Journal Prob Solving 1

Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

Contemporary Educational Psychology 42 (2015) 116

Contents lists available at ScienceDirect

Contemporary Educational Psychology


j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / c e d p s y c h

Models of learning progress in solving complex problems: Expertise


development in teaching and learning
Min Kyu Kim *
Research Laboratory for Digital Learning, College of Education and Human Ecology, Ohio State University, 310M Ramseyer Hall, 29 West Woodruff Avenue,
Columbus, OH 43210, USA

A R T I C L E

I N F O

Article history:
Available online 28 March 2015
Keywords:
Learning progress
Cognitive change
Problem solving
Mental models
Latent class

A B S T R A C T

This study proposes that learning is a process of transitioning from one stage to another stage within a
knowledge base that features concepts and relations. Drawing on the theories of expertise, this study
explored two different models of learning progress (i.e., three- and two-stage models) in the context of
classroom learning and identied a model that was a good t to the data. Participants in this investigation included 136 students and 7 professors from six different universities in the United States. In order
to detect and validate stages of learning progress in participants written responses to an ill-structured
and complex problem scenario, this study utilized Exploratory Factor Analysis (EFA) and the Continuous Log-Linear Cognitive Diagnostic Model (C-LCDM) method (Bozard, 2010). The results demonstrate
that the three latent classes matched the three stages of the three-stage model. This study provides an
account of a diagnostic model of learning progress and associated assessment methods, yet further studies
are required to investigate different conditions.
2015 Elsevier Inc. All rights reserved.

1. Introduction
Learning is fundamentally about the change in knowledge and
skills needed to solve problems (Bransford, Brown, & Cocking, 2000;
Spector, 2004). Over the past few decades, many studies have addressed how people learn through complex problem solving in
diverse disciplines, including cognitive psychologies (Sinnott, 1989),
applied psychologies (Zsambok & Klein, 1997), and educational psychologies (Jonassen, 1997; Spiro, Feltovich, & Coulson, 1996), yet
understanding changes in the ability to solve a problem within the
context of classroom teaching remains a challenge due to the uncertain and complex nature of problems (Choi & Lee, 2009) and
changes that take place in the short term.
The theories of mental models and expertise development can
help address these issues. The theory of mental models explains that
problem solving involves a process of building mental representations of a problem situation (Johnson-Laird, 1983). Mental models
are holistic structural representations of the facts, concepts, variables, objects, and their relationships within a problem situation
(Dochy, Segers, Van den Bossche, & Gijbels, 2003; Jonassen, Beissner,
& Yacci, 1993; Segers, 1997). Cognitive change takes place when
learners confront unfamiliar and challenging situations (diSessa,

* Department of Educational Studies, College of Education and Human Ecology,


Ohio State University, 310M Ramseyer Hall, 29 West Woodruff Avenue, Columbus,
OH 43210, USA. Fax: 1-614-292-7020.
E-mail address: [email protected]
http://dx.doi.org/10.1016/j.cedpsych.2015.03.005
0361-476X/ 2015 Elsevier Inc. All rights reserved.

2006; Festinger, 1962; Piaget, 1964) or a pre-structural lack of knowledge (Biggs & Collis, 1982). When striving to resolve problem
situations, learners experience changes in their mental representations whereby the problem situations are recognized, dened, and
organized (Seel, 2003, 2004).
Learners possibly experience qualitatively different levels of
knowledge structure when engaged in problem solving. In line with
the idea that children experience qualitatively distinct but sequential knowledge states (Piaget, 1964), developmental psychologists
have seen that learning and development evolve as the learner constructs a qualitatively distinct knowledge structure (Alexander, 2003,
2004; Flavell & Miller, 1998; Siegler, 2005; Siegler, Thompson, &
Opfer, 2009; Vygotsky, 1934/1978; Werner, 1957). A number of experimental studies have demonstrated that qualitatively different
cognitive stages take place when learners respond to problems in
both the short term and the long term (e.g., Alexander & Murphy,
1998; Chen & Siegler, 2000; Opfer & Siegler, 2004; Siegler et al., 2009;
Vosniadou & Vamvakoussi, 2008).
Expertise studies have sought evidence to explain the development of expertise. As Fig. 1 illustrates, traditional expert-novice
studies dene an expert as one who consistently and successfully
performs in a specic, selected domain due to a highly enhanced,
ecient, and effective long-term memory (Ericsson & Kintsch, 1995;
Flavell, 1985; Simon & Chase, 1973). In addition, contemporary
studies of expertise do not restrict expertise development to chunking and pattern recognition. For example, Ericsson (2003, 2005, 2006)
suggested that expertise is developed by deliberate practice in which
learners engage in appropriate and challenging tasks carefully chosen

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

Fig. 1. The focus of expertise development studies.

by masters, devote years of practice to improve their performance, and rene their cognitive mechanisms through selfmonitoring efforts, such as planning, reection, and evaluation. From
this point of view, expertise development in a particular domain
requires long-term devotion (e.g., ten or more years) to disciplined and focused practice (Ericsson, 2003, 2005, 2006).
This study addresses two major limitations often found in the
traditional approach. One is that traditional studies tend to contrast two extremes, experts and novices, resulting in lack of
developmental focus (Alexander, 2004, p. 278). Missing accounts
of the middle stages leave the developmental process somewhat
unclear. Furthermore, Alexander (2004) argued that the educators
cannot condently apply the ndings of research conducted in a
non-academic setting.
In response to these problems, Alexander (2003, 2004) developed a Model of Domain Learning (MDL) to explain multiple stages
of expertise development in the educational context and validated the model (Alexander & Murphy, 1998; Alexander, Sperl, Buehl,
Fives, & Chiu, 2004). In spite of the unique value of the model, it
still assumes that long-term changes during schooling are required to master a domain.
The current study shifts the focus of expertise development to
problem-solving situations (i.e., task level) in the classroom and examines how short-term changes can lead to expertise. Several studies
(e.g., Chi, 2006; Newell & Simon, 1972) have suggested that an individuals understanding of a problem situation reects levels of
expertise in solving that problem. Based on the same idea, Gobet
and Wood (1999) tested an explicit knowledge model of learning
and expertise for the design of computer-based tutoring systems.
Concerning problem solving in the classroom, recent studies have
investigated differences between and changes in expertise (Ifenthaler,
Masduki, & Seel, 2009; Schlomske & Pirnay-Dummer, 2008; Spector
& Koszalka, 2004). For example, Ifenthaler et al. (2009) investigated longitudinal changes using hierarchical linear modeling (HLM)
techniques. However, a common limitation of these studies is a lack
of developmental focus (Alexander, 2004, p. 278), either simplifying the difference to expert vs. novice or falling short of explicitly
modeling stages of expertise development.
Using conrmatory analysis, the current study explored potential models of learning progress in solving ill-structured, complex
problem situations in a classroom setting. A valid model can provide

better insight into the cognitive development associated with solving


complex problems, insight that is essential to adapt teaching and
learning to individual differences in learning with precision and condence (Grow-Maienza, Hahn, & Joo, 2001; Hattie, 2009; Stigler &
Stevenson, 1991). The following research questions guided the study:
1. Which theories explain learning progress as expertise development in solving a complex problem?
2. What are some potential models of learning progress that are
both theoretically rigorous and viable?
3. Which model best explains the stages of learning progress?
2. Learning progress in solving ill-structured and
complex problems
2.1. Ill-structured and complex problems
Problems in professional and daily life are often characterized
by their uncertainty and complexity. Ill-structuredness refers to the
vagueness and unknown features of a problem, while complexity
refers to the large number of problem variables and the dynamic
relationships among them (Chi & Glaser, 1985; Eseryel, Ifenthaler,
& Ge, 2013; Jonassen, 1997; Kitchner, 1983; Sinnott, 1989; Wood,
1983). According to Alexander (1997), who distinguished topical
knowledge from domain knowledge, an ill-structured and complex
problem situation, especially in the context of teaching, can be associated with both. The topical knowledge of a complex problem
situation might be specic and deep in a particular context (e.g.,
identifying the complex nature of global warming and its connection to frequent wild res in California), while its domain knowledge
might cover a wide range of knowledge within a eld (e.g., ecology)
(Schraw, Dunkle, & Bendixen, 1995; Shin, Jonassen, & MaGee, 2003).
The context of the problem might be unknown or vague to a certain
degree. Due to the problems ill-dened and complex nature and
the different levels of topical and domain knowledge among the students, responses to the problem could reveal diverse perspectives
and conicting solutions (Chi & Glaser, 1985; Jonassen, 1997, 2000;
Kitchner, 1983; Shin et al., 2003; Wood, 1983).
Many models have accounted for ill-structured problem solving
as a procedural mental activity (Jonassen, 1997; Pretz, Naples, &
Sternberg, 2003; Sinnott, 1989). Pretz et al. (2003) identied seven

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

primary activities: (a) recognizing the problem, (b) dening and representing the problem, (c) developing a solution strategy, (d)
organizing ones knowledge about the problem, (e) allocating mental
resources for solving the problem, (f) monitoring ones progress
toward the goal, and (h) evaluating the solution. These problemsolving activities can typically be summarized into two main phases:
(a) problem representation and (b) solution development. Concerning the rst phase, Newell and Simon (1972) explained that a
problem solver conceptualizes the problem space in which all of
the possible conditions of a problem exist. Studies of expertise have
demonstrated clear distinctions among the mentally represented
problem spaces of experts and novices (Chase & Ericsson, 1982; Chi,
Glaser, & Farr, 1988; Ericsson, 2005; Ericsson & Staszewsli, 1989;
Spector, 2008; Spector & Koszalka, 2004). The focus of the current
study was to model the stages of building a problem space (i.e.,
problem representation).
2.2. Learning progress as changes in an individuals problem space
Learning progress can be dened as a series of gradual or sudden
changes in a learners understanding (i.e., problem space) that are
facilitated by instruction. The theory of mental models accounts for
these cognitive changes. Mental models are cognitive artifacts that
a learner constructs in order to understand a given problem situation (Anzai & Yokoyama, 1984; Craik, 1943; Greeno, 1989;
Johnson-Laird, 2005a, 2005b; Kieras & Bovair, 1984; Mayer, 1989;
Norman, 1983; Seel, 2003, 2004). A problem situation is mentally
represented in a learners mind when he or she is involved in the
problem-solving process. Mental models depend primarily on a learners prior knowledge (i.e., prior mental models), and they change
over time (Seel, 1999, 2001). Mental model changes in a learning
situation are not simple shifts without direction but transformations that lead closer to a pre-dened goal. In this way, such changes
can be considered progress. Thus, progress in mental modeling involves learning-dependent and developmental transitions between
preconceptions and causal explanations (Anzai & Yokoyama, 1984;
Carley & Palmquist, 1992; Collins & Gentner, 1987; Johnson-Laird,
1983; Mayer, 1989; Seel, 2001, 2003, 2004; Seel & Dinter, 1995; Shute
& Zapata-Rivera, 2008; Smith, diSessa, & Roschelle, 1993; Snow,
1990).
Mental models improve over time as a learner develops mastery
in a given problem situation (Johnson-Laird, 2005a, 2005b; Seel,
2003, 2004). Progress in mental models might involve both quantitative and qualitative change. For instance, a mental model might
be enlarged when a learner imports a new concept into an existing model (e.g., A sharp increase in the world population could be
a major cause of global warming.). Or a mental model might fundamentally change to adapt to a new situation (e.g., The current wild
re rate is not that high because this frequency has been observed
in the past. Therefore, global warming might not be associated with
wild res.).
2.3. 3S knowledge structure dimensions: surface, structural, and
semantic dimension
Studies have shown that mental models progress as structural
knowledge heightens (Johnson-Laird, 2005a, 2005b; Jonassen et al.,
1993). When problem solving is dened as a mental activity that
relies on structurally represented mental models (Dochy et al., 2003;
Segers, 1997), the assessment of problem-solving knowledge and
skills must be sensitive to the structural characteristics of the knowledge base (Gijbels, Dochy, Van den Bossche, & Segers, 2005).
Some have argued that knowledge structure consists of (a)
surface, (b) structural, and (c) semantic dimensions (Ifenthaler, 2006;
Kim, 2012b; Pirnay-Dummer, 2006; Spector & Koszalka, 2004). For
example, Kim (2012b) demonstrated a theoretical basis for these

dimensions, which had been conrmed in various studies on analogy


(Gentner & Medina, 1998; Holyoak & Koh, 1987; Simon & Hayes,
1976) and linguistic comprehension (Bransford & Franks, 1972;
Bransford, Barclay, & Franks, 1972; Bransford & Johnson, 1972;
Bransford & Johnson, 1972; Fodor, Bever, & Garrett, 1974; Gentner
& Medina, 1998; Katz & Postal, 1964; Kintsch & van Dijk). First,
surface dimension includes descriptive information of knowledge
components (i.e., concepts and their relations). This type of information is compatible with the surface level of mental models (i.e.,
the relevant objects and aspects of the context) (Holyoak & Koh,
1987; Simon & Hayes, 1976). According to linguistic comprehension studies, surface features, such as the relationships among nouns
in a text, characterize the shape of linguistic representations when
re-represented as mental models (Fodor et al., 1974; Katz & Postal,
1964).
Second, structural dimension includes the complexity and cohesiveness of a mental model as a whole. This dimension refers to
a deep level of well-organized knowledge within a particular context,
including underlying causal principles, key variables, and their connections (Bransford & Franks, 1972; Bransford & Johnson, 1972;
Gentner & Medina, 1998; Katz & Postal, 1964; Kintsch & van Dijk,
1978).
Semantic dimension includes understanding of concepts and their
relations in a knowledge structure. While surface (e.g., the number
of concepts) and structural features (e.g., the complexity of a knowledge structure) refer to the generic information of an entire structure,
semantic features are the individual concepts and propositional relations of a particular pair of concepts. The semantic dimension
includes principle variables that emerge from information integrated into the whole structure (Bransford et al., 1972; Bransford
& Franks, 1972; Bransford & Johnson, 1972; Katz & Postal, 1964; Kim,
2013; Kintsch & van Dijk, 1978).
Some studies have traced conceptual changes in response to a
problem situation using parameters to quantify knowledge structure dimensions (Ifenthaler et al., 2009; Ifenthaler & Seel, 2005;
Schlomske & Pirnay-Dummer, 2008). Two major limitations have
been found in the literature. First, no comprehensive account of
learning status has incorporated multiple knowledge structure dimensions. Second, and more seriously, many studies have analyzed
learning progress based on the assumption of linear growth, resulting in an oversimplied theoretical model of learning progress.
The current study proposed that the three dimensions are not
consistent across all learners but reect how individuals develop
their own mental models for solving a problem. The congurations of knowledge dimensions describing the various stages of
learning progress are likely to be as diverse as the learners
themselves.
3. Models of learning progress
Conceptual models of learning progress can be derived from
studies of expertise development. Recent studies have focused on
evolving expertise through learning and instruction (e.g., Kim, 2012a,
2012b; Alexander, 2003, 2004; Chi, 2006; Ericsson, 2003, 2005;
Ericsson & Simon, 1980, 1993). Building on expertise development models, the current study suggests two potential candidate
models that account for learning progress in complex problem
solving: a three-stage model and a two-stage model (Table 1).
3.1. Three-stage model
Alexander (2003, 2004) introduced a multi-stage model of expertise development that focuses on academic domain learning. The
Model of Domain Learning (Alexander, 2003, 2004; Alexander
et al., 2004) features three stages: Acclimation, Competence, and
Prociency-Expertise. This three-stage model of expertise was

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

Table 1
Models of learning progress in the development of expertise.
Level

Three-stage model

Two-stage model

1
2
3

Acclimation
Competence
Prociency

Misconception
Conception

empirically identied using analytic methods such as cluster analysis (Alexander & Murphy, 1998; Alexander et al., 2004). While
Alexanders Model of Domain Learning theorizes that expertise develops in the long term, the current study deploys the three-stage
model to explain short-term cognitive changes while solving a
problem in a classroom setting. The premise that qualitatively distinct changes take place remains the same in both the short term
and the long term (Siegler et al., 2009; Vosniadou & Vamvakoussi,
2008; Vygotsky, 1934/1978; Werner, 1957).
Table 2 describes the knowledge structure that characterizes each
stage according to the current study. The acclimation stage refers
to the initial point at which a learner becomes familiar with an unfamiliar problem. Accordingly, learners at this stage typically have
a pre-structural lack of knowledge (Biggs & Collis, 1982). Another
type of acclimating learner is one who has only limited and fragmented knowledge that is neither cohesive nor interconnected.
Learners become more competent in understanding a problem
through teaching and learning. For some competent learners, knowledge structures contain a small number of concepts and relations
that appear to be structured well but still require a better sense of
what is important in a particular situation. Other competent learners internalize a larger number of principles that have been taught
and exhibit a good semantic dimension, but poorly organized in contextual concepts.
The last stage is prociency-expertise. In this stage, procient learners, with increasing experience, construct sucient
contextual information and organize their knowledge base for a
problem situation that includes some domain-specic principles.
They conceptualize a sucient problem space in all three dimensions that accommodates the real features of a given problem
situation. The probability of resolving a problem markedly increases. Procient learners sometimes represent a relatively small
but ecient knowledge structure in which sucient key concepts
are well organized (good structural and semantic dimensions with
a surface dimension that is lacking). These knowledge structures
are in accord with the idea that experts sometimes create mental
models that contain an optimal rather than maximum number
of concepts and relations (Glaser, Abelson, & Garrison, 1983).
3.2. Two-stage model
The two-stage model is another possible candidate to explain
learning progress in the classroom setting where a particular problem
is presented. Some studies of conceptual change provide accounts
that suggest why only two stages are likely to occur.

Some have argued that conceptual change often requires shifts


in an entire knowledge base (Chi, 2008; Vosniadou & Vamvakoussi,
2008). From this point of view, a student with theoretical framework A (an inaccurate structure) shifts to theoretical framework B
(an expected structure). This shift from one model to another might
abruptly happen after an initially slower process. Studies have
claimed that radical conceptual changes usually happen at the end
of a slow process (Hatano & Inagaki, 1994; Vosniadou, 2003).
Vosniadou and Vamvakoussi (2008) contended that the slow and
gradual enrichment of knowledge is largely unconscious and that
the enrichment mechanism leads to conceptual changes in the long
run. For example, a recent experiment conducted by Siegler et al.
(2009) conrmed that stage transitions can take place suddenly, after
a given mental model has been in place for some time. They called
this transition a logarithmic-to-linear shift. In their experiment, children estimated the position of a number in a line to represent
numeric magnitude. Their estimations showed probabilistic patterns, which moved from being stable to approximating to the actual
value. This shift occurred abruptly after applying an initial approach numerous times. In another instance, some third-grade
students in a science class strongly believed that the earth is at
based on their naive theory. Even after instruction, they refused to
accept that the earth is spherical and continued to justify
their initial belief. Even some adults, interestingly, retain this
belief (one global society still argues that the earth is at;
http://theatearthsociety.org/forum/). After further instruction and
additional lessons, they might begin to doubt their naive theory and,
at some point, build a new one (knowledge base) that integrates
new evidence and principles to support a spherical earth.
This transition from a naive theory to an informed theory can
abruptly take place at some point during a slow and gradual process
of learning and instruction. From this point of view, the immature
knowledge status might have two causes: (a) a pre-structural lack
of knowledge (Biggs & Collis, 1982) and (b) a strong misconception. Middle stages are likely to exist, but they might have short
duration and occur just before an appropriate knowledge structure takes shape. If this situation is true, measuring middle stages
would prove dicult.
4. Methods
4.1. Participants
Participants in this investigation included 136 students and 7 professors in the area of Educational Technology. All students were
enrolled in an introductory educational technology course at a large
university in the southeastern United States. The course, designed
to equip pre-service teachers to apply learning technologies in classroom teaching, consisted of two components: (a) introducing
technology tools related to National Educational Technology Standards (NETS) and (b) designing technology-supported/enhanced
lesson activities to meet NETS. The course was delivered through
four sessions: one by a full-time instructor (T2) and three by three

Table 2
Three-stage model of learning progress.
Stage

Three dimensions of knowledge structure (3S)

Acclimation

(a) All dimensions (surface, structural, and semantic) are quite dissimilar to expert models; or (b) knowledge structures have similar
surface dimension to expert models but are missing some structural and semantic dimensions.
(a) Structural dimension might appear to be mastered because mental models, which consist of a small amount of contextual and
abstract knowledge, are likely to look cohesive and connected; or (b) student and expert models are highly similar in semantic
dimension but dissimilar in surface and structural dimensions.
(a) Structural dimension shows sucient complexity while surface dimension is adequate, but not enough to guarantee a semantic
t; (b) knowledge structures are well-featured in all dimensions (surface, structure, and semantic); or (c) a signicant number of
principles (semantic) create a cohesive structure (structural) but with a small number of concepts (surface).

Competence

Prociency-expertise

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

Table 3
Participant demographics.

Total
Gender
Class year

Course taken

Teacher experience

Teaching career

a
b

Female
Male
Freshman
Sophomore
Junior
Senior
None
ET-relateda
Otherb
Less than 2 years
No
Missing
Very interested
Interested
Somewhat
Not so
Not at all

Total

T1

T2

T3

T4

N/A

136
113 (83.1%)
23 (16.9%)
9 (6.6%)
38 (27.9%)
52 (38.2%)
37 (27.2%)
56 (42.2%)
8 (5.9%)
72 (52.9%)
6 (4.4%)
127 (93.4%)
3 (2.2%)
33 (24.3%)
19 (14.0%)
32 (23.5%)
33 (24.3%)
19 (14.0%)

13
11
2
0
5
4
4
3
2
8
0
13

4
4
2
2
1

76
62
14
6
16
31
23
37
4
35
5
70

20
7
18
20
11

17
14
3
1
8
3
5
9
0
8
0
16

3
0
6
4
4

25
22
3
2
8
13
2
6
2
17
1
24

5
8
5
5
2

5
4
1
0
1
1
1
1
0
4
0
4

1
0
1
2
1

Courses that are related to Educational Technology.


Education-related courses not directly about educational technology.

respective doctorate-level teaching assistants majoring in Instructional Technology (T1, T3, and T4). The full-time instructor had six
years teaching experience in middle schools and had taught the
course for over six years. For quality control, the instructor supervised the sessions taught by the three doctoral students.
Most participating students were undergraduate pre-service
teachers who had considered teaching in K-12. As Table 3 shows,
female students comprised 83% of the study participants, while 17%
were male. The majority of students were juniors and seniors (65%),
followed by sophomores (28%) and freshmen (7%). For 42% of the
students, the course was their rst related to educational technology, while only 6% had taken previous courses on the use of
educational technology. However, over half of students had taken
previous education-related courses (52%). In contrast to their coursework experience, over 93% of the students had no teaching
experience. Only 38% indicated a positive interest in being a teacher
even though they were all taking a course designed for preservice teachers. The same proportion of students (38%) indicated
little interest in teaching in K-12 schools.
Participants also included seven professors teaching at six different universities in the United States. They participated in the study
to build a reference model for a given problem situation. In order
to recruit these professors, we rst created a pool of potential candidates who were members of a scholarly association related to
Educational Technology and then made a short list of professors
based on pre-set criteria: (a) professors in Instructional Technology or related elds; (b) professors teaching a course titled
Instructional Design or Technology Integration in Learning; (c)
professors who research technology-integration in classroom learning; and (d) professors whose doctorates were received at least three
years ago. Seventeen professors were invited to participate in the
study, and seven professors were willing to contribute their expertise to the research.

4.2. Data collection


4.2.1. Problem task
This study collected participants responses to an ill-structured
and complex problem task, a simulated situation that participants
were asked to evaluate. The simulated situation was a failed attempt
to adapt a technology (i.e., tablet PC) to classroom teaching. Designed to elicit participants knowledge, the questions asked them
to describe, in explicit detail, the concepts, issues, factors, and variables likely to have contributed to the result of the project:
introducing tablet PCs had very little effect on the instructional practices employed in the classroom (see Appendix A).
In the third week of the semester, the instructors asked participating students to respond to the problem situation as an in-class
problem-solving activity. The participants wrote their responses using
natural language, which was required for two reasons. First, using
natural language enables individuals to verbalize their understanding of a problem situation and some feasible solutions (Kim, 2013).
Second, using natural language is more likely to provide a reliable
foundation whereby a descriptive knowledge representation can be
elicited (Kim, 2012b, 2013; Pirnay-Dummer & Ifenthaler, 2010;
Pirnay-Dummer, Ifenthaler, & Spector, 2010).
4.2.2. Reference model
Using Delphi survey procedures, seven professors established a
reference model against which student models could be compared (Goodman, 1987; Hsu & Sandford, 2007; Okoli & Pawlowski,
2004). The Delphi survey included three rounds to rene the reference model (see Table 4). In the rst round, the professors created
their own responses to the problem. These initial responses were
consolidated for use in the second round and also used as individual cases for analysis. Next, the consolidated responses and a list
of identied concepts were sent to the panel. The experts were asked

Table 4
Delphi procedure.
Round

Activity

R1. Brainstorming
R2. Narrowing down/ranking

Collect and consolidate all responses from experts


Send rened nal version of consolidated lists, including statements and used concepts
Ask experts to add comments if they disagree with or have different opinions about a statement
Ask experts to rank key statements and concepts
Send each panelist ranked statements and concepts summarized by the investigators
Ask for revision of judgments or specication of reasons for remaining outside the consensus

R3. Renement
Note. Adapted from Kim (2013, p. 960).

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

to comment on the listed statements and concepts and rank them.


After gathering the second-round surveys, the researcher created
a nal list of ranked statements and concepts. Based on the summary,
a draft of the reference model was created. In the nal round, the
results of the second survey were sent to the panel and revised according to their comments, where necessary. The outcome of these
procedures was a written reference model containing 23 key concepts identied by the panel (see Appendix B). The reference model
was believed to include most of the concepts and relations required to form a knowledge structure representing the problem
situation.
4.3. Data analysis
4.3.1. Procedure
This study attempted to validate the models of learning progress in problem solving using empirical data gathered from written
responses. This process entailed transforming lexical data into numerical data that described the knowledge structure dimensions
and then analyzing the transformed data using a particular set of
analytic techniques.
Data transformation was conducted using a three-step concept
mapping technique (Clariana, 2010; Curtis & Davis, 2003; Kim, 2012a,
2013; Narayanan, 2005; Novak & Caas, 2006; Spector & Koszalka,
2004; Taricani & Clariana, 2006): (a) elicit concepts and relations
in a text, (b) construct concept maps drawn by graph theory (Rupp,
Sweet, & Choi, 2010a; Schvaneveldt, Durso, Goldsmith, Breen, &
Cooke, 1985; Wasserman & Faust, 1994), and (c) compare the concept
maps to a reference model.
The rst step was to elicit judgments about concept relations in
students written responses to the problem. This study made judgments in compliance with Kims (2012a, 2013) Semantic Relation
(SR) approach, which obtains semantic relations from the syntactic structure of a written response and then creates a concept map
that better represents a students internal mental representation.
The second step was to construct concept maps. This study used
the network analysis software NetMiner (http://www.netminer.com)
to draw concept maps and generate concept map parameters. This
tool automatically rendered the concept map parameters based on
indicators suggested for educational diagnostics. The nal step was
to compare student concept maps to a reference model. This comparison yielded similarity measures that were later used for model
validation. Similarity measures were calculated using a similarity
analysis tool developed by Kim (2012a) using the C++ programming language. For a more detailed account of this concept map
method, see Kim (2013).
The similarity measures were continuous variables ranging from
0 to 1 (see details in the following section). The similarity measures were rst reviewed using SPSS 21 (http://www-01
.ibm.com/software/analytics/spss/) to determine whether they violated the multivariate normality assumption and whether there were
outliers. Next, the validated data were analyzed using Exploratory
Factor Analysis (EFA) in order to discover any relations between the
similarity measures and the 3S knowledge structure dimensions (i.e.,
latent factors). Lastly, the models were validated with the Continuous Log-Linear Cognitive Diagnostic Model (C-LCDM) method
(Bozard, 2010). R software (http://www.r-project.org/) was used to
run EFA, and M-Plus version 6.0 (https://www.statmodel.com/) was
used for C-LCDM analysis.
4.3.2. Similarity measures as indicators of levels of understanding
The human mind is not easily observed, but cognitive activity
can be indirectly inferred from externalized representations, such
as written responses to a problem situation. Drawing on the aforementioned concept mapping technique, participants written
responses were transformed into concept maps (see examples in

Fig. 2), whereby a list of parameters might be obtained. In terms


of graph theory, Ifenthaler (2010) introduced parameters that might
be used to diagnose knowledge representation in an educational
setting. After reviewing these parameters, using network analysis
methods (Coronges, Stacy, & Valente, 2007; Ifenthaler, 2010;
Wasserman & Faust, 1994), the current study proposed a new set
of parameters that could demonstrate knowledge structure dimensions (see Table 6). For example, the parameter average degree
indicates the average number of relations (links) to or from a concept
(node), ranging from 0 to g-1 (g being the total number of nodes).
As average degree increases, the knowledge structure is considered more complex.
Evaluation of a students concept map is often accomplished by
comparing it with a reference model, which is usually elicited from
an experts response (Coronges et al., 2007; Curtis & Davis, 2003;
Goldsmith & Kraiger, 1997; Taricani & Clariana, 2006). This comparison assesses similarity measures by overlaying network patterns
with the concept map parameters (Coronges et al., 2007; Monge &
Contractor, 2003). Similarity measures at each measurement occasion can indicate how close a learner model is to a reference model.
The current study proposed eleven similarity measures applicable
to the study of cognitive change (see Table 5).
To compare parameters, two types of similarity formulas were
applied: (a) numerical similarity and (b) conceptual similarity. For
most of the similarity measures (i.e., number of concepts, number
of relations, average degree, density of graphs, mean distance, diameter, connectedness, and subgroups), the numerical similarity
formula was used because the parameters were numerical values.
On the whole, a similarity formula assumes that each half of a
pair is equally signicant. In the case of a concept model comparison, the reference model and student model are not equivalent in
terms of maturity. A reference model acts as a standard toward which
a student model is expected to progress. A reference model is likely
to contain a greater number of concepts and relations and is comprised of a larger and more complex knowledge structure than a
novice model (Chi et al., 1988; Spector & Koszalka, 2004). Thus, a
modied formula was used for those measures because an optimal
value, rather than a greater value, indicated a good condition (see
Table 5). When f1 was less than f2 (f1 < f2), the original numerical similarity formula was used so that

s = 1

f1 f2
max ( f1, f2 )

where f1 is the frequency of a student model and f2 is the frequency of a reference model. The similarity ranged from 0 to 1, 0 s 1.
Otherwise, if f1 was not less than f2 (f1 f2), the similarity value was
set to 1 because the student value was greater than the reference value, indicating that the student model exceeded the reference
model in the relevant criteria.
For the three similarity measures (concept, proposition, and balanced semantic matching score), conceptual similarity was applied
because these measures concern the proportion of fully identical
elements between two concept maps. The conceptual similarity
method used in this study applied Tverskys (1977) similarity
formula, which assumes that the similarity of object A and object
B is not merely a function of the features common to A and B but
relies on the unique features of each object. As in the case of numerical similarity, an adjustment was made. Just as a picture
resembles an object rather than the inverse, a student model resembles the reference model, which is more salient. In this
asymmetric relation, the features of the student model are weighted
more heavily than the reference model features (Colman & Shar,
2008; Tversky & Shar, 2004). When the conceptual similarities were
calculated using Tverskys (1977) formula, in this study was
weighted more heavily than ( = 0.7 and = 0.3).

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

Fig. 2. Examples of concept maps elicited from participants written responses.

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

Table 5
Similarity measures.
Similarity measure

1. Number of concepts
2. Number of relationsa
3. Average degree

Denition

Compare the number of concepts


(nodes) in two models
Compare the number of links (edges)
in two models
Compare the average number of
degrees in two models

4. Density of graphs

Compare the density of two models

5. Mean distance

Compare the mean distances in two


models
Compare the largest geodesics in two
models
Compare the ratios of paired nodes
that reach each other in each graph
Compare the number of cohesive
subgroups in each graph

6. Diametera
7. Connectedness
8. Subgroups

9. Concept matchinga

10. Propositional matchinga


11. Balanced semantic
matchinga
a
b
c
d

s=

Compare semantically identical


concepts, including contextual and
principle variables
Compare fully identical propositions
(edges) between two concept maps
Compare the balances calculated by
dividing Propositional matching by
Concept matching

Parameters comparedb
Technical denition

Operationalization

The total number of nodes (vertices)

The overall number of conceptsd

The total number of links (edges)

The overall number of relations of paired


conceptsd
As the number of incoming and outgoing
relations grows, the complexity of the
cognitive structure is considered higher.d
The density of a concept map indicates how
cohesive it is.

The average number of links of a node, ranging


from 0 and g-1 (g being the total number of
nodes)
The density of a graph is the proportion of
possible lines that are actually present in the
graph.
The average geodesicc distance between any
pair of nodes in a network
The length of the largest geodesic between any
pair of nodes (1 to g-1)
Ratio of pairs, nodes that can reach each other
in the graph.
Subsets of actors among which there are
relatively strong, direct, intense, frequent, or
positive ties.
Qualitative comparison

Indicates how close the concepts are to one


another.
Represents how broad the understanding of a
domain is.d
Describes the extent to which the concepts are
connected.d
The more complex a cognitive structure, the
more subgroups will intermediate the entire
network.d

Qualitative comparison
Based on the values of two similarities:
Concept and Propositional matching

These similarity measures are adopted from Pirnay-Dummer and Ifenthaler (2010).
Similarity measures are calculated by comparing the parameters that are suggested in Wasserman and Faust (1994).
The shortest path between two nodes is referred to as a geodesic.
These parameters are also introduced by Ifenthaler (2010).

f ( A B)
f ( A B) + i f ( A B) + i f (B A )

Each of the similarity measures was assumed to be associated


with at least one of the three dimensions (i.e., 3S: Surface, Structural, and Semantic) of knowledge structure. Presumably, the
similarity measures demonstrate how close a students knowledge structure features are to the reference model.
4.3.3. Validation method: continuous log-linear cognitive diagnostic
model (C-LCDM)
This study adapted Latent Class Models (LCM) to identify the
stages of learning progress. The main assumption was that the qualitative stages of learning progress can be labeled latent classes due
to their psychometric characteristics. LCMs consist of corresponding latent classes associated with the other stages so that on any
given occasion of measurement, each individual has an array of latent
class memberships (Collins & Cliff, 1990; Collins et al., 1994; Heinen,
1996; Kaplan, 2008; Rost & Langeheine, 1997).
As a special case of constrained LCM, Log-linear Cognitive Diagnostic Model (LCDM) allows latent class models to place linear
restrictions on the log-linear parameters when both observed variables and latent predictor variables are categorical, in particular
dichotomous (Agresti, 2007; Rupp, Templin, & Henson, 2010b). According to Templin (2004), latent predictor variables, often called
attributes, are the atomic components of ability, the specic skills
that together comprise the latent space of general ability (p. 8).
Attribute variables have a binary continuum: mastery vs. nonmastery. An individual attribute is considered mastery when its
probability is higher than 0.5 (Rupp et al., 2010b). LCDMs are used
to determine whether an individual masters the attributes required to answer items correctly. The attributes that have or have
not been mastered constitute an individuals mastery prole, which

ultimately denes his/her latent class membership (Henson, Templin,


& Willse, 2009).
The current study posited that the 3S dimensions of a students knowledge structure are knowledge attributes. The status
of each knowledge attribute is either mastery or non-mastery.
The 3S attributes of knowledge structure meet the assumption of
LCDM. However, this study did not directly use LCDM because it used
continuous observed variables (i.e., similarity measures) instead of
dichotomous variables.
Bozard (2010) proposed the Continuous LCDM in order to accommodate continuous models. She dened C-LCDM using the
following equation, which is similar to the CFA model:

xij = i 0 + i1 j1 + i 2 j 2 + i12 j1 j 2 + eij ,


where xij symbolizes the response to the observable variable i by
person j; i indicates factor loading (i.e., attribute loading) to the
item, which takes three types of the following values: intercept = 0, main effect = 1, two-way interactions = 2, etc.; jk symbolizes
latent variable value (i.e., attribute value) for individual j and latent
variable k (k = 2 in equation 1 above); and eij represents the uniqueness of the individual in observed indicator variables (residuals).
C-LCDMs contain a set of attribute mastery proles in that the
set of k latent attributes can be considered a latent class model with
2k classes (Henson et al., 2009; Templin, 2004). In the context of
modeling learning progress, the three dimensions of knowledge
structure (i.e., k = 3 latent attributes: surface, structural, and semantic) are attributes in C-LCDMs. Each dimension (attribute) of
knowledge structure can be proled as mastered/obtained (1) or
non-mastered/not obtained (0). Those mastery proles provide stages
(i.e., 23 = 8 latent classes) of learning progress at which an individual might be classied. Based on the three Models of Learning
Progress discussed earlier, these eight latent classes are associated with the stages of each model (see Table 6).

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

Table 6
Class-to-prole table.
Latent class

Class 1 (C1)
Class 2 (C2)
Class 3 (C3)
Class 4 (C4)
Class 5 (C5)
Class 6 (C6)
Class 7 (C7)
Class 8 (C8)

3S attribute

Stages of learning progress

S1

S2

S3

Three-stage model

Two-stage model

0
0
0
0
1
1
1
1

0
0
1
1
0
0
1
1

0
1
0
1
0
1
0
1

Acclimation
Competence
Competence
Prociency
Acclimation

Prociency
Prociency

Misconception

Conception
Misconception

Conception
Conception

Note: 3S attributes involve the three dimensions of knowledge structure: Surface


(S1), Structural (S2), and Semantic (S3). 0 = absent/non-mastered and
1 = present/mastered.

For instance, class 4 has a mastery prole (i.e., S1: 0, S2: 1, and
S3: 1) that is matched to the procient learner in the three-stage
model, and conception in the two-stage model. These matches were
determined by theoretical assumptions about knowledge status at
each level. For example, a procient learner represents a relatively small number of concepts and relations (S1: 0 = surface feature
is absent) but an ecient knowledge structure (S2: 1 = structure
feature is present) in which sucient key concepts are wellstructured (S3: 1 = semantic feature is present).

as continuous variables (0 s 1). As Table 7 presents, most similarity measures showed biased distributions. The similarity measures
M9 to M11 were distributed in the lower area of the similarity band,
ranging from 0 to 0.68, with lower means ranging from 0.05 to 0.24,
while the distributions of M3, M5, M6, and M7 ranged from 0.15
to 1.00, with means ranging from 0.73 to 0.82. M1, M2, M4, and M8
had means between 0.31 and 0.40. These descriptive results suggest
that the similarity measures likely indicate different constructs, which
the current study assumes to be the different dimensions of knowledge structure.
Next, as shown in Table 8, correlations among the similarity measures were calculated. Very high correlations (r > .9) were identied
among variables M1, M2, M4, and M8. Strong correlations also
emerged between two paired variables, M4 and M8 (r = 0.85) and
M10 and M11 (r = 0.86). These high correlations might provoke multicollinearity concerns for a general linear model, such as multivariate
regression; however, they were less problematic for C-LCDM because
LCDMs allow multiple items (i.e., observed variables) to measure
the same set or similar sets of attributes (Rupp et al., 2010b). Consequently, we retained all 11 measures for further analysis. The
overall KaiserMeyerOlkin (KMO) value was 0.7, within the range
of good (between 0.7 and 0.8), even though two individual measures were below the recommended value of > 0.5 (M7 = 0.42 and
M14 = 0.42). Barletts test of sphericity, x2 (55) = 2570.92, p < .01, indicated that correlations between variables were signicantly large;
therefore, factor analysis was appropriate.

5. Results
5.2. The relations between similarity measures and 3S knowledge
structure dimensions

5.1. Descriptive statistics


Outlier analysis for the similarity measures was conducted rst.
As a result, three student cases were eliminated, resulting in a sample
size of 140, including the seven experts. Then, we examined the descriptive statistics and correlations of the eleven similarity measures

Table 7
Descriptive statistics of similarity measures.

M1. Concept
M2. Relation
M3. Average degree
M4. Density
M5. Mean distance
M6. Diameter
M7. Connectedness
M8. Subgroups
M9. Concept matching
M10. Propositional matching
M11. Balanced matching

Minimum

Maximum

Mean

SD

140
140
140
140
140
140
140
140
140
140
140

0.04
0.02
0.42
0.05
0.28
0.14
0.15
0.00
0.05
0.00
0.00

0.96
0.98
1.00
0.96
0.99
1.00
1.00
1.00
0.57
0.28
0.68

0.35
0.31
0.82
0.40
0.73
0.75
0.73
0.30
0.24
0.05
0.18

0.18
0.20
0.14
0.19
0.17
0.22
0.25
0.20
0.08
0.04
0.15

We anticipated that the eleven similarity measures would be associated with three latent factors: the surface, structural, and
semantic dimensions of knowledge structure. In order to test this
assumption, we ran an initial analysis to determine the number of
factors we had to extract from the data. The principle component
analysis calculated eigenvalues for each component and produced
two types of scree plots (see Fig. 3).
Three factors on the left graph had eigenvalues over Kaisers
criterion of 1, and both graphs suggested that these three factors
were above the cut-off point of inexion, explaining 76% of the
variance. Consequently, three factors were retained for nal
analysis.
Table 9 shows the factor loadings after oblique rotation. The similarity measures that clustered on the same factors suggest that factor
1 represents surface, factor 2 structural, and factor 3 semantic. For instance, M4. Density, clustered on factor 1 (surface), is
dened in graph theory as the proportion of existing links to all possible lines in the graph (Wasserman & Faust, 1994). Density is
associated with the number of existing links (i.e., a surface feature).

Table 8
Correlations of the similarity measures.

M1
M2
M3
M4
M5
M6
M7
M8
M9
M10
M11

M1

M2

M3

M4

M5

M6

M7

M8

M9

M10

M11

1
.97*
.60*
.95*
.64*
.65*
.01
.92*
.54*
.38*
0.16

1
.71*
.85*
.61*
.62*
.16
.92*
.56*
.43*
.21*

1
.39*
.65*
.62*
.51*
.63*
.49*
.37*
.24*

1
.59*
.63*
-.20*
.85*
.46*
.27*
0.08

1
.94*
.36*
.71*
.43*
.21*
0.08

1
.21*
.72*
.46*
.22*
0.08

1
.14
.06
.14
0.13

1
.50*
.37*
.15

1
.61*
0.26*

1
.86*

n = 140.
* Correlation is signicant at the 0.05 level (2-tailed).

10

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

Fig. 3. Number of factors to extract by Scree plots.

Among the similarity measures, three measures (M3. Average


degree, M7. Connectedness, and M9. Concept matching) loaded
on multiple factors. For example, the variance of M3. Average
degree was explained by surface and structural features. Interestingly, M7. Connectedness was negatively related to surface while
positively associated with structural. In addition, communality
analysis veried sample-size (N = 140) adequacy with values over
0.5, except M7, based on the suggestion that sample sizes between
100 and 200 are large enough to conduct factor analysis as long as
there are relatively few factors and the communality values are above
0.5 (Guadagnoli & Velicer, 1988).

Table 9
Summary of exploratory factor analysis.
Similarity measure

Communality

M1. Concept
M2. Relation
M3. Average Degree
M4. Density
M5. Mean Distance
M6. Diameter
M7. Connectedness
M8. Subgroups
M9. Concept matching
M10. Propositional matching
M11. Balanced matching

99.50
94.71
51.72
90.47
99.50
89.90
22.85
88.17
49.65
99.50
77.99

Oblique rotated factor loadings


Surface
1.01
0.96
(0.31)
1.00

(0.35)
0.82
(0.30)

Structure

Semantic

0.42
0.96
0.86
0.55
0.46
1.01
0.95

Note: Factor loadings in parentheses are excluded when the 0.4 cut-off value is applied.

Table 9 presents factor loadings, the absolute values of which were


all over 0.3, a reliable cut-off point. However, Steven (2002) claimed
that the signicance of factor loading is determined by the sample
size and the variance explained by a particular factor. According to
his suggestion, with a sample size of 140, factor loadings should be
greater than 0.4 and explain around 16% of the variance in a similarity measure. When we applied the cut-off value of 0.4, all but
three factors had single loadings on one item.
Having found that three factors (surface, structural, and semantic dimension) were associated with the similarity measures, we
created conrmatory models known as Q-matrix models. Like all
LCDMs, C-LCDM requires a substantive theoretical model whereby
researchers can interpret statistical classications as meaningful
latent classes. The Q-matrix is a hypothetical model that denes a
limited relationship between a set of attributes and a set of test items
(Templin, 2004). In the case of the current study, the value of each
similarity measure is dependent on mastering the designated attributes (required knowledge dimensions).
Table 10 describes the hypothesized relationships between the
11 similarity measures and the three dimensions of knowledge structure (i.e., attributes). In the table, an attribute is coded 1 when it
is required to raise the value of the measure. Two Q-matrices were
designed, one for each cut-off point. Using the 0.4 cut-off value, the
mastery associations in parentheses were excluded (i.e., for measures M3, M7, and M9). For convenience, this Q-matrix model was
called Main effect only (Model 1). Using the more relaxed 0.3 cutoff value, the Q-matrix had multi-loading on measures M3, M7, and
M9 and was called Interaction effect (Model 2). Also noteworthy

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

Table 10
Similarity measures associated with attributes of knowledge structure (Q-matrix).
Similarity measure

M1. Concept
M2. Relation
M3. Average degree
M4. Density
M5. Mean distance
M6. Diameter
M7. Connectedness
M8. Subgroups
M9. Concept matching
M10. Propositional matching
M11. Balanced matching

3S attribute

Table 12
The estimated nal class counts and proportions.
Latent class

Surface (S1)

Structure (S2)

Semantic (S3)

1
1
(1)
1
0
0
(1)
1
(1)
0
0

0
0
1
0
1
1
1
0
0
0
0

0
0
0
0
0
0
0
0
1
1
1

11

Class 1 (C1)
Class 2 (C2)
Class 3 (C3)
Class 4 (C4)
Class 5 (C5)
Class 6 (C6)
Class 7 (C7)
Class 8 (C8)

Q-matrix

Estimated classication (count)

S1

S2

S3

Model 1

Model 2

0
0
0
0
1
1
1
1

0
0
1
1
0
0
1
1

0
1
0
1
0
1
0
1

0.288 (38)

0.427 (62)

0.285 (40)

0.274 (37)

0.418 (60)

0.308 (43)

Stages of learning progress

Note: According to the factor loadings (see Table 9), the mastery associations in parentheses (i.e., for measures M3, M7, and M9) were excluded (cut-off value: 0.4).

is that the patterns of mastering attributes dened an individuals


latent classes (see Table 6).
5.3. Investigation of the models of learning progress using C-LCDM
Model estimation was conducted for Main effect only (Model 1)
and Interaction effect (Model 2) in order to obtain model-t indices.
Then, in order to select the best model, AIC and BIC indices were
compared because the likelihood ratio test tends to be unreliable
when using mixture analysis.
As Table 11 shows, Model 2, a complex model, had slightly better
AIC and BIC values, that were little smaller than the Model 1 values.
However, because the differences were so small, we decided to use
the simpler model (i.e., Model 1).
M-plus reported the estimated posterior probabilities and the
most likely latent class for each respondent. Table 12 describes the
posterior probability results, including the nal counts and the proportions of the latent classes.
Three latent classes were identied: 1, 3, and 7. Based on the
learning progress models, each class was matched to the appropriate stage (acclimation, competence, and prociency from
the three-stage model, and misconception and conception from
the two-stage model). Although the two-stage model was explained with the identied latent classes, the three-stage model
generated the most matches, including the middle stage, competence. Accordingly, we concluded that the three-stage model better
accounts for the multiple stages of expertise development.
Table 13 lists the parameter estimates of the Main effect only (Model
1). These parameter estimates could be used for standardizing a C-LCDM
if the sample size were much larger than the one in this study. To illustrate how these parameter values might be applied, we calculated
the parameter estimates associated with M1, concept similarity. M1 measures only one surface dimension. The intercept for M1 was estimated
to be 0.420, meaning that an individual who has not mastered the surface
dimension has an average correct response of 0.420. The main effect
for M1 for the surface dimension was estimated to be 0.167, meaning
that an individual who has mastered the attribute has an average correct
response of 0.587 (0.420 + 0.167). However, because this study piloted

Table 11
Comparison of relative t of C-LCDMs.
Model

Log-likelihood

AIC

BIC

Main effect only (model 1)


Interaction effect (model 2)

1078.495
1093.882

2076.990
2095.763

1959.324
1960.448

Note: Model 2 allowed interaction effects for M3 and M9.


AIC, Akaikes information criterion; BIC, Bayesian information criterion.

Latent class

Three-stage model

Two-stage model

Class 1 (C1)
Class 2 (C2)
Class 3 (C3)
Class 4 (C4)
Class 5 (C5)
Class 6 (C6)
Class 7 (C7)
Class 8 (C8)

Acclimation
Competence
Competence
Prociency
Acclimation

Prociency
Prociency

Misconception

Conception
Misconception

Conception
Conception

Note: The boldfaced characters indicates the identied classes and matched stages.

the C-LCDM with a small number of samples (N = 140), the interpretation of the parameter estimates should be limited to this study.
Lastly, based on the three-stage model, we investigated the associations between the similarities and the three identied latent classes:
class 1 (Acclimation); class 3 (Competence); and class 7 (Prociency).
As Table 14 shows, a series of Spearman rank-order correlations indicated that the ordered classes had a signicant positive relationship with
all similarity measures. Group mean difference tests proved that there
were signicant differences among the groups in all similarity measures except M11 (Balanced matching) (F (2, 130) = 1.97, p > .05).
6. Conclusions
6.1. Discussions and implications
The primary goal of this study was two-fold. One was to model
different levels of learning progress in the context of teaching and
Table 13
Parameter estimates for main effect only (model 1).
Parameter

Estimate

Standard error

M1_intercept
M1_main effect of surface
M2_intercept
M2_main effect of surface
M3_intercept
M3_main effect of structure
M4_intercept
M4_main effect of surface
M5_intercept
M5_main effect of structure
M6_intercept
M6_main effect of structure
M7_intercept
M7_main effect of structure
M8_intercept
M8_main effect of surface
M9_intercept
M9_main effect of semantic
M10_intercept
M10_main effect of semantic
M11_intercept
M11_main effect of semantic

0.420
0.167
0.384
0.178
0.772
0.109
0.462
0.157
0.671
0.145
0.676
0.179
0.678
0.124
0.377
0.173
0.664
0.423
0.522
0.476
0.645
0.465

0.037
0.020
0.038
0.017
0.015
0.011
0.038
0.025
0.019
0.013
0.025
0.019
0.026
0.027
0.041
0.027
0.006
0.000
0.004
0.000
0.013
0.000

Note: M1 (Concept), M2 (Relation), M3 (Average degree), M4 (Density), M5 (Mean


distance), M6 (Diameter), M7 (Connectedness), M8 (Subgroups), M9 (Concept matching), M10 (Propositional matching), and M11 (Balanced matching).

12

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

Table 14
Means, standard deviations, Spearmans rho, and group difference statistics.

M1. Concept
M2. Relation
M3. Average degree
M4. Density
M5. Mean distance
M6. Diameter
M7. Connectedness
M8. Subgroups
M9. Concept matching
M10. Propositional matching
M11. Balanced matching

Class 1
Mean (SD)

Class 3
Mean (SD)

Class 7
Mean (SD)

rs(2)

Sig.

.194 (.080)
.130 (.057)
.655 (.101)
.270 (.122)
.515 (.108)
.485 (.157)
.547 (.282)
.119 (.065)
.199 (.075)
.032 (.036)
.154 (.170)

.290 (.070)
.252 (.079)
.850 (.092)
.326 (.087)
.777 (.104)
.802 (.138)
.825 (.197)
.256 (.070)
.245 (.055)
.042 (.034)
.171 (.139)

.587 (.140)
.563 (.161)
.923 (.079)
.618 (.154)
.870 (.084)
.929 (.112)
.759 (.185)
.551 (.164)
.274 (.087)
.066 (.059)
.217 (.138)

.829**
.873**
.725**
.691**
.779**
.756**
.266**
.880**
.332**
.223**
.175*

178.69
180.81
91.43
102.02
134.19
109.61
19.18
175.92
11.20
6.59
1.97

.000
.000
.000
.000
.000
.000
.000
.000
.000
.002
.144

Note: rs(2) denotes Spearmans rank order correlations with 2 degrees of freedom. Standard deviations appear in parentheses next to means.
* p .05.
** p .01.

learning when solving ill-structured, complex problems. Theoretical discussions about expertise development underpin the models
of learning progress, including two-stage and three-stage models.
The other goal was to investigate a model of learning progress that
better ts the data obtained from 133 students and 7 expert professors. The validation analysis using C-LCDM demonstrated that
the data contained three latent classes that exactly corresponded
to the three-stage model (see Fig. 4): (a) acclimation, (b) competence, and (c) prociency.
Although the two-stage model was also explained by these three
latent classes, the study suggests that the three-stage model provides a better framework for measuring learning progress in complex
problem solving. In the three-stage model, the middle stage overcomes the limitation associated with contrasting extremes, such as
conception (i.e., experts) and misconception (i.e., novice).
The three levels of learning progress explain cognitive changes
over a short period of practice, such as in-class problem solving, that
correspond to Alexanders (2003, 2004) three levels of domain learning that develop over a relatively long period of time (i.e., featured
acclimation cluster, competence cluster, and prociency cluster). Interestingly, whether in the short term or the long term, the three
stages suggest that learners are likely to experience three qualitatively distinct levels in learning and development (Siegler et al., 2009;
Vygotsky, 1934/1978; Werner, 1957).
The current study observed two key conditions that relate to the
stages of learning progress: (a) teaching and learning and (b) prior
knowledge. The research context implies that students prior experience played an essential role in determining their levels of
understanding. Since data were collected early in the semester, participants in this study responded to the problem without instruction

Fig. 4. The measurable three stages of learning progress modied from Kim (2012b).

in related content. Moreover, all participants were undergraduates, most of whom were non-education majors. In this limitedinstruction condition, association with teaching experience showed
a signicant chi square value, yet there was no signicant association between levels of learning progress and student demographics:
school year, interest in becoming a teacher, and prior coursework.
Still, we withheld its generalized interpretation because some
of the expected cell values were less than ve frequencies (see
Appendix C).
We also had to consider the ill-structured and complex nature
of a given problem (e.g., its unknown elements and high number
of interactions between associated factors; see Choi & Lee, 2009;
Eseryel et al., 2013; Jonassen, 2014; Kim, 2012b). The nature of the
problem explains the reason there was no other latent class pertaining to the prociency stage. For example, due to the problems
unknown elements and high complexity, the participants rarely
conceptualized a knowledge structure featuring all dimensions (i.e.,
Class 8).
Interestingly, all seven experts were identied as procient, but
unexpectedly, their knowledge structure did not perfectly match the
reference model in all three dimensions. The current study began
with the assumption that experts in a domain would build similar
understandings of a problem despite its ill-structuredness and high
complexity. However, the results showed that their prociency was
built on their own belief, forming somewhat different knowledge
structures that emphasized some concepts and facts in the reference model. Although they constructed a reference model in
collaboration with each other, the individual expert models did not
include most of the internal and external factors associated with a
given problem situation (c.f., Jonassen, 2014).
These ndings imply that we might observe different sets of latent
classes within the three-stage model. For example, complex yet wellstructured problems are likely produce common and convergent
answers (Bogard et al., 2013; Jonassen, 2014; Shin et al., 2003). These
answers might represent two types of prociency: (a) latent class
4 (a cohesive knowledge structure with a signicant number of principles (semantic) but a small number of concepts (surface)); or (b)
latent class 8 (a well-featured knowledge structure at all levels
(surface, structural, and semantic)).
Lastly, the three-stage model of learning progress and proposed measurement could guide the development of technologyenhanced assessment applications. Current demands for studentcentered learning require prompt and precise understanding of
student changes, both cognitive and non-cognitive (Lee & Park, 2007).
Enhanced knowledge about students is critical to helping students engage in their own learning. For example, one such
application might be an in-class response system. If a teacher

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

projected a problem case to the whole class, the students could


submit written responses to the system. The system could then
analyze the students knowledge structures, levels of understanding, types of expertise, and missing or inaccurate concepts. Finally,
all assessment results could be delivered to a teacher in the form
of organized and visible data. Another example might be an advanced Intelligent Tutoring System (ITS). Using an ITS for science
education, students could try to identify the complex nature of global
warming and how it might be affecting recent wild res in California and then design an environmental structure to resolve that
problem. In this situation, students could explore a virtual space
and conduct research guided by an agent. Once the student verbally explained his or her understanding of a complex problem
situation to an agent, the response could be assessed by the embedded technology. Drawing on the assessment results, the agent
could automatically deliver formative feedback and adaptive instructional support (e.g., different instructional modes and levels
of explanation depending on each students level of learning
progress).
6.2. Limitations and suggestions
The suggested methods and ndings of this study open pathways to future research. Some potential avenues are explained below.
First, as a basis for detecting and validating changes in conceptual
development, these models and methods are applicable to a wide
range of areas: conceptual change in a problem-solving situation,
linguistic comprehension, evaluation of scientic argumentations,
expertise modeling, and longitudinal studies of learning progress.
For example, investigating longitudinal changes in learning can help
us evaluate instructional effectiveness and determine proper educational support for individual students. Collins and Wugalter (1992)
pointed out that psychological research and theory is increasingly
turning to longitudinal studies to monitor development over a period
of time and test measurement methods using latent variables. The
C-LCDM method is in its infancy. In fact, as of this study, there was
only one reference. For longitudinal analysis, C-LCDM method should
be further developed by incorporating Markov chain models, which
are used to predict the probabilities of movement through stages
over a specic time interval (transition probabilities), similar to latent
transition analysis (Collins & Cliff, 1990; Collins et al., 1994; Martin,
Velicer, & Fava, 1996; Velicer, Martin, & Collins, 1996). Alternatively, we could gather multiple cross-sectional data in time and track
changes of learning levels for descriptive purposes.
Second, the ultimate goal of applying C-LCDMs to the validation of learning progress models is to determine parameters in the
C-LCDM model that can be generalized. In other words, based on
the identied parameters, a students stages might easily and quickly
be estimated using their responses. That diagnostic algorithm could
be embedded in an assessment technology. Devising an assessment technology adapted for the complex and dynamic structure
of mental models is essential because assessment is a fundamental step in feedback, revision, and reection on learning (Pellegrino,
Chudowsky, & Glaser, 2001). That technology could enable a teacher
to have a better sense of student learning and provide elaborate feedback and support. For example, McKeown (2009) used HIMATT with
40 actual classroom teachers. They managed to use that technology to diagnose student understanding and provide instructional
support to different individuals even though the tool was new to
them.
Third, this study discussed the assessment model and methods
for diagnosing domain knowledge as an internally represented
problem situation. Scholars generally accept that human cognition includes meta-cognition. Therefore, future studies should
investigate the extent to which metacognition inuences or interacts with the proposed stages of learning progress. Furthermore,

13

meta-cognition might progress through qualitatively different stages,


just as general cognition presumably does.
Considering that this study is only the rst attempt to suggest
a framework for measuring learning progress and to validate the
model with data, more study is needed to conclude how many stages
of knowledge structure might characterize learning progress. In addition, there are diverse theoretical accounts of the mental stages.
Therefore, scholars should conduct further studies to test various
conditions.
6.3. Closing thoughts
Along with the theoretical models of learning progress, this study
provides tools and methods for future research: (a) a set of parameters quantifying the attributes of knowledge structure; (b) a set
of similarity measures applicable to the study of cognitive changes;
and (c) a statistical approach (i.e., C-LCDM) to diagnosing the stages
of learning progress. These models and methods for understanding levels of mastery of complex problems should help teachers
develop learning environments that provide instructional support
and feedback catered to the needs of individual learners. Thus, future
studies should investigate instructional and feedback strategies associated with each developmental stage of learning progress.
Instructional models based on diagnostic assessment could lead to
the development of diverse instructional applications, such as intelligent tutoring systems.
Appendix A
Case study
Directions: Read the case study described below and then prepare
a response to the questions below (written response with at least
350 words is required for each question):
Assume that you have been involved in evaluating a media implementation project in an urban inner middle school. At the
beginning of the school year, all of the students assigned to
four subject area teachers (math, language arts, social studies
and science) in the seventh grade at the middle school were
given tablet PCs (laptop computers also equipped with a stylus/
pen and a touchscreen that can be written upon) and were also
given wireless internet access at home and at school for a entire
year.
The students took the tablet PCs home every evening and
brought them to class every day. The teachers were also provided with tablet PCs 24/7 (24 hours a day, every day of the
week) for the entire year. The teachers and students were trained
on how to use the tablet PCs. Moreover, all of the curriculum
materials (textbooks, workbooks, student study guides, teacher
curriculum guides, some activities, tests, etc.) were installed
on the tablet PCs or were accessible through the tablet
PCs.
Your job as one of the evaluators for the project was to examine
how this innovation (providing teachers and students with tablet
PCs 24/7) changed the way instruction was presented in the
classrooms of the four teachers. Results indicated that the innovation had very little effect on the manner in which instruction
took place in the teachers classrooms.
1. Based on what you have learned about the use of technology in
education, describe what concepts, issues, factors, and variables are likely to have contributed to the fact that the
introduction of the tablet PCs had very little effect on the instructional practices that were employed in the classes.
2. Describe the strategies that could have been employed to help
mitigate the factors that you think contributed to the minimal
effect the tablet PCs had on instructional practices. When

14

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

you are answering this question, use the concepts, factors,


and variables you described in question 1 or add other assumptions and information that would be required to solve this
problem.
Appendix B
Reference model to question 1
Technology implementations usually begin with an identied
instructional need. Instructional need was likely not fully identied
due to insucient study of how instructional practices in the classroom were being conducted already without the technology.
One big issue is dening what a successful integration or change
in instructional practice actually is. While teachers in the situation
may have felt that they knew this already, the assumptions
inherent in a design situation need to be articulated and checked
if the assumptions are not to distort the design space by which
instructional practices are manipulated. Teachers did not have
enough professional development using the technology in classroom teaching and learning, on ways to integrate use into their
teaching, and best practices with regard to effective educational
use.
Teacher professional development that discusses not just
technical know-how but also pedagogy could help teachers realize
how to do things differently that takes full advantage of the
affordances of the tablets. Training as a professional development
should be extensive including teacher beliefs and attitude. Teacher
beliefs play a role in adopting new practices and changing
their instructional practice. Teachers may not believe that students learn with laptops, and thus do not use laptops in their
instruction. The only support teachers had during implementation was technical support; Teachers lacked a mentor who could
assist them as instructional issues arose throughout the year.
Mentoring on additional and advanced uses of the technology in
the classroom is critical for teachers to increase their skills and
maintain their motivation in utilizing the technology. In addition,
a mentor could help teachers to maintain the belief that these efforts
will have positive results. There are concerns that the environment does not support change. An ongoing supportive environment
where teachers initially learn how to use the technology, how to
use the technology with their content, and how to continue to
develop their expertise in the technology and incorporating it to
the classroom is critical.
Environment could include a culture that does not support the
desired performance.
For example, the lack of incentives to make effective use of a new
technology could also contribute to lack of use. The intervention
seems to have been applied to this community rather than involving teachers from the beginning as collaborators in its design
and modication. Teachers were not involved in the decision to
implement the new media; thus, they did not fully buy into the
plan.
Appendix C
Cross-tabulation of school year and classes
School year

Freshman
Sophomore
Junior
Senior
Total
*p .05.

Class
Acclimation

Competence

Prociency

3
8
17
10
38

4
20
23
15
62

2
9
12
10
33

Total

x2 (df)

9
37
52
35
133

1.914 (6)

.927

Cross-tabulation of interest and classes


Interest in
becoming
a teacher

Total x2 (df)

Class

Acclimation Competence Prociency

Not interested
5
at all
Not so
9
interested
Somewhat
9
Interested
Interested
5
Very interested 10
Total
38

10

17

14

32

14

32

8
16
62

6
7
33

19
33
133

2.813 (8) .946

*p .05.

Cross-tabulation of teaching experience and classes


Teaching
experience

Total x2 (df)

Class

Acclimation Competence Prociency

None
33
Yes but less
5
than 2 years
Total
38

59
1

32
0

9
37

60

32

133

9.032 (2) .011

*p .05.

Cross-tabulation of prior coursework and classes


Prior
coursework

Total x2(df)

Class

Acclimation Competence Prociency

None
20
Other education 16
courses
Instructional
2
Design courses
Total
38

22
36

12
19

54
71

62

33

133

3.224 (4) .521

*p .05.

References
Agresti, A. (2007). An introduction to categorical data analysis. Hoboken, NJ: Wiley
Inter-Science.
Alexander, P. A. (1997). Mapping the multidimensional nature of domain learning:
The interplay of cognitive, motivational, and strategic forces. In M. L. Maehr &
P. R. Pintrich (Eds.), Advances in motivation and achievement (Vol. 10, pp. 213250).
Greenwich, CT: JAI Press.
Alexander, P. A. (2003). The development of expertise: The journey from acclimation
to prociency. Educational Researcher, 32(8), 1014.
Alexander, P. A. (2004). A model of domain learning: Reinterpreting expertise as a
multidimensional, multistage process. In D. Y. Dai & R. J. Sternberg (Eds.),
Motivation, emotion, and cognition: Integrative perspectives on intellectual
functioning and development (pp. 273298). Mahwah, NJ: Lawrence Erlbaum
Associates, Publishers.
Alexander, P. A., & Murphy, P. K. (1998). Proling the differences in students
knowledge, interest, and strategic processing. Journal of Educational Psychology,
90, 435447.
Alexander, P. A., Sperl, C. T., Buehl, M. M., Fives, H., & Chiu, S. (2004). Modeling domain
learning: Proles from the eld of special education. Journal of Educational
Psychology, 96(3), 545557. doi:10.1037/0022-0663.96.3.545.
Anzai, Y., & Yokoyama, T. (1984). Internal models in physics problem solving. Cognition
and Instruction, 1, 397450.
Biggs, J., & Collis, K. (1982). Evaluating the quality of learning: the SOLO taxonomy.
New York: Academic Press.
Bogard, T., Liu, M., & Chiang, Y. V. (2013). Thresholds of knowledge development in
complex problem solving: a multiple-case study of advanced learners cognitive
processes. Educational Technology Research and Development, 61(3), 465503.
<http://doi.org/10.1007/s11423-013-9295-4>.
Bozard, J. L. (2010). Invariance testing in diagnostic classication models (Unpublished
masters thesis). University of Georgia.
Bransford, J. D., Barclay, J. R., & Franks, J. J. (1972). Sentence memory: A constructive
versus interpretive approach. Cognitive Psychology, 3, 193209.
Bransford, J. D., Brown, A. L., & Cocking, R. R. (Eds.), (2000). Learning and transfer.
In How people learn: Brain, mind, experience, and school (pp. 3178). Washington,
DC: National Academy Press.
Bransford, J. D., & Franks, J. J. (1972). The abstraction of linguistic ideas. Cognitive
Psychology, 2, 331350.

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

Bransford, J. D., & Johnson, M. K. (1972). Contextual prerequisites for understanding:


Some investigations of comprehension and recall. Journal of Verbal Learning and
Verbal Behavior, 11(6), 717726, doi:10.1016/S0022-5371(72)80006-9.
Carley, K., & Palmquist, M. (1992). Extracting, representing and analyzing mental
models. Social Forces, 70, 215225.
Chase, W. G., & Ericsson, K. A. (1982). Skill and working memory. In G. H. Bower (Ed.),
The psychology of learning and motivation (Vol. 16, pp. 158). New York: Academic
Press.
Chen, Z., & Siegler, R. S. (2000). Intellectual development in childhood. In R. Sternberg
(Ed.), Handbook of intelligence (pp. 92116). New York: Cambridge.
Chi, M. T. H. (2006). Two approaches to the study of experts characteristics. In K.
A. Ericsson, N. Charness, P. Feltovich, & R. Hoffman (Eds.), Cambridge handbook
of expertise and expert performance (pp. 121130). Cambridge University Press.
Chi, M. T. H. (2008). Three types of conceptual change: Belief revision, mental model
transformation, and categorical shift. In S. Vosniadou (Ed.), International handbook
of research on conceptual change (pp. 334). New York: Routledge.
Chi, M. T. H., & Glaser, R. (1985). Problem solving ability. In R. J. Sternberg (Ed.), Human
abilities: An information processing approach. New York: W.H. Freeman.
Chi, M. T. H., Glaser, R., & Farr, M. (Eds.), (1988). The nature of expertise. Hillsdale,
NJ: Erlbaum.
Choi, I., & Lee, K. (2009). Designing and implementing a case-based learning
environment for enhancing Ill-structured problem solving: Classroom
management problems for prospective teachers. Educational Technology Research
and Development, 57(1), 99129. doi:10.2307/25619959.
Clariana, R. B. (2010). Multi-decision Approaches for Eliciting Knowledge Structure.
In D. Ifenthaler, P. Pirnay-Dummer, & N. M. Seel (Eds.), Computer-Based Diagnostics
and Systematic Analysis of Knowledge (pp. 4159). Springer US. Retrieved
September 11, 2011, from: <http://link.springer.com/chapter/10.1007/
978-1-4419-5662-0_4>.
Collins, A., & Gentner, D. (1987). How people construct mental models. In D. Holland
& N. Quinn (Eds.), Cultural models in language and thought (pp. 243265).
Cambridge: Cambridge University Press.
Collins, L. M., & Cliff, N. (1990). Using the longitudinal Guttmann simplex as a basis
for measuring growth. Psychological Bulletin, 108, 128134.
Collins, L. M., Graham, J. W., Rousculp, S. S., Fidler, P. L., Pan, J., & Hansen, W. B. (1994).
Latent transition analysis and how it can address prevention research questions.
NIDA Research Monograph, 142, 81111.
Collins, L. M., & Wugalter, S. E. (1992). Latent class models for stage-sequential
dynamic latent variables. Multivariate Behavioral Research, 27(1), 131157.
Colman, A. M., & Shar, E. (2008). Tversky, Amos. In N. Koertge (Ed.), New dictionary
of scientic biography (Vol. 7, pp. 9197). Farmington Hills, MI: Charles Scribners
Sons.
Coronges, K. A., Stacy, A. W., & Valente, T. W. (2007). Structural comparison of
cognitive associative networks in two populations. Journal of Applied Social
Psychology, 37(9), 20972129. doi:10.1111/j.1559-1816.2007.00253.x.
Craik, K. J. W. (1943). The nature of explanation. Cambridge UK: Cambridge University
Press.
Curtis, M. B., & Davis, M. A. (2003). Assessing knowledge structure in accounting
education: An application of Pathnder Associative Networks. Journal of
Accounting Education, 21(3), 185195. doi:10.1016/S0748-5751(03)00024-1.
diSessa, A. A. (2006). A history of conceptual change research. In R. K. Sawyer (Ed.),
Cambridge handbook of the learning science (pp. 265281). New York: Cambridge
University Press.
Dochy, F., Segers, M., Van den Bossche, P., & Gijbels, D. (2003). Effects of problem-based
learning: A meta-analysis. Learning and Instruction, 13(5), 533568. doi:10.1016/
S0959-4752(02)00025-7.
Ericsson, K. A. (2003). The acquisition of expert performance as problem solving:
Construction and modication of mediating mechanisms through deliberate
practice. In J. E. Davidson & R. J. Sternberg (Eds.), The psychology of problem solving
(pp. 3183). New York: Cambridge University Press.
Ericsson, K. A. (2005). Recent advances in expertise research: A commentary on the
contributions to the special issue. Applied Cognitive Psychology, 19, 233241.
Ericsson, K. A. (2006). The inuence of experience and deliberate practice on the
development of superior expert performance. In K. A. Ericsson, N. Charness, P.
Feltovich, & R. R. Hoffman (Eds.), Cambridge handbook of expertise and expert
performance (pp. 685706). Cambridge, UK: Cambridge University Press.
Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological Review,
102(2), 211245. doi:10.1037/0033-295X.102.2.211.
Ericsson, K. A., & Staszewsli, J. J. (1989). Skilled memory and expertise: Mechanisms
of exceptional performance. In D. Klahr & K. Kotovsky (Eds.), Complex information
processing: The impact of Herbert A. Simon (pp. 235267). Hillsdale, NJ: Lawrence
Erlbaum.
Ericsson, K., & Simon, H. (1980). Verbal reports as data. Psychological Review, 87(3),
215251. doi:10.1037/0033-295X.87.3.215.
Ericsson, K., & Simon, H. (1993). Protocol analysis: Verbal reports as data (2nd ed.).
Boston: MIT Press. ISBN 0262050293.
Eseryel, D., Ifenthaler, D., & Ge, X. (2013). Validation study of a method for assessing
complex ill-structured problem solving by using causal representations.
Educational Technology Research and Development, 61(3), 443463. doi:10.1007/
s11423-013-9297-2.
Festinger, L. (1962). Cognitive dissonance. Scientic American, 207(4), 93102.
Flavell, J. H. (1985). Cognitive development (2nd ed.). Englewood Cliffs, NJ:
Prentice-Hall.
Flavell, J. H., & Miller, P. H. (1998). Social cognition. In W. Damon, D. Kuhn, & R. S.
Siegler (Eds.), Handbook of child psychology: Vol. 2. Cognition, perception, and
language (pp. 851898). New York: Wiley.

15

Fodor, J. A., Bever, T. G., & Garrett, M. F. (1974). The psychology of language: An
introduction to psycholinguistics and generative grammar. New York: McGraw-Hill.
Gentner, D., & Medina, J. (1998). Similarity and the development of rules. Cognition,
65, 263297.
Gijbels, D., Dochy, F., Van den Bossche, P., & Segers, M. (2005). Effects of problem-based
learning: A meta-analysis from the angle of assessment. Review of educational
research, 75(1), 2761.
Glaser, E. M., Abelson, H. H., & Garrison, K. N. (1983). Putting knowledge to use:
Facilitating the diffusion of knowledge and the implementation of planned change
(1st ed.). San Francisco: Jossey-Bass.
Gobet, F., & Wood, D. (1999). Expertise, models of learning and computer-based
tutoring. Computers & Education, 33(23), 189207.
Goldsmith, T. E., & Kraiger, K. (1997). Applications of structural knowledge assessment
to training evaluation. In J. K. Ford (Ed.), Improving training effectiveness in
organizations (pp. 7395). Mahwah, NJ: Lawrence Erlbaum Associates.
Goodman, C. M. (1987). The Delphi technique: A critique. Journal of Advanced Nursing,
12(6), 729734.
Greeno, J. G. (1989). Situations, mental models, and generative knowledge. In D. Klahr
& K. Kotovsky (Eds.), Complex information processing (pp. 285318). Hillsdale, NJ:
Lawrence Erlbaum Associates, Publishers.
Grow-Maienza, J., Hahn, D., & Joo, C. (2001). Mathematics instruction in Korean
primary schools: Structures, processes, and a linguistic analysis of questioning.
Journal of Educational Psychology, 93(2), 363376.
Guadagnoli, E., & Velicer, W. F. (1988). Relation of sample size to the stability of
component patterns. Psychological Bulletin, 103(2), 265275.
Hatano, G., & Inagaki, K. (1994). Young childrens naive theory of biology. Cognition,
50(13), 171188. doi:10.1016/0010-0277(94)90027-2.
Hattie, J. (2009). Visible learning: A synthesis of over 800 meta-analyses relating to
achievement. New York: Taylor & Francis Group.
Heinen, T. (1996). Latent class and discrete latent trait models: Similarities and differences.
Thousand Oaks, CA: Sage Publication.
Henson, R. A., Templin, J. L., & Willse, J. T. (2009). Dening a family of cognitive
diagnosis models using log-linear models with latent variables. Psychometrika,
74(2), 191210.
Holyoak, K. J., & Koh, K. (1987). Surface and structural similarity in analogical transfer.
Memory & Cognition, 15, 332340.
Hsu, C., & Sandford, B. (2007). The Delphi technique: Making sense of consensus.
Practical Assessment Research & Evaluation, 12(10).
Ifenthaler, D. (2006). Diagnose lernabhngiger Vernderung mentaler Modelle.
Entwicklung der SMD-Technologie als methodologische Verfahren zur relationalen,
strukturellen und semantischen Analyse individueller Modellkonstruktionen. Freiburg:
Universitts-Dissertation.
Ifenthaler, D. (2010). Relational, structural, and semantic analysis of graphical
representations and concept maps. Educational Technology Research &
Development, 58(1), 8197. doi:10.1007/s11423-008-9087-4.
Ifenthaler, D., Masduki, I., & Seel, N. M. (2009). The mystery of cognitive structure
and how we can detect it: Tracking the development of cognitive structures over
time. Instructional Science, doi:10.1007/s11251-009-9097-6.
Ifenthaler, D., & Seel, N. M. (2005). The measurement of change: Learning-dependent
progression of mental models. Technology, Instruction, Cognition, and Learning,
2(4), 321340.
Johnson-Laird, P. N. (1983). Mental models: Towards a cognitive science of language,
inference, and consciousness. Cambridge: Cambridge University Press.
Johnson-Laird, P. N. (2005a). Mental models and thoughts. In K. J. Holyoak (Ed.), The
Cambridge handbook of thinking and reasoning (pp. 185208). Cambridge
University Press.
Johnson-Laird, P. N. (2005b). The history of mental models. In K. I. Manktelow & M.
C. Chung (Eds.), Psychology of reasoning: Theoretical and historical perspectives
(pp. 179212). Psychology Press.
Jonassen, D. H. (1997). Instructional design models for well-structured and illstructured problem-solving learning outcomes. Educational Technology Research
and Development, 45(1), 6594.
Jonassen, D. H. (2000). Toward a design theory of problem solving. Educational
Technology Research and Development, 48(4), 6385.
Jonassen, D. H. (2014). Assessing Problem Solving. In J. M. Spector, M. D. Merrill, J.
Elen, & M. J. Bishop (Eds.), Handbook of Research on Educational Communications
and Technology (pp. 269288). Springer New York. Retrieved January 17, 2015,
from: <http://link.springer.com/chapter/10.1007/978-1-4614-3185-5_22>.
Jonassen, D. H., Beissner, K., & Yacci, M. (1993). Structural knowledge: Techniques
for representing, conveying, and acquiring structural knowledge. Structural
knowledge: Techniques for representing, conveying, and acquiring structural
knowledge.
Kaplan, D. (2008). An overview of Markov chain methods for the study of
stage-sequential development processes. Developmental Psychology, 44(2),
457467.
Katz, J. J., & Postal, P. M. (1964). An integrated theory of linguistic descriptions.
Cambridge: M.I.T. Press.
Kieras, D. E., & Bovair, S. (1984). The role of a mental model in learning to operate
a device. Cognitive Science, 8, 255273.
Kim, M. (2012a). Cross-validation study on methods and technologies to assess mental
models in a complex problem-solving situation. Computers in Human Behavior,
28(2), 703717. doi:10.1016/j.chb.2011.11.018.
Kim, M. (2012b). Theoretically grounded guidelines for assessing learning progress:
Cognitive changes in ill-structured complex problem-solving contexts. Educational
Technology Research and Development, 60(4), 601622. doi:10.1007/s11423-0129247-4.

16

M.K. Kim/Contemporary Educational Psychology 42 (2015) 116

Kim, M. (2013). Concept map engineering: Methods and tools based on the semantic
relation approach. Educational Technology Research and Development, 61(6),
951978. doi:10.1007/s11423-013-9316-3.
Kintsch, W., & van Dijk, T. A. (1978). Toward a model of text comprehension and
production. Psychological Review, 85(5), 363394. doi:10.1037/0033295X.85.5.363.
Kitchner, K. S. (1983). Cognition, metacognition, and epistemic cognition:
A three-level model of cognitive processing. Human Development, 4, 222232.
Lee, J., & Park, O. (2007). Adaptive instructional systems. In J. M. Spector, M. D. Merill,
J. van Merrienboer, & M. P. Driscoll (Eds.), Handbook of research for educational
communications and technology (pp. 469484). Routledge: Taylor & Francis Group.
Martin, R. A., Velicer, W. F., & Fava, J. L. (1996). Latent transition analysis to the stages
of change for smoking cessation. Additive Behaviors, 21(1), 6780.
Mayer, R. E. (1989). Models for understanding. Review of Educational Research, 59(1),
4364.
McKeown, J. O. (2009). Using annotated concept map assessments as predictors of
performance and understanding of complex problems for teacher technology
integration (Unpublished doctoral dissertation). Tallahassee, FL: Florida State
University.
Monge, P. R., & Contractor, N. S. (2003). Theories of communication networks. New
York: Oxford University Press.
Narayanan, V. K. (2005). Causal mapping: An historical overview. In V. K. Narayanan
& D. J. Armstrong (Eds.), Causal mapping for research in information technology
(pp. 119). Hershey, PA: Idea Group Publishing.
Newell, A., & Simon, H. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice
Hall.
Norman, D. (1983). Some observations on mental models. In D. Gentner & A. L. Stevens
(Eds.), Mental models (pp. 714). Hillsdale, NJ: Erlbaum.
Novak, J. D., & Caas, A. J. (2006). The origins of the concept mapping tool and the
continuing evolution of the tool. http://citeseerx.ist.psu.edu/viewdoc/summary?doi
=10.1.1.106.3382 Retrieved May 6, 2012.
Okoli, C., & Pawlowski, S. D. (2004). The Delphi method as a research tool: An example,
design considerations and applications. Information & Management, 42(1), 1529.
doi:10.1016/j.im.2003.11.002.
Opfer, J. E., & Siegler, R. S. (2004). Revisiting preschoolers living things concept: A
microgenetic analysis of conceptual change in basic biology. Cognitive Psychology,
49, 301332.
Pellegrino, J. W., Chudowsky, N., & Glaser, R. (Eds.), (2001). Knowing what students
know. Washington, DC: National Academy Press.
Piaget, J. (1964). Development and learning. In R. E. Ripple & V. N. Rockcastle (Eds.),
Piaget rediscovered (pp. 720). Ithaca, NY: Cornell University.
Pirnay-Dummer, P. (2006). Expertise und modellbildung MITOCAR. Freiburg:
Universitts-Dissertation.
Pirnay-Dummer, P., & Ifenthaler, D. (2010). Automated knowledge visualization and
assessment. In D. Ifenthaler, P. Pirnay-Dummer, & N. M. Seel (Eds.), Computerbased diagnostics and systematic analysis of knowledge. New York: Springer.
Pirnay-Dummer, P., Ifenthaler, D., & Spector, J. (2010). Highly integrated model
assessment technology and tools. Educational Technology Research & Development,
58(1), 318. doi:10.1007/s11423-009-9119-8.
Pretz, J. E., Naples, A. J., & Sternberg, R. J. (2003). Recognizing dening, and
representing problems. In J. E. Davidson & R. J. Sternberg (Eds.), The psychology
of problem solving (pp. 330). New York: Cambridge University Press.
Rost, J., & Langeheine, R. (Eds.), (1997). Applications of latent trait and latent class models
in the social sciences. New York: Waxmann.
Rupp, A. A., Sweet, S., & Choi, Y. (2010a). Modeling learning trajectories with epistemic
network analysis: A simulation-based investigation of a novel analytic method for
epistemic games. Presented at the annual meeting of the International Society
for Educational Data Mining (EDM), Pittsburgh, PA.
Rupp, A. A., Templin, J. L., & Henson, R. A. (2010b). Diagnostic measurement: Theory,
methods, and applications. New York: The Guilford Press.
Schlomske, N., & Pirnay-Dummer, P. (2008). Model based assessment of learning
dependent change during a two semester class. In Kinshuk, D. Sampson, & M.
Spector (Eds.), Proceedings of IADIS International Conference Cognition and
Exploratory Learning in Digital Age 2008 (pp. 4553). Freiburg, Germany.
Schraw, G., Dunkle, M. E., & Bendixen, L. D. (1995). Cognitive processes in well-dened
and ill-dened problem solving. Applied Cognitive Psychology, 9, 116.
Schvaneveldt, R. W., Durso, F. T., Goldsmith, T. E., Breen, T. J., & Cooke, N. M. (1985).
Measuring the structure of expertise. International Journal of Man-Machine Studies,
23, 699728.
Seel, N. M. (1999). Semiotics and structural learning theory. Journal of Structural
Learning and Intelligent Systems, 14(1), 1128.
Seel, N. M. (2001). Epistemology, situated cognition, and mental models: Like a bridge
over troubled water. Instructional Science, 29(45), 403427.
Seel, N. M. (2003). Model-centered learning and instruction. Technology, Instruction,
Cognition, and Learning, 1(1), 5985.

Seel, N. M. (2004). Model-centered learning environments: Theory, Instructional


design, and ethics. In N. M. Seel & S. Dijkstra (Eds.), Curriculum, plans, and processes
in instruction design: International perspectives (pp. 4974). Hillsdale, NJ: Lawrence
Erlbaum Associates, Inc.
Seel, N. M., & Dinter, F. R. (1995). Instruction and mental model progression:
Learner-dependent effects of teaching strategies on knowledge acquisition and
analogical transfer. Educational Research and Evaluation, 1(1), 435.
Segers, M. (1997). An alternative for assessing problem-solving skills: The overall
test. Studies in Educational Evaluation, 23(4), 373398.
Shin, N., Jonassen, D. H., & MaGee, S. (2003). Predictors of well-structured and
ill-structured problem solving in an astronomy simulation. Journal of Research
in Science Teaching, 40(1), 727.
Shute, V. J., & Zapata-Rivera, D. (2008). Using an evidence-based approach to assess
mental models. In D. Ifenthaler, P. Pirnay-Dummer, & J. M. Spector (Eds.),
(pp. 2341). Understanding models for learning and instruction: Essays in honor
of Norbert M. Seel. New York: Springer.
Siegler, R. S. (2005). Childrens learning. American Psychologist, 60, 769
778.
Siegler, R. S., Thompson, C. A., & Opfer, J. E. (2009). The logarithmic-to-linear shift:
One learning sequence, many tasks, many time scales. Mind, Brain, and Education,
3, 143150.
Simon, H. A., & Chase, W. G. (1973). Skill in chess. American Scientist, 61, 394
403.
Simon, H. A., & Hayes, J. R. (1976). The understanding process: Problem isomorphs.
Cognitive Psychology, 8, 165190.
Sinnott, J. D. (1989). A model for solution of ill-structured problems: Implications
for everyday and abstract problem solving. In J. D. Sinnott (Ed.), Everyday problem
solving: Theory and applications (pp. 7299). New York: Praeger.
Smith, J. P., diSessa, A. A., & Roschelle, J. (1993). Misconceptions reconceived: A
constructivist analysis of knowledge in transition. The Journal of the Learning
Sciences, 3(2), 115163.
Snow, R. E. (1990). New approaches to cognitive and conative assessment in
education. International Journal of Educational Research, 14(5), 455473.
Spector, J. M. (2004). Problems with problem-based learning: Comments on
model-centered learning and instruction in Seel (2003). Technology, Instruction,
Cognition and Learning, 1(4), 359374.
Spector, J. M. (2008). Expertise and dynamic tasks. In H. Qudrat-Allah, J. M. Spector,
& P. I. Davidsen (Eds.), Complex decision making: Theory and practice. Berlin:
Springer-Verlag.
Spector, J. M., & Koszalka, T. A. (2004). The DEEP methodology for assessing learning
in complex domains. Syracuse, NY: Syracuse University. Final report to the National
Science Foundation Evaluative Research and Evaluation Capacity Building.
Spiro, R. J., Feltovich, P. J., & Coulson, R. L. (1996). Two epistemic world-views:
Pregurative schemas and learning in complex domains. Applied Cognitive
Psychology, 10, s51s61.
Steven, J. P. (2002). Applied multivariate statistics for the social sciences (4th ed.).
Hillsdale, NJ: Erlbaum.
Stigler, J. W., & Stevenson, H. W. (1991). How Asian teachers polish each lesson to
perfection. American Educator, 15, 1220.
Taricani, E. M., & Clariana, R. B. (2006). A technique for automatically scoring
open-ended concept maps. Educational Technology Research & Development, 54(1),
6582. doi:10.1007/s11423-006-6497-z.
Templin, J. L. (2004). Generalized linear mixed prociency models (Unpublished doctoral
dissertation). University of Illinois at Urbana-Champaign.
Tversky, A. (1977). Features of similarity. Psychological Review, 84, 327352.
Tversky, A., & Shar, E. (2004). Preference, belief, and similarity: Selected writings.
London: MIT Press.
Velicer, W. F., Martin, R. A., & Collins, L. M. (1996). Latent transition analysis for
longitudinal data. Addition, 91(Suppl.), s197s209.
Vosniadou, S. (2003). Exploring the relationships between conceptual change and
intentional learning. In G. M. Sinatra & P. R. Pintrich (Eds.), Intentional conceptual
change (pp. 377406). Mahwah, NJ: Erlbaum.
Vosniadou, S., & Vamvakoussi, X. (2008). The framework theory approach to the
problem of conceptual change. In S. Vosniadou (Ed.), International handbook of
research on conceptual change (pp. 334). New York: Routledge.
Vygotsky, L. (1934/1978). Mind in society. Cambridge, MA: Harvard University Press.
Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications.
Cambridge University Press.
Werner, H. (1957). The concept of development from a comparative and organismic
point of view. In D. B. Harris (Ed.), The concept of development: An issue in the
study of human behavior (pp. 125148). Minneapolis: University of Minnesota.
Wood, P. K. (1983). Inquiring systems and problem structures: Implications for
cognitive developments. Human Development, 26, 249265.
Zsambok, C. E., & Klein, G. (Eds.), (1997). Naturalistic decision making. Mahwah, NJ:
Lawrence Erlbaum Associates.

You might also like