blueprints for complex learning: the 4c/id- · pdf fileblueprints for complex learning: the...

Blueprints for Complex Learning: The 4C/ID-Model

Jeroen J. G. van MerriënboerRichard E. ClarkMarcel B. M. de Croock

This article provides an overview descriptionof the four-component instructional designsystem (4C/ID-model) developed originally byvan Merriënboer and others in the early 1990s(van Merriënboer, Jelsma, & Paas, 1992) forthe design of training programs for complexskills. It discusses the structure of trainingblueprints for complex learning and associatedinstructional methods. The basic claim is thatfour interrelated components are essential inblueprints for complex learning: (a) learningtasks, (b) supportive information, (c)just-in-time (JIT) information, and (d)part-task practice. Instructional methods foreach component are coupled to the basiclearning processes involved in complexlearning and a fully worked-out example of atraining blueprint for “searching forliterature” is provided. Readers who benefitfrom a structured advance organizer shouldconsider reading the appendix at the end ofthis article before reading the entire article.

The instructional design enterprise is a bitlike an ocean liner—huge, slow, ponderous, andrequiring large amounts of energy and a greatdeal of time to move it even one degree off itscurrent path. Recent discussions and develop-ments in the field concern rapid technologicaland societal changes and the resulting need forvery complex knowledge at work (Berryman,1993; Cascio, 1995); new constructivist designtheories for problem solving (Jonassen, 1994;Reigeluth, 1999a; Schwarz, Brophy, Lin, &Bransford, 1999); arguments for new contextand technology-based design (Driscoll & Dick,1999; Kozma, 2000; Richey, 1998); two decadesof systematic design research and developmentby John Anderson (1983, 1993; Anderson &Lebiere, 1998), and innovative work on “firstprinciples of instruction” by designer-researcherDavid Merrill (2000). These welcome discus-sions have at least one important goal in com-mon-the gradual evolution of design theory toaccommodate complex learning. Future designtheory should support the development of train-ing programs for learners who need to learn andtransfer highly complex cognitive skills or “com-petencies” to an increasingly varied set of real-world contexts and settings. In addition,adequate design for complex skills helps over-come findings that under some conditions, in-adequate design may cause learning problems(Clark, 1988).

The 4C/ID-model proposed in this article ad-dresses at least three deficits in previous instruc-tional design models. First, the 4C/ID-modelfocuses on the integration and coordinated per-formance of task-specific constituent skillsrather than on knowledge types, context or

ETR&D, Vol. 50, No. 2, 2002, pp. 39–64 ISSN 1042–1629 39

AAH GRAPHICS, INC. / (540) 933-6210 / FAX 933-6523 / 05-30-2002 / 12:29

presentation-delivery media. Second, the modelmakes a critical distinction between supportiveinformation and required just-in-time (JIT) in-formation (the latter specifies the performancerequired, not only the type of knowledge re-quired). And third, traditional models use eitherpart-task or whole-task practice; the 4C/ID-model recommends a mixture where part-taskpractice supports very complex, “whole-task”learning.

Novices learn complex tasks in a very dif-ferent way than they do simple tasks. Evidencefor this claim can be found in research on learn-ing concepts (Corneille & Judd, 1999), verbal in-formation (Pointe & Engle, 1990), mathematics(Wenger & Carlson, 1996), visual comparisontasks (Pellegrino, Doane, Fischer, & Alderton,1991) and a variety of complex work skills (Ack-erman, 1990), among others. Most designmodels emphasize instruction in relativelysimple learning tasks and assume that a large,complex set of interrelated tasks are achievableas “the sum of the parts”—by sequencing astring of simplified, component task proceduresuntil a complex task is captured. There is over-whelming evidence that this does not work (seevan Merriënboer, 1997, for an in-depth discus-sion of these issues). Existing design modelsmost often assume that knowledge of simpletask performance, once acquired, transfersreliably to novel future problems despite consid-erable evidence to the contrary (e.g., Clark &Estes, 1999; Perkins & Grotzer, 1997).

These relatively new insights about complexlearning are presented in a design theorydeveloped originally by van Merriënboer andothers in the early 1990s (van Merriënboer,Jelsma, & Paas, 1992). The complete design sys-tem and its psychological backgrounds aredescribed in van Merriënboer (1997; see also vanMerriënboer & Dijkstra, 1996, for its theoreticalbasis). This article presents an overview of themost recent version of the design theory, called4C/ID. It is a version of the model that currentlyprovides the basis for the development of com-puter-based design tools in a European projectcalled ADAPTIT (Advanced Design Approachfor Personalized Training—Interactive Tools).

An overview of the 4C/ID-model is given inthree parts. First, the elements of complex learn-

ing that must be accommodated in design aredescribed conceptually, using a concrete ex-ample of the skills necessary to search for docu-ments in a computerized database. Second, adescription is presented of the four “blueprintcomponents” (4C) that support complex learn-ing, namely (a) learning tasks; (b) supportive in-formation; (c) JIT information, and (d) part-taskpractice. Instructional methods are illustratedfor each component. Finally, the use of themodel for designing adaptive instruction is dis-cussed and some empirical studies that supportthe effectiveness of the model are brieflyreviewed. We will also briefly discuss cognitivetask analysis as a method for capturing advanceexpertise as content for complex training.

COMPLEX LEARNING

Complex learning is always involved withachieving integrated sets of learning goals—multiple performance objectives. It has little todo with learning separate skills in isolation, butit is foremost dealing with learning to coordinateand integrate the separate skills that constitutereal-life task performance. Thus, in complexlearning the whole is clearly more than the sumof its parts because it also includes the ability tocoordinate and integrate those parts. As an il-lustration, Figure 1 provides a simple descrip-tion of the constituent skills that make up themoderately complex cognitive skill, “searchingfor relevant research literature.” A well-designed training program for complex learningwill not aim at trainees’ acquiring each of theseconstituent skills separately, but will instead tryto achieve that the trainees acquire the ability touse all of the skills in a coordinated and in-tegrated fashion while doing real-life literaturesearches.

The skills hierarchy in Figure 1 depicts thetwo fundamental types of relations betweenconstituent skills that must be taken into accountwhen designing a training program (cf. Gagné’s“learning hierarchy,” Gagné, Briggs, & Wager,1992). First, there is a horizontal relationship be-tween coordinate skills that is indicated from leftto right. This relationship can be temporal (e.g.,you first select an appropriate database and then


40 ETR&D, Vol. 50, No. 2

formulate the search query for the selecteddatabase), simultaneous (e.g., you concurrentlyformulate a search query and perform the searchuntil you have a relevant and manageable list ofresults), or transposable (e.g., determining therelevant field of study and determining therelevant period of time can be done in any orderor even simultaneously). The second type ofrelation is the vertical relationship, which is in-dicated from bottom-to-top between child skillson a certain level and their parent skill one levelhigher. This relationship signifies that con-stituent skills lower in the hierarchy enable orare prerequisite to the learning and performanceof skills higher in the hierarchy (e.g., you mustbe able to operate a search program in order tobe able to perform a search). In an intertwinedhierarchy, additional relations between con-stituent skills that are important for trainingdesign may be added. For instance, similarityrelations may indicate constituent skills that areeasily mixed up.

Figure 1 also illustrates a typical charac-teristic of complex learning outcomes. Namely,for expert task performers, there are qualitativedifferences between constituent skills involved.

Some constituent skills are performed in a vari-able way from problem to problem situation.For instance, formulating a search query invol-ves problem solving and reasoning in order tocope with the specific requirements of each newsearch. Experts can effectively perform suchconstituent skills because they have highly com-plex cognitive schemata available that help themto reason about the domain and to guide theirproblem solving. Thus, schemata enable anotheruse of the same knowledge in a new problemsituation, because they contain generalizedknowledge, or concrete cases, or both, that canserve as an analogy.

Other constituent skills lower in the hierar-chy may be performed in a highly consistentway from problem to problem situation. For in-stance, operating the search program is a con-stituent skill that does not require reasoning orproblem solving. Experts can effectively per-form such constituent skills because theirschemata contain rules that directly associateparticular characteristics of the problem situa-tion to particular actions. In other words, rulesenable the same use of identical, situation-specific knowledge in a new problem situation.

Figure 1 Skills hierarchy for the moderately complex skill “searching for relevant researchliterature.” Nonrecurrent skills are represented in roman font, recurrent skills in italics.Double horizontal arrows with a solid line represent a simultaneous relationship;double horizontal arrows with a dotted line represent a transposable relationship(see text).


BLUEPRINTS FOR COMPLEX LEARNING 41

Experts may even reach a level of performancewhere they operate the search program fullyautomatically (unconsciously, without mental ef-fort). Conscious control is then no longer re-quired because the rules have becomeautomated. The result is that trained experts areable to focus their attention on other, non-automatic constituent skills while operating thesearch program.

Training programs for complex learningshould pay attention not only to the coordina-tion and integration of constituent skills, but alsoto these qualitative differences in desired exit be-havior of constituent skills. In order to identifythese qualitatively different performance objec-tives, constituent skills are classified as eithernonrecurrent or recurrent. For nonrecurrent(novel, effortful) constituent skills the desiredexit behavior varies from problem to problemsituation, and is guided by cognitive schematathat steer problem-solving behavior (cognitivestrategies) and allow for reasoning about thedomain (mental models). For recurrent (routine)constituent skills the desired exit behavior ishighly similar from problem to problem situa-tion, and is driven by rules that link particularcharacteristics of the problem situation to par-ticular actions. For instance, constituent skillsthat may be classified as recurrent for searchingfor relevant research literature are usingthesauri, using Boolean operators, and operat-ing the search program (in Figure 1, these recur-rent skills are printed in italics). Thisclassification is particularly important becausethe dominant learning processes for non-recurrent constituent skills are fundamentallydifferent from those for recurrent constituentskills.

For the nonrecurrent aspects of a complexskill and the complex skill as a whole, the mainlearning processes that must be promoted are re-lated to schema construction. From the viewpointof practice, learners should be encouraged tomindfully abstract away from the concrete ex-periences that are offered to them. Schemata arethen actively (re-)constructed in order to makethem more in agreement with concrete experien-ces. This process of induction is central to thedesign of learning tasks that offer the concreteexperiences in a training program for complex

learning. From the viewpoint of informationpresentation, learners should be encouraged toconnect newly presented information to alreadyexisting schemata, that is, to what they alreadyknow. This way, schemata are (re-)constructedand embellished with the new information thatis relevant to learning and performing the skill.This process of elaboration is central to the designof information that helps learners to perform thenonrecurrent aspects of a complex skill. It iscalled “supportive information.”

For the recurrent aspects of a complex skill,the main learning processes that must bepromoted are related to what is called rule auto-mation. Automation is mainly a function of theamount and quality of practice that is providedto the learners and eventually leads toautomated rules that directly control behavior.Rules are formed in two processes: First compila-tion, which embeds specific knowledge or infor-mation in the rules (proceduralization) andchunks rules together that are consistently ap-plied in the same order (composition), andsecond strengthening, which increases thestrength of a rule each time it is successfully ap-plied (Anderson, 1983, 1993, Anderson &Lebiere, 1998). The processes of compilationand, especially, subsequent strengthening arecentral to the design of part-task practice, whichoffers additional practice for selected recurrentconstituent skills in a training program for com-plex learning. From the viewpoint of informa-tion presentation, it is important to presentto-be-proceduralized information preciselywhen learners need it during practice. Thisprocess of restricted encoding of new informationinto cognitive rules is central to the design of in-formation that helps learners to learn and per-form the recurrent aspects of a complex skill. It iscalled JIT information and presented to learnersduring their work on learning tasks and duringpart-task practice.

To sum up, a training program for complexlearning must pay attention to the integrationand coordination of all skills that constitute acomplex cognitive skill (i.e., integrated objec-tives), and concurrently promote schema con-struction for nonrecurrent aspects and ruleautomation for recurrent aspects of the complexskill. By doing so, the training program aims at


42 ETR&D, Vol. 50, No. 2

transfer of learning—the ability to apply thecomplex cognitive skill in a wide variety of newreal-life situations. The familiar aspects thatlearners encounter in transfer situations can bedealt with thanks to the availability of rules,which also free up cognitive resources that maybe used to handle the unfamiliar aspects of thetransfer tasks. This process of rule-based trans-fer complies with the “component fluencyhypothesis” (e.g., Carlson, Khoo, & Elliot, 1990).Furthermore, unfamiliar aspects can be dealtwith thanks to the availability of complex cogni-tive schemata, which may be interpreted inorder to guide the problem-solving process andto reason about the domain. This process ofschema-based transfer produces reasonably ef-fective behavior for unfamiliar aspects of a prob-lem situation. The combination of the twotransfer processes is believed to allow for reflec-tive expertise because complex schemata mayalso be used to monitor and evaluate one’s ownperformance, including a reflection on thequality of solutions reached by the application ofrules. This complies with the “understandinghypothesis” (e.g., Ohlsson & Rees, 1991; see vanMerriënboer, 1997, for a complete discussion).Building on these assumptions, the next sectiondescribes the main blueprint components oftraining programs aimed at complex learning.

THE FOUR BLUEPRINT COMPONENTS

The basic message of the 4C/ID-model is thatenvironments for complex learning can alwaysbe described in terms of four interrelatedblueprint components. These components arebased on the four categories of learning proces-ses that are central to complex learning:

1. Learning Tasks: concrete, authentic, whole-task experiences that are provided to learnersin order to promote schema construction fornonrecurrent aspects and, to a certain degree,rule automation by compilation for recurrentaspects. Instructional methods primarily aimat induction, that is, constructing schematathrough mindful abstraction from the con-crete experiences that are provided by thelearning tasks.

2. Supportive Information: information that is

supportive to the learning and performanceof nonrecurrent aspects of learning tasks. Itprovides the bridge between learners’ priorknowledge and the learning tasks. Instruc-tional methods primarily aim at elaboration,that is, embellishing schemata by estab-lishing nonarbitrary relationships betweennew elements and what learners alreadyknow.

3. JIT Information: information that is prereq-uisite to the learning and performance ofrecurrent aspects of learning tasks. Instruc-tional methods primarily aim at compilationthrough restricted encoding, that is, embed-ding procedural information in rules. JIT in-formation is not only relevant to learningtasks but also to:

4. Part-task Practice: practice items that areprovided to learners in order to promote ruleautomation for selected recurrent aspects ofthe whole complex skill. Instructionalmethods primarily aim at rule automation,including compilation and subsequentstrengthening to reach a very high level ofautomaticity.

Component 1: Learning Tasks

A sequence of learning tasks is the backbone ofevery training program aimed at complex learn-ing (see Figure 2, which represents the learningtasks as circles). The learning tasks are typicallyperformed in a real or simulated task environ-ment and provide whole-task practice: ideally,they confront the learners with all constituentskills that make up the whole complex skill. It isimportant to stress that learning tasks shouldengage learners in activities that require them towork with the constituent skills, this as opposedto activities in which they have to study generalinformation about or related to the skills. For thenonrecurrent aspects of the complex skill andthe complex skill as whole (which is always non-recurrent), learning tasks promote schema con-struction by inductive processing. That is, thelearning tasks stimulate learners to constructcognitive schemata by mindfully abstractingaway from the concrete experiences that thelearning tasks provide. Learning processes like



generalization and discrimination subsequentlyreconstruct schemata to make them more in ac-cordance with new experiences. To-be-con-structed schemata come in two forms: (a) mentalmodels that allow for reasoning in the domain be-cause they reflect the way in which the learningdomain is organized, and (b) cognitive strategiesthat guide problem solving in the domain be-cause they reflect the way problems may be ef-fectively approached.

Task classes. It is clearly impossible to providehighly complex learning tasks right from thestart of the training program because this wouldyield excessive cognitive overload for thelearners, which impairs learning and perfor-mance (Sweller, van Merriënboer, & Paas, 1998).Thus, learners will typically start their work onrelatively simple learning tasks and progresstoward more complex tasks. Complexity is af-fected by the number of constituent skills in-volved, the number of interactions between

constituent skills, and the amount of knowledgenecessary to perform the constituent skills. Taskclasses are used to define simple-to-complexcategories of learning tasks and to steer theprocess of selection and development of suitablelearning tasks (see the dotted lines around thecircles in Figure 2). Task classes and not the in-dividual learning tasks define the basic se-quence of a training program developedaccording to the 4C/ID-model. Learning taskswithin a particular task class are equivalent inthe sense that the tasks can be performed on thebasis of the same body of knowledge (i.e., men-tal models and cognitive strategies). A morecomplex task class requires more knowledge ormore elaboration of knowledge for effective per-formance (cf., The Elaboration Theory,Reigeluth, 1999b; Reigeluth & Stein, 1983). Thebasic idea is to use a whole-task approach wherethe first task class refers to the simplest versionof whole tasks that experts encounter in the realworld. For increasingly more complex task clas-

Figure 2 A graphical view on the four components: (a) learning tasks, (b) supportiveinformation, (c) just-in-time (JIT) information, and (d) part-task practice.


44 ETR&D, Vol. 50, No. 2

ses the assumptions that simplify task perfor-mance are relaxed. The final task class repre-sents all tasks, including the most complex onesthat professionals encounter in the real world.

A simple illustration of this simplifying-as-sumptions approach1 might be given for themoderately complex skill, searching for litera-ture (see also Figure 1). The following task fac-tors can be identified that determine howcomplex it is to perform this skill: (a) the clear-ness of the concept definitions within or be-tween domains (ranging from clear to unclear);(b) the number of articles that are written aboutthe topic of interest (ranging from small tolarge); (c) the number of domains in whichrelevant articles have been published and hence,the number of databases that need to be sear-ched (ranging from one familiar database tomany databases that are relevant for the topic ofinterest); (d) the type of search (ranging from asearch on titles and key words to abstracts andfull text); and (e) the number of search terms andBoolean operators used (ranging from fewsearch terms to many search terms that are inter-connected with Boolean operators). Given thesefactors, the assumptions for the first, simplesttask class can be defined as follows: A categoryof learning tasks that confronts learners withsituations in which the search is performed in adomain in which the concepts are clearlydefined, on titles and keywords in one particulardatabase, with only few search terms and yield-ing a limited number of relevant articles. Themost complex task class is defined as a categoryof learning tasks that confronts learners withsituations where concept definitions within orbetween domains are unclear and in which full-text searches have to be performed in severalrelevant databases, with many search terms in-terconnected by Boolean operators in order tolimit the otherwise large number of relevant ar-ticles. Additional task classes of an intermediatecomplexity level can be added in between byvarying one or more of the task factors.

Once the task classes are defined, the learn-ing tasks can be selected and developed for each

class. For instance, one could ask an experiencedlibrarian to come up with concrete cases inwhich a successful search has been performedon titles in one particular database, with onlyfew search terms and yielding a limited numberof highly relevant articles (i.e., cases that fitwithin the first task class). The same is done forsubsequent, more complex task classes. Thecases that are selected for each task class formthe basis for the to-be-developed learning tasks.For each task class, enough cases are needed toensure that learners receive enough practice toreach mastery. It should be noted that the casesor learning tasks within the same task class arenot further ordered from simple to complex;they are considered to be equivalent in terms ofdifficulty. However, on this microsequencinglevel a high variability of the learning taskswithin the same task class is of utmost impor-tance (e.g., Gick & Holyoak, 1983; Paas & vanMerriënboer, 1994). They are best sequenced inrandom order and should differ from each otherin terms of the saliency of defining charac-teristics, the context in which the task has to beperformed, the familiarity of the task, or anyother task dimensions that also vary in the realworld. This high variability is necessary topromote the development of rich cognitiveschemata, which allow for schema-based trans-fer from the training program to the real world.

Learner support. While there is no increasing dif-ficulty for the learning tasks within one taskclass, they do differ with regard to the amount ofsupport provided to learners. Much support isgiven for learning tasks early in a task class, andno support is given for the final learning task ina task class. This process of diminishing supportas learners acquire more expertise is called“scaffolding.” It is repeated for each subsequenttask class, yielding a saw-tooth pattern of sup-port throughout the whole training program(see the filling of the circles in Figure 2). Ageneral framework of human problem solving(Newell & Simon, 1972) is used to distinguishsupport structures. According to thisframework, four elements are needed todescribe learners’ work on a learning task: (a)the given state that a learner is confronted with;(b) the criteria for an acceptable goal state; (c) a

1 More advanced approaches to the specification of taskclasses are described in van Merriënboer, 1997.



solution, that is, a sequence of operators thatenables the transition from the given state to thegoal state, and (d) a problem-solving process,which may be seen as the tentative applicationof mental operations in order to reach a solution.This framework is used to make a distinction be-tween product-oriented and process-orientedsupport. Product-oriented support only relatesto the first three elements: The given state, thegoal state, and the solution. Process-orientedsupport also takes the problem-solving processitself into account.

Product-oriented support is provided in alesser or higher degree by different types oflearning tasks. Highest product-oriented sup-port is provided by a case study or worked-outexample, which confronts the learner with agiven state, a desired goal state, and a solution,intermediate solutions, or both. In order toarouse learner interest, it may be desirable to usecase studies that describe a spectacular event,such as an accident, a success story, or a dis-puted decision that turned out all right. Typical-ly, learners have to answer questions thatprovoke deep processing and the induction ofmental models from the given examplematerials. By studying examples of intermediatesolutions, learners get a clear impression of howa particular domain is organized. At the otherextreme, no support is provided by a convention-al learning task, which provides only a givenstate and a desired goal state. Learners have tocome up with a solution themselves. Table 1presents some illustrations of other types oflearning tasks, roughly ordered from high tolow product-oriented support (see vanMerriënboer, 1997, for a complete description oftypes of learning tasks).

Process-oriented support is also directedtoward the problem-solving process itself.Highest process-oriented support is provided bya modeling example, which confronts the learnerwith an expert who is performing the task andsimultaneously explaining why the task is per-formed as it is performed. It is essential topresent an appropriate role model that hascredibility as well as expertise the observer cancomprehend. Thinking aloud may be very help-ful to bring the hidden mental problem-solvingprocesses of the expert into the open. As for case

studies, learners often have to answer questionsthat provoke deep processing and the inductionof cognitive strategies from the given modelingexample. By studying the modeling example,they get a clear impression of the systematicapproaches and rules of thumb that profes-sionals use.

Process-oriented support may also beprovided in the form of performance constraintsand performance support structures. Both arebased on a cognitive task analysis of strategicknowledge, which yields a description of cogni-tive strategies as systematic approaches to prob-lem solving (SAPs) that experts use to solveproblems in the domain of interest. An SAP dis-tinguishes the successive phases in a problem-solving process and the rules of thumb orheuristics that may be helpful to successfullycomplete each of the phases. Performance con-straints typically require learners to completeone phase satisfactorily before they may enterthe next phase. Performance support structures areless directive and typically take the form ofproblem-solving support. For instance, in orderto guide learners through the problem-solvingprocess, process worksheets that list the mainphases and useful rules of thumb for each of thephases may be provided to them. Or as a moreadvanced approach, computer-based learningtools may invite learners to approach the prob-lem at hand as an expert would do (for an ex-ample, see Dufresne, Gerace, Thibodeau-Hardiman, & Mestre, 1992).

Component 2: Supportive Information

Obviously, learners need information in order towork fruitfully on nonrecurrent aspects of learn-ing tasks and to genuinely learn from thosetasks. This supportive information provides thebridge between what learners already know andtheir work on the learning tasks. It is the infor-mation that teachers typically call “the theory”and which is often presented in study books andlectures. Because the same body of generalknowledge underlies all learning tasks in thesame task class, and because it is not knownbeforehand which knowledge is preciselyneeded to successfully perform a particular


46 ETR&D, Vol. 50, No. 2

learning task, supportive information is notcoupled to individual learning tasks but to taskclasses (see the supportive information in Figure2). The supportive information for each sub-sequent task class is an addition to or an elabora-tion of the previous information, allowinglearners to do things that could not be donebefore. Instructional methods for the presenta-tion of supportive information primarilypromote schema construction through elabora-tion, that is, helping students to establish non-

arbitrary relationships between newlypresented information elements and their priorknowledge. This process of elaboration yieldshighly complex schemata that should allow fordeep understanding.

As discussed in the previous section, the cog-nitive schemata that may help learners to per-form nonrecurrent aspects of a complex taskcome in two forms. (a) Mental models allow oneto reason within the learning domain, and (b)cognitive strategies allow one to systematically

Table 1 Examples of different types of learning tasks for the complex skill, searching forrelevant research literature. The learning tasks fit the first task class (see text) and areordered from high product-oriented support (case studies) to no support(conventional learning tasks).

Task Class: Learners are confronted with situations where concepts in the to-be-searched domain are clearly defined. Onlya small number of articles is written about the subject and articles are only written in one field of research. Therefore, thesearch only needs to be performed on titles of articles in one database from the particular field of research. Only a few searchterms are needed to perform the search and the search will yield a limited number of articles.

Learning Task Given(s) Goal(s) Solution Task Description

Case study + + + Learners receive a research question, a list with articles, and a search query used to produce the list of articles. They must evaluate the quality of the search query and the list of articles.

Reverse Predict + + Learners receive a list with articles and a search query used to produce the list of articles. They must predict possible research questions for which the list of articles and search query are relevant.

Imitation +Analog +Analog +Analog Learners have a worked-out example available of a + + Find research question, a list with articles, and a search

query used to produce the list of articles. They receive another research question and the goal to produce a list with a limited amount of relevant articles. By imitating the given example materials, they must formulate the search query, perform the search and make a selection of articles for the new research question.

A-specific goal + Define Find Learners receive a research question and a highly a-specific goal, for instance to come up with as many search queries as possible that might be relevant to the research question. They must formulate those search queries.

Completion + + Complete Learners receive a research question, the goal to produce a list with a limited number of relevant articles, and an incomplete search query. They must complete the search query, perform the search and make a selection of articles.

Conventional + + Find Learners receive a research question and the goal to produce a list with a limited number of relevant articles. They must formulate the search query, perform the search and make a selection of articles.

+ indicates that this element is provided to the learners.



approach problems in this domain and use rulesof thumb or heuristics that guide the problem-solving process. Supportive information reflectsboth types of schematic knowledge. For in-stance, it is known that Tiger Woods makes ex-tensive study of the layout of golf coursesaround the world (to develop mental models ofhow the world is organized) and of videotapesof his competitors (to develop cognitivestrategies of how to approach problems in thisworld). Thus even expert task performers fur-ther develop their mental models and cognitivestrategies in order to improve their perfor-mance. The same is true for learners in a trainingprogram aimed at searching for relevant re-search literature. In addition to working onlearning tasks as specified in Table 1, learnersmay, for instance, study how databases are or-ganized in order to develop useful mentalmodels, and they may study how expertlibrarians develop search queries in order todevelop more effective strategies themselves.

Mental models. Mental models are declarativerepresentations of how the world is organizedand may contain both general, abstractknowledge and concrete cases that exemplifythis knowledge. So, strong models allow forboth abstract and case-based reasoning. Mentalmodels may be viewed from different perspec-tives and can be analyzed as conceptual models,structural models, or causal models. First, con-ceptual models (what is this?) focus on how“things” are interrelated and allow for the clas-sification or description of objects, events or ac-tivities. For instance, knowledge about severaltypes of market stocks, and how these differfrom each other, helps financial analysts todetermine the risk associated with particularportfolios. Second, structural models (how is thisorganized?) describe how plans for reaching par-ticular goals are related to each other. Plans canbe distinguished in scripts (what happenswhen?) that focus on how events are related intime and help to understand and predict be-havior, and building blocks or templates (how isthis built?) that focus on how objects are relatedin space and help to understand or design ar-tifacts. For instance in the biology domain,knowledge about routine sequences of events

(i.e., scripts) occurring in a particular species ofbirds enables a biologist to predict and under-stand the ritual of mating behavior. In the com-puter-programming domain, knowledge aboutstereotyped patterns of programming code (i.e.,programming templates) and how these pat-terns fit together helps computer programmersto understand and develop programs. Third,causal models (how does this work?) focus on howprinciples affect each other and help to interpretprocesses, give explanations for events, andmake predictions. For instance, knowledgeabout how components of a chemical factoryfunction, and how each component affects othercomponents, helps process operators to diag-nose malfunctions. Mental models may alsocombine these three different perspectives andthus allow for qualitative reasoning in a par-ticular domain.

Central to mental models is the existence ofmany nonarbitrary relationships betweenknowledge elements. For the presentation ofsupportive information, it is of utmost impor-tance to stress those nonarbitrary relationships.Table 2 lists some popular instructional methodsthat help learners to identify relevant relation-ships. These methods can be used in an ex-pository fashion or in an inquiry fashion.Expository methods explicitly present the non-arbitrary relationships to learners. For instance,when learners study a particular piece ofmachinery one could explicitly indicate to themwhat the different parts of the machine are (seeMethod 1 in Table 2). Inquiry methods, on theother hand, ask the learners to “discover” therelationships. Thus, in the previous example oneshould ask the learners to identify the differentparts of the machine. Inquiry approaches aretime-consuming, but because they directly buildon learners’ prior knowledge they are very ap-propriate for interconnecting new informationand already existing cognitive schemata. It is aform of guided discovery, because the leadingquestions (e.g., which parts can be distinguishedin this machine?) help learners to identifyrelevant nonarbitrary relationships.

A particularly important relationship is theexperiential one, which relates general, abstractknowledge to concrete cases (see Method 2 inTable 2). The 4C/ID-model distinguishes the


48 ETR&D, Vol. 50, No. 2

presentation of general information (i.e., adidactical specification of conceptual, structuraland causal models) and concrete cases or casestudies that illustrate this information. If a con-ceptual perspective is taken, case studies maydescribe concrete objects, events, or situations.For models with a structural perspective, casestudies may be artifacts designed in order toreach particular goals. And for models with acausal perspective, case studies may illustratereal-life processes. Computer-based simulationsprovide a powerful approach to the presentationof case studies, because learners are then able tochange the settings of particular variables andstudy the effects of those changes on other vari-ables, that is, explore relationships (de Jong &van Joolingen, 1998). The goal of such“microworlds” is not primarily to practice thecomplex target skill, but to help learners con-struct mental models of how the world is or-ganized through active experimentation.

The 4C/ID-model furthermore distinguishesinductive and deductive strategies for present-ing supportive information. In an inductivestrategy one or more case studies are presentedas part of the supportive information; then the

general, abstract information is dealt with; andfinally the learning tasks are given. In a first typeof inductive strategy, the inductive-inquirystrategy, one presents one or more case studiesand then asks the learners to identify therelationships between pieces of information il-lustrated in the case. As described above, such aform of guided discovery is time consuming andshould be used only if there is enough instruc-tional time available, learners have no ex-perience with the skill whatsoever, and a deeplevel of understanding is required. A secondtype of inductive strategy is the inductive-ex-pository strategy: One starts with one or morecase studies and then explicitly presents therelationships between pieces of information thatwere illustrated in the cases. The 4C/ID-modelsuggests using this approach by default, becauseit is reasonable and time effective, and startingwith concrete, recognizable case studies workswell for learners with little prior knowledge. Thethird alternative is a deductive strategy, wherelearners work from the general, abstract infor-mation directly toward learning tasks that fulfillthe role of case studies. Here, one starts with ex-plicitly presenting relationships between pieces

Table 2 Ten popular instructional methods for the presentation of supportive information,stressing nonarbitrary relationships. Adding the prefix “Ask the learners to . . .”indicates an inquiry use of the method.

Instructional method Highlighted relationship(s)

1. (Ask the learners to . . .) analyze a particular idea Subordinate kind of or part of relationinto smaller ideas

2. (Ask the learners to . . .) present a well-known, Subordinate experiential relationfamiliar example or counterexample for a particular idea

3. (Ask the learners to . . .) present a more general idea Superordinate kind of or part of relationor organizing framework for a set of similar ideas

4. (Ask the learners to . . .) compare and contrast a set Coordinate kind of or part of relationof similar ideas

5. (Ask the learners to . . .) provide a description of a Subordinate kind of or part of relationparticular idea in its main features or characteristics

6. (Ask the learners to . . .) find an analogy for a Coordinate similarity relationparticular idea

7. (Ask the learners to . . .) explain the relative location Location relationof elements in space or time

8. (Ask the learners to . . .) rearrange elements and Location relationpredict effects

9. (Ask the learners to . . .) explain a particular state of affairs cause-effect or natural process relation10. (Ask the learners to . . .) make a prediction of future states cause-effect or natural process relation



of information (the theory) and then illustratesthis general information with one or more learn-ing tasks with maximum product-oriented sup-port (note that many instructors use this methodby default!). A problem is that learners withoutprior knowledge may have severe difficultieswith understanding the general information. Itshould thus be used only if instructional time islimited, learners have already some experiencewith the skill, and a deep level of understandingis not strictly necessary.

Cognitive strategies. Like mental models, cogni-tive strategies contain both general, abstractknowledge and concrete cases that exemplifythis knowledge. As mentioned before, cognitivestrategies may be analyzed as SAPs, describingthe successive phases in a problem-solvingprocess and the rules of thumb or heuristics thatmay be helpful to successfully complete each ofthe phases. Instructional methods for presentingcognitive strategies closely resemble methodsfor presenting mental models, and in particular,structural and causal models. For instance, onemight ask the learners to explain why one phaseshould precede another phase (see Method 7 inTable 2), predict the effects of rearrangingphases (Method 8), explain how the use of par-ticular rules of thumb brought a particular stateof affairs about (Method 9), or predict the effectsof the use of particular rules of thumb (Method10). For cognitive strategies, the experientialrelationship refers to concrete cases that take theform of modeling examples, which illustratehow the application of SAPs can help to reach asolution. The modeling examples provide ahinge between the supportive information(where they illustrate cognitive strategies) andthe learning tasks (where they may be seen aslearning tasks with maximum process-orientedsupport). As argued before, modeling examplesmay refer to an expert who is performing a non-trivial task and simultaneously explaining whyparticular decisions and actions are taken (e.g.,by thinking aloud). Preferably, the example isinterspersed with questions that require thelearners to think critically about the problem-solving process that is modeled. Because of thehighly abstract character of cognitive strategies,the 4C/ID-model prescribes only an inductive-

expository strategy for their presentation. Thus,as a rule one, should start with the presentationof one or more modeling examples and then ex-plicitly present the problem-solving phases andrules of thumb that are illustrated by those ex-amples.

Cognitive feedback. A final part of supportive in-formation relates to feedback that is provided onthe quality of performance. This so-called cogni-tive feedback (Balzer, Doherty, & O’Connor,1989; Butler & Winne, 1995) refers to the non-recurrent aspects of performance only andshould thus promote schema construction. Be-cause nonrecurrent performance is never “cor-rect” or “incorrect,” but only more or lesseffective, cognitive feedback is only providedafter learners have finished one or more learningtasks, or even after they have finished a wholetask class. Well-designed feedback shouldstimulate learners to reflect on the quality oftheir personal problem-solving processes andfound solutions, so that more effective mentalmodels and cognitive strategies can bedeveloped. This central role of reflection isshared with a cognitive apprenticeship model(Collins, Brown & Newman, 1989; see alsoKluger & DiNisi, 1998). Debriefing sessions,peer or expert critiques, and group discussionsoffer a valuable approach. Then, learners’ actualproblem-solving processes may be comparedand contrasted with presented SAPs, modelingexamples that illustrated those SAPs, or prob-lem-solving processes reported by otherlearners. Further, found solutions may be com-pared and contrasted with presented general in-formation, case studies that illustrated thisgeneral information, or solutions found for pre-vious problems or reported by other learners. Insuch discussions, inquiry methods as presentedin Table 2 may well be used as a form of “feed-back by discovery.”

Component 3: Just-in-Time Information

Whereas supportive information pertains to thenonrecurrent aspects of a complex skill, JIT in-formation pertains to the recurrent aspects, thatis, constituent skills that should be performed


50 ETR&D, Vol. 50, No. 2

after the training in a highly similar way overdifferent problem situations. JIT informationprovides learners with the step-by-stepknowledge they need to know in order to per-form the recurrent skills. They can be in the formof, for example, directions teachers or tutorstypically give to their learners during practice,acting as an “assistant looking over yourshoulder.” Because the JIT information is identi-cal for many learning tasks, which all require thesame recurrent constituent skills, it is typicallyprovided during the first learning task for whichthe skill is relevant (see JIT information in Figure2). For subsequent learning tasks, JIT informa-tion is quickly faded away as learners gain moreexpertise (a principle called fading). Instructionalmethods for the presentation of JIT informationprimarily promote compilation throughrestricted encoding of situation-specificknowledge into cognitive rules. The JIT informa-tion is specified at the entry level of the learners,that is, at a level that is suitable to present to thelowest-level ability learner. It is not critical thatthe information be embedded in existingschemata in declarative memory. Because ofthis, during presentation, no particular referencehas to be made to related knowledge structuresin long-term memory.

Rules that enable learners to correctly per-form the recurrent aspects of a complex skill areformed through practice, and this process isfacilitated when the information that is neces-sary for forming the rules is directly available inworking memory, precisely when learners needit. This concerns information that describes therules themselves (or procedures that combinethose rules) as well as information describingthe knowledge elements (i.e., facts, concepts,plans or principles—the same knowledge ele-ments that make up complex schemata) that areprerequisite to learning and to performing thoserules. For instance, when you are learning toplay golf, your instructor will preferably tell youhow to hold your club, how to take your stance,and how to make your swing out on the drivingrange while you are making your first drives,and not during a theory lesson in a classroom.The same is true for learners in a training pro-gram aimed at searching for relevant researchliterature. For the recurrent aspects of this com-

plex skill, such as operating the search program,procedural directions for operating the programare also best presented during practice, preciselywhen learners need it. The next section discussesthe design of such information displays.

Information displays. JIT information is or-ganized in small units, called information dis-plays. Organization in small units is consideredto be essential because only the presentation ofrelatively small amounts of new information atthe same time can prevent processing overloadduring practice. Information displays include adidactical specification of the rules that describecorrect performance as well as the knowledgethat is prerequisite to a correct application ofthose rules. For instance, a rule may state that“in order to start the machine, you must firstswitch it on” and also indicate that the on-offswitch is located on the back of the machine (i.e.,a fact that is prerequisite to a correct applicationof the rule). Or in the context of searchingrelevant research literature, a rule for operatingthe search program may state that “in order tosearch on keywords, select the choice FIELDS inthe menu SEARCH and enter the desired searchterms in the field that is labeled KW.” The sameinformation display may give a definition of theconcept FIELD (i.e., a concept that helps to under-stand the given rule). These examples makeclear that information displays may best be char-acterized as how-to instruction or rule-based in-struction (Fisk & Gallini, 1989).

A traditional approach to JIT informationpresentation lets learners memorize the infor-mation before they start to work on learningtasks, so that it may be activated in workingmemory when needed. This approach is notrecommended for the simple reason thatmemorization is a dull activity and has no ad-vantages to more active JIT approaches. As aregular approach, one should directly presentinformation displays when the learners needthis information to work on the recurrentaspects of a particular learning task. It is thusconnected to the first learning task for which it isrelevant, and for subsequent learning tasks itfades away. However, a requirement of this ap-proach is that the designer have some controlover the learning tasks that confront the learner;



otherwise it is not feasible to connect the infor-mation displays to the learning tasks. If trainingtakes place on the job, the designer often lacksthis control. Learning aids such as on-line helpsystems, checklists, and manuals then provide agood alternative. While the JIT information isnot directly presented when it is needed for thelearning tasks, it is at least easily available andreadily accessible. In the field of minimalism(Carroll, Smith-Kerker, Ford, & Mazur-Rimetz,1988) guidelines for the design of “minimalmanuals” are in full agreement with this ap-proach (e.g., van der Meij & Carroll, 1995).

Demonstrations and instances. Most of the ele-ments in information displays are general state-ments about the recurrent skill, or, generalities(Merrill, 1983, 1999). For instance, rules aregeneral in that they can be applied in a variety ofsituations, and prerequisite concepts are generalin that they refer to a category of objects orevents. It is often desirable to present examplesthat illustrate or exemplify those generalities.For rules, such examples are called demonstra-tions; for concepts, plans, and principles, theyare called instances. The 4C/ID-model suggestsproviding demonstrations and instances in thecontext of the learning tasks. This should allowlearners to place the recurrent skill in the contextof the whole task. Thus, demonstrations of therecurrent aspects of a complex skill ideally coin-cide with suitable learning tasks such as model-ing examples, and instances of prerequisiteknowledge elements ideally coincide withsuitable learning tasks such as case studies. Thisis a deductive-expository approach, where thegeneralities (i.e., the information displays) arepresented simultaneously with the examples (i.e.,demonstrations and instances), which are part ofthe same learning task as the information dis-play is connected to.

Two examples illustrate this principle. Sup-pose that a complex skill in the process controldomain requires the execution of a standardprocedure (i.e., a recurrent constituent skill) todetect possible out-of-bound situations. Con-nected to the first learning task for which therecurrent skill is relevant, an information dis-play will provide a general description of theprocedure as well as its prerequisite knowledge.

Then, the procedure is best demonstrated as partof a modeling example, where the learner’s at-tention is focused on the recurrent aspects thatthe demonstration is about. Another example isin the domain of computer programming. Sup-pose that completion tasks are used asking thelearners to complete increasingly larger parts ofpartial, well-structured computer programs.When a particular programming plan (i.e., astereotyped pattern of programming code, suchas an assignment plan, a looping plan, etc.) isused for the first time in the given part of a to-be-completed program, an information displaymay be presented with a rule describing whenand how to use this plan, as well as a generaldescription of the plan that is prerequisite to theapplication of the rule. Note that at the sametime, a concrete instance of this programmingplan is presented in the given part of the to-be-completed computer program. In CompletionASsignment COnstructor (CASCO), an intel-ligent tutoring system for teaching introductoryprogramming (e.g., van Merriënboer & Luur-sema, 1996), the programming code that ex-emplifies the information display (i.e., theinstance) is highlighted in the to-be-completedprogram.

Corrective feedback. A final part of JIT informa-tion relates to feedback that is provided on therecurrent aspects of performance. Like all JIT in-formation, this feedback should promote com-pilation. If rules that algorithmically describeeffective performance are not correctly applied,the learner is said to make an “error.” Correctivefeedback on such errors is preferably presentedimmediately after misapplication of a particularrule. It is necessary for the learner to preserve in-formation about the conditions for applying aparticular rule in working memory until feed-back (right-wrong) is obtained. Only then can arule be compiled that attaches the correct actionto its critical conditions. Obviously, any delay offeedback may hamper this process.

The 4C/ID-model does not propagate theidea of errorless learning. On the one hand, it ispractically impossible to prevent errors whenlearners are working on rich learning tasks. Buteven more important, for many recurrentaspects of a complex skill it is considered impor-


52 ETR&D, Vol. 50, No. 2

tant that learners learn to recognize their own er-rors and how to recover from them. Well-designed feedback should then inform thelearner why there was an error and provide asuggestion or hint of how to reach the goal. Sucha hint will often take the form of an example ordemonstration. It is important not to simply givethe correct action because this does not allow forpractice, which is critical for compilation tooccur. Furthermore, it may be necessary to indi-cate to the learner how to recover from theresults of the error that has been made.

Component 4: Part-task Practice

Learning tasks are designed in such a way thatthey primarily promote schema construction,but they also facilitate compilation for recurrentaspects of the complex skill. This process isdriven by the repeated practice of recurrent con-stituent skills in the learning tasks. Often, learn-ing tasks provide enough opportunity topractice both the nonrecurrent and the recurrentaspects of the complex skill. This is possible be-cause one can take care of the different nature ofunderlying learning processes for recurrent andnonrecurrent constituent skills in the context ofinformation presentation. JIT informationpresentation aims at restricted encoding ofnewly presented information in rules; suppor-tive information presentation aims at elabora-tion of existing schemata with new information.However, if a very high level of automaticity ofparticular recurrent aspects is required, thelearning tasks may provide insufficient repeti-tion to provide the necessary amount ofstrengthening. Only then, it is necessary to in-clude additional part-task practice for thoseselected recurrent aspects in the training pro-gram (see part-task practice in Figure 2). But ingeneral, an overreliance on part-task practice isnot helpful to complex learning.

Part-task practice promotes the compilationof procedures or rules and especially their sub-sequent strengthening, which is a very slowprocess that requires extensive amounts of prac-tice items. Well-known examples of part-taskpractice are drilling children on multiplicationtables and playing scales on musical instru-ments. In training design, part-task practice is

typically applied for recurrent constituent skillsthat are critical in terms of safety, for instance,detecting dangerous air traffic situations from aradar screen in the context of air traffic control.But if available instructional time allows, it mayalso be used for recurrent constituent skills withrelations in the skills hierarchy indicating thatthey (a) enable the performance of many otherskills higher in the hierarchy (which is a centralidea of Gagné’s learning hierarchy, Gagné et al.,1992), or (b) have to be performed simultaneous-ly with many other coordinate skills. It is criticalto start part-task practice within an appropriatecognitive context because it has been found to beeffective only after exposure to a simple versionof the whole complex skill (Carlson et al., 1990;Schneider & Detweiler, 1988). One should thusidentify the first task class for which perfor-mance of the recurrent aspect is required, andinitiate part-task practice during this taskclass—preferably after case studies or otherlearning tasks with ample learner support havealready been worked on. This allows learners toidentify the activities that are required to in-tegrate the recurrent aspect in the learning tasks.The following sections briefly discuss the designof practice items for part-task practice andmethods for overtraining.

Practice items. Compared to the specification oflearning tasks, the specification of practice itemsfor part-task practice is a pretty straightforwardprocess. For learning tasks, simple-to-complextask classes first guide the process of selectingconcrete cases, which are then transformed intomeaningful learning tasks that require thelearners to perform several constituent skills in acoordinated fashion. For part-task practice,however, there is only one relevant recurrentconstituent skill or objective whose effective per-formance can be algorithmically described interms of rules. Practice items should then invitelearners to repeatedly perform the recurrentconstituent skill. The saying, practice makes per-fect, is actually true for part-task practice. It isimportant that the whole set of practice items bedivergent, meaning that it be representative forall situations that can be treated by the rules.This is necessary to develop a broad set of situa-tion-specific rules that may subsequently yield



optimal rule-based transfer to new problemsituations.

Only for highly complex algorithms, repre-sented by large rule sets, it may be necessary towork from simple to complex practice items.The whole algorithm is then decomposed intoparts, and learners are extensively trained oneach part separately before they begin to prac-tice the whole recurrent skill. This form of se-quencing is fundamentally different fromsequencing learning tasks. In order to facilitateschema construction, a whole-task approach isused to sequence simple-to-complex task clas-ses; the learning tasks within the same task classexhibit a high variability, and each learning taskrequires the integration and coordination of con-stituent skills involved. In contrast, breaking thetask down in parts that are separately trainedand then gradually combined toward the wholetask (i.e., a part-whole approach) yields a lowervariability and facilitates a rapid automation ofrules.

With regard to learner support, there is also astriking difference between support for learningtasks and support for practice items in part-taskpractice. The performance of recurrent aspectscannot be described as the tentative applicationof mental operations in order to find a solution(i.e., problem solving). Applying the rules simp-ly is the solution and ensures that the desiredgoal state is reached. It is thus the application ofrules that is of importance, rather than thesearch for a solution. So, performance supportfor part-task practice takes the form of proceduresupport. Special practice items may be relevant ifalgorithms leave learners error prone, or if dif-ferent algorithms may easily be mixed up. Forinstance, a well-known strategy for orderingpractice items is the recognize-edit-produce se-quence (REP, Gropper, 1983), which starts withitems that require learners to recognize whichrules to apply, continues with items for whichlearners have to edit incorrect applications of therules, and ends with conventional items forwhich learners have to apply the rules in orderto produce the solution. Performance constraintsfor part-task practice may take the form of“training wheel interfaces” (Carroll et al., 1988)which resemble the use of training wheels onchildren’s bikes. If particular rules leave learners

error prone, one may make the actions related tothose rules “unreachable” for the learners earlyin the training process. Such a training-wheelsapproach may also be used to support the learn-ing of recurrent aspects during whole-task prac-tice on learning tasks (e.g., see Leutner, 2000).

JIT information for part-task practice. JIT infor-mation is obviously not only relevant for learn-ing tasks, where it pertains to the recurrentaspects of the task, but also to part-task practice,where it relates to the single recurrent skill thatis practiced. Compared to JIT informationpresentation for learning tasks, one may nowcarry forward the principle of presenting JITeven further and provide the information that isrelevant for applying a particular rule and itsprerequisite knowledge precisely at the momentthat this one rule has to be applied by thelearner. This is known as single-step or step-by-step instruction (Landa, 1983). Furthermore,demonstrations of rule application and instan-ces of prerequisite knowledge cannot beprovided as part of a learning task (i.e., in theirwhole-task context), but are provided separatelyand simultaneously to the information displays.For instance, if part-task practice is provided fortraining special emergency procedures, infor-mation displays provide a step-by-step descrip-tion of the procedure and may defineprerequisite concepts such as “alarm limit” or“emergency setting.” A demonstration shouldclearly indicate the desired outcome of the pro-cedure, the materials and other pieces of equip-ment that will be manipulated (i.e., provideconcrete instances of the prerequisiteknowledge), and should show the actual execu-tion of the procedure using these materials.Feedback on the quality of performance shouldbe provided during practice, ideally, immedi-ately after performing a particular step in a pro-cedure or applying a particular rule.

Overtraining. Part-task practice as discussedabove will lead to accurate performance of arecurrent skill. However, extensive amounts ofovertraining may be necessary to make the skillfully automatic. Then, the main underlyinglearning process is no longer compilation butstrengthening. For skills that need to be per-


54 ETR&D, Vol. 50, No. 2

formed highly automatically, the ultimate goalis not always highest accuracy. More often, thegoal is to obtain acceptable accuracy, combinedwith high speed and the ability to perform theskill together with other skills, and ultimately, inthe context of the whole task. In order to reachthis, the recurrent skill (which is already beingperformed to the required level of accuracy) isfirst practiced under speed stress. After speedcriteria have been reached, the skill is practicedunder time-sharing conditions simultaneouslywith other effort-demanding skills. And finally,the skill is practiced in the context of the wholetask. So, performance criteria gradually changefrom (a) accuracy, to (b) accuracy combined withspeed, to (c) accuracy combined with speedunder time-sharing conditions or high overallworkload (Salisbury, Richards, & Klein, 1985).

Relatively short, spaced periods of part-taskpractice or overtraining (i.e., distributed prac-tice) yield better results than long, concentratedperiods of part-task practice (i.e., massed prac-tice). Therefore, part-task practice is best inter-twined with the learning tasks because thisprovides distributed practice and also enablesthe learners to relate the recurrent constituentskill to the whole complex skill. The same prin-ciple of intermix training is applied if part-taskpractice is provided for more than one recurrentconstituent skill. Practice on those skills is alsointertwined in order to distribute practice and tofacilitate the perception of interrelationships be-tween constituent skills (cf., Schneider, 1985).

This section ends the description of the fourcomponents, which have been graphically inter-connected to each other in Figure 2. Table 3 givesa simplified blueprint of a training program forsearching for relevant research literature in atextual format. This blueprint also illustrates thefour components. First, three task classes aredescribed, each containing several types of learn-ing tasks. The task classes show an increase incomplexity while the learning tasks within eachtask class show a decrease in learner support.Second, supportive information is specified foreach task class. In the first task class, an induc-tive-expository strategy is illustrated: Learnersfirst receive a modeling example before theystart to study supportive information that was il-lustrated in the modeling example. In the second

task class, an inductive-inquiry approach is il-lustrated: Learners work on a case study andhave to identify and discover the relationshipsbetween different plans (i.e., templates forsearch queries using Boolean operators) that areillustrated in the case. In the third task class, adeductive approach is illustrated: Supportive in-formation is made available before learners startto work on the first learning task. At the end ofthe second and third task class, learners receivecognitive feedback on their work on a conven-tional learning task. The third component, JIT in-formation, is specified for each learning taskwhere it is relevant. And fourth and last, it isspecified where to initiate additional part-taskpractice. In this blueprint, it is assumed thatlearners receive some additional part-task prac-tice in the use of Boolean operators, startingparallel to the first learning task that illustratestheir use.

DISCUSSION

In this article, we presented a description of thefour blueprint components that are the basicbuilding blocks for training programs for com-plex learning designed according to the 4C/ID-model, and their major theoretical foundations incognitive psychology. The four blueprint com-ponents refer to (a) learning tasks; (b) supportiveinformation; (c) JIT information, and (d) part-taskpractice. These components and their associatedinstructional methods were described in detailand an example of a training blueprint was givenfor the moderately complex skill searching forrelevant research literature (see Table 3).

The 4C/ID-model should be used to developtraining programs for complex skills and whentransfer is the overarching learning outcome.Such training programs have a typical length ofweeks, months or even years. The model is notdeveloped for teaching conceptual knowledgeor procedural skills per se. It also is not very use-ful for designing very short programs that onlytake an instructional time of hours or a few days(e.g., traditional lesson design or design of shortworkshops). If the model is used, a blueprint aspresented in Table 3 may not always provideenough detail to start the actual development of



Table 3 Simplified example of a training blueprint for the moderately complex skill, searchingfor relevant research literature.

Task Class 1: Learners are confronted with situations where the concepts in the to-be-searched domain are clearly defined.Only a small number of articles is written about the subject and articles are only written in one field of research. Therefore,the search needs only to be performed on titles of articles in one database from the particular field of research. There are onlya few search terms needed to perform the search and the search will yield a limited number of articles.

Supportive Information: Modeling exampleLearners watch an expert who performs a literature search and explains his or her actions while doing so.

Supportive Information: Presentation of cognitive strategies• Systematic approach to problem solving (SAP) of the four phases involved in performing a literature

search: (a) selecting an appropriate database, (b) formulating a search query, (c) performing the search, and (d) selecting results.

• SAPs for quickly scanning the relevance of scientific articles.Supportive Information: Presentation of mental models• Conceptual model of literature search concepts• Structural model of how databases are organized and can be used.• Conceptual model of different types of scientific articles and how they are organized.

Learning Task 1.1: Case studyLearners receive three worked-out (good) examples of literature searches. Each example describes a different research questions in the same subject matter domain, the search query and the produced list of articles. The learners have to study the examples and explain why the different search queries produced the desired results.Learning Task 1.2: Completion JIT* information presentationLearners receive a research question and an incomplete • Procedures for operating search query that produces a list containing irrelevant the search programitems. They must refine the search query using • Procedures for using additional search terms, perform the search and select a thesaurusthe relevant articles.Learning Task 1.3: Conventional JIT information presentationLearners receive a research question. They have to • Procedures for operatingperform a literature search for the 10 most relevant the search program articles. (fading)

• Procedures for using a thesaurus (fading)

Task Class 2: Learners are confronted with situations where the concepts in the to-be-searched domain are clearly defined. Alarge number of articles is written about the subject, but only in one field of research. Therefore, the search needs only to beperformed on titles of articles in one database from the particular field of research. However, many search terms need to beinterconnected with Boolean operators to limit the otherwise large number of articles the search can yield.

Supportive Information: Case studyLearners receive three worked-out (good) examples of literature searches. Each example contains an elaborate search query in which Boolean operators are used.

Table continues *JIT = Just-in-time


56 ETR&D, Vol. 50, No. 2

Table 3 continued

Supportive Information: Inquiry for mental models• Learners are asked to identify templates of search queries describing Boolean combinations of search termsthat can be used to make search queries more specific.Learning Task 2.1: Imitation + constraint JIT information Part-task PracticeLearners have a worked-out example of a research presentation • Applying Boolean question available, a list of articles and an elaborate • Syntax for specifying operators (continueBoolean search query to produce the list of articles. Boolean search queries as needed)They receive a similar research question, and a goal to produce a list with a limited number of relevant articles. By imitating the given example they must formulate the search query, perform the search and select relevant articles. They can only perform the search after the search query is approved.Learning Task 2.2: Completion JIT information Learners receive a research question and a list of presentationsearch terms. They have to formulate a search • Syntax for specifying query by combining the given search terms Boolean search queries using Boolean operators. (fading)Learning Task 2.3: Conventional JIT information Learners receive a research question. They have presentationto perform a literature search for the 10 most • Syntax for specifying relevant articles. Boolean search queries

(fading)

Supportive Information: Cognitive feedbackLearners receive feedback on their approach to solve the problem in Learning Task 2.3.

Task Class 3: Learners are confronted with situations where the concepts in the to-be-searched domain are not clearlydefined. Identical terms are used for different concepts, and identical concepts are described with different terms. A largenumber of articles is written about the subject and articles are written in several fields of research. Therefore, next tosearching on titles of articles, the search also needs to be performed on abstracts and texts. Also, databases from differentfields of research have to be searched. Many search terms need to be interconnected with Boolean operators to make sure thatall relevant articles (using different terminology) are found and that irrelevant articles (using the same terminology asrelevant ones) are excluded.

Supportive Information: Presentation of cognitive strategies• SAP for determining the number of databases to search and whether to also search on abstracts and full

texts.Supportive Information: Presentation of mental models• Structural model of templates of search queries describing Boolean combinations of search terms that

can be used to search for articles about ill-defined subjects.• Conceptual model of different types of databases for different fields of study, describing structure,

special search requirements, etc.

Learning Task 3.1: Completion + Reverse JIT information Part-task PracticeLearners receive a research question and an elaborate presentation • Applying Boolean search query. They have to predict which databases • Procedures for searching operators (continueshould be used and then perform the query. They specific databases as needed)then have to refine the query and select relevant articles.Learning Task 3.2: Conventional JIT information Learners receive a research question. They have to presentationperform a literature search for the 10 most relevant • Procedures for searching articles. specific databases (fading)

Supportive Information: Cognitive feedbackLearners receive feedback on their approach to solve the problem in Learning Task 3.2.



instructional materials. Especially for thedevelopment of computer-based or self-instruc-tional materials, more concrete and specificdescriptions may be needed of the supportiveand JIT information that should be made avail-able to the learners, as well as the practice itemsthat make up part-task practice and the supportstructures that are available for those practiceitems and the learning tasks. According to the4C/ID-model, a process of cognitive taskanalysis is then needed to flesh out the blueprint(van Merriënboer, 1997).

With respect to the design of individualizedor adaptive training programs it should benoted that it is not always desirable to specify allaspects of the blueprint beforehand. In ourliterature example (Table 3), the order forpresenting learning tasks, practice items, anddifferent types of information is fixed and iden-tical for all learners. However, to be able toadapt instruction to differences in learnerprogress, it is required that sequencing andtiming for the presentation of information andpractice opportunities can be dynamically ad-justed. The 4C/ID-model allows for this. For ex-ample, instead of designing a fixed sequence oflearning tasks with learner support diminishingat a fixed rate, one could design sets of learningtasks where each set contains several versions ofthe same learning task but with differentamounts of learner support. During trainingeither a human tutor or a computer-based sys-tem can select and present learning tasks withan optimal amount of learner support, based onlearner performance on previous learning tasks.Only when the required level of performance fora particular task class has been reached does thelearner continue to the next task class. In moreadvanced training environments this dynamicapproach can be taken one step further. InCASCO, (van Merriënboer & Luursema, 1996),fuzzy-logic algorithms are used to model learnerskill development and to dynamically generatenew learning tasks that best suit individuallearner needs.

With regard to instructional methods, the4C/ID-model typically applies a mix of con-structivist and instructivist approaches. Thebasis for the design of a training program iswhole-task practice, offering nontrivial, realistic

and increasingly more authentic task classes andlearning tasks to the learners. Schema construc-tion by induction and mindful abstraction fromconcrete cases are assumed to be key learningprocesses, reflecting a strong constructivist ap-proach. With regard to the presentation of infor-mation, an inductive-inquiry strategy or(guided) discovery approach also reflects a con-structivist viewpoint. This strategy is prescribedfor the presentation of supportive informationfor which deep understanding is required—provided that the available training time allowsfor it. But for the reason of instructional efficien-cy, the 4C/ID-model also has some clear instruc-tivist features. For the presentation of JITinformation, a deductive-expository strategy isrecommended. For the presentation of suppor-tive information, an inductive-expositorystrategy is recommended by default, and adeductive-expository strategy is recommendedif the available time for instruction is limited andlearners already have some relevant experience.Thus, in order to make the training process moreefficient, it is sometimes necessary to providelearners with prespecified, general knowledgethat may be helpful and offer guidance to solvethe problems in a particular domain.

It is important to realize that blueprintsdeveloped according to the 4C/ID-model markthe transition from the design phase to theproduction or development phase. The modeldoes not provide detailed guidelines for thisdevelopment phase. Important elements such asoverviews of content structure, summaries, tran-sitions and so forth are not dealt with. The mainreason for this is that the guidelines for thedevelopment of learning environments and theproduction of instructional materials are oftenmedia specific. After a blueprint for a trainingprogram has been finished, a final decisionshould be made as to the primary and secondarymedia that will be used. This selection of mediais influenced by many factors not discussed inthe 4C/ID-model, such as constraints to imple-ment the design (time and money), special taskrequirements, and characteristics of the targetgroup. After the final media selection, special-ized instructional design models that providemedia-specific guidelines for materials develop-ment should be consulted.


58 ETR&D, Vol. 50, No. 2

While the 4C/ID-model does not includeguidelines for final media selection, it does limitthe available media options for each of the fourcomponents (see van Merriënboer, 1997). This isbecause each of the four components cor-responds to another category of learning proces-ses, and particular learning processes are bestsupported by particular media. According to the4C/ID-model, the primary medium is alwaysrelated to performing the learning tasks and willthus typically involve a real or simulated taskenvironment. Consequently, most instructionalsystems based on the 4C/ID-model can be char-acterized as problem-based, simulation-based,simulator-based, case-based, or scenario-basedlearning environments. The secondary mediaare related to supportive information (suitablemedia include books, hypertext systems, lec-tures, etc.); JIT information (including on-linehelp systems, job-aids, pop-up menus, balloonhelp, etc.), and part-task practice (includingdrill-and-practice computer programs, part-tasktrainers, etc.).

With regard to the effectiveness of developedtraining programs, the most important claim isthat the 4C/ID-model helps to develop trainingprograms that lead to higher transfer perfor-mance than conventional instruction, and thatthis effect increases as transfer tasks differ morefrom the original training tasks. This predictionhas been tested in several domains by compar-ing performance after training (and especially,transfer performance) of training strategiesdeveloped according to the 4C/ID-model withconventional real-world strategies or strategiesdeveloped according to other models. For ex-ample, in a range of studies in the computerprogramming domain, including classroomstudies (van Merriënboer, 1990a, 1990b) andcomputer-based training studies (Schuurman,1999; van Merriënboer & de Croock, 1992; vanMerriënboer, Schuurman, de Croock, & Paas,2002), 4C/ID strategies yielded higher transferperformance than control strategies, and this su-periority became more evident on far transferproblems for which learners had to design andconstruct new computer programs that requiredsolutions not encountered before. Also in otherdomains including statistical analyzing (Paas,1992, 1993), computer numerically controlled

programming (Paas & van Merriënboer, 1994),and fault management in process industry (deCroock, 1999; de Croock, van Merriënboer, &Paas, 1998; Jelsma, 1989) training strategies thatfollowed 4C/ID principles were designed andtested, and results supported the main predic-tions of the 4C/ID-model.

At present, more studies are being carried outthat investigate in greater detail importantaspects of the 4C/ID-model, including thetiming of information presentation (Kester,Kirschner, van Merriënboer, & Bäumer, 2001),modalities of information presentation (Tabbers,Martens, & van Merriënboer, 2001), and optimalstep sizes in process worksheets for learningtasks (Nadolski, Kirschner, van Merriënboer, &Hummel, 2001). It is our strong conviction thatonly research-based models and methodologiesmay be strong enough to move the huge, slow,and ponderous ocean liner of instructionaldesign a few degrees off its current path.

Jeroen J. G. van Merriënboer[[email protected]] is Professor ofEducational Technology at the Open University ofthe Netherlands, Heerlen, The Netherlands. Richard E. Clark is Professor of Education at theRossier School of Education, University of SouthernCalifornia. Marcel B.M. de Croock is Assistant Professor ofEducational Technology at the Open University ofthe Netherlands, Heerlen, The Netherlands. Part of the work presented here was carried outwithin the ADAPTIT project, sponsored by theEuropean Commission under contractIST-1999-11740. Correspondence concerning this article should beaddressed to Jeroen J. G. van Merriënboer, OpenUniversity of the Netherlands, EducationalTechnology Expertise Center (OTEC), P.O. Box 2960,NL-6401 DL Heerlen, The Netherlands.

REFERENCES

Ackerman, P.L. (1990). A correlation analysis of skillspecificity: Learning, abilities and individual dif-ferences. Journal of Experimental Psychology: Learning,Memory and Cognition, 16, 883–901.

Anderson, J.R. (1983). The architecture of cognition.Cambridge, MA: Harvard University Press.

Anderson, J.R. (1993). Rules of the mind. Hillsdale, NJ:Lawrence Erlbaum Associates.

Anderson, J.R., & Lebiere, C. (1998). The atomic com-



ponents of thought. Mahwah, NJ: Lawrence ErlbaumAssociates.

Balzer, W.K., Doherty, M.E., & O’Connor, R. (1989). Ef-fects of cognitive feedback on performance.Psychological Bulletin, 106, 410–433.

Berryman, S.E. (1993). Learning in the workplace. In L.Darling-Hammond (Ed.), Review of research in educa-tion (Vol. 19, pp. 343–401). Washington, DC:American Educational Research Association.

Butler, D.L., & Winne, P.H. (1995). Feedback and self-regulated learning: A theoretical synthesis. Review ofEducational Research, 65, 245–281.

Carlson, R.A., Khoo, H., & Elliot, R.G. (1990). Com-ponent practice and exposure to a problem-solvingcontext. Human Factors, 32, 267–286.

Carroll, J.M., Smith-Kerker, P.L., Ford, J.R., & Mazur-Rimetz, S.A. (1988). The minimal manual. Human-Computer Interaction, 3, 123–153.

Cascio, W.F. (1995). Whither industrial and organiza-tional psychology in a changing world of work?American Psychologist, 50, 928–939.

Clark, R.E. (1988). When teaching kills learning: Re-search on mathemathanics. In H. Mandl, E. DeCorte,N. Bennett, & H.F. Friedrich (Eds.), Learning and in-struction: European research in an international context(Vol. 2, pp. 1–22). Oxford, UK: Pergamon Press.

Clark, R.E., & Estes, F. (1999). The development ofauthentic educational technologies. EducationalTechnology, 39(2), 5–16.

Collins, A., Brown, J.S., & Newman, S. (1989). Cogni-tive apprenticeship: Teaching the craft of reading,writing, and mathematics. In L.B. Resnick (Ed.),Knowing, learning, and instruction: Essays in honor ofRobert Glaser. Hillsdale, NJ: Lawrence Erlbaum As-sociates.

Corneille, O., & Judd, C.M. (1999). Accentuation andsensitization effects in the categorization of multi-faceted stimuli. Journal of Personality and SocialPsychology, 77, 927–941.

de Croock, M.B.M. (1999). The transfer paradox: Trainingdesign for troubleshooting skills. Unpublished PhDThesis. Enschede, The Netherlands: University ofTwente.

de Croock, M.B.M., van Merriënboer, J.J.G., & Paas,F.G.W.C. (1998). High vs. low contextual inter-ference in simulation-based training oftroubleshooting skills: Effects on transfer perfor-mance and invested mental effort. Computers inHuman Behavior, 14, 249–267.

de Jong, T., & van Joolingen, W.R. (1998). Scientific dis-covery learning with computer simulations of con-ceptual domains. Review of Educational Research, 68,179–201.

Driscoll, M., & Dick, W. (1999). New researchparadigms in instructional technology: An inquiry.Educational Technology Research and Development,47(2), 7–18.

Dufresne, R.J., Gerace, W.J., Thibodeau-Hardiman, P.,& Mestre, J.P. (1992). Constraining novices to per-form expertlike problem analyses: effects on schema

acquisition. The Journal of the Learning Sciences, 2,307–331.

Fisk, A.D., & Gallini, J.K. (1989). Training consistentcomponents of tasks: Developing an instructionalsystem based on automatic-controlled processingprinciples. Human Factors, 31, 453–463.

Gagné, R.M., Briggs, L.J., & Wager, W.W. (1992). Prin-ciples of instructional design (4th ed.). New York, NY:Holt, Rinehart and Winston.

Gick, M., & Holyoak, K.J. (1983). Schema inductionand analogical transfer. Cognitive Psychology, 15, 1–38.

Gropper, G.L. (1983). A behavioral approach to in-structional prescription. In C.M. Reigeluth (Ed.), In-structional design theories and models (pp. 101–161).Hillsdale, NJ: Lawrence Erlbaum Associates.

Jelsma, O. (1989). Instructional control of transfer.Enschede, The Netherlands: Bijlstra & vanMerriënboer.

Jonassen, D. (1994). Thinking technology: Toward aconstructivist design model. Educational Technology,34(4), 34–37.

Kester, L., Kirschner, P.A., van Merriënboer, J.J.G., &Bäumer, A. (2001). Just-in-time information presen-tation and the acquisition of complex cognitiveskills. Computers in Human Behavior, 17, 373–391.

Kluger, A., & DiNisi, A. (1998). Feedback interven-tions: Toward the understanding of a double-edgedsword. Current Directions in Psychological Science,7(3), 67–72.

Kozma, R. (2000). Reflections on the state of education-al technology. Educational Technology Research andDevelopment, 48(1), 5–15.

Landa, L.N. (1983). The algo-heuristic theory of in-struction. In C.M. Reigeluth (Ed.), Instructionaldesign theories and models (pp. 163–211). Hillsdale, NJ:Lawrence Erlbaum Associates.

Leutner, D. (2000). Double-fading support-a trainingapproach to complex software systems. Journal ofComputer Assisted Learning, 16, 347–357.

Merrill, M.D. (1983). Component display theory. InC.M. Reigeluth (Ed.). Instructional design theories andmodels (pp. 278–333). Hillsdale, NJ: LawrenceErlbaum Associates.

Merrill, M.D. (1999). Instructional transaction theory:Instructional design based on knowledge objects. InC.M. Reigeluth (Ed.), Instructional design theories andmodels: A new paradigm of instructional theory (Vol. 2).Mahwah, NJ: Lawrence Erlbaum Associates.

Merrill, M.D. (2000, October). First principles of instruc-tion. Paper presented at the international conferenceof the Association for Educational Communicationsand Technology (AECT), Denver, CO.

Nadolski, R.J., Kirschner, P.A., van Merriënboer, J.J.G.,& Hummel, H.G.K. (2001). A model for optimizingstep size of learning tasks in competency-based mul-timedia practicals. Educational Technology Researchand Development, 49(3), 87–103.

Newell, A., & Simon, H.A. (1972). Human problem solv-ing. Englewood Cliffs, NJ: Prentice Hall.


60 ETR&D, Vol. 50, No. 2

Ohlsson, S., & Rees. E. (1991). The function of concep-tual understanding in the learning of arithmetic pro-cedures. Cognition and Instruction, 8, 103–179.

Paas, F.G.W.C. (1992). Training strategies for attainingtransfer of problem-solving skill in statistics: A cog-nitive load approach. Journal of Educational Psychol-ogy, 84, 429–434.

Paas, F.G.W.C. (1993). Instructional control of cognitiveload in the training of complex cognitive tasks. Un-published PhD Thesis. Enschede, The Netherlands:University of Twente.

Paas, F.G.W.C., & van Merriënboer, J.J.G. (1994).Variability of worked examples and transfer ofgeometrical problem solving skills: A cognitive loadapproach. Journal of Educational Psychology, 86, 122–133.

Pellegrino, J.W., Doane, S.M., Fischer, S.C., & Alder-ton, D. (1991). Stimulus complexity effects in visualcomparisons: The effects of practice and learningcontext. Journal of Experimental Psychology: HumanPerception and Performance, 17, 781–791.

Perkins, D.N., & Grotzer, T.A. (1997). Teaching intel-ligence. American Psychologist, 52, 1125–1133.

Pointe, L.B., & Engle, R.W. (1990). Simple and complexword spans as measures of working memorycapacity. Journal of Experimental Psychology: Learning,Memory and Cognition, 16, 1118–1133.

Reigeluth, C.M. (Ed.) (1999a). Instructional designtheories and models: A new paradigm of instructionaltheory (Vol. 2). Mahwah, NJ: Lawrence Erlbaum As-sociates.

Reigeluth, C.M. (1999b). The elaboration theory:Guidance for scope and sequence decisions. In C.M.Reigeluth (Ed.), Instructional design theories andmodels: A new paradigm of instructional theory (Vol. 2).Mahwah, NJ: Lawrence Erlbaum Associates.

Reigeluth, C.M., & Stein, F. (1983). The elaborationtheory of instruction. In C.M. Reigeluth (Ed.), In-structional design theories and models (pp. 335–381).Hillsdale, NJ: Lawrence Erlbaum Associates.

Richey, R. (1998). The pursuit of useable knowledge ininstructional technology. Educational Technology Re-search and Development, 46(4), 7–22.

Salisbury, D.F., Richards, B.F., & Klein, D. (1985).Designing practice: A review of prescriptions andrecommendations from instructional designtheories. Journal of Instructional Development, 8(4), 9–19.

Schneider, W. (1985). Training high-performanceskills: Fallacies and guidelines. Human Factors, 27,285–300.

Schneider, W., & Detweiler, M. (1988). The role of prac-tice in dual-task performance: Toward workloadmodeling in a connectionist/-control architecture.Human Factors, 30, 539–566.

Schuurman, J.G. (1999). On the proper treatment of learn-ing and transfer: A study on introductory computerprogramming. Unpublished PhD Thesis. Enschede,The Netherlands: University of Twente.

Schwartz, D., Brophy, S., Lin, X., & Bransford, J. (1999).

Flexibly adaptive instructional design: A case studyfrom an educational psychology course. EducationalTechnology Research and Development, 47(2), 39–60.

Sweller, J., van Merriënboer, J.J.G., & Paas, F.G.W.C.(1998). Cognitive architecture and instructionaldesign. Educational Psychology Review, 10, 251–296.

Tabbers, H.K., Martens, R.L., & van Merriënboer, J.J.G.(2001, August). The modality effect in multimedia in-struction. Paper presented at the 23rd Annual Con-ference of the Cognitive Science Society (Cogsci),Edinburgh, Scotland.

van der Meij, H., & Carroll, J.M. (1995). Principles andheuristics for designing minimalist instruction. Tech-nical Communication, 42, 243–261.

van Merriënboer, J.J.G. (1990a). Strategies forprogramming instruction in high school: Programcompletion vs. program generation. Journal of Educa-tional Computing Research, 6, 265–287.

van Merriënboer, J.J.G. (1990b). Teaching introductorycomputer programming: A perspective from instructionaltechnology. Enschede, The Netherlands: Bijlstra &van Merriënboer.

van Merriënboer, J.J.G. (1997). Training complex cogni-tive skills: A four-component instructional design modelfor technical training. Englewood Cliffs, NJ: Educa-tional Technology Publications.

van Merriënboer, J.J.G., & de Croock, M.B.M. (1992).Strategies for computer-based programming in-struction: Program completion vs. program genera-tion. Journal of Educational Computing Research, 8,365–394.

van Merriënboer, J.J.G., & Dijkstra, S. (1996). The four-component instructional design model for trainingcomplex cognitive skills. In R.D. Tennyson & F.Schott (Eds.), Instructional design: Internationalperspectives. Theory and research (Vol. 1, pp. 427–445).Hillsdale, NJ: Lawrence Erlbaum Associates.

van Merriënboer, J.J.G., Jelsma, O., & Paas, F.G.W.C.(1992). Training for reflective expertise: A four-com-ponent instructional design model for training com-plex cognitive skills. Educational Technology Researchand Development, 40(2), 23–43.

van Merriënboer, J.J.G., & Luursema, J.J. (1996). Im-plementing instructional models in computer-basedlearning environments: A case study in problemselection. In T.T. Liao (Ed.), Advanced educationaltechnology: Research issues and future potential (NATOASI Series F, Vol. 145, pp. 184–206). Berlin, Ger-many: Springer Verlag.

van Merriënboer, J.J.G., Schuurman, J.G., de Croock,M.B.M., & Paas, F.G.W.C. (2002). Redirectinglearners’ attention during training: Effects on cogni-tive load, transfer test performance and training ef-ficiency. Learning and Instruction, 12, 11–37.

Wenger, J.L., & Carlson, R.A. (1996). Cognitive se-quence knowledge: What is learned? Journal of Ex-perimental Psychology: Learning, Memory andCognition, 22, 599–619.



Appendix Blueprints for Complex Learning: 10-point summary

Complex Learning

1. The performance of a complex cognitive skill is made up by many interrelated constituent skills that may eachshow qualitatively different performance. Learning a complex cognitive skill is primarily concerned withlearning to perform its constituent skills in a coordinated and integrated fashion. The constituent skills aredepicted in a skills hierarchy-a structural model of the complex cognitive skill showing how the performancesof constituent skills at the same level of generality are ordered in time (horizontal relationships) and how con-stituent skills at a lower level enable the performance of skills at a higher level (vertical relationships).

2. Nonrecurrent constituent skills are performed in a different way from problem to problem situation. Expertscan effectively perform nonrecurrent skills because they have highly complex cognitive schemata availablethat guide their problem solving (cognitive strategies) and allow them to reason about their domain (mentalmodels). Therefore, training for the nonrecurrent aspects of the complex skill and for the complex skill as awhole should aim at schema construction. Learners are stimulated to construct new schemata by mindfullyabstracting away from concrete whole-task experiences (induction) and to connect newly presented informa-tion to already existing schemata (elaboration).

3. Recurrent constituent skills are performed in a similar way from problem to problem situation. Experts caneffectively perform recurrent skills because their schemata contain rules that link particular characteristics ofthe problem to particular actions. Therefore, training for the recurrent aspects of the complex skill should aimat rule automation. Learners should be presented repeated practice to allow them to embed specificknowledge or information into rules and to chunk rules together that are consistently applied in the sameorder (compilation). For reaching high levels of automaticity, additional part-task practice should beprovided to increase the strength of the rules each time they are applied (strengthening).

4. Training programs for complex cognitive skills primarily aim at transfer of learning-the ability to perform thecomplex cognitive skill in a wide variety of new real-life problem situations. The familiar aspects in the newtransfer situations can be dealt with because of the availability of automated rules, which also free up cogni-tive resources that may be used to handle the unfamiliar aspects of the transfer task (component fluencyhypothesis). These unfamiliar aspects of the transfer task can be dealt with thanks to the availability of com-plex cognitive schemata, which may be interpreted in order to reason about the domain and to guide the prob-lem-solving process (understanding hypothesis). The result is called reflective expertise.

Appendix continues


62 ETR&D, Vol. 50, No. 2

Appendix continued

Blueprint Component 1: Learning Tasks

5. Learning tasks are concrete, authentic “whole-task experiences” that are presented to learners to promote schema construction through induction and, to a certain degree, rule automation by compilation forthe recurrent aspects.

6. Learning tasks are organized in a simple-to-complex sequence of task classes, i.e., each task class is defined to contain learning tasks that require the same body of knowledge for their performance. Learning tasks within the same task class are sequenced for high variability.

7. Learning tasks within the same task class start with high build-in learner support, which disappears at the end of the task class (a process called scaffolding). Learner support can be either product-oriented (e.g., use of case studies or completion problems) or process-oriented (e.g., use of modeling examples or process worksheets).

Blueprint Component 2: Supportive Information

8. Supportive information helps to learn and perform nonrecurrent aspects of learning tasks. It primarily promotes schema construction through elaboration. It consists of mental models, cognitive strategies and cognitive feedback. In an inductive strategy, case studies and modeling examples are presented first as part of the supportive information, where after general models and strategies are either inquired (inductive-inquiry or guided discovery) or presented (inductive-expository). In a deductive strategy, general models and strategies are presented first and then illustrated by the first learning tasks. Supportive information remains available to the learners throughout the task class. Cognitive feedback is presented after one or more learning tasks.

Blueprint Component 3: JIT Information

9. JIT information is prerequisite to the learning and performance of recurrent aspects of learning tasks. It promotes restricted encoding of new information into rules. It consists of information displays, demonstrations/instances, and corrective feedback. It is presented with the learning task in which the recurrent skill for which it is needed is practiced first and is then quickly faded away as learners acquire more expertise.

Appendix continues



Appendix continued

Blueprint Component 4: Part-task Practice

10. Part-task practice for selected recurrent aspects that require a very high level of automaticity consists of a sequence of practice items that is provided to learners to promote rule automation (especially, strengthening). Each sequence contains a divergent set of practice items for all situations that the underlying rules can deal with. For complex rule-sets, snowballing and REP-sequences may be applied. JIT information is specified like it is for recurrent aspects of learning tasks. Part-task practice sessions are best intermixed with learning tasks and with each other.

Circle Your Calendar Now

The AECT Annual ConferenceNovember 12–16, 2002

Wyndham Anatole HotelDallas, Texas

Back in the Saddle Again!


64 ETR&D, Vol. 50, No. 2

blueprints for complex learning: the 4c/id- · pdf fileblueprints for complex learning: the...

Documents