The ICAP Framework Linking Cognitive Engagement, 2014

This article was downloaded by: [George Mason University]
On: 18 December 2014, At: 03:59

Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer
House, 37-41 Mortimer Street, London W1T 3JH, UK
Educational Psychologist
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/hedp20
The ICAP Framework: Linking Cognitive Engagement

to Active Learning Outcomes
a a
Michelene T. H. Chi & Ruth Wylie
a
Mary Lou Fulton Teachers College, Arizona State University
Published online: 28 Oct 2014.
Click for updates
To cite this article: Michelene T. H. Chi & Ruth Wylie (2014) The ICAP Framework: Linking Cognitive Engagement to Active
Learning Outcomes, Educational Psychologist, 49:4, 219-243, DOI: 10.1080/00461520.2014.965823
To link to this article: http://dx.doi.org/10.1080/00461520.2014.965823
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of
the Content. Any opinions and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied
upon and should be independently verified with primary sources of information. Taylor and Francis shall
not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other
liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
EDUCATIONAL PSYCHOLOGIST, 49(4), 219–243, 2014
Copyright Ó Division 15, American Psychological Association
ISSN: 0046-1520 print / 1532-6985 online
DOI: 10.1080/00461520.2014.965823
The ICAP Framework: Linking Cognitive

Engagement to Active Learning Outcomes
Michelene T. H. Chi and Ruth Wylie
Mary Lou Fulton Teachers College
Arizona State University
Downloaded by [George Mason University] at 03:59 18 December 2014
This article describes the ICAP framework that defines cognitive engagement activities on
the basis of students’ overt behaviors and proposes that engagement behaviors can be
categorized and differentiated into one of four modes: Interactive, Constructive, Active, and
Passive. The ICAP hypothesis predicts that as students become more engaged with the
learning materials, from passive to active to constructive to interactive, their learning will
increase. We suggest possible knowledge-change processes that support the ICAP hypothesis
and address the limitations and caveats of the hypothesis. In addition, empirical validation
for the hypothesis is provided by examining laboratory and classroom studies that focus on
three specific engagement activities: note taking, concept mapping and self-explaining. We
also consider how ICAP can be used as a tool for explaining discrepant findings, dictate the
proper choice of a control condition, and evaluate students’ outputs. Finally, we briefly
compare ICAP to existing theories of learning.
Educators have long recognized that, although students can and negative reactions to teachers, classmates, academics,
learn from receiving information passively, they fare much and so on. We are not focusing on the precursor stages of
better by learning actively. “Active learning” is typically engagement embodied within the motivational perspective,
defined by educational researchers as learning that requires or the large-grained behavioral and the emotional perspec-
students to engage cognitively and meaningfully with the tives; instead, we are focusing on the amount of cognitive
materials (Bonwell & Eison, 1991), to get “involved with engagement that can be detected by smaller grained behav-
the information presented, really thinking about it (analyz- ioral activities while students learn. Although there is some
ing, synthesizing, evaluating) rather than just passively research on cognitive engagement, it emphasizes broad
receiving it” (King, 1993, p. 2). Thus, “active” students are notions such as thoughtfulness and willingness to exert the
considered to be cognitively “engaged” (Corno & Madi- necessary effort to succeed and master complex skills and
nach, 1983), and we use those terms synonymously. ideas (Fredricks, Blumenfeld, & Paris, 2004).
Despite the fact that “active learning” is defined as In this article, the term “learning activities” is used as a
engaging cognitively, most of the research on engagement broad term, referring to the large collection of instructional or
considers it from the motivational perspective (Blumenfeld, learning tasks from which teachers or educational designers
Kempler, & Krajcik, 2004; Pintrich & De Groot, 1990; can choose for students to do (e.g., reading, solving problems,
Zimmerman, 1990), the behavioral perspective, or the emo- learning to understand charts and diagrams, etc.). We use the
tional perspective. When engagement is discussed in moti- term “engagement” or “engagement activities” to refer to the
vational terms, it tends to mean the precursor attitude or way a student engages with the learning materials in the con-
interest in getting involved with the learning materials. text of an instructional or learning task, reflected in the overt
Behavioral engagement refers generally to the notion of behavior the student exhibits while undertaking an activity,
participating and addresses large-grained measures, such as such as summarizing at the end of each paragraph, either
how often students attend class or do homework, whereas orally or in written form. We refer to summarizing as an
emotional engagement encompasses measures of positive “engagement activity or behavior” that the student voluntarily
undertakes while learning in the context of an instructional
task. A teacher can, however, design tasks that elicit more or
Correspondence should be addressed to Michelene T. H. Chi, Mary Lou
Fulton Teachers College, Arizona State University, P.O. Box 871811, Mail less engagement from students (e.g., embedding a prompt at
Code 1811, Tempe, AZ 85287-1811. E-mail: [email protected] the end of each paragraph to remind students to summarize).
220 CHI AND WYLIE
In short, although “active learning” is a great idea for article differs from our prior publication in that Chi (2009)
overcoming “passive learning,” we have identified three cited evidence primarily from laboratory experiments mea-
concrete practical challenges that teachers may face when suring performances on a variety of tasks including, but not
developing lessons that promote “active learning.” First, limited to, learning. In this article, we cite studies that focus
broad recommendations such as engage students cogni- on three explicit engagement activities (note taking, con-
tively, encourage meaningful learning, and get students to cept mapping, and self-explaining). In addition, we cite
think about it do not tell teachers how to create activities classroom studies in support of the ICAP hypothesis (that
that overcome “passive learning.” Second, teachers have I>C>A>P), showing the relevance of the hypothesis to
few criteria to use in deciding which are the best “active classroom learning.
learning” activities to design and implement. Third, there In the third section, we consider several caveats or fac-
are no guidelines for teachers regarding how to best modify tors that may override or violate the ICAP hypothesis. We
their favorite existing assignments in order to optimize also consider new applications of ICAP as a tool for
“active learning.” Although many others have also tackled explaining discrepant findings, choosing control conditions,
the challenge of creating more “active learning” environ- coding student responses, and guiding instructional design.
ments (e.g., Bonwell & Eison, 1991; Bonwell & Suther- We also briefly compare this theoretical framework with
land, 1996; Fink, 2013; Meyers & Jones, 1993; Rosenthal, other frameworks in the literature.
1995; Rowles, 2013; Rubin & Herbert, 1998) the frame-
work we introduce in this article can remediate these practi-
cal challenges by providing specific domain-general SECTION 1: A THEORETICAL FRAMEWORK FOR
guidelines for creating more engaging lessons as well as DIFFERENTIATING ENGAGEMENT ACTIVITIES
modifying existing activities to increase engagement, using ACCORDING TO THE MODE OF OVERT
simple overt measures of behavior to assess the level to BEHAVIORS
which students are engaged. Moreover, the framework is
empirically grounded and supported by evidence from a The ICAP framework, referred to in previous publications
variety of learning activities, domains, and student ages. as DOLA for Differentiated Overt Learning Activities (Chi,
This article has three major sections. The first section 2009; Menekse, Stump, Krause, & Chi, 2013), consists of a
describes our ICAP framework that defines engagement in taxonomy that differentiates four modes or categories of
terms of overt behaviors that students can undertake and engagement, based on the overt behaviors displayed or
teachers can see. In brief, we propose that there are differ- undertaken by students. Each mode of engagement corre-
ent modes or categories of “active learning,” corresponding sponds to several different types of behaviors and to differ-
to different overt behaviors that elicit different knowledge- entiable knowledge-change processes. Based on the
change or learning processes. Moreover, we propose that differential knowledge-change processes and the resulting
learning activities and their resulting overt engagement changes in knowledge they produce, the ICAP framework
behaviors can be differentiated into one of four modes: generates the hypothesis that predicts different levels of
interactive, constructive, active, or passive. Each mode cor- learning outcomes (see Figure 1). In this section, we
responds to a different set of underlying knowledge-change describe the four modes, the assumptions of the ICAP
processes, to be elaborated later. Based on the set of knowl- framework, the knowledge-change processes and the
edge-change processes, each mode predicts a different level expected changes in knowledge for each mode, the cogni-
of learning such that the Interactive mode of engagement tive outcomes, the ICAP hypothesis, and its predicted learn-
achieves the greatest level of learning, greater than the Con- ing outcomes.
structive mode, which is greater than the Active mode,
which in turn is greater than the Passive mode A Taxonomy of Four Modes of Overt Engagement
(I>C>A>P). Thus, the ICAP hypothesis predicts different Behaviors
levels of learning for different modes of overt behaviors.
Higher levels imply learning with deeper understanding. Students’ engagement with learning materials can be opera-
Although a preliminary version of this framework has been tionalized by the overt (or lack of overt) behaviors they
presented in the literature (Chi, 2009), this article expands undertake while learning. Although far from perfect, overt
the framework by explicating the underlying assumptions behaviors are a good proxy to reflect different modes of
more concretely and fully and by clarifying the knowledge- engagement that teachers can use to ascertain whether a stu-
change processes and the associated changes in knowledge dent is in fact engaged in a specific mode for a given activ-
for each category of activities. ity. Specifically, students’ overt behavioral manifestations
In the second major section, the article describes the can be characterized and differentiated into four behavioral
learning outcomes of each mode of overt activity and pro- modes: passive, active, constructive, and interactive. By
vides empirical evidence that supports the predictions made our definition, then, the mode at which students engage can
by the ICAP hypothesis. The evidence we provide in this be seen either during instruction or while students are
THE ICAP FRAMEWORK 221
P Frameworrk
ICAP Passive mode of engagement. In our taxonomy we
define a passive mode of engagement as learners being ori-
Taxonomyy of four moodes, their d
deﬁnions, aand assump
pons ented toward and receiving information from the instruc-
Iteracve,, Construcve, Acve, Passive tional materials without overtly doing anything else related
Knowledge-change processes & resulng ch hanges in kn
nowledge to learning. For example, paying attention and listening to a
Storing, In
negrang, In
nferring, Co-inferring lecture without overtly doing anything else (i.e., listening
Cognive outcomes without taking notes) is a passive engagement behavior.
ply, Transfer, Co-createe
Recall, App We acknowledge that it is possible for students to be
ICAP hypo
othesis covertly processing the materials deeply while listening to
Interacve > Construcve > Acve > Passive a lecture (or observing a video), even though overtly
appearing only to be passively engaged (this point is elabo-
FIGURE 1 The ICAP framework consists of a taxonomy of four modes
of activities, their definitions and assumptions; hypothesized knowledge-
rated later).
change processes and the resulting changes in knowledge; projected cogni- In the literature, studies that examine students’ interac-
tive outcomes; and the ICAP hypothesis of learning outcomes. tions with computer-based learning environments often dis-
criminate students’ engagement in a binary way, in terms
undertaking specific instructor-designed learning activities of whether students are oriented toward and paying atten-
(see Table 1). The top row of Table 1 shows the four modes tion (i.e., zoning out or being off-task). In these studies
of engagement behaviors, behaviors with their characteris- (e.g., see D’Mello, Olney, Williams, & Hays, 2012), “being
tic descriptors such as receiving, manipulating, generating, engaged” and looking at the instructional materials map to
and dialoguing. The left column shows three different our passive mode, which is the lowest level of engagement
instructional contexts: listening to a lecture, reading a text, in our framework.
and observing a video. The cells of Table 1 list examples of To differentiate our use of the term passive (as well as
engagement activities within each mode in the context of the terms active, constructive, and interactive), these terms
the three instructional tasks. are italicized, whereas the general dichotomy in the litera-
Table 1 shows that for any instructional task, students ture of “active” and “passive learning” are in quotes. See
can engage in different modes of activities or different Table 1, column 1, for other passive activities in the context
activities within the same mode. For example, while read- of other instructional tasks.
ing a text or watching a video, students can be engaged in a
passive, active, constructive, or interactive way depending Active mode of engagement. Learners’ engagement
on what overt behaviors they do while reading or while with instructional materials can be operationalized as active
watching a video. As depicted in Table 1, while reading, if some form of overt motoric action or physical manipula-
students can be reading silently, reading and underlining, tion is undertaken. Using this criterion to characterize
reading and taking notes in one’s own words, or reading active, laboratory studies in the psychology and cognitive
and asking a friend questions about the materials. Similarly, science literature provide numerous examples of active
while reading, students can engage in one of several types overt behaviors, such as rotating objects to inspect parts of
of activities that all fit into a single (e.g., constructive) it more carefully (James et al., 2002), looking for objects in
mode, such as self-explaining, creating concept maps, and an environment from a description of it (Kastens & Liben,
comparing and contrasting. Thus, instructional tasks are 2007), and so forth. For various performance measures,
orthogonal to engagement mode. active behaviors excel passive behaviors.
TABLE 1
Examples of Learning Activities by Mode of Engagement
PASSIVE Receiving ACTIVE Manipulating CONSTRUCTIVE Generating INTERACTIVE Dialoguing
LISTENING to a lecture Listening without doing Repeating or rehearsing; Reflecting out-loud; Defending and arguing a
anything else but oriented Copying solution steps; Drawing concept maps; position in dyads or small
toward instruction Taking verbatim notes Asking questions group
READING a text Reading entire text passages Underlining or highlighting; Self-explaining; Asking and answering
silently/aloud without doing Summarizing by copy-and- Integrating across texts; comprehension questions
anything else delete Taking notes in one’s own with a partner
words
OBSERVING a video Watching the video without Manipulating the tape by Explaining concepts in the Debating with a peer about the
doing anything else pausing, playing, fast- video; Comparing and justifications;
forward, rewind contrasting to prior Discussing similarities &
knowledge or other differences
materials
222 CHI AND WYLIE
In the context of learning measures, undertaking active student copied it, then the student was being active, because
activities have also been shown to exceed passive activities, the diagram was already present.
such as when students manipulate some parts of the learn- Constructive activities include activities such as drawing
ing materials, by pointing to or gesturing at what they are a concept map (Biswas, Leelawong, Schwartz, & Vye,
reading or solving (Alibali & DiRusso, 1999), pausing and 2005; Novak, 1990); taking notes in one’s own words
rewinding parts of a video tape (in order to review certain (Trafton & Trickett, 2001); asking questions (Graesser &
selected parts of the tape; Chi, Roy, & Hausmann, 2008), Person, 1994); posing problems (Mestre, 2002); comparing
underlining certain text sentences that they think are impor- and contrasting cases (Schwartz & Bransford, 1998); inte-
tant (Katayama, Shambaugh, & Doctor, 2005), copying grating two texts (Rouet, Britt, Mason, & Perfetti, 1996), or
some of the problem solution steps (VanLehn et al., 2007), integrating text and diagrams (Butcher, 2006), or integrat-
mixing certain chemical amounts in a hands-on laboratory ing across multimedia resources (Bodemer, Ploetzner,
(Yaron, Karabinos, Lange, Greeno, & Leinhardt, 2010), Feuerlein, & Spada, 2004); making plans (Pea & Kurland,
choosing a justification from a menu of options (Conati & 1984); inducing hypotheses and causal relations (Suthers &
VanLehn, 2000), and so forth. Hundhausen, 2003); drawing analogies (Chinn & Malhotra,
By restricting active activities to those that require some 2002); generating predictions (Schauble, Glaser, Duschl,
form of motoric behaviors that cause focused attention while Schulze, & John, 1995); reflecting and monitoring one’s
manipulating, we are distinguishing them from overt activi- understanding and other self-regulatory activities (Azevedo
ties that are carried out mindlessly. For example, if students et al., 2006); constructing time lines for historical phenom-
are asked to read passages out loud in their entirety (Oakhill, ena (Dawson, 2004), self-explaining (Chi et al., 1989), and
Cain, & Bryant, 2003), then this reading out loud activity is so forth. These behaviors can all be classified as construc-
closer to passive than active because attention was not tive because they satisfy the criterion that learners generate
focused on some specific passages or sentences. However, it some sort of external output such as notes, hypotheses, jus-
is not always clear-cut when focusing is or is not involved tifications, questions, predictions, self-evaluations, and
while being active. For example, if students read some parts time lines, all of which contain additional ideas that go
of the passages with greater emphasis (such as in a louder beyond the original learning materials.
voice), then it would appear that focused attention is involved. Our use and operational definition of constructive or
Sometimes it is ambiguous whether an activity should be clas- generative is more specific and concrete than the ways the
sified as passive or active, unless one can determine whether it terms “constructive” and “constructionism” are used in the
causes focused attention. broader literature. In the broader literature, “constructive”
has several meanings. One meaning reflects a theoretical
Constructive mode of engagement. Our taxonomy perspective that students should “construct” their own
defines constructive behaviors as those in which learners understanding rather than “being told” or “instructed”
generate or produce additional externalized outputs or by a teacher (Bruner, 1961; Papert, 1980; Piaget, 1930).
products beyond what was provided in the learning materi- In this context, “constructionism” is contrasted with
als. Thus, a characteristic descriptor of the constructive “instructionism” (Kafai, 2006). A second use of construc-
mode is generative. To meet the criteria for constructive, tion contrasts “being told” or “direct instruction” with
the outputs of generative behaviors should contain new “discovery learning,” in which students construct the rules
ideas that go beyond the information given; otherwise such and relations they need. A third use of “constructivism”
behaviors are merely active/manipulative. For example, in aligns more closely with our definition, in referring to it as
a constructive behavior such as self-explaining, learners are “knowledge construction by the individual,” or “social co-
articulating what a text sentence or a solution step means to constructions of knowledge” by two or more individuals.
them (Chi, Bassok, Lewis, Reimann, & Glaser, 1989), by This view does align with our view of constructive in
generating inferences that are not explicitly stated in the assuming that “knowledge is actively constructed by the
text sentence, or providing justification for the step. Both learner” (Loyens & Gijbels, 2008, p. 352). However, our
the inferences and the justification go beyond the provided definition is still more concrete because we operationalize
information. If the student’s self-explanation is verbatim to it explicitly as generating outputs that go beyond the infor-
what was read, then the student is only repeating or self- mation presented.
explaining actively rather than constructively, because no Besides being more concrete and explicit, by focusing
new information is provided. our definition from the learner’s perspective, we also un-
For another example, suppose a student is reading a confound the role of the instructor from the role of the
worked-out example solution of a physics problem in a learners. For example, it is sometimes assumed that the
text. If this example solution does not have any diagrams in “constructivist” perspective is experiencing a backlash in
it and the student draws a free-body diagram, then the stu- recent educational literature because at times learners bene-
dent would have constructed an output. On the other hand, fit from instructors’ direct scaffolding (Kirschner, Sweller,
if the solution example had a diagram already and the & Clark, 2006; Mayer, 2004). However, such findings are
consistent with our framework because instructional scaf- when they consist of mutual exchanges of ideas between
folding allows students to be more generative in that the two individuals resulting in new ideas that neither individ-
scaffolding often requires the students to respond in a way ual knew initially nor could generate alone. Such mutual
that is usually constructive. For example, Chi, Siler, Jeong, exchanges of ideas imply that both partners make substan-
Yamauchi, and Hausmann (2001) argued that tutoring is tive contributions to the topic under discussion, such as
effective because many of the tactics used by tutors can be defending and arguing a position (Schwarz, Neuman, &
reframed as prompting or encouraging tutees to construct Biezuner, 2000), criticizing each other by requesting justifi-
knowledge, either through the use of content-free prompts cation (Okada & Simon, 1997), asking and answering each
or scaffolding prompts such as pumping for “what else,” other’s questions (Webb, 1989), explaining to each other
hinting (“So, it’s kind of leaving out the lungs there?”), and (Roscoe & Chi, 2007), and elaborating on each other’s con-
fill-in-the-blank kinds of requests (such as “OK, and it goes tributions (such as clarifying, building upon, correcting,
from the atrium to the [pause]?”). These prompts encourage etc.; Hogan, Nastasi, & Pressley, 1999). This is an ideal
students to respond more constructively because they conception of an interactive or joint dialogue pattern in that
require more than a yes/no response (which we consider to both partners make substantive contributions.
be an active response only). In addition to being constructive, a dialogue must have a
sufficient frequency of turn taking to meet our definition of

Interactive mode of engagement. The term interactive. Two students who take turns giving minilec-
“interaction” has been used loosely in the literature to refer tures to each other, even if they are being constructive, will
both to interpersonal activities and to characteristics of likely not reap the same benefits as two students who fre-
human–computer systems (i.e., if a learning environment or quently interject to ask each other questions, make clarifica-
computer-based system expects a response from the user, tions, and so forth. We hypothesize that by frequently
and provides feedback to that response, then often it is taking turns, it will be easier for students to incorporate
labeled an interactive system). In layman’s terminology, to their partners’ understanding of the domain and to make
engage means to interact with a device, a learning environ- adjustments to their own mental model, because such
ment, or system without differentiating whether the interac- dynamic and ongoing exchanges have the dual advantages
tions of the leaner is active or constructive. In this article, of allowing for more frequent revisions on smaller compo-
we operationalize interactive behaviors to dialogues that nents of knowledge. We have made the same argument for
meet two criteria: (a) both partners’ utterances must be pri- why self-explaining facilitates learning, because it fosters
marily constructive, and (b) a sufficient degree of turn tak- ongoing integration of minute pieces of new knowledge
ing must occur. We do not restrict who the partners can be, (Chi, de Leeuw, Chiu, & Lavancher, 1994, p. 474), and
provided that the criteria are met. Examples include a there is some evidence in support of our claim of the impor-
learner talking with another person who can be a peer, a tance of taking frequent turns (Chi & Kang, 2014).
teacher, a parent, or computer agent (assuming the com- Another type of dialoguing that is not very interactive is
puter agent responds in a content-relevant way). when one partner dominates and generates most of the sub-
Because being interactive requires that each person of stantive contributions and the other partner merely agrees
the group contributes constructively, this means that if two or contributes with backchannel responses (e.g., “mmm,”
people interact only in some physical or motoric way with- “uh-huh,” etc.). In such cases, the dominant partner or the
out discourse, such as two students copying each other’s “speaker” may be learning more than his or her “listening”
homework solutions, then they are interacting only actively partner because the “speaker” is being constructive,
and not constructively. The available evidence agrees with whereas the “listener” is simply being active by agreeing
our interpretation, in that when two people work together, and affirming to some of the “speaker’s” contributions. We
learning seems to occur when there is verbal discussion have referred to this pattern of dialoguing as individual dia-
rather than only motoric or physical interactions (Milrad, logue (Chi & Menekse, in press). This may explain why
2002). Therefore, for now, we restrict our definition of some collaborative learning studies found no advantage of
interactive behaviors to discourse or dialoguing (Salomon collaboration over individual learning (Barron, 2003; Yetter
& Perkins, 1998). et al., 2006). Thus, individual dialogue pattern tend to pro-
Are all dialogues the same, and are they all equally inter- mote more learning for the dominant speaker, whereas both
active and beneficial to learning? As previously indicated, partners can benefit from joint dialogue pattern.
according to the ICAP framework, dialogues are truly inter- The same operational definition of interactions can be
active only if each speaker’s utterances generate some applied to the level of interactivity of a computer-based sys-
knowledge beyond what was presented in the original tem (such as an intelligent tutoring system or a learning envi-
learning materials and beyond what the partner has said; ronment). Because such systems often require a response
thus, both partners need to be constructive. It is in this sense from the student, are they all truly interactive? No, because
that interactive subsumes constructive. Damon (1984) and our definition of interaction is learner centered and applies to
Rafal (1996) similarly defined dialogues as truly interactive the kind of responses that a learner makes. Suppose a
224 CHI AND WYLIE
student’s response to a computer-based learning environment responses is necessary to confirm the actual mode of an
consists of selecting an answer from a menu of choices; activity. For example, in dialogues, some dyads might
selecting is only active in our taxonomy in that the student engage in individual dialogue (in which only one partner is
does not generate a product. Therefore, responses to a so- being constructive) instead of in joint dialogue, where both
called interactive computer-based system can also be classi- partners are constructive. As a result, an analysis of the con-
fied as passive, active, constructive, or interactive, depending tent of the products or the behavioral transcriptions is nec-
on the level of cognitive engagement required of the learners. essary to correctly classify an activity. From the teacher’s
perspective, such an analysis might inform how to revise
Assumptions Underlying the ICAP Taxonomy the activity so that it aligns more with her intention.
Our taxonomy of four modes of overt behaviors requires Advantages of externalized outputs. Although we
eight assumptions that are described next. acknowledge that students can be generative covertly with-
out externalizing any outputs, we also assume that there are
Content relevant. First, the classification into the four practical, cognitive, learning, and epistemic advantages to
modes of activities assumes that students are doing activi- externalizing outputs as static products (e.g., notes, dia-
ties that are content relevant. For example, if students are grams). From a practical standpoint, externalized static out-
gesturing as part of an active task (such as when teachers puts provide performance data for teachers and researchers
ask students to point at the important equations on the to analyze and verify that students are completing the tasks
whiteboard), if students point instead at random figures on and doing them in the way intended by the design of the
the whiteboard, then their behavior would not be considered tasks. From a cognitive perspective, one advantage of exter-
a beneficial active one. Similarly, when students are con- nalizing outputs is that it can help overcome cognitive load
structing or interacting, their outputs or dialogues must be (Sweller, 1994). For example, it is difficult to generate a
content relevant for the activity to be considered truly con- large or complex concept map in memory, so offloading it
structive or interactive. externally onto a tablet or paper reduces the memory
requirement. From a learning perspective, the externalized
Intended versus enacted. Our taxonomy starts with outputs become new materials that a student can further
classifying the intended overt behaviors as designated by examine to infer new knowledge or learn from errors.
an activity. However, asking students to carry out a behav- Moreover, the process of externalizing provides students
ior does not guarantee that the expected behavior will actu- the opportunity to monitor whether in fact they know the
ally be carried out in the intended way. For example, if a materials. From an epistemic perspective, an advantage to
student is asked to be constructive by generating a sum- externalizing outputs is that it commits a student to that
mary, the student can reduce the task to an active one by knowledge and gives her a sense of ownership of that
summarizing using a delete-and-copy strategy (i.e., delete knowledge (Kitchner, 1983).
the irrelevant sentences and copy the relevant ones). To
know accurately what mode of behavior a student is carry- Greater probability. Our taxonomy assumes that it is
ing out, teachers need to judge the students’ products from only more likely that the overt behaviors are an approxima-
an ICAP perspective (i.e., evaluate the summary). tion to how the students are actually engaging with the
Besides a mismatch between the intended engagement learning materials most of the time. For example, while lis-
mode and the enacted engagement mode (the enacted tening to a lecture, if students are doing nothing else
behavior), there can also be a misalignment between the besides orienting and receiving, then it is more likely that
overt display of behavior and the covert processes. For they are engaging passively while listening to the lecture.
example, a student may appear to be only actively involved We are not claiming that they are engaged passively the
in learning (such as underlining key concepts in a book) entire time that they are listening (e.g., at times students
and yet in reality the student is covertly constructive (such could be covertly self-explaining while appearing to be pas-
as thinking deeply about and self-explaining the material). sive.) Moreover, even though students may not always
Nevertheless, we rely on overt manifestations as our mea- complete an activity in the way it was intended, we assume
sure of engagement not only because it is the best noninva- that students are more likely to engage with the learning
sive measure available but also because it provides activity at the prescribed level.
concrete criteria for teachers to use when designing class-
room and homework activities. This design feature of ICAP Independence. Although this point was articulated
is presented in the third section of this article. earlier in the context of defining our taxonomy, this impor-
tant assumption is reiterated here briefly. Engagement
Analyzing the outputs. Because there may be a dis- activities are orthogonal to instructional tasks, in that a stu-
crepancy between the intended engagement mode and the dent can undertake various engagement activities while
enacted engagement mode, a content analysis of learners’ learning, regardless of how instruction is delivered. Thus,
the mode of engagement, referring to what students are Learning is typically conceived of as causing some
actually doing while being instructed, is independent of the changes, and these changes can occur at various grain sizes,
instruction itself. such as systemic changes in terms of people’s participation
in social or cultural activities; behavioral changes in terms
Hierarchy. From the way the behavior of each mode is of what people do such as eating a healthier diet; knowl-
operationally defined, there is a hierarchy in our taxonomy edge changes in terms of what people know, causing them
so that a higher mode subsumes a lower mode. That is, to be able to solve some problems and explain some phe-
interactive behaviors subsume constructive behaviors, and nomena; and cellular changes in terms of changes in the
constructive behaviors subsume active, and active behav- activation patterns of their brain’s neurons. These various
iors subsume passive. For example, we pointed out earlier grain sizes of change require different measurements to
that to meet our definition of interactive, both partners must detect whether changes have occurred. Because the ICAP
be constructive. Likewise, to be constructive, such as draw- framework is intended to understand how cognitive engage-
ing a diagram, one must also be active (i.e., engaging in the ment can enhance an individual student’s learning of com-
motoric act of drawing). Finally, being active, such as plicated school subject matters, the appropriate grain size
underlining a sentence in a text, requires that the learner to talk about learning is the knowledge level; thus we use
focuses on and orients toward the underlined sentences, the terms “changes in knowledge” and “knowledge-change
thus active behavior includes passive. processes” to reflect our assessment of students’ under-
standing after learning and the processes by which they
Intermode boundary. The boundaries between the acquired this knowledge.
modes of activities are not meant to be totally rigid. Obvi- By our definition, knowledge-change processes are
ously some activities may be difficult to classify and fall on dynamic processes that students engage in while learning
the boundary between two categories, because accurate new information. We simplify and postulate that there are
classification also depends on the actual cognitive processes four broad types of knowledge-change processes that are
carried out by the student. One of the most difficult activi- relatively distinct from each other, and each can be associ-
ties to classify is problem solving, because it depends on ated with a passive, active, constructive, or interactive
how a specific student is doing it and the context in which it mode of behavior. We first summarize the four types of
was taught. For example, solving an algebra problem can knowledge-change processes by providing a single-word
be classified as an active behavior when the student label for each.
matches the conditions of application of an equation and
merely applies the formula of an equation (e.g., plug and Store: new information is stored in an isolated way
chug). However, solving a novel or more difficult problem (passive).
may be classified as constructive if a student has to rederive Integrate: new information activates relevant prior
an equation, or reconceptualize certain components of a knowledge and while storing, new information is inte-
problem, such as reconceptualize two masses of a physics grated with activated prior knowledge (active).
problem (one block on top of another block) as a single Infer: new information is integrated with activated
compound body. There are ways to determine more pre- prior knowledge, and new knowledge is inferred from
cisely the students’ cognitive processes when undertaking activated and integrated knowledge (constructive).
an activity in a controlled laboratory setting, but this cannot Co-Infer: Each learner infers new knowledge from
be done easily in a classroom setting. activated and integrated knowledge and iteratively
In summary, when these assumptions are met, we assert infers knowledge with new inputs from conversational
that the ICAP taxonomy is a useful and powerful way to partner(s) (interactive).
classify engagement activities.
Processes and changes for passive. In a passive
Knowledge-Change Processes Underlying Each mode of engagement, the knowledge change processes can
Mode of Engagement and the Expected Changes be postulated to be isolated “storing” of the received infor-
in Knowledge mation in an encapsulated manner. Storing information in
an isolated way means that new knowledge is not integrated
In Chi (2009), we briefly introduced the possible processes with existing or prior knowledge. The consequence of iso-
underlying each mode of engagement. In this article, we lated storing of information is that the newly learned encap-
elaborate and refer to these engagement processes as sulated knowledge is “inert” (Whitehead, 1929), meaning
“knowledge-change processes,” then describe (a) what they that access to that knowledge is possible only when a spe-
are, (b) the assumptions we make with respect to these cific cue to activate it is given, such as an exam question
knowledge-change processes, (c) the resulting changes in expected from a specific chapter (Brown & Palincsar, 1989,
knowledge, and (d) the expected cognitive and learning p. 394), or the same context is provided. An example of
outcomes as a result of the knowledge changes. encapsulated knowledge can be seen in a protocol snippet
226 CHI AND WYLIE
collected by Scardamalia (1992; cited also in Chi et al., can participate in individual or joint dialogue patterns,
1994). A student had read about how blood clots. When the but only the joint patterns are truly interactive. In these
student was asked to explain how a cut heals, the student interactive joint dialogues, each member of the dyad must
first responded in the following way, based on the materials be constructive while interacting, thereby engaging in the
she had read: “When you get a cut it bleeds. In your blood cognitive processes of activating, integrating, and infer-
there are things called platelets. Platelets made a shield on ring. But in addition, in joint dialogues, each speaker’s stor-
your cut, it is called a scab, it protects your skin when it is ing, activating, integrating, and inferring processes further
healing.” So the student obviously had stored the newly benefit from the contributions provided by a peer, such as
learned information. After this answer, the student was asked their elaborations, feedback, suggestions, and perspectives.
to use her own theory to answer the same question. Here is Next is a hypothetical example of such co-inferring
the student’s second response to the same question: “My the- processes.
ory is that when you get a cut the blood vessel that got cut Suppose Alice (Learner A) and Bob (Learner B) are both
dies and the heart stops sending blood to that vessel until it exposed to base instructional information that they both
heals.” Thus, the student’s old ideas were not integrated encode (Knowledge 1) and are collaboratively learning.
with the newly learned knowledge, suggesting that the new Alice starts the interaction by activating her own prior
ideas were stored in an isolated or encapsulated way. knowledge relevant to Knowledge 1 (Schema A) and makes
an inference from Schema A (Inference 1). Alice shares
Processes and changes for active. When students Inference 1 with Bob, which is information that goes
engage in an active way, such as highlighting key sentences beyond what was originally presented in Knowledge 1. Bob
when reading on a computer screen, such manipulations then activates his own schema (Schema B) relevant to
can be interpreted as ways of emphasizing certain parts of Knowledge 1 plus Inference 1, and based on his Schema B,
the learning materials. Such emphases may cause learners makes an inference (Inference 2) that goes beyond the orig-
to activate a body of knowledge (such as a skeletal or par- inal Knowledge 1 C Inference 1. Alice can now make an
tial schema or a mental model) that is relevant to the inference and contribution based on Inference 2, plus the
emphasized sentences. In addition, once a relevant schema original presented Knowledge 1, and her own Inference 1.
is activated, learners can assimilate or integrate new infor- So in continued interactions, partners benefit from the infer-
mation into the activated schema, allowing the learner to ences of their partners, and each subsequent inference can
fill gaps in the schema, thereby making it more complete incorporate the inferences made by the other partner as
(Bartlett, 1958). Gap filling is a process that is easily mod- well as integrating it with one’s own contributions and
eled computationally and has been explored extensively inferences. Thus, the advantage of such co-inferring pro-
(Conati & Carenini, 2001; VanLehn, Jones, & Chi, 1992). cesses is that each partner can benefit from the inferences
of the other partner in a cumulative and spiraling way.
Processes and changes for constructive. Construc- In general, the knowledge-change processes for each
tive behaviors often require the processes of “inferring” higher mode subsume the processes for the lower mode, in
(e.g., induce, deduce, and abduce). Inferring also includes the same way that the behavior for the higher modes sub-
revising, repairing, reorganizing, and reflecting. Inferring sumes the behavior of the lower modes.
can be thought of generally as a process of elaborating,
such as adding more details or qualifications. Revising and Assumptions Underlying Knowledge-Change Processes
repairing can be thought of as the processes of changing
what was initially stored but is incorrect. Reorganizing can Just as we made assumptions regarding the classification of
be changing hierarchical relations, or pattern of relation- overt engagement behaviors, we also made a number of
ships. In contrast, reflecting may require a student to evalu- assumptions regarding the possible underlying knowledge-
ate what she or he does or does not understand in order to change processes as involving storing information in an
carry out a constructive activity such as generating a isolated way, activating prior relevant knowledge, integrat-
question. ing new information with prior knowledge, and/or making
Finally, constructive activities also occur when learners inferences from it. Of course, this can be done in various
integrate various parts of the learning materials (including permutations and repetition of storing, integrating, and
comparing, combining/linking, discriminating/contrasting inferring, either doing it alone or with a peer. Our main
information from disparate sources or different paragraphs assumption is not whether we have postulated accurately
within a passage, reasoning analogically, etc). Thus, the the knowledge-change processes for each mode but rather
knowledge-change processes of constructive activities can that they are relatively different for each mode. Besides this
involve multiple types of inferring processes. main assumption that engaging behaviors elicit these dis-
tinct fundamental knowledge-change processes, there are
Processes and changes for interactive. As three other important assumptions about the nature of
described previously, individuals interacting in dialogues knowledge-change processes.
Distinct from learning processes. The first assump- the way information was initially learned and the changes
tion is that knowledge-change processes underlying in knowledge resulting from it.
engagement are distinct from the cognitive processes of In Table 2, the first row (labeled Example activities)
learning in the context of a specific instructional/learning gives two examples of behaviors in each mode, the second
task itself, such as solving a problem or understanding a row postulates the underlying knowledge-change processes
diagram or reading a passage. For example, the cognitive for each mode of engagement, and the third row postulates
processes involved in the learning task of problem solving the expected changes to the knowledge as a result of the
includes creating and searching a problem space, whereas engagement processes undertaken.
the processes of understanding a diagram might include
perceiving the spatial relations among the components, A Hypothesized Order of Learning Outcomes:
mentally rotating the angles in the diagram, and the pro- Interactive>Constructive>Active>Passive (ICAP)
cesses of reading include decoding and identifying word
meaning. These processes, unique to the specific learning The preceding section described the underlying knowledge-
tasks of solving problems, understanding diagrams, and change processes postulated for each mode of activity
reading, are distinct from the knowledge-change processes (summarized in the second row of Table 2), and the associ-
of engagement activities such as storing, integrating, and ated knowledge changes expected for each mode (the third
inferring while engaged in self-explaining or drawing a dia- row of Table 2). On the basis of these knowledge changes,
gram. Our assumption is that the knowledge-change pro- different cognitive and learning outcomes will result. In
cesses of storing, integrating, and inferring, supplement this section, we describe the expected relative cognitive
these task-specific learning processes. and learning outcomes for each engagement mode, result-
ing in a hypothesis about learning that is referred to as
ICAP (as shown in the fourth and fifth rows of Table 2).
Same set of knowledge-change processes. The
For passive activities, because new knowledge is
second assumption is that, although the types of activities
encoded in an isolated or encapsulated way during learning,
within each mode may be different (e.g., in reading a text,
the outcome is that such inert knowledge can be retrieved
the various active engagement activities could be underlin-
and recalled, but only when relatively the same cue or con-
ing text sentences or summarizing by deleting irrelevant
text is given. Such isolated storing is adequate for learning
sentences), their underlying knowledge-change processes
that does not require integration with prior knowledge, such
are the same. That is, underlining text sentences causes the
as learning a specific procedure of how to operate an ATM
learner to focus attention on the underlined sentences in the
machine. Knowledge stored in an isolated way can be
same way that deleting irrelevant sentences when produc-
retrieved and reused later when the same context is pro-
ing a text summary causes the learner to focus attention on
vided (such as seeing another similar ATM machine). It
the remaining sentences of the summary. Although the
can also be retrieved when using episodic tags. For exam-
activities of underlining and summarizing are different,
ple, a student might be able to solve a problem like the one
the knowledge-change process of focusing attention is the
the teacher worked out at the blackboard if the student was
same. As another example, for constructive activities such
reminded that it is the problem demonstrated last Tuesday.
as self-explaining and drawing a concept map, the knowl-
For active activities in which students manipulate infor-
edge-change processes for both activities include integrat-
mation while learning, this manipulation causes them to
ing and inferring new knowledge. It is precisely the
activate the relevant prior knowledge corresponding to the
similarity in the knowledge-change processes for each
emphasized information (such as a related schema). This
mode of engagement that allows us to generalize our pre-
then allows them to assimilate and fill gaps in their schema,
dictions of learning outcomes as a function of the mode of
making their activated schema more complete and strength-
engagement across various activities, to be explained next.
ened. Students can more readily retrieve the knowledge and
apply it in new context, such as when solving problems
Cognitive outcomes resulting from the knowledge- somewhat similar to what they have learned, explaining
change processes. Finally, we assume that different similar concepts, and so forth. In short, learning with active
knowledge-change processes cause different changes in engagement can be quite substantial because significant
one’s knowledge, resulting in different cognitive outcomes, knowledge completion has occurred. One could say that
such as being able to recall, apply, transfer, and co-create. they have achieved, at minimum, a shallow understanding.
For example, if a student has only learned the material in For constructive activities, because learners typically
an encapsulated way (i.e., the information was stored in an generate inferences and relations for conceptual knowledge,
isolated way while learning), then the student can only and rationales and justifications for procedural knowledge,
recall facts. The cognitive outcome for each of the four not only does their knowledge or schema become more
modes is elaborated in the next section along with the learn- complete and strengthened, but in fact their schema can be
ing outcomes. Thus, we attribute the outcome of learning to more enriched and coherent because it can be substantially
228 CHI AND WYLIE
TABLE 2
Example Activities, Knowledge-Change Processes, Knowledge Changes, Cognitive Outcomes, and Learning Outcome by Mode of Engagement
CATEGORY Characteristic PASSIVE Receiving ACTIVE Manipulating CONSTRUCTIVE Generating INTERACTIVE Dialoguing
Example activities Listening to explanations; Taking verbatim notes; Self-explaining; Discussing with a peer;
Watching a video Highlighting sentences Comparing and contrasting Drawing a diagram with a
partner
Knowledge-change Isolated “storing” processes in “Integrating” processes in “Inferring” processes include: “Co-inferring” processes
processes which information is stored which the selected & integrating new information involve both partners taking
episodically in encapsulated emphasized information with prior knowledge; turns mutually creating. This
form without embedding it activates prior knowledge & inferring new knowledge; mutuality further benefits
in a relevant schema, no schema, & new information connecting, comparing & from opportunities &
integration can be assimilated into the contrasting different pieces processes to incorporate
activated schema. of new information to feedback, to entertain new
infer new knowledge; ideas, alternative
analogizing, generalizing, perspectives, new directions,
reflecting on conditions of a etc.
procedure, explaining why

something works.
Expected changes New knowledge is stored, but Existing schema is more New inferences create new New knowledge and
in knowledge stored in an encapsulated complete, coherent, salient, knowledge beyond what was perspectives can emerge
way. and strengthened. encoded, thus existing from co-creating knowledge
schema may become more that neither partner knew.
enriched; procedures may be
elaborated with meaning,
rationale and justifications;
and mental models may be
accommodated; and schema
may be linked with other
schemas.
Expected cognitive Recall: knowledge can be Apply: knowledge can be Transfer: knowledge of Co-create: knowledge and
outcomes recalled verbatim in applied to similar but non- procedures can be applied to perspectives can allow
identical context (e.g., reuse identical contexts (i.e., a novel context or distant partners to invent new
the same procedure or similar problems or problem; knowledge of products, interpretations,
explanation for identical concepts that need to be concepts permit procedures, and ideas.
problems or concepts). explained) interpretation &
explanations of new
concepts.
Learning outcomes: ICAP Minimal understanding Shallow understanding Deep understanding, potential Deepest understanding,
for transfer potential to innovate novel
ideas
revised (accommodated), and perhaps more interconnected Finally, for interactive activities, the changes in knowl-
or linked with other schemas, facilitating transfer. Thus the edge as a consequence of reciprocally constructive or co-
knowledge structure changes substantially. One example of inferring interaction is that each peer’s alternative perspec-
accommodation is the change from conceiving of the tives, guidance, and challenges will improve and expand
human circulatory system as a “single-loop” to a “double- each other’s knowledge in a cyclical dynamic way. In suc-
loop” model (Chi et al., 1994). Learners can achieve cessful dialoguing, the contributions of each peer incorpo-
such accommodation changes in their knowledge structure rate the inferences of their partner, potentially resulting in
from constructively engaging with the materials (e.g., new knowledge that neither partner could have created
self-explaining). alone. The resulting changes in knowledge are that novel
In the context of conceptual domains, more enriched and and innovative ideas and perspectives may emerge that nei-
interconnected understanding can provide generalizations ther peer originally knew and could not have generated
of explanations, reasoning analogically, and so forth. In the while working independently. So one could say that learn-
context of procedural domains, providing justifications and ers in this mode have achieved the deepest understanding
rationales allows the procedures to be transferred and used and can perhaps co-create innovations.
in a new context. Thus, knowledge-change processes from As previously described, the different learning outcomes
constructive activities can deepen one’s understanding of that we have postulated for each mode can be mapped or
the materials to facilitate transfer. Transfer is difficult to translated into differing levels of learning, such as minimal
achieve, as pointed out in a 2012 special issue on transfer understanding, shallow understanding, deep understanding,
(Goldstone & Day, 2012). and deepest understanding (as shown in the last row of
Table 2). Thus, an interactive mode of engagement (I) can (Menekse et al., 2013). Aside from reading a short passage
enhance learning more than a constructive (C) mode, which of background knowledge, the four modes or conditions
is better than an active (A) mode, which in turn is better consisted of reading a text passage (passive); reading and
than a passive (P) mode. In short, we hypothesize that the highlighting important sentences within the text (active);
relative learning levels will be ordered more or less in interpreting a graph that described the information con-
the direction of I>C>A>P. Although we could generate tained in the text passage but without having read the text
the ICAP hypothesis from the hierarchical nature of the passage (constructive); and interpreting the same graph
behavior modes alone, postulating reasonable underlying jointly with a peer (interactive), again without having read
knowledge-change processes for each mode and then see- the text passage. The learning outcomes, comparing pretest
ing their hierarchical nature provide further validity to the and posttest for each condition, are shown in Figure 2.
hypothesized order of the I>C>A>P hypothesis. These results support the ICAP hypothesis and show that
learning improves significantly at a rate of around 8% to
10% across each mode.
SECTION 2: EMPIRICAL VALIDATION OF THE ICAP
HYPOTHESIS Two Studies With Three Modes of Engagement
To recap, the ICAP framework defines and differentiates In addition to the study we conducted that included all four
different modes of overt behaviors, postulates the underly- modes of engagement, we found two studies the instruc-
ing knowledge-change processes, and the learning out- tional conditions of which can be interpreted to fall across
comes for each mode. The learning outcomes can be three modes of engagement according the ICAP frame-
predicted to vary from minimal, to shallow, to deep, and work. One lab study (Coleman, Brown, & Rivkin, 1997),
deepest. The ICAP hypothesis postulates that learning is within the domain of evolutionary biology, consisted of a
enhanced to the greatest degree with interactive behaviors, three-person learning context in which each student took
followed next by constructive behaviors, followed next by one of three different roles and the roles rotated. Specifi-
active behaviors, and followed last by passive behaviors. cally, the three roles were listening to another student
Because the entire ICAP framework and its associated (which we categorize as being passive because the overt
ICAP hypothesis is conceptually derived, it is essential to behavior is receptive in nature), summarizing (which we
seek empirical validation of the hypothesis. categorize as active because they report that the protocols
In our prior paper (Chi, 2009), evidence in support of the revealed that in many of the summaries participants did not
hypothesis consisted of a random assortment of mostly lab- go beyond the given text; Coleman et al., 1997, p. 360), or
oratory studies that examined a variety of tasks that partici- explaining the materials (constructive). Although the study
pants could undertake, using a variety of outcome did include an additional factor of whether or not students
measures. The prior paper included studies that can be prepared their summaries and explanations with the intent
interpreted to be making pairwise comparisons between to teach, we do not consider this an interactive activity
two modes of engagement. In this article, we present four because while teaching, their partners were instructed only
types of additional evidence in support of the ICAP hypoth- to listen and not ask questions. Thus we averaged the scores
esis. First, we present the results of a study from our lab across this factor for summarize-to-self and summarize-to-
that compared all four modes of learning, as the strongest other to create an overall summarize score (and repeated
test of the ICAP hypothesis. Second, we present two studies this process for the explaining-to-self and explaining-to-
from the literature that compared three of the four ICAP other conditions). We plotted these new averages, and the
modes. Third, we present additional studies that made pair- results show the outcomes of students in each of these three
wise comparisons between engagement modes (passive, roles, for two measures of near transfer and one measure of
active, constructive, and interactive) for three specific far transfer. Figure 3 shows that students did increasingly
engagement activities: taking notes, creating concept maps, better from passive to active to constructive.
and self-explaining. For example, we look at the learning A second study that can also be reinterpreted to have
differences between self-explaining enacted in an active manipulated three of the four modes of active learning was
way compared to self-explaining enacted in a constructive conducted in the domain of plate tectonics (Gobert & Clem-
mode. Finally, we present evidence from classroom studies. ent, 1999). They compared learning gains among students
Thus, all the supporting evidence presented in this article who studied by drawing diagrams from text (constructive),
comes from studies that were not cited in the prior paper. writing summaries of the text (active), or only read the text
(passive). As predicted by the ICAP hypothesis, the results
A Study With Four Modes of Engagement showed that the constructive (diagram) group did better on
both measures of spatial and causal knowledge than the
We have undertaken a study in our lab that included all four active (summary) group did, which in turn did better than
modes of engagement in the domain of materials science the passive reading group.
230 CHI AND WYLIE
interface. Taking notes is also somewhat similar to summa-

rizing. In this section, we consider all these related activi-
ties (underlining, highlighting, and summarizing) as
instances of note taking.
Previous studies have revealed a number of ways to
operationalize the engagement activity of taking notes. Stu-
dents may be asked to copy notes verbatim in their entirety,
a passive activity because attention is not focused on some
emphasized parts of the notes. Or students may summarize
a text or lecture in a copy-and-paste way (in which case we
would classify it as active because the copied parts can be
focused on), or take notes in their own words (which is
often constructive because the summary may include con-
FIGURE 2 Pretest and posttest performance by mode of engagement
(Menekse, Stump, Krause, & Chi, 2013). clusions and inferences). In addition, recent technological
innovations have supported students in collaborative note
taking (interactive). As expected, the manner in which stu-

Pairwise Comparisons Across Two Modes dents take notes influences their learning. For example, a
for Three Engagement Activities lab study that can be classified as passive versus active,
demonstrated that active note taking in which students
As previously mentioned, we take a different approach to underlined the most important sentence in each paragraph
pairwise comparisons of studies. Instead of comparing two was more effective than simply passively reading the para-
different activities undertaken in two different modes, as graphs (Rickards & Friedman, 1978; see Peper & Mayer,
was done in Chi (2009), we use the ICAP taxonomy to 1986, for similar results with a video-watching task). In
compare how a specific activity is implemented, as was comparing constructive to passive, Trafton and Trickett
done in Fonseca and Chi (2011). Here we focus on three (2001) saw a correlation between constructively taking
activities: note taking, concept mapping, and self-explain- notes using an online notepad and improved problem-solv-
ing. These three were chosen because students often engage ing performance. Namely, students who chose to take notes
in them while learning, and many studies have been carried did better than those that passively studied the material.
out testing their effectiveness by manipulating how they are Further evidence supporting the ICAP hypothesis can be
implemented. We can thus verify whether the levels of seen in a study that compared constructive to active note
learning outcomes are in the predicted directions according taking. For example, taking free-form notes (constructive)
to the ICAP hypothesis. The studies on the three engage- is more effective than cutting and pasting short segments of
ment activities are summarized in Tables 3 to 5 and the target text (active; Bauer & Koedinger, 2007). Finally,
described next. note taking can also be done interactively. Kam et al.
(2005) developed a collaborating note-taking system that
Note taking. In learning from text, taking notes is a led students to take better notes compared to students who
common learning strategy that many students undertake on took notes alone. Students who worked collaboratively
their own, even without a teacher’s advice (Kobayashi, were more likely to include reflections and pose questions
2006). Several other activities are related to taking notes, about the material than students who worked alone. How-
such as underlining the most important sentences per para- ever, the small sample size in their classroom study was
graph or highlighting the key sentences using a computer insufficient to observe any significant learning differences
8
5
# Correct
4 Average Listen (passive)

Average Summarize (acve)
3
Average Explain (construcve)
2
0
Near Transfer 1 (max = 10) Transfer 2 (max = 10) Transfer (max = 3)
Assessment Type
FIGURE 3 Near and far transfer posttest performance by mode of engagement (Coleman, Brown, & Rivkin, 1997).
TABLE 3
Pairwise Comparisons Between Different Modes of Note Taking
Passive Active Constructive Interactive
Passive No known studies

Active Note taking > No note taking for far Highlighting notes D Pasting Notes
transfer (Peper & Mayer, 1986) (Bauer & Koedinger, 2007)
Underlining > Reading (Rickards & Summarize after each segment D
Friedman, 1978) Summarize throughout lecture
(Peper & Mayer, 1986)
Constructive Taking notes with notepad > Not Free-form notes > Cutting/Paste No known studies
taking notes (Trafton & Trickett, Bauer & Koedinger, 2007)
2001)
Interactive No known studies No known studies Collaboratively taking notes > No known studies
Individually taking notes, in
the quality of the notes, but
no learning difference (Kam
et al., 2005)
Note. The light gray shading refers to the diagonal cells with activities implemented in the same mode; the dark gray shading refers to cells with redundant
studies as the non-shaded ones.
between the interactive (collaborative) and constructive encourage students to physically select but not generate
(individual) conditions. or integrate the information. Similarly, Peper and Mayer
The ICAP hypothesis can also explain results from (1986) found no difference between students who took
studies that show no differences between different types notes by summarizing after segments of a lecture and
of note-taking styles that fit the same ICAP category those who took notes throughout the lecture. Assuming
(e.g., both conditions are active or both are construc- that they took similar type of copy–paste notes, the
tive). Bauer and Koedinger (2007) found no learning ICAP framework predicts this lack of difference because
differences between students who used a computer inter- both conditions can be considered active. However, to
face to highlight segments of a text and those who used accurately interpret the results, we would need to exam-
copy and paste to select segments of the text to create ine the product of note taking. This and other limitations
notes. This finding is expected according to the ICAP are discussed in Section 3. These studies are shown in
hypothesis because both are active activities that the diagonal cells of Table 3.
TABLE 4
Pairwise Comparisons Between Different Modes of Concept Mapping

Active Copy map > Read map (Willerman No known studies
& Mac Harg, 1991)
Constructive Correcting concept map > Reading Building a concept map from No known studies
text (Chang, Sung, & Chen, 2002) generating > Constructing a
Concept maps C Lecture > Lecture map via selection (Yin,
only for higher level material and Vanides, Ruiz-Primo,
low PK students (Schmid & Ayala, & Shavelson, 2005)
Telaro, 1990)
Concept maps > Study C
Discussion for ESL students
(Chularut & DeBacker, 2004)
Concept maps > Read C Discuss
(Guastello, Beasley, & Sinatra,
2000)
Interactive Lecture C Collaboratively creating No known studies Collaboratively building Collaborative build concept
concept map > Lecture for maps > individually map D Collaboratively build
science content assessment building (Czerniak & with 2 additional resources
(Czerniak & Haney, 1998) Haney, 1998; Okebukola & (van Boxtel, van Der Linden
Jegede, 1998) & Kanselaar, 2000)
232 CHI AND WYLIE
TABLE 5
Pairwise Comparisons Between Different Modes of Self-Explaining (SE)

Active SE worked example with menu > Study example No known
only (Conati & Van Lehn, 2000) studies
Constructive Explaining sentences > Repeating sentences No known Explaining own answers D Explaining
(O’Reilly et al., 1998) studies other’s answers (Pillow, Mash,
Rereading with SE > Rereading only (Griffin et al., Aloian, & Hill, 2002)
2008)
Prompted SE > Rereading twice (Chi et al., 1994)
Explaining others’ solution > Watching other solution
(Pine & Messer, 2000)
Studying incomplete examples > Studying completed
examples (Atkinson et al., 2003; Stark, 1999)
Interactive No known studies No known Explain other group members No known
studies contributions > Self-explain own studies

contributions in groups (Kramarski &
Dudai, 2009)
Explain with partner > Explain alone
(Hausmann et al., 2008)
Concept mapping. Concept maps are graphical repre- passively). In addition, there are studies that compare
sentations of knowledge where concepts are represented as constructive to active version of concept mapping. For
nodes and are connected through labeled relations. For example, Yin, Vanides, Ruiz-Primo, Ayala, and Shavel-
example, in a concept map of an ecosystem, two nodes may son (2005) found that students who were asked to gener-
be “wolves” and “deer” and may be connected by an arrow ate the labels for the links on a concept map
labeled “prey upon.” Creating such maps requires selecting (constructive) did better in total accuracy and map struc-
relevant concepts, arranging them hierarchically, and deter- ture complexity than students who were asked to select
mining the relationships between them. They are often used the links to complete the concept map (active).
in classrooms as aids to understanding lectures and texts Likewise, collaboratively building concept maps (inter-
(Coffey et al., 2003). More than 500 studies have investi- active) results in better learning than just listening to a lec-
gated the effectiveness and applications of concept map- ture without building any maps (passive; Czerniak &
ping (e.g., Novak, 1990; Novak & Gowin, 1984; Haney, 1998). Collaboratively building maps is also better
O’Donnell, Dansereau, & Hall, 2002). Two meta-analyses than individually creating maps, supporting our prediction
(Horton, McConney, Gallo, Woods, Senn, & Hamelin, that interactive is better than constructive (Czerniak &
1993; Nesbit & Adesope, 2006) have found a positive effect Haney, 1998; Okebukola & Jegede, 1988).
of concept mapping on knowledge acquisition when com- There are also studies that show that when concept map-
pared to other learning activities. ping is carried out at the same mode of engagement, the
Concept mapping can be implemented in a number of results show no differences in learning gains. For example,
ways. Students can be asked to copy a concept map (an van Boxtel, van der Linden, and Kanselaar (2000) imple-
active way of engaging), which has been shown to be mented two versions of interactive concept mapping. In
more effective for learning than reading a concept map one condition, after students received class instruction on a
(passive; Willerman & Mac Harg, 1991). Building or topic, they were asked to collaboratively create a concept
correcting a concept map can both be categorized as a map with no additional resources. In the second condition,
constructive activity because students must generate a after receiving instruction, they were asked to collabora-
map based on learning materials or revise a map while tively create a concept map with two additional textbook
correcting. Several studies have shown that generating chapters available. The results showed that students learned
or correcting concept maps is better than a variety of significantly in both conditions, but, as predicted by the
passive activities such as reading a text (Chang, Sung, & ICAP hypothesis, there were no differences between condi-
Chen, 2002), listening to a lecture (Schmid & Telaro, tions. These studies are shown in Table 4.
1990), or whole-class discussions (Chularut & DeBacker,
2004; Guastello, Beasley, & Sinatra, 2000; we classify Self-explaining. Self-explaining is the activity of
whole-class discussions as passive because the majority explaining an idea or concept aloud to oneself as one learns, or
of students in class discussions are participating one can think of it as trying to make sense of the learning
materials. Twenty years of research has consistently supported of a worked example by having students explain each
the finding that students learn better when they explain to them- example step to themselves, such as justifying how a solu-
selves the materials they are studying, called the self-explanation step follows from a prior step (Chi et al., 1989; Renkl
tion effect (Chi, 2000). This is because self-explaining helps et al., 1998). Other, more indirect constructive methods
students clarify an idea by elaborating on it, or inferring new include providing incomplete worked-out examples (Stark,
ideas. The self-explanation effect has been studied across age 1999), thus requiring students to infer the omitted steps (a
groups, domains, and instructional formats, such as reading a form of self-explaining), resulting in more transfer than
text passage or studying a worked-out example (Bielaczyc, simply studying completed worked-out examples (a more
Pirolli, & Brown, 1995; Chi et al., 1994; Ferguson-Hessler & passive mode of engagement). Incomplete examples can
de Jong, 1990; Hausmann & Chi, 2002; McEldoon, Durkin, & also involve fading (taking away) some of the example
Rittle-Johnson, 2013; Renkl, Stark, Gruber, & Mandl, 1998; steps, especially as the learner gets more proficient at solv-
Siegler, 1995; Wong, Lawson, & Keeves, 2002). ing problems (Atkinson, Renkl, & Merrill, 2003). Thus,
Self-explaining can be carried out in a conceptual fading provides an opportunity for the learner to generate
domain when students read a text passage, or it can be car- her own steps (more constructive) and this has been shown
ried out in a procedural domain when students study a to be more effective than not fading (more passive).
worked-out example. Worked-out examples are problem Self-explaining can be implemented as an active activity
statements with line-by-line solution steps; they can be by having students choose an explanation or elaboration
demonstrated by a teacher on a whiteboard or embedded in among a set of options. An example of an active way of
textbooks. In this section, we examine studies that involve self-explaining is a computer-based example-studying task
self-explanation using texts, worked-out examples, and that asks students to select a justification for each solution
other related contexts. step from a menu of choices (Conati & VanLehn, 2000).
Self-explaining is most often implemented in a construc- Such active ways of studying examples have been shown to
tive way because students are encouraged to make sense of be superior to a passive way, in which students simply stud-
the learning materials by generating inferences or by inte- ied examples without any prompts to justify or explain by
grating new knowledge with prior knowledge. For concep- choosing.
tual domains, many studies have compared self-explaining Kramarski and Dudai (2009) compared interactive ver-
implemented in a constructive way with other passive activ- sus constructive forms of self-explaining by training ninth-
ities. For example, O’Reilly, Symons, and MacLatchy-Gau- grade students to generate self-explanations to prompts
det (1998) tested college students’ recall and recognition such as, “What is my conclusion?” and “Is my explanation
ability for factual knowledge of the human circulatory sys- clear?” as they solved math problems in groups of four
tem when students were asked to either explain what the (constructive). In the interactive condition, students were
sentence means to them (i.e., what new information it pro- prompted with questions that focused on building on and
vides, and how it relates to what they already know, stan- responding to the other group members’ contributions, such
dard self-explaining constructive prompts) or repeat each as “How can I respond to my friend regarding the correct-
sentence on the computer screen until the next fact ness of his/her explanation?” and “How can I modify my
appeared (physically repeating is an active condition friend’s solution and explanation?” The interactive groups
because repeating causes more focused attention but does scored significantly higher in their mathematical accuracy
not generate new knowledge). As predicted, they found that and problem transfer scores than the self-explain groups. A
recall and recognition were significantly higher for the self- similar study had students studying worked-out examples
explanation group when compared to the repetition group. in the context of solving physics problems using an intelli-
Other studies comparing the same two conditions showed gent tutoring system. Students either worked alone at the
that self-explaining was better than reading the text twice computer to solve the problems and were prompted to
for both meta-comprehension (Griffin, Wiley, & Thiede, explain the solution steps presented on the screen (the con-
2008) and learning (Chi et al., 1994). structive condition) or worked in dyads at the computer to
Self-explaining can also be used in a broader context solve the problems and were prompted to generate joint
than reading from a text. Pine and Messer (2000) for exam- explanations to the solution steps (the interactive condi-
ple, asked children to “explain” how the instructor was able tion). The jointly explaining dyads outperformed the solo
to balance a beam (a constructive activity) compared with solvers on multiple measures: They answered faster, fin-
asking the children simply to “watch” (without making any ished more problems in the allotted time, entered more cor-
comments) the instructor demonstrate how to balance the rect entries, displayed a lower error rate, requested fewer
beam (passive). Children in the constructive “explain” con- hints when solving problems with the ITS, and so on (Haus-
dition improved in their ability to balance the beam signifi- mann, van de Sande, & VanLehn, 2008).
cantly more than children in the passive “watch” condition. With respect to comparing versions of self-explaining
For procedural domains, self-explaining has been tradi- implemented in the same mode, Pillow, Mash, Aloian, and
tionally implemented in a constructive way in the context Hill (2002) asked 4- to 5-year-old children to predict
234 CHI AND WYLIE
misinterpretation of ambiguous pictures. In one condition, promotes better learning than having students participate in
the children were prompted to explain their own misinter- a whole-class discussion (passive; Chularut & DeBacker,
pretations (explain-own condition is constructive), and in 2004; Guastello et al., 2000). As mentioned earlier, we
the other condition they were asked to explain the misinter- assume that a whole-class discussion is constructive only
pretations of a puppet viewing similar pictures (explain- for the few students who are involved in the discussion. For
puppet condition is also constructive). Both conditions are the majority of students who are silently listening to the dis-
clearly constructive, and neither condition received feed- cussion, the activity is passive.
back. As predicted by ICAP, there were no significant dif- A classroom study by Hendricks (2001) found that
ferences between the two groups in their final scores on the discussing cause–effect relations with a peer is superior
posttests. These studies are shown in Table 5. For more to observing the teacher identify cause–effect relations.
detailed analyses (with effect sizes) and comparisons of Again, this is consistent with ICAP’s prediction because
self-explaining implemented in different modes, see Fon- discussing cause–effect relations with a peer is an inter-
seca and Chi (2011). active activity, whereas watching the teacher identify
cause–effect relations is only passive. Similarly, if stu-
Classroom Studies dents are interacting in cooperative groups (interactive),
this leads to greater learning than listening to a tradi-

There are also many classroom studies whose interventions tional lecture (passive; Ebert-May, Brewer, & Allred,
can be reinterpreted using the ICAP framework. These 1997).
studies are summarized in Table 6, with the cells indicating Two classroom studies have contrasted an interactive
the modes involved in the pairwise contrasts. For example, mode with an active mode. One study found that learning
a classroom study comparing students who took notes using using the jigsaw strategy (in which each student within a
a partial, scaffolded outline to those who were given a group develops expertise on a subtopic and explains it to
handout with an almost verbatim transcript of the lecture the other students within his or her group, thus being inter-
showed that students who used the partial outline to take active) promotes more learning than classrooms where indi-
notes did significantly better on both immediate and reten- vidual students gather information for themselves (active;
tion tests than those who were given the comprehensive Doymus, 2008). Another study found peer tutoring (inter-
handout (Russell, Caris, Harris, & Hendricson, 1983). This active) to be better than filling out guided notes (active;
finding is predicted by the ICAP hypothesis in that taking Mastropieri, Scruggs, Spencer, & Fontana, 2003). Like-
notes with a partial outline is more comparable to an active wise, an interactive self-explaining group tends to per-
task because students must copy additional notes from the formed better than a constructive self-explain-only group.
lecture to complete the handout, whereas having a complete In Kramarski and Dudai’s (2009) study, students who self-
transcript of the lecture requires no copying and reviewing explained and received feedback from each other (thus
it is thus passive. Other studies have compared and found interactive) performed better than students who self-
that having students build concept maps (constructive) explained but did not receive feedback (constructive only).
TABLE 6
Pairwise Contrasts of Classroom Studies Involving Comparisons Between a Mix of Passive, Active, Constructive, and Interactive Engagement
Activities
Passive
Active Taking scaffolded notes > Comprehensive
handout (Russell et al., 1983)
Constructive Building concept maps > Whole class Guided prompts C Solve > Solve (Whitten
discussions (Chularut & DeBacker, 2004; & Rabinowitz, 2010)
Guastello, Beasley, & Sinatra, 2000) Compare C Contrast C Write C Solve >
Solve (Docktor et al., 2010)
Interactive Cooperative groups > Traditional lecture Peer tutoring > Filling out guided notes SE C Receive Feedback > Self-
(Ebert-May, et al., 1997) (Mastropieri et al., 2003) explanation without feedback
Student–student discussion C small-group Jigsaw groups > Individuals gathering (Kramarski & Dudai, 2009)
activities C feedback > Traditional lecture information (Doymus, 2008)
(Deslauriers, Schelew, Wieman, 2011) Generating with partner > Selecting with
Discussing cause-effect with peer partner (Zheng & Linn, 2013)
> Observing teacher identify cause-effect
relationships (Hendricks, 2001)
There are also a number of classroom studies comparing classroom intervention studies. In sum, the empirical evi-
constructive learning activities with active activities. For dence from both laboratory and classroom studies provide
example, two studies compared problem-solving alone support for the ICAP hypothesis.
(assuming this tends to be an active activity when solving
in a plug-and-chug way) to solving problems with an added
constructive activity (such as compare and contrast, Dock- SECTION 3: CAVEATS, NEW TOOL, AND
tor, Strand, Mestre, & Ross, 2010; and responding to guided COMPARISONS WITH OTHER THEORIES
prompts, Whitten, 2011). The results showed that adding
the constructive activity improves learning. In this third major section, we consider several caveats and
Classroom studies also tend to include multiple activities factors that need to be addressed. We also entertain ways
as one intervention. For the purpose here, we classify the that ICAP can serve as a tool for (a) explaining discrepant
intervention according to the highest activity mode. For findings, (b) determining proper control conditions for
example, in the study by Deslauriers, Schelew, and Wieman experimental studies, (c) developing rubrics to code learn-
(2011), they compared traditional listening to lecture ing data, and (d) guiding instructional design. Finally, we
(passive) to an intervention consisting of predicting, reason- briefly compare ICAP with other learning theories.
ing, solving, critiquing, and peer discussion. We classify

this intervention as an interactive one because it includes
peer-to-peer discussion. As predicted, the intervention pro- Factors and Considerations That May Override
moted more learning than listening to traditional lectures. the ICAP Hypothesis
Classroom studies also illustrate a key point of the ICAP
framework, namely, simply asking students to work The ICAP taxonomy of a learner’s overt behaviors is a
together does not automatically make an activity interac- gross first cut at predicting learning outcomes as a function
tive. As mentioned in the first section, to be interactive, stu- of what students are asked to do. However, our analyses
dents must work together on a constructive task. In a and predictions are based on only one aspect of a learning
classroom study, Zhang and Linn (2013) compared the situation, the aspect of what students do to engage with the
effects of two conditions on learning chemical reactions. In materials. Clearly there are alternative factors that influence
the first condition, students worked in pairs to draw mole- how well students learn. In this section, we consider three
cules linking the atoms with lines (drawing is a generative factors that can override the predictions of the ICAP
activity). In the second condition, pairs of students physi- hypothesis.
cally selected static images that represented the correct
molecular structure. Even though both conditions involved Assessment measures. An important factor that
working with a partner, only the generate condition can be may lead to unexpected results according to the ICAP
considered interactive according to the subsumptive nature hypothesis is the type of measures used to assess learning.
of the ICAP framework; that is, interactive activities In general, performance on shallow or easy assessment
require partners to be constructive and go beyond the pro- questions typically is not sensitive to various intervention
vided material. Thus, this comparison can be reinterpreted efforts, and this is true for ICAP activities as well. This is
as comparing interactive to active and, as expected by the because shallow questions that require only recall of infor-
ICAP hypothesis, the interactive (drawing in pairs) condi- mation can be answered with passive receiving type of
tion showed greater learning gains than the active (select in activities. For example, Aleven and Koedinger (2002) com-
pairs) condition. pared students who were prompted to explain in addition to
solving geometry problems to those who solved the same
Summary problems without prompts to explain. They found no
advantage for explanation on easy-to-guess items but sig-
This section reviewed studies in the literature to see if their nificant benefits of explanation for hard-to-guess and trans-
results validate the predictions of the ICAP hypothesis. fer problems. Similarly, for the study (shown in Figure 2)
First, we reported results from a study conducted in our lab in which we manipulated all four conditions within the
that compared all four modes and identified two studies in same study (Menekse et al., 2013), we found that perfor-
the literature that contained three conditions that can be mance on easy multiple-choice questions was the same for
reinterpreted using ICAP. Results from all three studies val- all levels of engagement activities, but condition differen-
idate our predictions. Second, we showed that findings from ces emerged on more challenging assessment items.
studies that compared the same engagement activity imple- Thus, when the learning results are nonsignificant, one
mented in different modes also support the ICAP hypothe- problem may be that the assessment measures are too shal-
sis; that is, depending on how (or which mode) an activity low; thus, they are not sensitive enough to detect differen-
is implemented influenced the learning outcomes. Third, ces in learning outcomes as a function of the level of
we validated the prediction of the ICAP hypothesis in student engagement.
236 CHI AND WYLIE
Domain. A second factor that can lead to violations of Task differences within a mode. The four engage-
the ICAP hypothesis is the topic domain. Although rare, ment modes are based on a student’s overt behavior. Typi-
not all topics or skills necessarily show increased learning cally the activities within each mode are distinct, so that
benefits with increased cognitive engagement. We next pro- pairwise comparisons of two modes usually compare two
vide two examples and an explanation for why this might different activities (as was done in Chi, 2009) or the same
be the case. activity implemented in two different ways (resulting in dif-
One type of topic for which increased cognitive engage- ferent modes), as shown in Tables 3, 4, and 5. In our analy-
ment may not be helpful is simple procedural domains for ses of the studies in Tables 3 to 5, we have shown that the
which the rules are arbitrary and cannot be logically same task, implemented with variations that do not change
deduced. For example, the English article system (teaching its mode, produces equivalent learning outcomes.
students when to use a, an, the, or no article) is governed Although it is often the case that when two activities fall
by a complicated set of human-made rules and exceptions within a single cell (i.e., use the same mode of engage-
exist. In a computer-based tutoring system developed to ment), nonsignificant learning differences are obtained;
teach students how to select the correct article, students however, ICAP cannot always make accurate predictions
were asked to select either the correct article or the article about learning outcomes from undertaking two different
and the rule (explanation) for why that article was correct engagement activities that can be classified into the same
(e.g., “The noun is made definite by a prepositional mode. This is because a number of factors can affect the
phrase”). There were no learning differences between the complexity of different tasks or activities within a mode
two conditions (Wylie, 2011). (e.g., the goal or knowledge to-be-learned of a task, the
There are two ways to interpret this result. The simple time it takes to carry out the task, the procedure or process
one is that both are active selection tasks, so one would not by which a task is carried out, etc.). We illustrate next with
expect a learning difference. A second interpretation is also an example.
possible, if we assume that selecting an explanation is The task of interacting with a peer to invent and the task
closer to being constructive than selecting the answer alone, of comparing and contrasting two cases while working in
thus making the result counterintuitive. This second inter- pairs are both interactive activities because they require stu-
pretation suggests that self-explaining article usage may dents to produce new knowledge: either inventing a rule or
not be a useful way to learn about the appropriate article to method in the invent case or come up with similarities and
use because there is no way to rationalize or work through differences in the compare-and-contrast case. These two
the reasons for when a particular article is used. In contrast, conditions were manipulated by Chi, Dohmen, Shemwell,
in the many examples of effective self-explanation dis- Chase, and Schwartz (2012) when they asked sixth-grade
cussed earlier, it makes sense to ask a why or a how ques- students to learn about horizontal projectile motion. In the
tion, because presumably with more elaborations and invent condition, students were asked to work in pairs to
inferences, one can generate a reason. The same may be invent a single method to figure out where to put the final
true for topics that are so challenging that students cannot destination as a result of a set of projectile trajectories that
bootstrap themselves by self-explaining into deeper or were shot at different speeds and from different heights. In
more correct understand because they lack the relevant the compare–contrast condition, students were asked to
schema to interpret the new information correctly. One work in pairs to compare and contrast sets of the same pro-
such example in the science domain is the concept of jectile trajectories. Students in the invent condition showed
emergence, because emergence is a novel idea that many greater learning gains compared to students in the com-
students do not have relevant or related knowledge to pare–contrast condition, even though both conditions are
understand. Therefore, constructive activities, such as interactive.
answering prompted questions, are able to help students A finer grained analysis of the task, however, suggests
learn only the basic detailed information about the microle- why the two interactive activities result in different learning
vel behavior of the molecules and not the interlevel causal gains. An analysis of the worksheets shows that the two
relationships of how behaviors of the molecules at the engagement activities cause different ways of processing
microlevel collectively explain the behavior at the the cases. In the compare–contrast condition, students paid
macrolevel pattern (Chi, Roscoe, Slotta, Roy & Chase, attention to single factor or feature (e.g., speed, height,
2012). This is because this collective interlevel relation- etc.). This is revealed in comments such as, They go in dif-
ship is an emergent one, which is foreign to students ferent speeds or They start at different heights. As coded by
who are used to sequential causal chain relationships. the authors, a double feature comment would involve two
Thus, for domains or topics in which students cannot features, such as more speed D further distance, and a triple
bootstrap themselves into deeper understanding due to a statement would relate three features. The worksheet analy-
lack of relevant schemas, or for topics for which no ses showed that only 10% of the compare–contrast group
deeper rationales exist, the predictions made by ICAP produced triple-feature statements, whereas 100% of the
may not hold. invent group did. Conversely, the compare–contrast group
produced on average 3.0 single-feature statements, whereas questions is a constructive way of summarizing, and they
the invention groups produced no single-feature statements. compared that condition to a passive task of reading out
In short, the activity of making comparisons across two loud. Thus, it is not surprising that learning is better in the
cases biased the students to compare features individually, constructive summarizing case (Mastropieri et al., 2001).
whereas the invent activity encouraged students to make On the other hand, Coleman et al. compared summarizing
more holistic statements and connect features with each by selecting only the most important sentences, an active
other, which is necessary in order to invent the correct task, to summarizing by explaining, which is a constructive
method. task. Therefore, it is not surprising that summarizing
This study demonstrates that although both learning through selection (active) is worse for learning than sum-
activities are interactive, their differing task structures marizing through explanation (constructive; Coleman et al.,
biased the way students performed them, resulting in differ- 1997). Thus, an activity such as summarizing cannot be
ent qualitative ways of learning, which leads to different claimed as beneficial in an absolute way; its utility for
amounts of learning. Notice, however, that the ICAP learning depends on how it is implemented and to what
hypothesis focuses its predictions across modalities, which appears activity it is compared. In short, what appears to be
may not be violated in this case, in that both inventing and discrepant results in the literature is completely systematic
comparing and contrasting (both interactive) should pro- from ICAP’s point of view.
duce greater learning outcomes than an active or passive As the preceding sections show, in scrutinizing such
activity. studies, it is apparent that whether note taking is an effec-
In summary, it is possible that different levels of learn- tive learning activity depends on three factors: (a) how it is
ing outcomes are achieved for activities within the same implemented, that is, what are the instructions/directions
mode of engagement. One explanation is that some activi- given to the student, thus making it an active, constructive,
ties are more cognitively demanding than other activities, or interactive activity; (b) what the alternative (or often the
even if they both belong to the same mode. This is because control) condition it is contrasted with (i.e., if it was imple-
overt behavior is a gross analysis of engagement behavior mented as a constructive activity, then was it compared
and not a cognitive analysis of a given task activity. How- with another passive, active, or constructive activity); and
ever, even so, the prediction may still hold across modes of (c) how students actually carried out the activity. In short,
activities. ICAP can highlight the systematicity that underlies appar-
ent contradictions between studies.
ICAP Can Further Serve as a Tool to Explain, Dictate,
Evaluate, and Design Using ICAP to dictate the choice of a control condi-
tion in research design. Besides explaining discrepant
In this section, we describe the contribution of the ICAP results in the literature, ICAP can also be used as a tool for
framework and hypothesis as a tool for both research pur- determining areas that are understudied and, most impor-
poses and instructional design. tantly, for deciding control conditions. As Tables 3 to 5
show, the majority of the studies on interventions compared
Using ICAP to explain discrepant findings about a constructive mode with a passive mode. Far fewer studies
learning activities. In a preceding section of this article, have compared the constructive mode with active, interac-
we stated that one way of validating the ICAP hypothesis tive, or other constructive modes.
was to examine studies using three specific learning activi- With respect to choosing meaningful control conditions,
ties (note taking, concept mapping, and self-explaining) the most stringent comparison of an intervention is to com-
when each activity was implemented according to different pare it with another intervention that employs an engagement
modes of engagement. Under its lens, we can provide an activity of the same mode. For example, to claim that insert-
interpretation for seemingly discrepant findings. By dis- ing a metacognitive prompt is a useful intervention (Beal &
crepant findings, we mean that for a given learning activity, Stevens, 2010), one needs to compare it with another type of
such as note taking, some research suggests that it is helpful prompt that also elicits generative responses so that both
for learning, whereas others report that it is not. To illus- activities are constructive, rather than comparing it with
trate, consider the following two studies. In one study (Mas- reading academic advice that does not elicit a response. Only
tropieri et al., 2001), the results showed that note taking by by comparing two interventions that fit in the same mode of
answering who–what questions was beneficial, whereas engagement can one draw legitimate conclusions about the
Coleman et al. (1997) concluded that note taking by sum- effectiveness of a specific type of intervention.
marizing was not as beneficial. Using the ICAP framework, That is, suppose one finds that Intervention A is better
we can resolve these discrepant findings by first categoriz- than Intervention B. Before claiming that it is the specific
ing the learning activity and then comparing it to the processes behind Intervention A that caused the learning
alternative condition used in each study for comparison. In gains, it is important to examine the engagement mode of
the Mastropieri et al. publication, answering who–what the two interventions. For example, if Intervention A is
238 CHI AND WYLIE
constructive and Intervention B is active than the learning constitutes an “active learning” activity? Moreover, how do
benefits may not be due to the specific nature of Interven- teachers decide which activity is relatively better for foster-
tion A but may only lead to a more general conclusion that ing student learning? For example, is think-pair-share better
Intervention A is more cognitively engaging than Interven- than developing critiques? Finally, research provides few
tion B. guidelines for teachers to tell them how to modify their cur-
rent homework and seatwork assignments to encourage stu-
Using ICAP to evaluate students’ outputs. So far, dents to learn more effectively by being active.
we have discussed only the mode of an activity, as intended Our framework, on the other hand, can provide specific
by the designer of the activity, such as a teacher. For exam- guidelines for how to create lessons that incorporate overt
ple, if a teacher asks students to draw concept maps, then behaviors that are associated with higher levels of engage-
presumably this is a constructive task. However, in many ment and their associated knowledge-change processes. We
cases, this output (the concept maps) needs to be evaluated have evaluated whether the ICAP framework and ICAP
to determine whether in fact the students treated it as a con- hypothesis are beneficial for professional development by
structive task. This is the distinction mentioned earlier conducting several small-scale training workshops in which
between intended versus enacted. For example, suppose in we introduced teachers to the framework and evaluated
such an activity, the teacher populated or provided a set of whether they could successfully use the framework when
six concepts as starting concepts to be used as “nodes” in designing lessons. Initial results are promising and show
the concept map. Upon evaluating or coding of the that teachers are able to both understand the framework and
students’ maps, it becomes apparent that some students cre- correctly apply it when designing new lessons.
ated concept maps that only contained the six concepts that We have also developed an online module containing
were provided, whereas other students created maps that information about ICAP. Specifically, the module can scaf-
had many other concept nodes. This suggests that the stu- fold teachers during the design process and encourage them
dents who used only the six provided nodes did not create to consider not only the specific task but how to design bet-
any new outputs beyond what was provided (thus they were ter assessments that measure deep knowledge as well. For
only active), whereas the other students who produced addi- example, a teacher who normally delivers instruction via a
tional concept nodes were more constructive. Thus, the passive mode such as lecturing will learn from the module
ICAP framework can be used to design a rubric to code and to provide students with a guided notes worksheet thus cre-
evaluate students’ outputs. And often an accurate classifica- ating an active activity, or the teacher may learn to create a
tion of an activity’s mode can only be done after such constructive activity by prompting students to build a con-
evaluations. cept map during the lecture. Similarly, a teacher who previ-
ously considered an interactive activity to be any task a
Using ICAP as a guide for instructional design. In student did not complete alone could learn from the module
addition to the research contributions, the ICAP framework to ensure that interactive activities not only involve work-
has strong practical implications as teachers and other ing in groups but also require students to be constructive or
instructional designers can use it to choose, modify, or generative while completing the tasks.
design tasks for students to perform. Thus, even though With the explosion of computer-based instruction, this
engagement behaviors are strictly defined from the same set of challenges of how to design student activities also
learners’ perspective, clearly students can be encouraged to confronts designers of computer-based learning environ-
engage in certain ways through the careful design of learn- ments. Because some forms of activities (such as selecting an
ing activities. answer from a menu of choices) are easier to implement in a
In the current depictions in the literature, even though computer-based learning environment than others (such as
“active learning” (along with the related concept generating a free-form response that needs to be verified),
“constructive learning”; Cobb, 1994; Jonassen, 1991; designers of learning environments also need clear-cut rec-
Pelech, 2010; Wilson, 1997) may be ambiguous, “active ommendations about the trade-offs between the cost of imple-
learning” does have instructional implications. It suggests mentation and effectiveness of students’ activities for
that teachers can “encourage the learner to engage in mak- learning. In this article, even though we focus primarily on
ing sense of the material” (Mayer, 2008, p. 17). But are teachers’ practices, our work has direct implications for
there concrete methods recommended for teachers to adopt designers of all learning environments.
and adapt? King (1993) explicitly offered 12 getting
involved and really thinking about it activities for “active Comparison With Other Theories
learning” (e.g., posing problems, thinking analogically,
developing critiques, etc.). However, these 12 examples do To summarize, the ICAP framework consists of four modes
not specify what qualifies exactly as an “active learning” of overt engagement activities, passive, active, construc-
activity. And if a teacher chooses not to use one of the 12 tive, and interactive, the associated knowledge-change pro-
examples, how would she know what alternative activity cesses for each mode, the associated resulting changes in
knowledge, with the concomitant improvement in learning additional inferences, and so on, in order to understand
as a function of the knowledge changes and cognitive out- what having a double loop in circulation might mean. Con-
comes. In other words, the ICAP framework lays out a pos- structing in ICAP does not mean that the learner discovers
sible causal chain, leading from engagement activities to the fact that “The heart has a double loop in circulation.”
the knowledge-change processes underlying the engage- Constructing in ICAP is a means to achieve or learn with
ment activities, resulting in changes in knowledge itself understanding.
and the use of that changed knowledge. We further clarified
the ICAP hypothesis and described the circumstances under Cognitive load theory. In cognitive psychology, cog-
which it can be violated. In addition, we showed how it can nitive load refers to the load imposed on working memory
explain discrepant findings, help determine appropriate from an information-processing framework. Because
control conditions, and improve teachers’ design of learn- humans have a limited working memory capacity, the
ing activities. amount of information that can be processed concurrently
In general, ICAP is a theory of cognitive engagement during complex learning activities can overload the finite
with a behavioral metric. However, two caveats must be amount of working memory capacity. Cognitive load the-
stated. First, the knowledge-change processes associated ory provides explicit empirically based guidelines for the
with each mode of engagement are hypothetical. We have best ways to design instructional materials, with the goal
not carried out studies to verify that these knowledge-change of decreasing extraneous cognitive load during learning.
processes are in fact taking place when students engage in For example, cognitive load theory suggests integrating
one mode over another. The second caveat is that there may diagrams and text information rather than presenting them
be other theoretical interpretations of ICAP, viewing cogni- separately (i.e., split attention effect). In general, cognitive
tion that relies much less on representations and memory. load theory suggests that instruction should be designed to
However, we cannot derive an ICAP hypothesis from such focus the learner’s attention toward the germane materials
an alternative view, other than the behavioral view, based on (Sweller, 1994).
the hierarchical nature of the overt activities of each mode. The obvious difference between cognitive load theory
Thus, we welcome any theoretical lens that can derive the and the ICAP framework is that ICAP is concerned with
ICAP hypothesis. In this final section, we highlight how eliciting higher modes of activities from learners, instead of
ICAP differs from existing learning theories (constructivism, changing and reducing the load imposed by an activity. So
cognitive load theory, and Bloom’s revised taxonomy), in the difference is one of focusing on the learners’ activities
terms of its theoretical stance and interpretation. versus the load of instructional activities. Moreover,
ICAP’s predictions seem to be the opposite of predictions
Constructivism. Constructivism is a broad framework from the load theory. For example, cognitive load theory
that assumes that the responsibility of learning should states that the greater the load in the to-be-processed pre-
reside increasingly with the learner (Von Glasersfeld, sented materials, the more difficult a task becomes, and
1989). Moreover, constructivism emphasizes the impor- therefore the less resulting learning. However, ICAP makes
tance of the learner being actively involved in the learning the opposite prediction. That is, as the mode of engagement
process, unlike previous educational viewpoints where the goes from passive to interactive, the activity becomes more
responsibility rested with the instructor to teach and where effortful (i.e., imposes more load). Although we can infer
the learner played a passive, receptive role. Thus ICAP is that this is true by the hypothetical knowledge-change pro-
very similar to constructivism in this sense of focusing cesses involved, direct evidence can be gathered from ask-
more on the actions of the learner than the instructor. Con- ing students in a self-report. McEldoon (2014) asked
structivism translates to instruction by encouraging a vari- students to rate their perceived cognitive load (or working
ety of learner-controlled or learner-centered activities such memory demand) after completing a worked example that
as discovery, hands-on, experiential, collaborative, project- included instructional explanations of calculating an analy-
based, and task-based learning. These activities have ana- sis of variance. One condition completed it actively by
logs in the active, constructive, and interactive modes of determining final values within the worked example and
learning. So the basic difference between constructivism copying or paraphrasing instructional explanations. The
and the ICAP framework is that ICAP differentiates in a other condition completed the worked example construc-
more concrete and fine-grained way (both in terms of the tively by generating intermediate values and generating
behaviors and in terms of the learning outcomes) the activi- self-explanations prior to receiving instructional explana-
ties that constructivism has promoted. Furthermore, per- tions. She found that students in the constructive condition
haps constructivism is misinterpreted, but the notion of reported significantly higher levels of cognitive load or
“constructing an understanding,” let’s say of the sentence working memory demand than those in the active condi-
“The heart has a double loop in circulation,” is interpreted tion. For one of the lessons, the constructive condition
in ICAP as the way (if doing it constructively) to integrate also had significantly higher learning gains. Thus, with
that sentence with one’s prior knowledge, making respect to cognitive engagement, ICAP may make the
240 CHI AND WYLIE
opposite prediction from cognitive load theory, in that the processes, however, refer to the processes of learning or
more load or effort is used to process learning materials, what is referred to here as knowledge-change processes.
the more learning is achieved. However, at times ICAP Third, ICAP proposes that co-construction through interac-
and cognitive load theory predict congruent results; tion can achieve the greatest learning outcomes, whereas
namely, if increased engagement corresponds to increased Bloom’s taxonomy does not distinguish between intraper-
germane cognitive load (e.g., when students are asked to sonal and interpersonal cognition and, hence, his learning
generate self-explanations, they are being constructive, outcomes can be achieved at all levels without interaction
according to ICAP, and by explaining they are also with others.
increasing germane load).
Bloom’ s taxonomy. It is a daunting task to compare CONCLUSION

and contrast Bloom’s taxonomy with ICAP. We are basing
our comparison with Bloom’s taxonomy as described in the The ICAP framework and hypothesis provide specific,
2001 revised version (Anderson & Krathwohl, 2001). The operationalized definitions of engagement activities that
goal of Bloom’s taxonomy is to first classify learning objec- can easily be applied to a number of learning environments.
tives and then to design instructional activities and assess- The ICAP hypothesis predicts that as activities move from
ment that align with the objectives. A teacher’s learning passive to active to constructive to interactive, students
objectives must first be classified into one of six categories undergo different knowledge-change processes and, as a
of cognitive processes: remember, understand, apply, ana- result, learning will increase. This hypothesis has been vali-
lyze, evaluate, and create. Suppose an instructional objec- dated through a number of classroom and laboratory stud-
tive is to “differentiate between rational numbers and ies, and in this article, we examined its validity in three
irrational numbers.” “Differentiate” or “distinguish specific tasks: note taking, self-explaining, and creating
between” are assumed to require the cognitive processes of concept maps. In addition, we use the framework to clarify
analyze, according to Anderson and Krathwohl (2001), discrepant findings in the literature, propose a number of
because analyze “involves breaking material into its con- useful applications of the framework, and compare and con-
stituent parts and determining how the parts are related to trast ICAP to existing cognitive and learning theories.
one another and to an overall structure” (p. 79). Once Essentially, ICAP is a hypothesis about the relative level of
teachers have classified the learning objective as analyze, learning associated with each of these four modes of stu-
they must design instruction that requires students to ana- dent engagement, with the advantage being that it can
lyze, such as use examples and non-examples to help stu- detect regularity across a large corpus of data in the litera-
dents form the proper categories of rational and irrational ture and can be used to inform both classroom and labora-
numbers. In addition, assessment items must be designed to tory studies.
align with the objective of analyze, such as designing a test
question that asks students to classify each number on a list
of real numbers as either a rational or an irrational number. ACKNOWLEDGMENTS
The major characteristic difference between Bloom’s
taxonomy and the ICAP taxonomy is that Bloom’s taxon- Comments and edits from the reviewers and Clark Chinn
omy focuses its users on their instructional goals and how are greatly appreciated. Special thanks also go to Bryan
to measure whether the goal has been achieved, whereas Henderson, Katherine McEldoon, Glenda Stump, and Kurt
ICAP focuses its users on the means for achieving the VanLehn for their comments.
instructional goals. Because one framework focuses on
ends and the other on means, the two frameworks are com-
plementary. Minor differences also exist. First, ICAP is FUNDING
more parsimonious in that the framework applies to learn-
ing of various forms of knowledge, whether it is factual, The authors are grateful for support from the Institute of
conceptual, or procedural; and the categories of learner Education Sciences (Award #R305A110090) for the project
engagement activities can be easily distinguished by com- Developing Guidelines for Optimizing Levels of Students’
paring information generated by the learner(s) vis-a-vis the Overt Engagement Activities.
information provided by the learning environment. Second,
Bloom’s cognitive processes refer to the processes
involved in carrying out the assessment task, such as the
REFERENCES
task of analyzing that requires decomposing into constitu-
ent units, similar to various task-specific cognitive pro- Aleven, V., & Koedinger, K. (2002). An effective metacognitive strategy:
cesses underlying other problem-solving tasks and Learning by doing and explaining with a computer-based cognitive
activities mentioned in this article. ICAP’s cognitive tutor. Cognitive Science, 26, 147–179.
Alibali, M., & DiRusso, A. (1999). The function of gesture in learning to Chi, M. T. H. (2009). Active-Constructive-Interactive: A conceptual
count: More than keeping track. Cognitive Development, 14, 37–56. framework for differentiating learning activities. Topics in Cognitive
Anderson, L. W., & Krathwohl, D. R. (2005). A taxonomy for learning, Science, 1, 73–105.
teaching, and assessing. London, England: Longman. Chi, M. T. H., Bassok, M., Lewis, M. W., Reimann, P., & Glaser, R.
Atkinson, R. K., Renkl, A., & Merrill, M. M. (2003). Transitioning from (1989). Self–explanations: How students study and use examples in
studying examples to solving problems: Effects of self-explanation learning to solve problems. Cognitive Science, 13, 145–182.
prompts and fading worked-out steps. Journal of Educational Psychol- Chi, M. T. H., de Leeuw, N., Chiu, M. H. & Lavancher, C. (1994). Elicit-
ogy, 95, 774–783. ing self-explanations improves understanding. Cognitive Science, 18,
Azevedo, R., Greene, J. A., Moos, D. C., Winters, F. I., Cromley, J. G., & 439–477.
Godbole-Chaudhuri, P. (2006). Is externally-regulated learning by a Chi, M. T. H., & Kang, S. (2014). The learning benefits of dialogue videos
human tutor always effective in facilitating learning with hypermedia? over monologue videos: Implications for designing online videos. Manu-
In S. Barab, K. Hay, & D. Hickey (Eds.), Proceedings of the 7th Interna- script submitted for publication.
tional Conference on Learning Sciences (pp. 16–22). Mahwah, NJ: Chi, M. T. H., & Menekse, M. (in press). Dialogues as interactions. In
Erlbaum. C. Asterhan & L. Resnick (Eds.), Socializing intelligence through
Barron, B. (2003). When smart groups fail. Journal of the Learning Scien- academic talk and dialogue. Washington, DC: AERA.
ces, 12, 307–359. Chi, M. T. H., Roscoe, R. D., Slotta, J. D., Roy, M., & Chase, C. C. (2012).
Bartlett, F. C. (1958). Thinking: An experimental and social study. London, Misconceived causal explanations for emergent processes. Cognitive
England: Allen & Unwin. Science, 36, 1–61.
Bauer, A., & Koedinger, K. R. (2007). Selection-based note-taking appli- Chi, M. T. H., Roy, M., & Hausmann, R. G. M. (2008). Observing tutorial
cations. CHI 07 Proceedings of the SIGCHI Conference on Human Fac- dialogues collaboratively: Insights about human tutoring effectiveness
tors in Computing Systems. New York, NY: ACM. from vicarious learning. Cognitive Science, 32, 301–341.
Beal, C. R., & Stevens, R. H. (2010). Improving students’ problem Chi, M. T. H., Siler, S. A., Jeong, H., Yamauchi, T., & Hausmann, R. G.
solving in a web-based chemistry simulation through embedded (2001). Learning from human tutoring. Cognitive Science, 25, 471–533.
metacognitive messages. Technology, Instruction, Cognition & Chinn, C. A., & Malhotra, B. A. (2002). Epistemologically authentic
Learning, 8, 255–271. inquiry in schools: A theoretical framework for evaluating inquiry tasks.
Bielaczyc, K., Pirolli, P. L., & Brown, A. L. (1995). Training in self-expla- Science Education, 86, 175–218.
nation and self-regulation strategies: Investigating the effects of knowl- Chularut, P., & DeBacker, T. K. (2004). The influence of concept mapping
edge acquisition activities on problem solving. Cognition and on achievement, self-regulation, and selfefficacy in students of English
Instruction, 13, 221–252. as a second language. Contemporary Educational Psychology, 29,
Biswas, G., Leelawong, K., Schwartz, D., & Vye, N. (2005). Learning by 248–263.
teaching a new agent paradigm for education software. Applied Artificial Cobb, P. (Ed.). (1994). Learning mathematics: Constructivist and interac-
Intelligence, 19, 363–392. tionist theories of mathematical development. Dordrecht, The Nether-
Blumenfeld, P. C., Kempler, T. M., & Krajcik, J. S. (2004). Motivation and lands: Kluwer Academic.
cognitive engagement in learning environments. In R. K. Sawyer (Ed.), Coffey, J. W., Carnot, M. J., Feltovich, P. J., Feltovich, J., Hoffman, R. R.,
The Cambridge handbook of the learning sciences (pp. 475–488). New & Ca~nas, A. J. (2003). A summary of literature pertaining to the use of
York, NY: Cambridge University Press. concept mapping techniques and technologies for education and perfor-
Bodemer, D., Ploetzner, R., Feuerlein, I., & Spada, H. (2004). The active mance support (Tech. rep. submitted to the US Navy Chief of Naval
integration of information during learning with dynamic and interactive Education and Training). Pensacola, FL: Institute for Human and
visualisations. Learning and Instruction, 14, 325–341. Machine Cognition.
Bonwell, C. C., & Eison, J. A. (1991). Active learning: Creating excite- Coleman, E. B., Brown, A. L., & Rivkin, I. D. (1997). The effect of
ment in the classroom. Washington, DC: School of Education and instructional explanations on learning from scientific texts. Journal of
Human Development, George Washington University. the Learning Sciences, 6, 347–365.
Bonwell, C. C., & Sutherland, T. E. (1996). The active learning continuum: Conati, C., & Carenini, G. (2001). Knowledge encapsulation: Studies on
Choosing activities to engage students in the classroom. New Directions the development of medical expertise. In 17th International Joint Con-
for Teaching and Learning, 1996(67), 3–16. ference on artificial intelligence (pp. 1301–1306). San Francisco, CA:
Brown, A. L., & Palincsar, A. S. (1989). Guided, cooperative learning and Morgan Kaufmann.
individual knowledge acquisition. In L. Resnick (Ed.), Knowing, learn- Conati, C., & VanLehn, K. (2000). Toward computer-based support of
ing, and instruction: Essays in honor of Robert Glaser (pp. 393–451). meta-cognitive skills: A computational framework to coach self-expla-
Hillsdale, NJ: Erlbaum. nation. International Journal of Artificial Intelligence in Education, 11,
Bruner, J. S. (1961). The act of discovery. Harvard Educational Review, 389–415.
31, 21–32. Corno, L., & Mandinach, E. (1983). The role of cognitive engagement in
Butcher, K. R. (2006). Learning from text with diagrams: Promoting men- classroom learning and motivation. Educational Psychologist, 18, 88–108.
tal model development and inference generation. Journal of Educational Czerniak, C. M., & Haney, J. J. (1998). The effect of collaborative concept
Psychology, 98, 182–197. mapping on elementary preservice teachers’ anxiety, efficacy, and
Chang, K.-E., Sung, Y.-T., & Chen, I.-D. (2002). The effect of concept achievement in physical science. Journal of Science Teacher Education,
mapping to enhance text comprehension and summarization. The Jour- 9, 303–320.
nal of Experimental Education, 71, 5–23. Damon, W. (1984). Peer education: The untapped potential. Journal of
Chi, M., Dohmen, I., Shemwell, J. T., Chase, C. C., & Schwartz, D. L. Applied Developmental Psychology, 5, 331–343.
(2012, April). Seeing the forest from the trees: A comparison of two Dawson, I. (2004). Time for chronology? Ideas for developing chronologi-
instructional models using contrasting cases. Paper presented at the cal understanding. Teaching History, 117, 14–24.
annual meeting of the American Educational Research Association, Deslauriers, L., Schelew, E., & Wieman, C. (2011). Improved learning in a
Vancouver, BC, Canada. large-enrollment physics class. Science, 332, 862–864.
Chi, M. T. H. (2000). Self-explaining expository texts: The dual processes D’Mello, S., Olney, A., Williams, C., & Hays, P. (2012). Gaze tutor: A
of generating inferences and repairing mental models. Advances in gaze-reactive intelligent tutoring system. International Journal of
Instructional Psychology, 5, 161–238. Human-Computer Studies, 70, 377–398.
242 CHI AND WYLIE
Docktor, J. L., Strand, N. E., Mestre, J. P., & Ross, B. H. (2010, October). Kastens, K. A., & Liben, L. S. (2007). Eliciting self-explanations improves
A conceptual approach to physics problem solving. In AIP Conference children’s performance on a field-based map skills task. Cognition and
Proceedings (Vol. 1289, p. 137). Portland, OR: American Institute of Instruction, 25, 45–74.
Physics. Katayama, A. D., Shambaugh, R. N., & Doctor, T. (2005). Promoting
Doymus, K. (2008). Teaching chemical equilibrium with the jigsaw tech- knowledge transfer with electronic note taking. Teaching of Psychology,
nique. Research in Science Education, 38, 249–260. 32, 129–131.
Ebert-May, D., Brewer, C., & Allred, S. (1997). Innovation in large lec- King, A. (1993). From sage on the stage to guide on the side. College
tures: Teaching for active learning. Bioscience, 47, 601–607. Teaching, 41, 30–35.
Ferguson-Hessler, M., & de Jong, T. (1990). Studying physics texts: Dif- Kirschner, P. A., Sweller, J., & Clark, R. E. (2006). Why minimal guidance
ferences in study processes between good and poor performers. Cogni- during instruction does not work: An analysis of the failure of construc-
tion and Instruction, 7, 41–54. tivist, discovery, problem-based, experiential, and inquiry-based teach-
Fink, L. D. (2013). Creating significant learning experiences: An inte- ing. Educational Psychologist, 41, 75–86.
grated approach to designing college courses. New York, NY: Wiley & Kitchner, K. S. (1983). Cognition, metacognition, and epistemic cognition.
Sons. Human Development, 26, 222–232.
Fonseca, B. & Chi, M. T. H. (2011). The self-explanation effect: A con- Kobayashi, K. (2006). Combined effects of note-taking/reviewing on
structive learning activity. In R. E. Mayer & P. A. Alexander (Eds.), The learning and the enhancement through interventions: A meta-analytic
handbook of research on learning and instruction (pp. 296–321). New review. Educational Psychology, 26, 459–477.
York, NY: Routledge/Taylor and Francis. Kramarski, B., & Dudai, V. (2009). Group-metacognitive support for
Fredricks, J. A., Blumenfeld, P. C., & Paris, A. H. (2004). School engage- online inquiry in mathematics with differential self-questioning. Journal
ment: Potential of the concept, state of the evidence. Review of Educa- of Educational Computing Research, 40, 377–404.
tional Research, 74, 59–109. Loyens, S. M. M., & Gijbels, D. (2008). Understanding the effects of con-
Gobert, J. D., & Clement, J. J. (1999). Effects of student-generated dia- structivist learning environments: Introducing a multi-directional
grams versus student-generated summaries on conceptual understanding approach. Instructional Science, 36, 351–357.
of causal and dynamic knowledge in plate tectonics. Journal of Mastropieri, M. A., Scruggs, T., Mohler, L., Beranek, M., Spencer, V.,
Research in Science Teaching, 36, 39–53. Boon, R. T., & Talbott, E. (2001). Can middle school students with seri-
Goldstone, R. L., & Day, S. (Eds.) (2012). New conceptualizations of ous reading difficulties help each other and learn anything? Learning
transfer of learning [Special issue]. Educational Psychologist, 47(3). Disabilities Research & Practice, 16, 18–27.
Graesser, A. C., & Person, N. K. (1994). Question asking during tutoring. Mastropieri, M. A., Scruggs, T. E., Spencer, V., & Fontana, J. (2003). Pro-
American Educational Research Journal, 31, 104–137. moting success in high school world history: Peer tutoring versus guided
Griffin, T. D., Wiley, J., & Thiede, K. W. (2008). Individual differences, notes. Learning Disabilities Research & Practice, 18, 52–65.
rereading, and self-explanation: Concurrent processing and cue validity Mayer, R. E. (2004). Should there be a three-strikes rule against pure dis-
as constraints on metacomprehension accuracy. Memory & Cognition, covery learning? The case for guided methods of instruction. American
36, 93–103. Psychologist, 59, 14–19.
Guastello, E. F., Beasley, T. M., & Sinatra, R. C. (2000). Concept mapping Mayer, R. E. (2008). Learning and instruction (2nd ed.). Upper Saddle
effects on science content comprehension of low-achieving inner-city River, NJ: Pearson Prentice Hall.
seventh graders. Remedial and Special Education, 21, 356–364. McEldoon, K. L. (2014). Supporting novice learning through focused
Hausmann, R. G. M., & Chi, M. T. H. (2002). Can a computer interface processing of worked examples and explanations. (Unpublished doctoral
support self-explaining? Cognitive Technology, 7, 4–14. dissertation). Vanderbilt University, Nashville, TN.
Hausmann, R. G. M., van de Sande, B., & VanLehn, K. (2008). Shall we McEldoon, K. L., Durkin, K. L., & Rittle-Johnson, B. (2013). Is self-expla-
explain? Augmenting learning from intelligent tutoring systems and nation worth the time? A comparison to additional-practice. British
peer collaboration. In B. P. Woolf, E. Aimeur, R. Nkambou, & S. Lajoie Journal of Educational Psychology, 83, 615–632.
(Eds.), Intelligent tutoring systems (pp. 636–645). Amsterdam, The Menekse, M., Stump, G., Krause, S., & Chi, M. T. H. (2013). Differenti-
Netherlands: IOS. ated overt learning activities for effective instruction in engineering
Hendricks, C. C. (2001). Teaching causal reasoning through cognitive classrooms. Journal of Engineering Education, 102, 346–374.
apprenticeship: What are results from situated learning? Journal of Edu- Mestre, J. P. (2002). Probing adults’ conceptual understanding and transfer
cational Research, 94, 302–311. of learning via problem posing. Journal of Applied Developmental Psy-
Hogan, K., Nastasi, B. K., & Pressley, M. (1999). Discourse patterns and chology, 23, 9–50.
collaborative scientific reasoning in peer and teacher-guided discussions. Meyers, C., & Jones, T. B. (1993). Promoting Active Learning. Strategies
Cognition and Instruction, 17, 379–432. for the College Classroom. San Francisco, CA: Jossey-Bass.
Horton, P. B., McConney, A. A., Gallo, M., Woods, A. L., Senn, G. J., & Milrad, M. (2002). Using construction kits, modeling tools and system
Hamelin, D. (1993). An investigation of the effectiveness of concept dynamics simulations to support collaborative discover learning. Educa-
mapping as an instructional tool. Science Education, 77, 95–111. tional Technology and Society, 5, 76–87.
James, K. H., Humphrey, G. K., Vilis, T., Corrie, B., Baddour, R., & Good- Nesbit, J. C., & Adesope, O. O. (2006). Learning with concept and knowledge
ale, M. A. (2002). “Active” and “passive” learning of three-dimensional maps: A metaanalysis. Review of Educational Research, 76, 413–448.
object structure within an immersive virtual reality environment. Behav- Novak, J. D. (1990). Concept mapping: A useful tool for science education.
ior Research Methods, Instruments, & Computers, 34, 383–390. Journal of Research in Science Teaching, 27, 937–949.
Jonassen, D. H. (1991). Evaluating constructivistic learning. Educational Novak, J. D., & Gowin, D. B. (1984). Learning how to learn. New York,
Technology, 31, 28–33. NY: Cambridge University Press.
Kafai, Y. B. (2006). Playing and making games. Games and Culture, 1, Oakhill, J. V., Cain, K., & Bryant, P. E. (2003). The dissociation of word
36–40. reading and text comprehension: Evidence from component skills. Lan-
Kam, M., Wang, J., Iles, A., Tse, E., Chiu, J., Glaser, D., . . ., Canny, J. guage and Cognitive Processes, 18, 443–468.
(2005, April). Livenotes: A system for cooperative and augmented note- O’Donnell, A. M., Dansereau, D. F., & Hall, R. H. (2002). Knowledge
taking in lectures. In Proceedings of ACM Conference on Human Fac- maps as scaffolds for cognitive processing. Educational Psychology
tors in Computing Systems (CHI ’05) (pp. 531–540). Portland, OR. Review, 14, 71–86.
Okada, T., & Simon, H. A. (1997). Collaborative discovery in a scientific Schwartz, D. L., & Bransford, J. D. (1998). A time for telling a time for
domain. Cognitive Science, 21, 109–146. telling. Cognition and Instruction, 16, 475–522.
Okebukola, P. A., & Jegede, O. J. (1988). Cognitive preference and learn- Schwarz, B. B., Neuman, Y., & Biezuner, S. (2000). Two wrongs may
ing mode as determinants of meaningful learning through concept map- make a right . . . if they argue together! Cognition and Instruction, 18,
ping. Science Education, 72, 489–500. 461–494.
O’Reilly, T., Symons, S., & MacLatchy-Gaudet, H. (1998). Brief research Siegler, R. (1995). How does change occur: A microgenetic study of num-
report: A comparison of self-explanation and elaborative interrogation. ber conservation. Cognitive Psychology, 28, 225–273.
Contemporary Educational Psychology, 23, 434–445. Stark, R. (1999). Lernen mit L€osungsbeisplielen: Einfluss unvollst€andiger
Papert, S. (1980). Mindstorms: Children, computers and powerful ideas. L€osungsbeispiele auf Beispielelaboration, Motivation and Lernerfolg
New York, NY: Basic Books. [Learning by worked-out examples: The impact of completion tasks on
Pea, R., & Kurland, M. (1984). On the cognitive effects of learning com- example elaboration, motivation and learning outcomes]. Bern, Switzer-
puter programming. New Ideas in Psychology, 2, 137–168. land: Huber.
Pelech, J. (2010). The comprehensive handbook of constructivist teaching: Suthers, D. D., & Hundhausen, C. D. (2003). An experimental study of the
From theory to practice. Charlotte, NC: Information Age. effects of representational guidance on collaborative learning processes.
Peper, R. J., & Mayer, R. E. (1986). Generative effects of note-taking dur- Journal of the Learning Sciences, 12, 183–218.
ing science lectures. Journal of Educational Psychology, 78, 34–38. Sweller, J. (1994). Cognitive load theory, learning difficulty, and instruc-
Piaget, J. (1930). The child’s conception of physical causality. Oxford, tional design. Learning and Instruction, 4, 295–312.
England: Harcourt Brace. Trafton, J. G., & Trickett, S. B. (2001). Note-taking for self-explanation
Pillow, B. H., Mash, C., Aloian, S., & Hill, V. (2002). Facilitating child- and problem solving. Human–Computer Interaction, 16, 1–38.
ren’s understanding of misinterpretation: Explanatory efforts and van Boxtel, C., van der Linden, J., & Kanselaar, G. (2000). Collaborative
improvements in perspective taking. The Journal of Genetic Psychology, learning tasks and the elaboration of conceptual knowledge. Learning
163, 133–148. and Instruction, 10, 311–330.
Pine, K. J., & Messer, D. J. (2000). The effect of explaining another’s VanLehn, K., Graesser, A. C., Jackson, G. T., Jordan, P., Olney, A., &
actions on children’s implicit theories of balance. Cognition and Instruc- Rose, C. P. (2007). When are tutorial dialogues more effective than
tion, 18, 35–52. reading? Cognitive Science, 31, 3–62.
Pintrich, P. R., & De Groot, E. (1990). Motivation and self regulation com- VanLehn, K., Jones, R. M., & Chi, M. T. H. (1992). A model of the self-
ponents of academic performance. Journal of Educational Psychology, explanation effect. The Journal of the Learning Sciences, 2, 1–59.
82, 33–40. Von Glasersfeld, E. (1989). Cognition, construction of knowledge, and
Rafal, C. T. (1996). From co-construction to takeovers: Science talk in a teaching. Synthese, 80, 121–140.
group of four girls. Journal of Learning Sciences, 5, 279–293. Webb, N. M. (1989). Peer interaction, problem solving, and cognition:
Renkl, A., Stark, R., Gruber, H., & Mandl, H. (1998). Learning Multidisciplinary perspectives. International Journal of Educational
from worked-out examples: The effects of example variability and Research, 13, 1–119.
elicited self-explanations. Contemporary Educational Psychology, 23, Whitehead, A. N. (1929). The aims of education. New York, NY:
90–108. Macmillan.
Rickards, J. P., & Friedman, F. (1978). The encoding versus the external Whitten, W. B., II. (2011). Learning from and for tests. In A. S. Benjamin
storage hypothesis in note taking. Contemporary Educational Psychol- (Ed.), Successful remembering and successful forgetting: A festschrift in
ogy, 3, 136–143. honor of Robert A. Bjork (pp. 217–234). New York, NY: Psychology
Roscoe, R. D., & Chi, M. T. H. (2007). Understanding tutor learning: Press.
Knowledge-building and knowledge-telling in peer tutors’ explanations Willerman, M., & Mac Harg, R. A. (1991). The concept map as an advance
and questions. Review of Educational Research, 77, 534–574. organizer. Journal of Research in Science Teaching, 28, 705–711.
Rosenthal, J. S. (1995). Active learning strategies in advanced mathematics Wilson, B. G. (1997, March). Reflections on constructivism and instruc-
classes. Studies in Higher Education, 20, 223–228. tional design, instructional development paradigms. Englewood Cliffs,
Rouet, J.-F., Britt, M. A., Mason, R. A., & Perfetti, C. A. (1996). Using NJ: Educational Technology.
multiple sources of evidence to reason about history. Journal of Educa- Wong, R. M. F., Lawson, M. J., & Keeves, J. (2002). The effects of self-
tional Psychology, 88, 478–493. explanation training on students’ problem solving in high-school mathe-
Rowles, C. J. (2013). Strategies to promote critical thinking and active matics. Learning and Instruction, 12, 233–262.
learning. Teaching in nursing: A guide for faculty, 258–279. Wylie, R. (2011). Examining the generality of self-explanation (Unpub-
Rubin, J., & Herbert, M. (1998). Peer teaching-model for active learning. lished doctoral dissertation). Carnegie-Mellon University, Pittsburgh,
College Teaching, 48, 26–30. PA.
Russell, I. J., Caris, T. N., Harris, G. D., & Hendricson, W. D. (1983). Yaron, D., Karabinos, M., Lange, D., Greeno, J. G., & Leinhardt, G.
Effects of three types of lecture notes on medical student achievement. (2010). The ChemCollective—virtual labs for introductory chemistry
Academic Medicine, 58, 627–636. courses. Science, 328, 584–585.
Salomon, G., & Perkins, D. N. (1998). Individual and social aspects of Yetter, G., Gutkin, T., Saunders, A., Galloway, A., Sobansky, R., & Song,
learning. Review of Research in Education, 23, 1–24. S. (2006). Individual practice for complex problem solving: A caution-
Scardamalia, M. (1992, July). The role of adaptation and understanding in ary tale. Journal of Experimental Education, 74, 137–159.
knowledge-building communities. Paper presented at conference of Yin, Y., Vanides, J., Ruiz-Primo, M. A., Ayala, C. C., & Shavelson, R. J.
NATO Advanced Science Institute, Kolymbari Crete, Greece. (2005). Comparison of two concept-mapping techniques: Implications
Schauble, L., Glaser, R., Duschl, R. A., Schulze, S., & John, J. (1995). for scoring, interpretation, and use. Journal of Research in Science
Journal of the learning students’ understanding of the objectives and Teaching, 42, 166–184.
procedures of experimentation in the science classroom. Journal of the Zhang, Z. H., & Linn, M. C. (2013). Learning from chemical visualiza-
Learning Sciences, 4, 131–166. tions: Comparing generation and selection. International Journal of Sci-
Schmid, R. F. & Telaro, G. (1990). Concept mapping as an instructional ence Education, 35, 2174–2197.
strategy for high school biology. Journal of Educational Research, 84, Zimmerman, B. J. (1990). Self-regulated learning and academic achieve-
78–85. ment: An overview. Educational Psychologist, 21, 3–17.

The ICAP Framework Linking Cognitive Engagement, 2014

Uploaded by

Copyright:

Available Formats

The ICAP Framework Linking Cognitive Engagement, 2014

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The ICAP Framework Linking Cognitive Engagement, 2014

Uploaded by

Copyright:

Available Formats

This article was downloaded by: [George Mason University]

On: 18 December 2014, At: 03:59

The ICAP Framework: Linking Cognitive Engagement

Click for updates

To link to this article: http://dx.doi.org/10.1080/00461520.2014.965823

PLEASE SCROLL DOWN FOR ARTICLE

The ICAP Framework: Linking Cognitive

PASSIVE Receiving ACTIVE Manipulating CONSTRUCTIVE Generating INTERACTIVE Dialoguing

sufficient frequency of turn taking to meet our definition of

procedure, explaining why

interface. Taking notes is also somewhat similar to summa-

taking (interactive). As expected, the manner in which stu-

4 Average Listen (passive)

Passive Active Constructive Interactive

Passive No known studies

Passive Active Constructive Interactive

Passive No known studies

Passive Active Constructive Interactive

Passive No known studies

studies contributions > Self-explain own studies

this leads to greater learning than listening to a tradi-

Passive Active Constructive Interactive

ing, solving, critiquing, and peer discussion. We classify

Bloom’ s taxonomy. It is a daunting task to compare CONCLUSION

You might also like