievement, with the more explicit lesson structure promoting better learning. Kallison found no aptitude-treatment interactions.

Kallison was careful to point out that although the sequence was altered, nothing got in the way of learning prerequisites. He modified sequence in such a way that learning hierarchies were not interfered with, only the reasonableness or "clarity" of the lesson organization: Where care was taken not to violate learning hierarchy principles, sequence could be disrupted, and it did not impact on learning, even with unclear presentation. As the learning task clearly involves intellectual skills, Gagné's principle of sequencing according to learning hierarchies was not violated. Although there is considerable evidence to validate learning hierarchies already, an unusual confirmation could be obtained by replicating Kallison's study with an additional condition of sequence modified in such a way as to violate learning hierarchy principles but maintain "clarity."

In another unusual test of the validity of the idea that learning tasks can be productively cast in a prerequisite relationship, Yao (1989) sought to test Gagné's assumption that in a validated learning hierarchy, some learners should be able to skip some elements based on their individual abilities. A valid learning hierarchy represents the most probable expectation of greatest learning for an entire sample. In a carefully designed experiment, Yao confirmed that some individuals could successfully skip certain prerequisites, and she found a treatment by ability interaction regarding the pattern of skipping in which certain forms of skipping can be less detrimental for high-ability learners than for lowability learners. However, as the theory predicts, the treatment that skipped prerequisites was less effective for both low- and bigh-ability learners (as a group).

18.5.4 Different Learning Outcomes Require Different External Conditions

In an effort to find evidence in support of this basic tenant of the conditions theory, we engaged in a survey of research, looking across a wide scope. The following research is presented in an effort to survey the evidence. The reader may find a dizzying variety of approaches and perspectives reflected. Studies and reviews on the following topics will be briefly presented to illustrate the variety of standpoints from which evidence may be found in general support of the conditions model: interaction between use of objectives and objective type, goal structure and learning task, advance organizers and learning task, presentation mode (e.g., visual presentation) and learning task, evoked cognitive strategies and learning outcomes, expertise and learning hierarchies, teacher thinking for different types of learning, adjunct questions and type of learning, feedback for different types of learning, and provided versus evoked instructional support for different types of learning. What follows, then, is a sample of studies that lend support-in varying ways from varying standpoints-to the theory that different instructional outcomes may best be achieved with differing types of instructional support.

18.5.4.1. Interaction of Use of Objectives and Objective Type. Hartley and Davies (1976) subjected to further examination a review by Duchastel and Merrill (1973) on the effects of providing learners with objectives. Although the original Duchastel and Merrill review found no effect, Hartley and Davies found that "behavioral objectives do not appear to be useful in terms of ultimate posttest scores, in learning tasks calling for knowledge and comprehension. On the other hand, objectives do appear to be more useful in higher-level learning tasks calling for analysis, synthesis, and evaluation" (p. 250). They also note a report by Yellon and Schmidt (1971) which pointed out a possible interference effect from informing students of objectives in problem-solving tasks by reducing the amount of reasoning required.

18.5.4.2. Goal Structure and Learning Task. Johnson and Johnson (1974) found in a review of research on cooperative, competitive, and individualistic goal structures that goal structure interacted with learning task. "Competition may be superior to cooperative or individualistic goal structures when a task is a simple drill activity or when sheet quantity of work is desired on a mechanical or skill-oriented task that requires little if any help from another person" (p. 220). They cite Chapman and Feder, 1917; Clayton, 1964; Clifford, 1971; Hurlock, 1927; Julian and Perry, 1967; Mailer, 1929; Miller and Hamblin, 1963; Phillips, 1954; Sorokin, Tranquist, Parten, and Zimmerman, 1930; and Tripplet, 1897. All findings do not clearly distinguish a grouping-by-outcomes (declarative/procedural) condition. For example, Smith, Madden, and Sobel, 1957; Yuker, 1955, found that memorization learning is also enhanced by cooperative work.

On the other hand, Johnson and Johnson pointed out: "When the instructional task is some sort of problem-solving activity, the research clearly indicates that a cooperative goal structure results in higher achievement than does a competitive goal structure" (p. 220). They cite Almack, 1930; Deutsch, 1949a; Edwards, DeViies, and Snyder, 1972; Gurnee, 1968; Husband, 1940; Jones and Vroom, 1964; Laughlin and McGlynn, 1967; O'Connel, 1965; Shaw, 1958; Wodarski, Hamblin, Buckholdt, and Feritor, 1971.

18.5.4.3. Visual Presentation Mode and Learning Task. Dwyer and Parkhurst (1982) present a multifactor analysis (three methods X four outcomes X three ability levels-reading comprehension). This analysis did not concentrate on different types of objectives, but apparently because different contents were used, the authors could draw this conclusion: "The results of this study indicated that (a) different methods of presenting programmed instruction are not equally effective in facilitating student achievement of all types of educational objectives" (p. 108). There were four measures, which were taken to represent four different types of learning outcome: (a) a drawing test involving generation of drawings given labels for parts of the heart such as aorta, pulmonary valve, and so forth; (b) identification test: a multiple-choice test of matching nature on various heart parts; (c) a terminology test consisting of 20 multiple-choice items on knowledge of facts, terms, and definitions; and (d) a comprehension test of 20 multiple-choice items that involved looking at the position of a given heart part during a specified moment in its functioning.

Analysis of the interactions among the different outcomes was not presented in the 1982 study; however, in what appears to be a follow-up study, Dwyer and Dwyer (1987) report the analyses of interactions. The authors conclude that "all levels of depth of processing are not equally effective in facilitating student achievement of different instructional objectives" (Dwyer & Dwyer, 1987, p. 264). In Dwyer's studies, tasks requiring "different levels of processing" appear to these reviewers as generally reflecting differing ways of eliciting declarative knowledge learning, yet meaningful differences among learning tasks were seen and reported by the authors of the studies.

18.5.4.4. Evoked Cognitive Strategies and Learning Outcomes. Kiewra and Benton (1987) report a study that investigated relationships among note taking, review of instructor's notes, and use of higher-order questions and their effect on learning of two sorts: factual and higher order. Subjects were college students in a college class setting. Half of the class was in a condition in which they took notes themselves and reviewed them, and the other half reviewed notes provided by the instructor. At the conclusion of the class, additional practice questions of a "higherorder" nature were provided to half of each group. An interaction between methodology and learning outcomes was reported. "Students who listed and reviewed the instructor's notes achieved more on factual items than did note takers, and . . . higber-order practice questions did not differentially affect test performance" (p. 186).

A study along lines similar to Kiewra and Benton's (1987) study was conducted by Shrager and Mayer (1989), in which some students were instructed to take notes and others were not so instructed, as both groups watched videotaped information. The researchers predicted that the "note taking would result in improved problem-solving transfer and semantic recafl but not verbatim recognition or verbatim fact retention for low-knowledge learners, but would have essentially no effects on test performance for high-knowledge leamers" (p. 263). This prediction was confirmed, supporting similar findings by Peper and Mayer (1978, 1986), who used the same design but different contents, automotive engines, and statistics. This study is somewhat confounded in treatment and learner characteristics. Degree of declarative knowledge and the stage of transition from declarative to procedural (Anderson, 1990) is often the distinction between novice and expert. Instead of indicating that declarative knowledge and procedural knowledge require different instructional conditions, the study may reveal, instead, that novice learners need more direct and explicit learning guidance in employing cognitive strategies that more knowledgeable learners will do on their own.

There is no doubt that properly applied to the proper task, the mnemonic keyword technique is a powerful one in assisting learning: "The evidence is overwhelming that the use of the keyword method, as applied to recall of vocabulary definitions, greatly facilitates performance. . . . In short, keyword methods effects are pervasive and of impressive magnitude" (Pressley, Levin & Delaney, 1982, pp. 70-7 1). The strategy, like many others, is a task-specific one: In other words, it makes no sense to apply it to other-than appropriate tasks. Levin (1986) elaborates on this principle and brings to bear an enormous amount of research from him and his associates on particular cognitive strategies (learning strategies) that have considerable power in improving learning.

18.5.4.5. Expertise and Learning Hierarchies. The utility and validity of learning hierarchies within authentic contexts has been studied by Dunn and Taylor (1990) and Dunn and Taylor (1994). In these studies, hierarchical analyses were performed on the activities of language arts teachers (1990) and medical personnel (1994). Development of expertise is encouraged to take place from "task-relevanf' experience, assisted by advice strategies developed from hierarchical analysis.

18.5.4.6. Adjunct Questions. Hamilton (1985) provides a review of research on using adjunct questions (see 30.61) and objectives in instruction. The review contains different sections on research with use of adjunct questions with different types of learning, leading to conclusions that vary with type of learning in question.

18.5.4.7. Practice. Some inconsistency is found in the results of studies looking at interaction of practice and types of learning. Hannafin and Colamaio (1987) found a significant interaction between practice and type of learning. Scores on practiced items were higher than nonpracticed items for each type of learning, but the effects were proportionately greatest for factual learning and least influential for procedural learning. However, in a study by Hannafin, Phillips, and Tripp (1986), opposite results were obtained, in which practice was more helpful for factual learning than for application learning. Slee (1989), in a review of interactive video research, noted that a lack of adequacy in lesson materials may confound these studies, as they both used the National Gallery of Art Tour videodisc, which was noted to have insufficient examples and practice available.

Rieber (1988) investigated effects of practice and animations on learning of two types: factual learning and application learning in a CBI lesson. The study looked at both immediate learning and transfer to other learning outcomes. Main-effect differences were not observed between either different elaboration treatments or practice. However, a significant interaction was found between learning outcome and transfer, in which the lesson promoted far transfer for factual information but did not facilitate far transfer for application learning. Another interaction was observed between practice and learning outcome, in which practice improved students application scores more than factual scores. As with studies by Hannafin and associates, unintended attributes of lesson materials may have confounded the study; in this case, as reported by the researcher, the lesson materials may have been too difficult.

18.5.4.8. Feedback for Different types of Learning. Getsie, Langer, and Glass (1985) provided a meta-analysis of research on feedback (reinforcement versus punishment) and discrimination learning. They concluded that punishment is an effective form of feedback for discrimination learning: "Punishment is clearly superior to reward only, with effect sizes ranging,from .10 to .31" (p. 20). The authors also concluded that reward is the least effective: "First, the most consistent finding is that compared to punishment or reward plus punishment, reward is the least efficient forrn of feedback during discrimination learning" (p. 20). Although discrimination learning was not compared with other forms of learning, we predict that this conclusion should not be generalized to other forms of learning (e.g., to provide punishment as feedback for practice in learning relational rules, as compared with informative feedback) or to other forms of feedback, such as levels of informational feedback.

Smith and Ragan (1993b) present a compilation of research and practice recommendations on designing instructional feedback for different learning outcomes. Using the Gagné types of learning construct as a framework, they present feedback prescriptions for different categories of learning task. They conclude that "questions regarding the optimal content of feedback ... really revolve around the issue of the match between the cognitive demands of the learning task; the cognitive skill, prior knowledge, and motivations of the learners; and constraints, such as time, within the learning environment" (p. 100).

An interesting insight into feedback and different types of learning is provided by a meta-analysis of research on feedback by Schimmel (1983). In attempting to explain the major inconsistencies in findings, Schimmel speculated such different characteristics of the instructional content as "different levels of difficulty in recall" (p. 11).

18.5.4.9. Provided vs. Evoked Instructional Support for Different Types of Learning. Husic, Linn, and Sloane (1989) report a study involving effects of different strategies for different types of learning. The content was learning to program in Pascal. Two different college classes were studied, a beginning class in which the learning task was characterized as "learning syntax" (perhaps analogous to rule using) and an advanced class that concentrated on "learning to plan and debug complex problems" (perhaps analogous to problem solving). The abstract of the report showed that:

Programming proficiency varied as a function of instructional practices and class level. Introductory students benefited from direct instruction, and AP students performed better with less direct guidance and more opportunities for autonomy. Characteristics of effective programming instruction vary depending on the cognitive demands of courses (Husic, Linn & Sloane, 1989, p. 570).

18.6 CONCLUSIONS

There are some conclusions we would draw from this review: 1. It appears that conditions models have a long history of interest in psychology, educational psychology, and instructional technology. This history illustrates work that may not be widely known among instructional technologists today; work that can be instructive as to the actual base and significance of the conditions approach. Perhaps we will see fewer erroneous statements in our literature about what is known regarding types of learning, learning hierarchies, and conditions of learning.

2. There appears to be continuing interest in this area, due to its utility in helping specify instructional strategies and also due to the sizable gaps and inconsistencies that exist in current formulations and research on and with them. We have described in this chapter many fruitful areas for further research.

3. We have reached a conclusion about the work of R. M. Gagné which we would like to share, and suggest that readers examine their own conclusions from reading. We find GagnCs work cast within so much that preceded it and which follows it to remain both dominating in its appeal and utility and, paradoxically, heavily flawed and in need of improvement. The utility and appeal of this work appears to derive greatly from the solid scholarship and cogent writing that Gagné brought to bear, as well as his willingness to change the fonnulation to keep up with changing times and new knowledge. Many of the gaps and flaws, in keeping with the paradox, appear to be a product of the very changes that he made to keep up with current interests. We believe those changes to be in the main beneficial, but see a clear need for systematic and rigorous scholarship on issues opened by those changes.

4. We still see utility in thinking of learning as more than one kind of thing, especially for practitioners. It is too easy, in the heat of practitioner's struggles, to slip to the assumption that all knowledge is declarative (as is so often seen in the learning outcomes statements of large-scale instructional systems) or all problem solving (as is so often assumed in the pronouncements of pundits and critics of public education), and, as a result, fail to consider either the vast arena of application of declarative knowledge or the multitude of prerequisites for problem solving. It is unhelpful to develop new systems of types of learning for the mere purpose of naming. Improvements in categorization schemes should be based on known differences in cognitive processing and required differences in external conditions.

5. There is substantial weakness in the tie from categories of learning to external conditions of learning. What is missing is the explication of the intemal conditions involved in acquisition of different kinds of learning. The research on transition from expert to novice and of artificial-intelligence research that attempts to describe knowledge of experts should be particularly fruitful in helping us fill this void. Perhaps this void is a result of failure to have a sufficient emphasis on qualitative research in our field.

6. There is research to support the conclusion that different external events of instruction lead to different kinds of learning, especially looking at the declarative/procedural level. What appears lacking is any systematic body of research directly on the central tenant, not just of conditions models but of practically anyone who would attempt to teach much less design instruction: What is the relationship between internal learner conditions and subsequent learning from instruction? Such a topic seems a far cry from studies that would directly inforrn designers as to procedures and techniques, yet such a great deal seems to hinge on that one question. With more insight into it, many quibbles and debates may disappear, and the work of translation into design principles may begin at a new level of efficacy.