an evaluation report of multimedia environments as cognitive learning tools
TRANSCRIPT
An evaluation report of multimedia environments as cognitive
learning tools
Norbert M. Seel*, Katharina Schenk
Freiburg Institute of Educational Science, Albert-Ludwigs University, Freiburg, Germany
Received in revised form 1 December 2001
Abstract
This article deals with the evaluation of a multimedia learning environment which has been developed and evaluated within the broader
context of a research project on the learning-dependent progression of mental models in economics. To carry out formative evaluations, we
have adapted a particular evaluation approach which allows and requires the implementation of specific evaluation instruments. The crucial
questions of our evaluation studies were the efficacy of a multimedia-based realization of the cognitive apprenticeship (CA) approach, the
diagnosis of mental model progression through the CA based instruction, and the effects of implemented metacognitive training. For the
assessment of the learning-dependent progression of the mental models, we developed and used a special diagnostic instrument for causal
diagrams, which are understood as reproductions of students’ mental models. In order to be able to meet statements about the practicability of
a multimedia based realization of CA, we measured the results of the tasks of learning during each different learning phase. Additionally,
several motivational variables and persistent learning strategies were measured. In this article, we will specify the adapted evaluation
instruments. Furthermore, we will report on the results of five replication studies and discuss the consequences for instructional design in
connection with the design of constructive learning environments.
q 2003 Elsevier Science Ltd. All rights reserved.
Keywords: Causal diagrams; Cognitive apprenticeship; Instructional design; Mental models; Metacognitive training
1. Introduction
There is considerable concern that student’s thinking
skills, motivational dispositions, and domain-specific
knowledge might be inadequate for them to lead fulfilling
lives in a global, information-rich, technology-oriented
world. Informed by recent theory and research on learning
and teaching, efforts to reform classroom instruction and
create learning environments that promote these ends are
underway. Hannafin (1992) argues that the improvement of
problem-solving abilities and other key skills requires
‘emergent technologies’ in order to design effective learning
environments that provide opportunities for reflective
thinking. However, this argument can be seen from different
points of view. One approach focuses on the improvement of
students’ technological literacy and advocates a new type of
understanding of information and communication technol-
ogy in educational settings. Another approach focuses on the
effective instructional design (ID) of multimedia environ-
ments as opposed to the technology itself. Obviously, these
perspectives are not mutually exclusive as each may be
considered within the context of ID.
ID is a theoretically sound educational technology for the
development, implementation, and evaluation of learning
environments that are adapted to learners, tasks, resources,
and contexts (Tennyson, Schott, Seel, & Dijkstra, 1997). An
analysis of the literature indicates that there is broad
consensus with regard to ID requirements and how to
evaluate various designs in different contexts (Weston,
McAlpine, & Bordonaro, 1995). Evaluation has been
considered a central and necessary part of instructional
planning from the very beginning of ID (Andrews &
Goodson, 1980; Cronbach, 1963), wherein a differentiation
is made between two kinds of evaluation (Scriven, 1967).
† formative evaluation aimed at the improvement of
instruction by means of feedback of information
concerning the effective use and outcomes; and,
0149-7189/03/$ - see front matter q 2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0149-7189(03)00003-X
Evaluation and Program Planning 26 (2003) 215–224
www.elsevier.com/locate/evalprogplan
* Corresponding author.
E-mail addresses: [email protected] (N.M. Seel), schenk@
ezw.uni-freiburg.de (K. Schenk).
† summative evaluation aimed at the measurement of the
degree to which intended results are achieved.
A further distinction is often drawn with regard to the
instruments of evaluation, insofar as quantitative or
qualitative data are to be assessed.
This article is concerned with the evaluation of a
particular multimedia learning environment which was
developed and evaluated as part of a comprehensive
research project focusing on the learning-dependent
progression of mental models. Insofar as the construction
of mental models presupposes constructive learning,
which occurs when learners actively construct meaningful
mental representations during instruction (Mayer, Moreno,
Boire, & Vagge, 1999), multimedia is considered to be
the most effective for the promotion of such constructive
learning. In the case of the given multimedia-learning
environment, the meaningful mental representation is a
coherent mental model of a dynamic model of macro-
economics and financial politics, respectively. In our
research, learning outcomes were evaluated with multiple
measuring instruments. On the one hand, students had to
produce causal diagrams of the problem situation; on the
other hand, student solutions to learning tasks and transfer
problems were tested.
In the following sections we will illuminate the
formative evaluation of the learning environment
‘Dynamic Systems of Economics’ (DSE) as applied in
several replication studies. An evaluation model will be
described with the focus on the following components:
methodology, data analysis, and interpretation. We will
conclude with a discussion of the main results of several
evaluation studies.
2. Model-based learning and instruction
Our research group has been involved in the develop-
ment and investigation of instructional intervention pro-
grams aimed at the improvement of model-based learning
and thinking for several years. The epistemological and
psychological foundations of this research rest on Seel’s
(1991) theory of mental models.
Mental models emerged in the 1980s as a theoretical
construct to encompass both situated cognition as well as
qualitative reasoning. Greeno (1989) argues that compre-
hension of and reasoning in specific situations necessarily
involves the use of mental models of different qualities.
Mental models are a central construct of symbolic models of
human cognition that presuppose the use and manipulation
of symbols. Following Wartofsky (1979), cognition takes
place while using mental representations in which individ-
uals organize symbols of experience or thought in such a
way that they effect a systematic representation of this
experience or thought as a means of understanding or
explaining it to others.
Mental models play a central and unifying role in
representing objects, states of affairs, sequences of
events, and the social and psychological actions of
daily life. They enable individuals to make inferences
and predictions, to understand phenomena, to decide
what action to take and to control its execution,
and, above all, to experience events by proxy (Johnson-
Laird, 1983, p. 397)
Craik (1943) introduced the idea of internal models to
psychology with the notion of a working model. According
to Craik, most cognitive theorists agree on the point that
mental models serve primarily to create situation-specific
plausibility. Due to an idealized reduction to relevant
characteristics of its original, a model is a concrete,
comprehensible, and feasible representation of non-obvious
or abstract objects. The representation of the objects’
attributes and components comes second to the represen-
tation of structural relationships. Mental models are not a
specific representational format such as images and
propositions, but rather higher-order cognitive constructions
(artifacts) which refer primarily to the content of mental
representations (Seel, 1991).
The functions of mental models, including structural
features, are defined on the basis of the objectives of the
model-constructing person. In physics and other sciences,
the term ‘model’ is always used in a functional sense.
‘Appearance models’ may serve to simplify a complex
phenomenon or represent structural relationships visually.
On the other hand, ‘derivative (thought) models’ (e.g.
Rutherford’s model of the atom) serve primarily to aid
analogical reasoning in exploring phenomena (e.g. quantum
mechanisms). Mental simulations occur when cognitive
operations simulate (in the sense of thought experiments)
specific transformations of objects that may occur in real-
life situations. In sum, mental models ‘run in the mind’s
eye’ to produce qualitative inferences with respect to the
situation to be cognitively mastered.
Although mental models may differ markedly in their
content, there is no evidence to suggest that they differ in
representational format or in the processes that construct
and manipulate them. What is at issue is how such
models develop as an individual progresses from novice
to expert, and whether there is any pedagogical
advantage in providing people with models of tasks
they are trying to learn (Johnson-Laird, 1989, p. 485)
In accordance with Snow (1990) we have identified the
learning-dependent progression of mental models as a
specific kind of transition mediating between student
preconceptions, which describe the initial states of the
learning process, and causal explanations, which describe
the desired end states of the learning process (Seel,
Al-Diban, & Blumschein, 2000). From the perspective of
instructional psychology, the guiding principle for
N.M. Seel, K. Schenk / Evaluation and Program Planning 26 (2003) 215–224216
influencing the construction of mental models has been
expressed by Mayer (1989) as follows: “Students given
model-instruction may be more likely to build mental
models of the systems they are studying and to use these
models to generate creative solutions to transfer problems”
(p. 47). This presupposes that the learner is sensitive to the
model-relevant characteristics of the learning environment,
such as the availability of certain information at a given
time, the way this information is structured and mediated,
and the ease with which it can be found in the environment.
An analysis of the relevant literature indicates that the
suggested instructional strategy to provide learners with a
designed conceptual model actually constitutes the main
trend of instructional research on mental models. According
to Carlson (1991), instruction can be designed to involve the
learner in an inquiry process in which facts are gathered
from data sources, similarities and differences among facts
noted, and concepts developed. In this process, the
instructional program serves as a facilitator of learning for
students who are working to develop their own answers to
questions. In this case, mental models are more proactive
and direct the learning experiences so that the result of
learning is dependent on the initial model, defined as the
learner’s ‘a priori understanding’ of the material to be
learned. On the other hand, instructional programs can
present concepts with clear definitions followed by clear
examples. A conceptual model may be presented before the
learning tasks in order to direct the learner’s comprehension
of the learning material. Over the past decades much
research applying this strategy has been done to provide
students with model-based instruction but several authors
(Royer, Cisero, & Carlo, 1993; Snow, 1990) have objected
that this kind of research has typically been done piece-
meal, in small-scale, specialized contexts. In order to
overcome these shortcomings we need a more comprehen-
sive instructional approach. Cognitive apprenticeship (CA)
(Collins, Brown, & Newman, 1989) provides a fundamental
basis for initiating and directing model-based learning.
Our research group started in 1994 with the development
of a multimedia environment aimed at an externally guided,
goal-oriented, and systematic influence upon the learners’
progression of mental models. CA (Collins et al., 1989) was
the only promising instructional strategy corresponding
with the idea of providing the students with model-
instruction in the aforementioned sense. There are six
instructional methods in CA: modeling, coaching, scaffold-
ing, articulation, reflection, and exploration. The instruc-
tional intervention of apprenticeship starts with the
presentation of an expert’s conceptual model of the tasks
to be accomplished, and then the students are coached and
scaffolded to adapt this model for their own solutions
(exploration) to the learning tasks designed. CA is based on
results of cognitive psychology and applies these results in a
prescriptive way in order to identify ‘ideal features’ of
learning environments. This approach prescribes in detail
what the learner has to do and in which sequence in order to
achieve particular objectives. However, the question as to
whether the CA approach may be appropriate for the design
of multimedia environments could not be answered at this
time. There are several studies (Casey, 1996; Chee, 1996;
Jarvela, 1995) that have investigated this. However, these
studies run parallel with our investigations and final
conclusions are not available. Therefore, we focused on
the issue of whether the preferred use and application
of multimedia technology allows a strict adaptation of
instructional regularities to individual regularities
of learning.
The research we have done in the past 6 years has
centered around two main topics:
1. the investigation of the learning-dependent progression
of mental models, more specifically of analogy models of
DSE; and,
2. how this progression can be guided or influenced through
a particular instructional intervention program designed
as a multimedia environment in accordance with
principles of CA. We focus on the second line of
research in this paper.
3. The evaluation model
We have adopted the evaluation approach of Ross and
Morrison (1997) with these main components: (1) needs
assessment, (2) methodology, (3) data analysis and
interpretation, and (4) dissemination results. The evaluation
of DSE focused on methodology and data analysis and
interpretation. Accordingly, we have realized:
† a program analysis in order to determine the content and
methods of their mediation within the multimedia
program;
† a participant’s analysis in order to determine the (groups
of) learners as well as the scope of the instructional
program;
† a specification of the evaluation design;
† the development of instruments of measurement; and,
† the implementation and control of the evaluation
design.
Following these methodological steps, the analysis and
interpretation of data was to be done in order to modify or
revise the instructional program or parts of it. Accordingly,
the data analysis centers around the formative evaluation of
instruction during the development phase for the purpose of
improvement.
3.1. Program analysis
The multimedia program is designed to explain the
dynamics of economic systems and to introduce the
monetary policy of the European Central Bank into
N.M. Seel, K. Schenk / Evaluation and Program Planning 26 (2003) 215–224 217
students’ discussion. In order to facilitate student construc-
tion of adequate mental models, several conceptual models,
especially a circuit model of economic systems, are
presented. According to the CA approach, effective learning
environments can be characterized by 18 features in four
broad dimensions: content, methods, sequencing, and the
sociology of teaching. We focused especially on methods,
which encompass the features of modeling, coaching,
scaffolding, articulation, reflection, and exploration.
Altogether, we integrated the relevant aspects of the CA
approach into a comprehensive model, as described in Fig. 1.
We realized CA in the following way (Seel et al., 2000).
In modeling, an expert demonstrates a solution to a problem,
and the students acquire an initial model of this process by
observing the expert’s approach. In coaching, the students
are supervised and given guidance as they try to find
solutions to a given task in an adaptive manner. In
scaffolding a special problem-solving heuristics is taught.
The realization of articulation and reflection proved to be
problematic within the multimedia program. Articulation is
the process of ‘thinking aloud’ while working on a task, and
reflection is the comparison of the problem solving
procedures applied by the learner and the expert respect-
ively. We realized both methods in the form of a ‘teach-
back’ procedure (Jih & Reeves, 1992; Sasse, 1991) in a
social learning situation. This procedure is based on the
‘constructive interaction’ between two communication
partners who share similar domain-specific knowledge.
In exploration, the final part of the apprenticeship
instruction, the learners have to solve transfer tasks—one
of them requires a ‘near transfer’ (i.e. the task remains in the
same subject matter domain of economics) the other one
requires a ‘far transfer’ from economics to ecology. On the
whole, the multimedia program realized the methods of CA
in the sequence illustrated in Fig. 2.
Additionally, two different instructional strategies aimed
at the improvement of analogical problem solving were
realized in modeling as well as in scaffolding:
† subsumption of analogous learning tasks under a schema
of a problem-solving structure, followed by its instantia-
tion through a detailed, worked-out example; and,
† induction of a more general problem-solving schema
from analogous learning tasks by the comparison of
different examples in order to extract structural
similarities.
3.2. Subjects
Taking curriculum constraints into account, the multi-
media program DSE is directed at 12th grade students of
German secondary schools (on average 18 years old). Two
pilot studies with university students indicated that the
program could also be used effectively with college students
in non-economic disciplines. In sum, more than 400
students have worked with DSE in various instructional
settings.
3.3. The evaluation design
In order to ensure that goals were being achieved and to
improve the multimedia program, the evaluation design
focused on formative evaluation as an iterative process of
trying out and revising of instruction during development.
However, formative evaluation can be conceived of in a
variety of ways (Weston et al., 1995). We implemented
formative evaluation as an iterative process in a series of
replication studies aimed at the gradual enhancement of
confidence. This process included empirically gathered
performance data and their interpretation.
Altogether, we realized five replication studies aimed at
(a) the formative evaluation of the multimedia program, and
(b) the diagnosis of mental models. The strategy of
formative evaluation applied related to the methods of the
Fig. 1. A view of the CA model (Seel et al., 2000).
N.M. Seel, K. Schenk / Evaluation and Program Planning 26 (2003) 215–224218
CA, and aimed at the gradual improvement of the
instruction (Table 1). This strategy can be sketched as
shown in Table 1.
This evaluation plan is based on the assumption that the
instruction can be considered effective with regard to the
achievement of goals if theoretically prognosed changes in
criterion variables can be observed. However, the efficacy of
the instruction may also depend on effects other than the
postulated ones, for example, on characteristics of the
instruction that are independent of the multimedia program
effects. This means, that measurable effects of an instruc-
tional intervention are not sufficient to justify the drawing of
conclusions about the model of efficacy applied.
3.4. Instruments of evaluation
A central goal of instruction is to improve student
performance, defined in terms of domain-specific knowl-
edge, skills, strategies, attitudes, and behavioral disposi-
tions. We assessed student domain-specific knowledge
using a test developed by Beck (1993). This test is
constructed in accordance with Bloom’s taxonomy of
cognitive objectives and permits statements about the
quality of knowledge in the field of economics. We assessed
the quantity and quality of learning and transfer tasks
accomplished within the learning environment. In coaching,
we measured the frequency and self-corrections of errors as
well as the time required to accomplish learning tasks,
whereas in scaffolding and exploration the frequency and
type of correct solutions were assessed. In order to assess the
learning-dependent progression of mental models, our
research group developed a special test procedure of causal
diagrams, which can considered to be a combination of
cognitive modeling and a particular structure-spreading
technique similar to concept mapping. Additionally, in two
studies we applied receptive interviews to assess the quality
of mental models, but this procedure proved to be
ineffective (Al-Diban, 2001). Finally, we also used protocol
analyses of the ‘teach-back’ phase of instruction.
Additionally, several motivational variables and persist-
ent learning strategies were measured, including: learner
achievement motivation, by means of a questionnaire
Table 1
Overview of the evaluation studies
Study Methods of CA implemented
Pilot study Modeling Coaching – – –
Study 1 Modeling Coaching Scaffolding – Exploration
Study 2 Modeling Coaching Scaffolding Computer-based, and individual articulation and reflection Exploration
Study 3 Modeling Coaching Scaffolding Articulation and reflection as teach-back procedure Exploration
Study 4 Modeling Coaching Scaffolding Articulation and reflection as teach-back procedure Exploration
Study 5 Modeling Coaching Scaffolding Articulation and reflection as training of metacognitive strategies Exploration
Fig. 2. The methods of CA as implemented in the multimedia program.
N.M. Seel, K. Schenk / Evaluation and Program Planning 26 (2003) 215–224 219
developed and validated by Rollett and Bartram (1977);
learner interests and attitudes, measured with a self-
constructed questionnaire; student perceptions of the
learning situation, also measured with a self-constructed
questionnaire to assess the ‘climate of multimedia-learning’
(Seel, 1980); and, student persistent learning strategies, with
a questionnaire by Wild and Schiefele (1994).
In addition to these external criteria we realized an
objective task analysis and a subjective task analysis to
determine whether the instruction was appropriated with
regard to curriculum and content criteria. Task analysis
procedures (Jonassen, Tessmer, & Hannum, 1999) pre-
suppose different criteria to achieve a sufficient differen-
tiation of the requirements and a classification of learning
tasks: (a) sub-task differences must be unambiguously
observable; (b) the task analysis should be derived
independently from a specific topic and should have general
validity; and (c) the analysis should be independent from
knowledge and dispositions of the application. In accord-
ance with these requirements, we applied a procedure of
problem and task analysis which Hacker, Sachse, and
Schroda (1998) developed especially for complex and
authentic tasks. With this procedure it was possible to
separate several dimensions, such as the complexity of
learning tasks (e.g. number of partial functions of a system
and of their relationships), the consistency of objectives
(number of goals, number of contradictory goals, degree of
concurrence of goals), the transparency of possible
solutions, degrees of freedom (variants of solutions),
dynamics (change of constraints), and the prior knowledge
necessary. By following this procedure, we were able to
characterize each learning task by means of a multi-
dimensional and objective profile of demands.
3.5. Implementation of the evaluation plan
The evaluation plan contains one pilot study with
university students and five replication studies with 12th
grade German students (Table 1). All studies were carried
out outside the school at the Laboratory for Multimedia
Research of the Technical University of Dresden (1994–
1998) and the University of Freiburg (1998 –2001),
enabling a strict control of external factors. The recruitment
of subjects took place on the basis of announcements in
schools, and the volunteers received a nominal fee. Subjects
were assigned to the different treatments randomly.
4. Results
This section starts with the report on findings of the tasks
analyses. In the next section we describe the main results of
different evaluation studies on the effectiveness of the
multimedia program in accordance with CA principles and
the development of mental models.
4.1. Results of task analyses
We realized a task analysis as part of the second
replication study in order to get information concerning the
degree of difficulty and the content-related quality of the
learning tasks in coaching, scaffolding, and exploration. We
asked three experts in the field of economics to analyze the
learning tasks, applying the aforementioned procedure of
Hacker et al. (1998). The experts’ judgments concerning the
dimensions of the task analysis obtained an average
agreement of at least 65% (Table 2).
The results indicate that the experts estimated the
dimension ‘complexity’, ‘dynamics’, and ‘transparency’ to
be most important for the task difficulty whereas the
dimensions ‘inconsistent goals’ and ‘degrees of freedom’
were rated as less important. The task analysis also
demonstrated that the instruction satisfied the apprentice-
ship features of increasing complexity and variety as well as
of increasing abstraction of the learning tasks in coaching,
scaffolding, and exploration. As a result of the expert’s task
analysis, three learning tasks of scaffolding and two of the
five exploration tasks were crossed from the list due to
evident weaknesses in content and overly strong demands
on knowledge.
In addition to the experts’ task analysis, we also asked
students to estimate the difficulty of the learning tasks
within the various apprenticeship methods. In the second
study, 84 subjects rated the difficulty of the coaching-tasks
as ‘not difficult’, whereas the learning tasks of scaffolding
and exploration were judged to be ‘more difficult’ and
‘difficult’ to solve. Within each method there were no
significant differences concerning the difficulty of the
learning tasks. However, there were significant differences
Table 2
Average agreements of experts and judgments on the relevance of the dimensions
Dimensions Inconsistency
of goals
Level of
complexity
Level of
transparency
Degree of
freedom
Level of
dynamics
Necessity of
prior knowledge
Agreement of
experts on average (%)
69.8 73.0 72.0 73.0 65.0 78.9
Average relevance
of the dimensions
1.7 4.3 3.3 2.3 4.0 2.6
N.M. Seel, K. Schenk / Evaluation and Program Planning 26 (2003) 215–224220
between these methods, coaching being the least difficult,
scaffolding more difficult, and exploration the most difficult
of the three. Interestingly, this increase of cognitive effort
involved in the solution of the learning tasks depending on
the method of apprenticeship instruction correlated with the
cognitive performance as measured with the knowledge test
of Beck (1993). There was a correlation of r ¼ 0:469 with
the measures in coaching but only a correlation of r ¼
20:74 with the scaffolding tasks and of r ¼ 20:379 with
the exploration tasks. This result corresponds with the
objective results of the various evaluation studies, which
indicated that the learners could not perform the learning
tasks in scaffolding and exploration as well as the coached
tasks.
4.2. Results of the evaluation studies
On the whole, the results of the five evaluations with
more than 400 subjects allow the statement that the CA
approach can be considered to be a sound framework for the
ID of environments aiming at constructivist learning.
Results of the second evaluation study indicated that the
attempts to install articulation and reflection in the multi-
media program failed insofar as the written statements and
comments required of the learners were revealed to be not
effective with regard to the reflective thinking about
problem-solving procedures which was intended. These
results correspond with Casey’s (1996) observation that
“interaction with the computer can provide a great deal of
information on why learners make the choices they make,
but computer-based intelligence seems pale in comparison
to the open peer dialogue we observed learners having
during the testing of CI” (p. 83).
As a result of these findings we implemented articulation
and reflection in the third study and later in the form of a
particular ‘teach-back’-procedure in accordance with Sasse
(1991). Moreover, in the fifth replication study we also
applied the method of ‘generative teaching’ by Kourilsky
and Wittrock (1992), but only with moderate effects. Both
of these alternatives for the realization of articulation and
reflection in a cooperative manner should be investigated in
more detail in further studies.
All replication studies agree on the point that the learners
come out on top in accomplishing the learning tasks of
coaching. Successful learners are characterized both by
fewer mistakes in task solutions of coaching and by longer
learning times for these tasks, indicating a well-planned
method in accomplishing the learning tasks. This appren-
ticeship method evidently aims at ‘controlled content-
oriented learning’ and narrows down the learners to imitate
the expert model they are provided in modeling.
In all of the studies we observed a significant decrease of
performance in scaffolding, which is characterized by a fading
of external guidance maintained. The learning outcomes of all
replication studies show that learners had difficulty in
developing and applying their own problem-solving
strategies. The significant decrease of performance from
coaching to scaffolding, which was observable in all of the
replication studies, indicates that the learners obviously
could not progress effectively from content- to process-
oriented learning in the sense of an increasingly self-
regulated accomplishment of analogous tasks.
In designing scaffolding we followed the statements of
Collins et al. (1989) and the standard procedures of
cognitive task analysis (Jonassen et al., 1999). Accordingly,
we organized the tasks in a structure that we considered to
be best for the solution of learning tasks by analogy. That is,
we provided the learners with a learning task which they
could solve easily and then we increased the difficulty of
tasks until the learners were no longer able to solve them on
their own. As an alternative, Hmelo and Guzdial (1996)
considered the redesign of learning tasks in such a way that
they support task performance with the help of a
‘supplantation’ (Salomon, 1979) of those cognitive oper-
ations that are involved in the task solutions. Moreover, task
performance can be supported with cognitive tools which
give advice to learners by representing a problem and then
manipulating the representation in the process of finding a
solution. These forms of scaffolding are taken by Hmelo and
Guzdial (1996) as examples of ‘glass-box scaffolding’
which aims at giving help to learners in situations which
they cannot master on their own. Moreover, the results of
our studies concerning the efficacy of scaffolding also
correspond with other investigations aimed at constructivist
learning.
Comparable with the learning results in scaffolding, the
subjects did not perform the two transfer tasks administered
in exploration well. One of the tasks remained within the
same subject matter domain of fiscal policy and thus aimed
at a ‘near transfer’ whereas the second task required the
transfer of freshly acquired knowledge into the different
subject matter area of ecology (i.e. ‘far transfer’). As in
scaffolding, the subjects achieved only average results on
these transfer tasks in the various replication studies. This
result generally corresponds with the literature on learning
transfer (Seel, 2000) insofar as the learners’ strong fixation
on content-oriented learning was consistently to the
detriment of learning effectiveness. Actually, with regard
to the accomplishment of transfer tasks, student perform-
ance in coaching was revealed to be a significant predictor.
From a didactic point of view, these results are not
satisfactory, especially with regard to the results of
regression analyses, according to which scaffolding did
not influence the requested transfer in exploration.
Jonassen, Marra, & Palmer (2003) point out with regard
to the efficacy of so-called constructivist learning environ-
ments that they often do not meet the expectations of
students (and teachers) with regard to their effects on
learning. This in turn may lead to motivational problems in
students (Jonassen & Tessmer, 1996/1997). With regard to
the cognitive effects of the instruction on learning and
transfer, the data of our replication studies correspond with
N.M. Seel, K. Schenk / Evaluation and Program Planning 26 (2003) 215–224 221
the argument of Jonassen and Marra. However, with regard
to motivational effects, we found consistently positive
learning emotions as well as explicitly good and persistent
motivation to learn. Actually, the findings of the various
replication studies show that there is a strong tendency
towards a positive motivation to learn and related interests
for these students in the domain of economics. Therefore, it
was not surprising that in the fifth study the learners’
content-oriented interests contributed to the variance of
performance in accomplishing the learning tasks of
scaffolding. Taking into account this finding as well as the
observation that enduring learning strategies evidently did
achieve greater effects than metacognitive training, we can
conclude that instructional interventions of a short duration
are strongly limited with regard to their properties.
In the pilot study and in two replications, we investigated
the efficacy of subsuming worked examples under a general
schema of solving problems vs. inducing such a schema
from worked examples. The data of these studies proved to
be inconsistent insofar as the strategy of subsumption was
superior to the strategy of induction in the pilot study with
regard to both the amount of mistakes and the time needed
for accomplishment of the learning tasks in coaching,
whereas in the second study a superiority of the inductive
strategy could be assessed with regard to the performances
in scaffolding. However, the fourth replication indicated
that there are no significant differences between the two
teaching strategies. Only in one case, namely in the
measured ‘procedural knowledge’ in performing the second
transfer task in exploration, was the inductive instruction
revealed to be more effective.
With regard to the measurement of the learning-
dependent progression of mental models, in all of the
various replication studies we could observe significant
changes in the accomplishment in producing causal
diagrams—understood here as external representations of
mental models. Much clearer were the results of regression
analyses, which indicated that the quality of the accom-
plishment of the learning tasks of increasing complexity
were influenced both by the learning experiences in the
course of instruction and the enduring learning strategies of
the subjects. The empirical findings of the fifth study show
significant learning progress in all experimental groups
depending on the complexity of their causal diagrams. We
interpret the substantial changes of causal diagrams
observed as a proof for the general effectiveness of the
instructional intervention. Our findings also support the
theoretical assumption that subjective causal models—as
measured with the help of causal diagrams—are situation-
dependent constructions. Remaining within the same
context of contents, the learners did construct causal
diagrams as cognitive artifacts correlating only minimally
with each other. Instructionally relevant are the results of
our investigations concerning the ‘adoption’ of the concep-
tual models the students were provided with in the
instruction. Our data contradict the assumption that students
adapt externally provided models and apply them to solve
tasks. In our replication studies the learners’ causal
diagrams showed only minor similarities with the concep-
tual models provided during instruction. Although con-
tingency coefficients indicate that the causal diagrams are
not fully independent of the conceptual models, the
correlation was not significant. Basically, we can agree
with Mayer’s (1989) verdict that “students given model-
instruction may be more likely to build mental models of the
systems they are studying and to use these models to
generate creative solutions to transfer problems” (p. 47), but
at the same time it is clear that the students do not adapt the
conceptual model provided through instruction one-to-one.
The results of the fifth evaluation study (Schenk, 2003)
did not indicate that metacognitive training had any effect
on the performance in the scaffolding and exploration part
of the learning program, but it did show significant
differences in different qualities of metacognitive knowl-
edge. The decisive predictors for the performances in
scaffolding were the students’ content-oriented interests.
When metacogitive training was implemented, several
learning strategies had a significant effect on the accom-
plishment of the transfer tasks in exploration. This
corresponds with the results of previous investigations
(Seel & Dinter, 1995). However, the amount of metacog-
nitive statements correlated positively with performances,
measured by the complexity of causal diagrams and
declarative knowledge in scaffolding.
5. Conclusions
Over the project’s life it was fruitful to realize and to
evaluate the methods of the CA approach with the help of a
step-by-step formative evaluation strategy. Parallel with
this, we investigated the learning-dependent progression of
mental models with the help of a specific diagnosis that
confirmed central assumptions of the theory of mental
models.
In the course of five replication studies, the suitability
of CA for the design of particular multimedia learning
environments could be proved in general—exceptionally
the methods of articulation and reflection, for which the
realization of a ‘teach-back’ procedure proved to be more
effective with regard to the required metacognitive
activities. Independent of two instructional strategies
(subsumption vs. induction) that were applied to design
the methods of modeling and scaffolding, the subjects
achieved excellent performance in the accomplishment of
the learning tasks administered in coaching. This consist-
ently positive result in all of our evaluation studies
indicates that the multimedia instruction was very effective
for the improvement of content-oriented learning. In
general, scaffolding aimed at fading instructional advice
can be considered the weak spot of the multimedia
instruction. Although the subjects would have liked more
N.M. Seel, K. Schenk / Evaluation and Program Planning 26 (2003) 215–224222
influencing control and alternatives during instruction,
they were not able to construct satisfactory problem
solutions or transfer their knowledge to other fields of
content. So far our results correspond with observations
and empirical results of other studies, such as Casey
(1996), Chee (1996) and Lajoie and Lesgold (1989),
which suggest that CA principles are suitable for the ID of
learning environments. Reservations include the obser-
vation that it was very difficult to implement articulation
and reflection in multimedia environments, and that
scaffolding and exploration, both of which aim to improve
self-regulated learning, were less effective than expected.
Taking into consideration the experiences we made in
designing the CA environment, we do not suggest the
realization of the methods articulation and reflection for
metacognitive improvement in a computer-based format
since it has not yet been possible to implement an intelligent
tutoring system in the program that might be able to guide
the learners’ self-verbalization and reflective thinking
adequately. Rather, we suggest the application of coopera-
tive procedures such as the ‘teach-back’ procedure in order
to realize articulation and reflection.
The weak learning results our subjects achieved in
scaffolding in the several replication studies actually
contradict its popularity in the literature (Hmelo &
Guzdial, 1996; Jonassen et al., 2003). This may be due
to the fact that the subjects of our studies were
constrained by the instructional program and did not
receive additional advice through a teacher as advocated
by Palincsar (1986), who considers such a dialogue to be
a solid basis for effective scaffolding. The multimedia
instruction was not capable of adapting the learning tasks
to the individual learner. For example, it cannot adapt
the difficulty of a learning task to the learners’ abilities
in such a way that a learner’s missing knowledge could
be compensated. Furthermore, the multimedia instruction
did not make appropriate ‘cognitive tools’ available to
support the learners in accomplishing the learning tasks.
As in a study by Niegemann (1995), the available
program resources (e.g. icons, glossary, cribs) were
evidently not sufficient and they were not used
effectively by the subjects of the replication studies.
The lesson learned from the students’ average perform-
ance in scaffolding is that this method should be
primarily oriented towards the idea of ‘glass-box
scaffolding’ in the future, with the aim of expanding
the opportunities of multimedia learning with appropriate
cognitive tools. We believe that the consistently average
achievement in exploration as well as in scaffolding can
be explained by enduring socialization effects of school-
ing. Schooling is targeted primarily at content-oriented
learning along with the acquisition of a lot of content
knowledge. For the development and practice of
cognitive skills, less resources are used and therefore
process-oriented learning in the sense of the development
of problem-solving skills can be seen as a negligible
quantity. This observation corresponds with the critics of
Krohne (1977) as well as with the major results of
TIMSS (Baumert, Bos, & Watermann, 1999).
The results of our studies demonstrate that causal
diagrams can be considered to be a suitable method of
assessing mental models as knowledge constructions of a
higher order that develop depending on learning experi-
ences. The pedagogical lesson learned from empirical
results with regard to significant effects on the complexity
of causal diagrams is that enduring learning strategies and
previous learning experiences may have greater influence
on the accomplishment of transfer tasks than temporary
instructional strategies, which perhaps need more time to
become effective than we could offer in our studies.
Konrad (1998) argues that successful self-regulated
learning—which was our goal in implementing metacogni-
tive training in the fifth study—presupposes that the learners
possess not only strategic knowledge but also broad world
knowledge. These learners can also fall back on specific
previous experiences in various learning situations. There-
fore, in future we should investigate in more detail whether
it would be effective to teach novice learners the application
of metacognitive strategies in the course of the instruction of
new subject matter.
In examining these evaluation results, we conclude that
learners constructed situation-bound problem solutions
independently of teaching strategies provided during
instruction, and that an effective design of successful
learning environments presupposes the provision of cogni-
tive tools which facilitate and support individual model-
building and revision aimed at problem solving. In this
sense, it is doubtful whether multimedia environments
which guide students on a predetermined ‘learning way’ are
actually suitable to overcome the ‘problems with problem-
based learning’ (Hoffman & Ritchie, 1997).
Acknowledgements
We gratefully acknowledge financial support for this
research from a generous grant provided by the German
Research Association (Deutsche Forschungsgemeinschaft)
with Grant-No. Se 399/4-1-3.
References
Al-Diban, S. (2001). Padagogische Diagnose mentaler Modelle. Freiburg:
Albert-Ludwigs-Universitat.
Andrews, D. H., & Goodson, L. A. (1980). A comparative analysis of
models of instructional design. Journal of Instructional Development,
3(4), 2–16.
Baumert, J., Bos, W., & Watermann, R. (1999) (2. Aufl.). TIMSS/III.
Schulerleistungen in Mathematik und den Naturwissenschaften am
Ende der Sekundarstufe II im internationalen Vergleich. Zusammen-
fassung deskriptiver Ergebnisse, Berlin: Max-Planck-Institut fur
Bildungsforschung.
N.M. Seel, K. Schenk / Evaluation and Program Planning 26 (2003) 215–224 223
Beck, K. (1993). Dimensionen der okonomischen Bildung. Messinstru-
mente und Befunde. Nurnberg: Universitat (Abschlussbericht zum
DFG-Projekt Wirtschaftskundlicher Bildungs-Test [WBT]. Normier-
ung und internationaler Vergleich).
Carlson, H. L. (1991). Learning style and program design in interactive
multimedia. Educational Technology Research and Development,
39(3), 41–48.
Casey, C. (1996). Incorporating cognitive apprenticeship in multi-media.
Educational Technology Research and Development, 44(1), 71–84.
Chee, Y. S. (1996). Mind bridges: A distributed, multimedia learning
environment for collaborative knowledge building. International
Journal of Educational Telecommunications, 2(2/3), 137–153.
Collins, A., Brown, J. S., & Newman, S. E. (1989). Cognitive apprentice-
ship: Teaching the crafts of reading, writing, and mathematics. In L. B.
Resnick (Ed.), Knowing, learning, and instruction (pp. 453–494).
Hillsdale, NJ: Erlbaum.
Craik, K. J. W. (1943). The nature of explanation. Cambridge: Cambridge
University Press.
Cronbach, L. J. (1963). Course improvement through evaluation. Teacher
College Record, 64, 672–683.
Greeno, J. G. (1989). Situations, mental models, and generative knowledge.
In M. Klahr, & K. Kotovsky (Eds.), Complex Information Processing
(pp. 285–318). Hillsdale, NJ: Erlbaum.
Hacker, W., Sachse, P., & Schroda, F. (1998). Design thinking—Possible
ways to successful solutions in product development. In P. Badke-
Schaub, H. Birkhofer, & E. Frankenberger (Eds.), Designers—The key
to successful product development. London: Springer.
Hannafin, M. J. (1992). Emerging technologies, ISD, and learning
environments: critical perspectives. Educational Technology Research
and Development, 40(1), 49–63.
Hmelo, C. E., & Guzdial, M. (1996). Of black and glass boxes: Scaffolding
for doing and learning. Proceedings of the Second International
Conference on the Learning Sciences, Charlottesville, VA: Association
for the Advancement of Computers in Education, pp. 128–133.
Hoffman, B., & Ritchie, D. (1997). Using multimedia to overcome the
problems with problem based learning. Instructional Science, 25(2),
97–115.
Jarvela, S. (1995). The cognitive apprenticeship model in a technologically
rich learning environment: interpreting the learning interaction.
Learning and Instruction, 5(3), 237–259.
Jih, H. J., & Reeves, T. C. (1992). Mental models: a research focus for
interactive learning systems. Educational Technology Research and
Development, 40(3), 39–53.
Johnson-Laird, P. N. (1983). Mental models: Towards a cognitive science
of language, inference, and consciousness. Cambridge: Cambridge
University Press.
Johnson-Laird, P. N. (1989). Mental models. In M. I. Posner (Ed.),
Foundations of cognitive science (pp. 469–499). Cambridge, MA: MIT
Press.
Jonassen, D. H., Marra, R., & Palmer, B. (2003). Epistemological
development: An implicit entailment of constructivist learning
environments. In N. M. Seel, & S. Dijkstra (Eds.), Curriculum, plans
and processes of instructional design: International perspectives.
Mahwah, NJ: Erlbaum, in press.
Jonassen, D. H., & Tessmer, M. (1996/1997). An outcomes-based
taxonomy for instructional systems design, evaluation, and research.
Training Research Journal, 2, 11–46.
Jonassen, D. H., Tessmer, M., & Hannum, W. H. (1999). Task analysis
methods for instructional design. Mahwah, NJ: Erlbaum.
Konrad, K. (1998). Kooperatives lernen bei studierenden: forderung
metakognitiver selbstaußerungen und (meta)kognitive profile. Unter-
richtswissenschaft, 26(1), 67–87.
Kourilsky, M., & Wittrock, M. C. (1992). Generative teaching:
an enhancement strategy for the learning of economics in
cooperative groups. American Educational Research Journal,
29(4), 861–876.
Krohne, H. W. (1977). Kognitive Strukturiertheit als Bedingung und Ziel
schulischen Lernens. Zeitschrift fur Entwicklungspsychologie und
Padagogische Psychologie, 9(1), 54–75.
Lajoie, S. P., & Lesgold, A. (1989). Apprenticeship training in the
workplace: computer-coached practice environment as a new form of
apprenticeship. Machine-Mediated Learning, 3, 7–28.
Mayer, R. E. (1989). Models for understanding. Review of Educational
Research, 59(1), 43–64.
Mayer, R. E., Moreno, R., Boire, M., & Vagge, S. (1999). Maximizing
constructivist learning from multimedia communication by minimiz-
ing cognitive load. Journal of Educational Psychology, 91(4),
638–643.
Niegemann, H. M. (1995). Zum einfluß von modeling in einer
computergestutzten lernumgebung: quasi-experimentelle untersuchung
zur instruktionsdesign-theorie. Unterrichtswissenschaft, 23(1), 75–87.
Palincsar, A. S. (1986). The role of dialogue in providing scaffolded
instruction. Educational Psychologist, 2(1/2), 73–98.
Rollett, B., & Bartram, M. (1977). Anstrengungsvermeidungstest (AVT).
Braunschweig: Westermann.
Ross, S. M., & Morrison, G. R. (1997). Measurement and evaluation
approaches in instructional design: Historical roots and current
perspectives. In R. D. Tennyson, F. Schott, N. M. Seel, & S. Dijkstra
(Eds.), Instructional design: International perspectives (Vol. 1) (pp.
327–351). Theory, research, and models, Mahwah, NJ: Lawrence
Erlbaum.
Royer, J. M., Cisero, C. A., & Carlo, M. S. (1993). Techniques and
procedures for assessing cognitive skills. Review of Educational
Research, 63(2), 201–243.
Salomon, G. (1979). Interaction of media, cognition and learning. San
Francisco: Jossey Bass.
Sasse, M. (1991). How to t(r)ap users’ mental models. In M. J. Tauber, & D.
Ackermann (Eds.), (Vol. 2) (pp. 59–79). Mental models and human–
computer interaction, Amsterdam: North-Holland.
Schenk, K. (2003). Effekte metakognitives Trainings auf Lernen und
Problemlosen. Dresden: Saxoprint.
Scriven, M. (1967). In R. W. Tyler, R. M. Gagne, & M. Scriven (Eds.),
Perspectives of curriculum evaluation (Vol. 1) (pp. 39–83). AERA
monograph series on curriculum evaluation, Chicago: Rand
McNally.
Seel, N. M. (1980). Lernerleben im Geschichtsunterricht der Sekundarstufe
I. Eine experimentelle Analyse. Munchen: Minerva.
Seel, N. M. (1991). Weltwissen und mentale Modelle. Gottingen: Hogrefe.
Seel, N. M. (2000). Psychologie des Lernens. Munchen: Reinhardt.
Seel, N. M., Al-Diban, S., & Blumschein, P. (2000). Mental models and
instructional planning. In J. M. Spector, & T. M. Anderson (Eds.),
Integrated and holistic perspectives on learning, instruction and
technology: Understanding complexity (pp. 129–158). Dordrecht,
NL: Kluwer.
Seel, N. M., & Dinter, F. R. (1995). Instruction and mental model
progression: learner-dependent effects of teaching strategies on knowl-
edge acquisition and analogical transfer. Educational Research and
Evaluation, (1), 4–35.
Snow, R. E. (1990). New approaches to cognitive and conative assessment
in education. International Journal of Educational Research, 14(5),
455–473.
Tennyson, R. D., Schott, F., Seel, N. M., & Dijkstra, S. (Eds.), (1997).
Instructional design: International perspectives (Vol. 1). Theory,
research, and models, Mahwah, NJ: Erlbaum.
Wartofsky, M. W. (1979). Models: Representation and the scientific
understanding. Dordrecht: Reidel.
Weston, C., McAlpine, L., & Bordonaro, T. (1995). A model for
understanding formative evaluation in instructional design. Educational
Technology Research and Development, 43(3), 29–48.
Wild, K. P., & Schiefele, U. (1994). Lernstrategien im studium. Ergebnisse
zur faktorenstruktur und reliabilitat eines neuen fragebogens. Zeitschrift
fur Differentielle und Diagnostische Psychologie, 15, 185–200.
N.M. Seel, K. Schenk / Evaluation and Program Planning 26 (2003) 215–224224