Instr Sci (2017) 45:439–468 DOI 10.1007/s11251-017-9411-7 ORIGINAL RESEARCH

Who benefits from a low versus high guidance CSCL script and why? Stephan Mende1 • Antje Proske1 • Hermann Ko¨rndle1 Susanne Narciss1

•

Received: 22 October 2015 / Accepted: 17 April 2017 / Published online: 21 April 2017  Springer Science+Business Media Dordrecht 2017

Abstract Computer-supported collaborative learning (CSCL) scripts can foster learners’ deep text comprehension. However, this depends on (a) the extent to which the learning activities targeted by a script promote deep text comprehension and (b) whether the guidance level provided by the script is adequate to induce the targeted learning activities effectively; both may be moderated by the learners’ prior knowledge. Inspired by the ICAP framework (Chi and Wylie in Educ Psychol 49:219–243, 2014), we designed a low (LGS) and a high guidance script (HGS) to support learners in performing interactive activities. These activities include generating outputs that go beyond the text, while simultaneously referring to the co-learner. In an experiment, 88 undergraduates were assigned randomly to either the LGS or HGS condition. After reading a text paragraph, LGS participants thought about discussion points for the upcoming collaborative discussion, while HGS participants were (a) prompted to generate outputs individually that go beyond the text and (b) exchange them with their co-learner to provide information about the co-learner’s comprehension state (awareness induction). Subsequently, dyads in both conditions discussed the paragraph in a chat to improve their text comprehension. Prior knowledge moderated the effect of the script guidance level on deep text comprehension: low prior knowledge learners benefitted from the HGS, whereas high prior knowledge learners profited from the LGS. Moderated mediation analyses revealed that these effects can be traced back to patterns of learning activities which differed regarding learners’ prior knowledge. Based on these results, possible directions for future research on CSCL scripting and ICAP are discussed. Keywords CSCL  Collaboration scripts  Prior knowledge  Text comprehension  Learning activities

& Stephan Mende [email protected] 1

Psychology of Learning and Instruction, Technische Universita¨t Dresden, Zellescher Weg 17, 01069 Dresden, Germany

123

440

S. Mende et al.

Introduction Deep text comprehension requires learners to construct meaning actively (Fiorella and Mayer 2016; Kintsch 2004; Wittrock 1989). To this end, computer-supported collaborative learning (CSCL) settings offer rich opportunities, but without further support, such settings do not always foster learners’ deep comprehension. Therefore, scripts are frequently applied to structure the CSCL process (Fischer et al. 2013a). The effects of CSCL scripts on domain-specific learning outcomes are mixed (Vogel et al. 2016). Current research discusses two central issues of script design, among others, which may account for the mixed results. The first issue regards the type of learning activities targeted by a script. It is generally assumed that positive script effects on desired learning outcomes are mediated through certain beneficial learning activities targeted by the script designer (Vogel et al. 2016). However, the majority of studies in the field of CSCL scripting have not examined whether and to what extent the scripts fostered the targeted learning activities and whether and to what extent these activities, in turn, truly fostered the learning outcomes (Deiglmayr et al. 2015; Fischer et al. 2013b; Vogel et al. 2016). This research gap should be addressed to get a clearer answer to the question of which type of activity a script should be designed for. The second issue relates to the level of guidance a script provides in relation to learner characteristics. In this regard, learners’ domain-specific prior knowledge may be a crucial moderator influencing the effects of the script guidance level on learning outcomes. This issue has been rarely considered so far (Kollar et al. 2006; Vogel et al. 2016). Based on the interactive–constructive–active–passive (ICAP) framework (Chi and Wylie 2014), we will discuss several learning activities in the following that might occur in CSCL, along with their assumed effects regarding deep text comprehension. According to the claim that so-called interactive activities are the optimal activity to foster deep comprehension (Chi and Wylie 2014), we subsequently present low and high guidance script components aimed at fostering those activities in CSCL. Based on a classification of CSCL script result patterns (Stegmann et al. 2011), we will then address the questions of how prior knowledge may moderate (a) the relationship between the learning activities discussed and deep text comprehension, and (b) the effect of the script guidance level on these learning activities. We then present our study, which is based on the central claim that an integrated analysis of the aspects mentioned above in terms of moderated mediation analysis (e.g., Hayes 2013) can contribute to an explanation of the underlying mechanisms (learners’ activities) and conditions (learners’ prior knowledge) of the mixed effects of CSCL scripts on learning outcomes (see Fig. 1 for an overview of the relationships addressed).

Learning activities in CSCL and their effects on deep text comprehension In CSCL various learning activities can occur (Vogel et al. 2016). The ICAP framework (Chi and Wylie 2014) allows classifying these activities in order to derive predictions regarding their effects on deep comprehension. ICAP distinguishes the following learning activity classes: interactive, constructive, active and passive (see Fig. 2 for an overview and examples from CSCL text learning). Passive activities are defined as the reception of learning information without executing any observable activities related to learning. Activities classified as active involve observable actions related to learning. Constructive activities, in addition to observable learning actions, include generating new outputs that

123

Who benefits from which CSCL script and why?

441

Fig. 1 Summary of the relationships examined

Fig. 2 Overview of the ICAP learning activity classes and related examples in collaborative text learning (see Chi 2009; Chi and Wiley 2014)

go beyond the learning material. Interactive activities include the characteristics of active and constructive activities, but they additionally involve referring to contributions of another person, for example, a co-learner or teacher (Chi and Wylie 2014). Inspired by these definitions, we propose to classify a specific CSCL learning activity into the ICAP framework by considering three criteria: (a) action criterion: engaging with the learning information in an overtly observable manner, (b) inference criterion: generating or inferring outputs that go beyond the learning information given and (c) reference

123

442

S. Mende et al.

criterion: referring to contributions of another person (e.g., co-learner, teacher) while dealing with the learning information. According to the hierarchical nature of the ICAP taxonomy (see Fig. 2) a higher activity class includes the features of all lower classes (Chi and Wylie 2014). That is, active activities fulfill the action criterion only, constructive activities fulfill the action and inference criteria and interactive activities fulfill the action, inference and reference criteria. Please note the terms active and interactive have also been used with other meanings in educational research. For example, in the memory literature, learning strategies are termed active that result in deep learning outcomes. By contrast, in the ICAP framework the term active describes not the expected outcome of a learning activity but rather the observable activity itself (Chi 2009). Other researchers regard transactive and interactive activities equally (e.g., Vogel et al. 2016). Transactive activities involve considering co-learners’ contributions while dealing with a learning task (Berkowitz and Gibbs 1983; Teasley 1997). Hence, transactive activities fulfill the action and reference but not necessarily the inference criterion. For example, if learners just restate text information given while referring to each other’s contributions the inference criterion is not fulfilled. By contrast, to be considered as an interactive activity, all three criteria of action, inference and reference must be fulfilled (Chi and Wylie 2014). How are the ICAP learning activities related to learning outcomes? Many studies in ICAP research rarely investigated directly whether learners actually performed the learning activities desired (Deiglmayr et al. 2015). Nevertheless, reviews investigating the ICAP hypothesis provide evidence that the depth of comprehension arising from the learning activities increases from passive to active through constructive up to interactive (Chi 2009; Chi and Wylie 2014; Fonseca and Chi 2011).

Script components to foster interactive learning activities Research shows that learners often have difficulties performing interactive activities spontaneously (Bromme et al. 2005; Chi and Menekse 2015; Kirschner and Erkens 2013; Kollar et al. 2006; Pfister 2005; Roscoe 2014; Weinberger 2011). The challenges associated with encouraging learners to go beyond the learning information given (inference criterion) and refer adequately to each other (reference criterion) can be addressed with CSCL scripts (e.g., Jeong and Hmelo-Silver 2016; Kollar et al. 2006). A CSCL script is defined as a computer-based external representation providing a set of rules to guide learners to a specific collaborative practice they would not perform spontaneously (Fischer et al. 2013a; Ha¨kkinen and Ma¨kitalo-Siegl 2007; Kollar et al. 2006; O’Donnell 1999). To this end, scripts can structure CSCL on a macrolevel by sequencing the learning process into different phases on a relatively global level without specifying concrete learning activities for the single phases. Scripts can also structure CSCL—and often in addition to macro-scripting—on a microlevel by delivering more specific scaffolds and instructions on how to perform concrete learning activities. Hence, micro-scripting involves a higher degree of guidance than macro-scripting (Fischer et al. 2013b; Weinberger et al. 2009).

Sequencing (macrolevel) CSCL needs to be structured to be successful. It should be determined at which time learners should process the learning content individually and when collaboratively (Bromme et al. 2005; Weinberger 2011). Sequencing collaborative learning into individual

123

Who benefits from which CSCL script and why?

443

thinking and collaborative discussion phases has been shown to benefit collaboration in terms of idea generation (Diehl and Stroebe 1987; Rietzschel et al. 2006). Furthermore, an individual study phase ensures that learners have enough time to relate their relevant prior knowledge to the learning material. This, in turn, may foster the access to and discussion of unshared knowledge during collaboration (e.g., Rummel and Spada 2005). Thirdly, prescribing an individual study phase before collaborative discussion can reduce the risk of cognitive overload due to the dual requirements of collaborative learning in terms of processing the learning content while simultaneously investing effort in communication and coordination with other learners (Gadgil and Nokes-Malach 2012; Janssen et al. 2010; Lam 2013; Weinberger 2011). Consequently, sequencing the CSCL process into individual study and collaborative discussion phases seems to be an adequate means to foster both the inference and reference criterion of interactive activities.

Process-oriented instruction (microlevel) The script principle of process-oriented instruction (Weinberger 2011) can be applied to guide learners in performing a specific learning activity. Prompting learners to self-explain in individual learning has proven to be a powerful method to encourage learners to externalize outputs which go beyond the learning material (Fonseca and Chi 2011; McNamara and Magliano 2009a) and foster deep comprehension across domains and instructional materials (Dunlosky et al. 2013). Learners, while explaining learning material to themselves, monitor their text understanding, identify omissions in the text or their own comprehension, and draw inferences to fill in these gaps (Chiu and Chi 2014). Hence, prompting learners to externalize self-explanations seems to be a promising approach for fostering the inference criterion of interactive activities.

Group awareness induction (microlevel) Awareness induction entails providing learners with an externalized representation of the co-learners’ knowledge and comprehension states (Weinberger 2011). This can foster cognitive group awareness (Janssen and Bodemer 2013). That means making the learner aware of (a) the opinions and ideas the co-learner has regarding the learning content and whether there is a shared understanding, (b) differences in understanding and conflicting points of view which need to be discussed and (c) unshared topic-relevant knowledge the partner may possess (Bromme et al. 2005; Engelmann et al. 2009; Gijlers and de Jong 2009; Janssen et al. 2007). Awareness induction reduces grounding costs and allows for referring effectively to the co-learner, for instance, in terms of extending the other’s reasoning or discussing different points of view (Engelmann et al. 2009; Gijlers and de Jong 2009; Janssen et al. 2007). Accordingly, supporting cognitive group awareness has been shown to improve CSCL in terms of transactivity and achievement (Janssen and Bodemer 2013). When responding to an open self-explanation prompt, learners explain in their own words how they understood the information read, state what is new to them, indicate difficulties in understanding, raise open questions and insert additional knowledge. In short, they externalize their current state of comprehension and knowledge regarding the learning material (Chi et al. 1994; Chi 2000). Hence, reading the co-learner’s response to such a prompt might be a suitable means to induce cognitive group awareness and, thus, to facilitate the reference criterion of interactive activities. In sum, combining the macro- and

123

444

S. Mende et al.

micro-script components mentioned above allows the creation of scripts with varying levels of guidance targeted at inducing interactive activities.

The role of prior knowledge In educational research the finding that—dependent on learners’ prior knowledge—the same level of instructional guidance can have different effects on the learning outcomes has been referred to as aptitude treatment interaction (Cronbach and Snow 1977) or expertise reversal effect (e.g., Kalyuga et al. 2003). Regarding the effects of the CSCL script guidance level on the learning outcomes desired, the role of prior knowledge has been rarely examined (Kollar et al. 2006; Vogel et al. 2016). To this end, we will apply an approach of Stegmann et al. (2011) that allows for a detailed examination of script guidance level effects. Stegmann et al. (2011) suggest differentiating between two issues of scripting. Firstly, a CSCL script can be functional or malfunctional, to the extent to which the learning activities targeted by the script promote the learning outcomes desired. If the targeted activities do not foster the learning outcomes, malfunctional scripting occurs (Stegmann et al. 2011). Secondly, it must be considered to what extent the guidance level provided by a script fosters the actual execution of the targeted learning activities. The guidance level may be too high when learners are already able to perform the activity in question on their own or would perform other more effective activities spontaneously instead. In this case, overscripting occurs and learners will spend more time handling the script, while time on task is decreased. Due to this reduced efficiency, the learning outcomes will also not be fostered in turn (Stegmann et al. 2011). If script guidance is too low in terms of providing insufficient information and scaffolds to support the execution of the learning activities targeted, under-scripting occurs. Hence, learners would hardly be able to execute the learning activities targeted and thought of to benefit learning outcomes (Stegmann et al. 2011). Translated to the issue of a potential interplay between the effects of the script guidance level and prior knowledge on deep text comprehension, we should address two questions: How may prior knowledge moderate (a) the effects of active, constructive and interactive activities on deep text comprehension, and (b) the effects of the script guidance level on the interactive activities targeted?

Prior knowledge and the effects of learning activities on deep text comprehension How may prior knowledge moderate the effects of the ICAP learning activities on deep comprehension? Inferences are necessary for integrating information to be learned into preexisting knowledge structures. Activities not fulfilling this criterion, such as active activities, are not expected to benefit deep comprehension (Chi and Wylie 2014). This is considered to be independent of prior knowledge (Cote´ et al. 1998; Kintsch 1998, 2004). By contrast, activities that fulfil the inference criterion, such as constructive and interactive activities are considered to be conducive to deep comprehension (Chi and Wylie 2014; King 1999; Kintsch 2004; McNamara and Magliano 2009b; Roscoe and Chi 2007). Research on text comprehension (Best et al. 2005; Kintsch 1998, 2004; McNamara and Magliano 2009b), instructional explanations (Renkl 1999; Wittwer and Renkl 2008) and collaborative learning (Cohen 1994; Nokes-Malach et al. 2012, 2015; VanLehn et al. 2007) provide indirect evidence, that the effects of constructive and interactive activities

123

Who benefits from which CSCL script and why?

445

on deep text comprehension may differ with respect to prior knowledge. This research suggests that if learners’ prior knowledge does not allow them to draw necessary inferences on their own, deep comprehension is fostered by providing additional learning resources (e.g., additional explanations). By contrast, if prior knowledge is sufficient, providing such additional resources has been shown to be ineffective or even detrimental. Constructive and interactive activities differ in the fulfillment of the reference criterion (Chi and Wylie 2014). That is, performing interactive activities in CSCL allows learners to use other learners’ externalized prior knowledge, ideas and perspectives not contained in the text as additional learning resources for drawing inferences (Chi and Wylie 2014; Deiglmayr and Schalk 2015; Nokes-Malach et al. 2015). Learners with high prior knowledge are more able to generate inferences on the basis of their preexisting knowledge structures, while learners with low prior knowledge are less so (Kintsch 2004; McNamara and Magliano 2009b). Hence, performing interactive activities might allow low prior knowledge learners to draw more of the inferences necessary to comprehend a given text deeply (cf. Deiglmayr and Schalk 2015). By contrast, high prior knowledge learners may profit less from interactive activities since they are able to draw more inferences based primarily on their own knowledge (constructive activities). In sum, interactive activities may be more beneficial for low than for high prior knowledge learners in terms of deep text comprehension.

Prior knowledge and the effects of the script guidance level on learning activities How might prior knowledge affect which script guidance level is adequate to induce interactive activities? To derive assumptions in this regard, we need to discuss how prior knowledge generally influences the likelihood of spontaneously drawing inferences (inference criterion) and referring to other learners’ contributions (reference criterion). First, prior knowledge is positively associated with the amount of inferences drawn during studying learning material (Best et al. 2005; Chan et al. 1992; Ertl et al. 2004; McNamara 2004; Webb 1989). Second, concerning the reference criterion, some researchers assume that higher prior knowledge is associated with an increasing ability to solve a task alone. This potentially lowers the probability of learners’ efforts to refer to colearners (Gadgil and Nokes-Malach 2012; Janssen et al. 2010). Thus, high prior knowledge learners may perform more constructive activities spontaneously than low prior knowledge learners (Nokes-Malach et al. 2012). By contrast, low prior knowledge learners may perform fewer activities involving inferences (i.e. constructive or interactive activities). Thus, a high guidance script providing relatively finegrained scaffolds concerning the inference criterion and the reference criterion may be more adequate for low than for high prior knowledge learners (cf. Fischer et al. 2013a; Kalyuga et al. 2003; Stegmann et al. 2011).

The present study The present study aims at examining the mediating role of learners’ activities and the moderating role of learners’ prior knowledge in a CSCL scenario with a low versus high guidance script aimed at fostering deep text comprehension outcomes through promoting interactive activities.

123

446

S. Mende et al.

We first examine to what extent prior knowledge may moderate the effect of the script guidance level on deep text comprehension (research question 1, see Fig. 1). Second, by means of moderated mediation analyses, we examine to what extent the script guidance level may affect deep comprehension mediated through the learning activities discussed and whether prior knowledge may moderate such mediated relationships (research question 2, see Fig. 1).

Method Participants and setting Ninety-two undergraduates of a German university participated in an experimental e-learning study about the human circulatory system. Students of medicine, biology or similar domains, as well as non-native speakers were excluded from participation. Four students were single-wise excluded post hoc because they did not follow the instructions in the pretest or the data were incomplete. The final sample contained 88 undergraduates (76.7% female, mean age: 22.8 years, SD = 3.6) participating either in the low (n = 45) or high (n = 43) guidance script condition. The sample included students of several subjects (53.4% psychology, 31.8% human/social sciences and 14.8% natural/engineering sciences).

Learning material Participants were provided with an expository text about the human circulatory system, translated and adapted from Chi et al. (2001). The text consisted of 1248 words, approximately evenly distributed over 22 paragraphs. The paragraphs were presented stepwise on the monitor within the e-learning environment. That is, each new paragraph was added to the bottom of the paragraphs finished already, which were colored gray, but stayed legible.

Design and procedure After an introduction into the e-learning system, participants’ demographic data and prior knowledge were gathered by means of an electronic pretest. All participants were told that they would learn with a text about the human circulatory system. They were instructed to understand the circulatory system in terms of its components, functioning and purpose (Jeong and Chi 2007). Students were grouped randomly in dyads by the e-learning system. These dyads were subsequently randomly assigned to either the low or the high guidance script condition.

The high guidance script condition (READ script) The high guidance script integrated the macro- and microlevel script components described in the introduction, forming the READ script. First, students were asked to read (R) the current text paragraph. Second, learners were requested to respond in writing to a selfexplanation prompt (see Appendix 1) adapted from Chi et al. (1994). The prompt was intended to induce comprehension processes going beyond the text (inference criterion),

123

Who benefits from which CSCL script and why?

447

and required learners to externalize (E) the results of this process. The learners’ externalizations in response to the prompt served as the basis for the third step of the individual study phase: In order to induce awareness (A), the externalization was delivered to the colearner explicitly and both learners were requested to read the co-learner’s externalization. This was intended to facilitate the reference criterion of interactive activities. After the individual study phase of reading, externalizing and awareness induction, learners were directed to the chat-based discussion phase (D). Here they were instructed to help each other to improve their text understanding. The collaborative discussion phase was not further scaffolded. These steps of the high guidance READ script were repeated for all paragraphs. In each discussion phase, the processed and current paragraphs, and the externalizations of both co-learners concerning the current paragraph were visible (see Fig. 3).

The low-guidance script condition The low guidance script condition followed the same procedure for every text paragraph as the high guidance script condition (individual study phase and collaborative discussion

Fig. 3 Example of the learning environment (discussion phase in the high guidance READ script condition)

123

448

S. Mende et al.

phase) on the macrolevel. However, after reading a paragraph individually, learners did not receive the micro-scaffolds of the high guidance READ script (prompted externalization and awareness induction). Instead, they were asked to think about discussion points for the following chat-based discussion phase without further scaffolds. When learners felt ready, they could proceed to the discussion, where they received the same instructions as under the high guidance script condition.

Measures Pretest We assessed the participants’ prior knowledge with the blood path drawing task from Jeong and Chi (2007). Students had to draw the blood path of the circulatory system on a human body outline and explain it in a textbox. Prior knowledge was coded using a predefined template consisting of circulatory system knowledge pieces, defined as topicrelevant idea units, e.g., ‘‘the heart is a pump’’ or ‘‘blood moves from the heart to the body.’’ Participants received one point for each written or drawn piece of knowledge (Jeong and Chi 2007). A second rater coded 25% of the data for inter-rater reliability (j = .89).

Posttest Students were requested to answer a posttest adapted from Chi et al. (2001) to measure deep text comprehension. The test included six knowledge-based inference multiple choice (MC) questions requiring the integration of text information with prior knowledge. Furthermore, it contained six transfer MC questions which required the application of the text information to health issues not addressed by the text. Examples of the items and answer options are provided in Appendix 2. The correct answers for all 12 questions were not explicitly stated in the text, nor were they derivable solely from the text information. The item difficulties ranged from .06 to .42, which is in accordance with the characterization of these questions by Chi et al. (2001) as extremely difficult. Each question consisted of four answer options, with either one or a maximal of two options being correct. A participant’s response to an item was considered to be correct only if all correct options and no incorrect option were selected. Percentages of correctly answered questions were computed for each participant.

Coding of learners’ individual externalizations in the high guidance script condition To conduct a treatment check, the content of the externalizations produced in the individual study phases of the READ script condition was coded based on the ICAP definitions of active and constructive learning activities (Chi 2009; Chi and Wylie 2014) and several published coding schemes of, for example, De Backer et al. (2014), Roscoe and Chi (2008) and Roscoe (2014). Accordingly, two decisions were made independently for each of the 22 externalizations per participant: (a) Does the externalization contain restated or paraphrased information already given in the instructional material indicative for active activity? (b) Does the externalization contain inferred, that is, new topic-relevant information not stated in the instructional material indicative for constructive activity? A second rater coded 20% of the individual externalizations. Cohen’s j indicated a very good

123

Who benefits from which CSCL script and why?

449

inter-rater reliability for active (j = .93) and for constructive activities (j = .81). The two resulting scores express the number of individual externalizations containing active and constructive activities, both with a range from 0 to 22.

Coding of the collaborative discussion activities The chat dialogues of the discussion phases were coded based on the definitions of the ICAP activity classes (Chi 2009; Chi and Wylie 2014), the coding schemes mentioned above and published operationalizations of transactive activities of, for example, Berkowitz et al. (2008), De Backer et al. (2014), Jeong and Chi (2007) and Noroozi et al. (2013). Three independent decisions were made per participant for each of the 22 discussion phases: (a) does the discussion phase contain indications for active activity? (b) does the discussion phase contain indications of inferencing without reference to any prior contribution of the co-learner (constructive activity)? (c) does the discussion phase contain indications of inferencing with reference to a prior contribution of the co-learner (interactive activity)? Consider the following two examples outlined from our coding of the chat-based discussions for illustration purposes: Learning text: ‘‘By the time blood reaches the veins, it is under much less pressure than it is in the arteries. Thus, the veins are not as strong or as flexible as the arteries. Many veins pass through skeletal muscles. During movements, these muscles contract, squeezing blood through the veins’’ (Chi et al. 2001, p. 519). Example 1: Learner 1: ‘‘Contractions with the help of skeleton muscles during movement’’ (rephrasing of text information given: active) Learner 2: ‘‘Blood is pressed through veins semi-passively by skeleton muscles’’ (inference of new information beyond the text given without referring to a colearner’s prior contribution: constructive) Example 2: Partner 1: ‘‘Ok, you simply need to know that the blood within the veins does not flow by itself, but requires the help of muscles—it partially flows upwards which is clearly more ‘‘effortful’’! :)’’ (constructive) Partner 2: ‘‘Exactly :-) I think that this is important for old people or humans with heart and circulatory problems in order to keep the blood running smoothly’’ (inference of new information beyond the text given, while referring to a co-learner’s prior contribution: interactive) Partner 1: ‘‘Yes, this makes sense, if the muscles are weaker, then the blood will not be pumped back quickly enough’’ (interactive) A second rater coded 20% of the chat protocols. Cohen’s j indicated good inter-rater reliability for active (j = .79) and for constructive activities (j = .73), and a very good agreement for interactive activities (j = .86). The final three scores express the number of discussion phases containing active, constructive and interactive activities, respectively. Each score ranged from 0 to 22.

Data analysis A moderated regression was performed to examine research question one. We conducted moderated mediation analyses also following a regression-based approach when

123

450

S. Mende et al.

addressing research question two (Hayes 2013). Because some of the research variables did not meet the prerequisite of normal distribution and homogeneity of variance, we performed bootstrap analyses to insure the validity of our results by applying the SPSS bootstrapping procedure (Afifi et al. 2007; Neal and Simons 2007). Thus, bootstrapped standard errors and bias-corrected and accelerated 95% bootstrap confidence intervals (1000 resamples) are reported for all regression weights. An effect is considered significant at a 5% significance level if the 95% bootstrap interval does not include zero (Afifi et al. 2007). All continuous predictors were centered prior to the analyses. Unstandardized regression coefficients are reported.

Results Descriptive and preliminary analyses Descriptive statistics for pretest and dependent variables are summarized in Table 1 and nonparametric correlations (Spearman’s rank-order coefficient rs) are shown in Table 2. No significant pre-group differences regarding age, sex, subject of study or prior knowledge occurred. The low and the high guidance script condition differed significantly for learning time (t(86) = -10.55, p \ .001). Hence, this variable was examined as a further mediator in addition to the main analyses to control for potential artefacts of learning time.

Treatment check: micro components of the high guidance script Participants of the high guidance script condition showed active activities in M = 14.91 (SD = 5.12) and constructive activities in M = 11.54 (SD = 4.49) of the individual Table 1 Descriptives of pretest and dependent variables (N = 88) Low guidance script (n = 45)

High guidance script (n = 43)

M

M

SD

SD

Age

22.47

3.24

23.09

Sexa

.22

.42

.26

3.90 .44

Learning timeb

57.89

14.32

94.70

18.27

Prior knowledge

12.93

7.79

7.50

13.37

Active activities (individual)

–

–

14.91

5.12

Constructive activities (individual)

–

–

11.54

4.49

–

–

14.04

4.74

Reading time externalizationc Active activities (discussion) Constructive activities (discussion) Interactive activities (discussion) Deep Posttestd a

11.67

6.71

1.72

1.40

3.84

3.17

.95

1.46

3.36

2.96

6.02

3.84

25.93

15.91

23.45

10.80

0 = female, 1 = male

b

Learning time in minutes

c

Average time spent on reading a co-learners’s individual externalization in seconds

d

Percent of MC items answered correctly

123

Who benefits from which CSCL script and why?

451

Table 2 Nonparametric correlations (Spearman’s Rho) of pretest and dependent variables (N = 88) Variable 1

Learning Timea

2

Prior knowledge

3

Active (Individual.)b

4

Constructive (Individual)b

5

Reading time (ext.)bc

6

Active (discussion)

7

Constructive (discussion)

8

Interactive (discussion)

9

Deep Posttestd

1 –

2

3b

4b

5b,

c

-.06

.16

-.25

.22

–

.05

.18

-.22

–

-.17 –

.39** -.23 –

6

7

-.44**

-.42**

9d

8 .33**

-.15

-.07

.15

.15

-.15

-.31*

-.63**

.00

.09

.19

.32*

.29

-.08

-.03

–

.49** –

.40**

-.48** -.27*

.15 -.01

.01

.17

–

.18 –

* p \ .05, **p \ .01, *** p \ .001 a

Learning time in minutes

b

Values refer only to the high guidance script condition (n = 43)

c

Time spent on reading a co-learner’s individual externalization in seconds

d

Percent of MC items answered correctly

externalization phases. They spent an average of 14.04 s (SD = 4.74) reading their colearner’s completed externalization (awareness induction phases), which seems reasonable considering the average length of 24.80 words (SD = 6.84) per externalization (cf. Trauzettel-Klosinski and Dietz 2012). Prior knowledge was not associated with any of the treatment check variables.

Effects of script guidance level and prior knowledge on deep text comprehension By conducting a moderated regression analysis we addressed research question one: How the script guidance level (low versus high) affects deep text comprehension and whether prior knowledge may moderate this relationship (see Table 3 for the statistical values). No significant main effect was found. Entering the interaction term between script guidance level and prior knowledge in a second step increased the explained variance significantly. This indicates a significant interaction effect between the script guidance level and prior knowledge. We computed the simple slopes for the script guidance level effects on deep text comprehension at different values of prior knowledge to explore this interaction effect further. The simple slopes showed that students with prior knowledge at one standard deviation below the mean who learned with the high guidance script outperformed participants who learned with the low guidance script (B = 7.09, BCa CI95% = .96–13.31), while there was no significant influence of the script guidance level at the mean. The effect of the script guidance level was significantly negative (B = -12.70, BCa CI95 % = -18.96 to -5.20) for participants with prior knowledge at one standard deviation above the mean, indicating that the low guidance script was more advantageous for this subgroup than the high guidance script in terms of deep text comprehension.

123

452

S. Mende et al.

Table 3 Regression results for the effects of the script guidance level and prior knowledge (step 1) and their interaction (step 2) on achievement in the deep text comprehension posttest (N = 88) Predictor

B

Boot SE

BCa CI95

%

R2 (F)

DR2 (DF)

Step 1 SCRGDLVa

-2.81

2.63

.76

.21

-2.81

2.43

1.42

.20

[.95, 1.73]

-1.30

.29

[-1.89, -.58]

Prior knowledge

[-7.90, 2.51]

.19*** (9.79)

[.33, 1.17]

Step 2 SCRGDLVa Prior knowledge SCRGDLV 9 Prior knowledge

[-7.67, 2.07]

.32*** (13.10)

.13*** (16.22)

SCRGDLV script guidance level. All continuous predictors were centered prior to analyses. Unstandardized regression coefficients are reported. Accelerated and bias-corrected bootstrap confidence intervals indicating significant regression weights are written in bold * p \ .05, **p \ .01, *** p \ .001 a

0 = low guidance script, 1 = high guidance script

Exploring why the script guidance level effects differed with respect to prior knowledge We conducted moderated mediation analyses to answer research question two and, thereby, explore the reasons why low prior knowledge learners profited from the high guidance script while high prior knowledge learners benefitted from the low guidance script in terms of deep text comprehension. To this end, we applied the SPSS macro PROCESS following the procedures recommended by Hayes (2013). The ICAP activities overlap conceptually per definition. Thus, the mediators were analyzed separately (cf. Preacher and Hayes 2008). Consequently we conducted a three-step procedure for each of the four mediators examined: (1) active, (2) constructive and (3) interactive activities in the discussion phases and (4) learning time (see Fig. 1): Analysis of the a-path: conducting a hierarchical regression to examine the effect of script guidance level (X) on a mediator (Me), in a first step, represents the estimation of the unconditional a-path. Prior knowledge (Mo) was included as a further predictor. Entering the interaction term between X and Mo in a second step to examine its effects on a mediator represents the estimation of the conditional a-path. By means of this (un-)conditional a-path analysis, it is examined whether and to what extent script guidance level affects a mediator (i.e., a certain learning activity or learning time) and whether prior knowledge moderates this relationship in terms of its magnitude and/or sign. Analysis of the b-path: conducting a hierarchical regression to examine the effect of a mediator (Me) on deep text comprehension (Y) while controlling for script guidance level (X), in a first step, represents the estimation of the unconditional b-path. Prior knowledge (Mo) was included as a further predictor. Entering the interaction term between Me and Mo, in a second step, to examine its effects on deep text comprehension (Y) represents the estimation of the conditional b-path. By means of this (un-)conditional b-path analysis, it is examined whether and to what extent a mediator (i.e., a certain learning activity or learning time) affects deep text comprehension and whether prior knowledge moderates this relationship in terms of its magnitude or size while controlling for the influence of script guidance level in both cases.

123

Who benefits from which CSCL script and why?

453

Analysis of the indirect path (a*b): Based on the (un-)conditional a-path and b-path analyses, mediation effects (unconditional indirect path) and moderated mediation effects (conditional indirect path) were estimated and tested for significance following a bootstrapping approach reporting bias-corrected bootstrap confidence intervals, as provided in PROCESS (Hayes 2013). This allows for conclusions as to whether the script guidance level affects deep text comprehension through a mediator or, in other words, because the script guidance level affects a mediator (i.e. a certain learning activity or learning time) which, in turn, affects deep text comprehension (mediation effect), while considering whether the magnitude and/or sign of such a potential mediation effect is dependent on learners’ prior knowledge (moderated mediation effect). In the following we will present the results of the moderated mediation analyses for each mediator. We will firstly report if a significant (moderated) mediation effect was found. If so, we will secondly consult the results of the a-path and b-path analyses to provide a detailed picture of how the reported effect is constituted (Hayes 2013): is there a moderated mediation effect regarding a certain mediator because the effect of the script guidance level on that mediator is moderated by prior knowledge, and/or because the effect of the mediator on deep text comprehension is moderated by prior knowledge in some way? To facilitate readability, statistical values are not presented in the text. The estimations of the (un-)conditional a- and b-paths are shown in Tables 4, 5, 6 and 7. Figure 4 provides a summary of the moderated mediation analyses. A comprehensive overview of the coefficients and bootstrapping results for the total effect (c-path), the direct effects (c0 paths) and the (un-)conditional a-paths, b-paths and a*b-paths (mediation effects) is presented in Appendix 3.

Active discussion activities A significant moderated mediation effect was found: Mediated via active discussion activities the script guidance level had a significantly positive effect on deep comprehension, but only for participants with prior knowledge at one standard deviation below the mean (Fig. 4, indirect path). Consulting the a- and b-path analyses helps to interpret this effect (Table 4): The high guidance script reduced active activities compared to the low guidance script for all learners irrespective of prior knowledge (see Fig. 4, a-path). These active activities were detrimental to deep text comprehension only for low prior knowledge learners (see Fig. 4, b-path). In sum, compared to the low guidance script, the high guidance script benefitted low prior knowledge learners’ deep text comprehension to the extent it reduced active activities that were disadvantageous to their deep comprehension.

Constructive discussion activities Results reveal a significant moderated mediation effect: Mediated through constructive discussion activities the script guidance level had a significantly negative effect on deep text comprehension, but only for participants with prior knowledge at one standard deviation above the mean (Fig. 4, indirect path). Consulting the a- and b-path analyses (Table 5) reveals that with the low guidance script the average and high but not the low prior knowledge learners showed more constructive activities than with the high guidance script (Fig. 4, a-path). However, constructive discussion activities were, in turn, only conducive to high prior knowledge learners’ deep text comprehension (Fig. 4, b-path). In sum, the low guidance script (as compared to the high guidance script) benefitted high

123

123

Active (discussion)

[-.18, .07]

[-11.88, -7.90] .14

.13

1.02

.00 (.14)

.52*** (30.03)

.05

-.08

-9.92 [-.21, .33]

[-.35, .16]

[-11.97, -7.81]

BCa CI95%

.30

.21

4.16

Boot SE

.20*** (6.87)

-.28

.74

-5.55

B

[-.93, .28]

[.32, 1.10]

[-15.66, 3.24]

BCa CI95%

Unconditional b-path (YMe.X)

.02

.31

.18

4.22

Boot SE

.07** (7.73)

.27*** (7.49)

.07

-.27

.77

-5.66

B

[.03, .11]

[-.94, .34]

[.42, 1.08]

[-16.22, 3.46]

BCa CI95%

Conditional b-path (YMe*Mo.X)

a

0 = low guidance script, 1 = high guidance script

* p \ .05, ** p \ .01, *** p \ .001

SCRGDLV script guidance level, X script guidance level, Me mediator, Mo prior knowledge, Y deep text comprehension, MeX effect of the script guidance level on the mediator, MeX*Mo interaction effect between the script guidance level and prior knowledge on the mediator, YMe.X effect of a mediator on deep text comprehension while controlling for the script guidance level; YMe*Mo.X interaction effect between the mediator and prior knowledge on deep text comprehension while controlling for the script guidance level. All continuous predictors were centered prior to analyses. Unstandardized regression coefficients are reported. Accelerated and bias-corrected bootstrap confidence intervals indicating significant regression weights are written in bold

DR (DF)

2

R2 (F)

Active (discussion) 9 Prior knowledge

.06

1.01

.52*** (45.44)

-.05

SCRGDLVa 9 Prior knowledge

-9.92

Prior knowledge

Boot SE

B

BCa CI95%

B

Boot SE

Conditional a-path (MeX*Mo)

Unconditional a-path (MeX)

SCRGDLVa

Predictor

Table 4 Regression results for the moderated mediation analysis with active discussion activities as mediator (N = 88)

454 S. Mende et al.

.51

.04

.33*** (20.66)

.10

-2.94 [.03, .16]

[-4.00, -1.80]

.49 .06

.05

.08

**

(12.03)

.41*** (19.57)

-.22

.21

-2.93

Boot SE

[-.34, -.09]

[.11, .30]

[-3.91, -1.94]

BCa CI95%

B

BCa CI95%

B

Boot SE

Conditional a-path (MeX*Mo)

Unconditional a-path (MeX)

.64

.21

3.14

Boot SE

.21*** (7.37)

.84

.68

-.34

B

[-.39. 2.31]

[.25, 1.10]

[-7.87, 6.41]

BCa CI95%

Unconditional b-path (YMe.X)

.05

.68

.19

3.22

Boot SE

.05* (5.92)

.26*** (7.33)

.13

.38

.62

-.05

B

[.01, .21]

[-.94, 1.98]

[.27, .95]

[-7.67, 6.36]

BCa CI95%

Conditional b-path (YMe*Mo.X)

a

0 = low guidance script, 1 = high guidance script

* p \ .05, ** p \ .01, *** p \ .001

SCRGDLV script guidance level, X script guidance level, Me mediator, Mo prior knowledge, Y deep text comprehension, MeX effect of the script guidance level on the mediator, MeX*Mo interaction effect between the script guidance level and prior knowledge on the mediator, YMe.X effect of a mediator on deep text comprehension while controlling for the script guidance level; YMe*Mo.X interaction effect between the mediator and prior knowledge on deep text comprehension while controlling for the script guidance level. All continuous predictors were centered prior to analyses. Unstandardized regression coefficients are reported. Accelerated and bias-corrected bootstrap confidence intervals indicating significant regression weights are written in bold

DR (DF)

2

R2 (F)

Constructive (discussion) 9 prior knowledge

Constructive (discussion)

SCRGDLVa 9 Prior knowledge

Prior knowledge

SCRGDLVa

Predictor

Table 5 Regression results for the moderated mediation analysis with constructive discussion activities as mediator (N = 88)

Who benefits from which CSCL script and why? 455

123

123

.05

.05 .10

.07

.74

Boot SE

[-.30, .10]

[-.03, .24]

[1.14, 4.08]

BCa CI95%

.44

.22

2.77

Boot SE

[-.13, 1.67]

[.26, 1.11]

[-10.24, .60]

BCa CI95%

.87

.72

-5.66

B

.01 (.88)

.06* (6.42)

.06

.41

.20

2.85

Boot SE

[-.23, -.01]

[-.01, 1.68]

[.35, 1.03]

[-11.34, .36]

BCa CI95%

a

0 = low guidance script, 1 = high guidance script

* p \ .05, ** p \ .01, *** p \ .001

SCRGDLV script guidance level, X script guidance level, Me mediator, Mo prior knowledge, Y deep text comprehension, MeX effect of the script guidance level on the mediator, MeX*Mo interaction effect between the script guidance level and prior knowledge on the mediator, YMe.X effect of a mediator on deep text comprehension while controlling for the script guidance level; YMe*Mo.X interaction effect between the mediator and prior knowledge on deep text comprehension while controlling for the script guidance level. All continuous predictors were centered prior to analyses. Unstandardized regression coefficients are reported. Accelerated and bias-corrected bootstrap confidence intervals indicating significant regression weights are written in bold

DR (DF)

2

.23*** (8.28)

.82

.72

-4.97

B

Conditional b-path (YMe*Mo.X)

.28*** (8.21)

.16** (5.12)

-.09

.10

2.65

B

Unconditional b-path (YMe.X)

R2 (F)

[-.05, .15]

[1.20, 4.10]

BCa CI95%

Conditional a-path (MeX*Mo)

-.11

.15** (7.26)

.73

Boot SE

2.65

B

Unconditional a-path (MeX)

Interactive (discussion) 9 prior knowledge

Interactive (discussion)

SCRGDLVa 9 Prior knowledge

Prior knowledge

SCRGDLVa

Predictor

Table 6 Regression results for the moderated mediation analysis with interactive discussion activities as mediator (N = 88)

456 S. Mende et al.

.47

.27

3.36

.07

.21

3.79

[-.23, .03]

[.29, 1.14]

[-8.09, 9.00]

-.16

.79

2.14

B

.01 (2.49)

.09** (10.96)

.01

.07

.17

3.74

Boot SE

[-.04, -.01]

[-.30, -.04]

[.42, 1.08]

[-7.35, 11.21]

BCa CI95%

a

0 = low guidance script, 1 = high guidance script

* p \ .05, ** p \ .01, *** p \ .001

SCRGDLV script guidance level, X script guidance level, Me mediator, Mo prior knowledge, Y deep text comprehension, MeX effect of the script guidance level on the mediator, MeX*Mo interaction effect between the script guidance level and prior knowledge on the mediator, YMe.X effect of a mediator on deep text comprehension while controlling for the script guidance level; YMe*Mo.X interaction effect between the mediator and prior knowledge on deep text comprehension while controlling for the script guidance level. All continuous predictors were centered prior to analyses. Unstandardized regression coefficients are reported. Accelerated and bias-corrected bootstrap confidence intervals indicating significant regression weights are written in bold

DR (DF)

2

.20*** (6.99)

-.09

.74

.68

BCa CI95%

.29*** (8.61)

[-1.69, .05]

[-.34, .68]

[30.53, 43.04]

Boot SE

R2 (F)

.58*** (38.73)

-.73

.17

36.91

B

Conditional b-path (YMe*Mo.X)

-.02

[-.71, .21]

[30.77, 43.52]

BCa CI95%

Unconditional b-path (YMe.X)

Learning time 9 prior knowledge

Learning time

.24

3.37

.57*** (55.86)

-.20

Prior knowledge

SCRGDLVa 9 Prior knowledge

36.91

Boot SE

B

BCa CI95%

B

Boot SE

Conditional a-path (MeX*Mo)

Unconditional a-path (MeX)

SCRGDLVa

Predictor

Table 7 Regression results for the moderated mediation analysis with learning time as mediator (N = 88)

Who benefits from which CSCL script and why? 457

123

458

S. Mende et al.

Fig. 4 Summary of the moderated mediation analyses (research question 2). Unmoderated a-paths and b-paths are indicated by solid lines and labelled with the according significant main effect. Moderated a-paths and b-paths (indicated by dotted lines) as well as a*b-paths are labelled with simple slopes, consonant with prior knowledge at one standard deviation below the mean (L), at the mean (M) and one standard deviation above the mean (H). Related results of the bootstrapping tests for significance are presented in Appendix 3. Note that no moderation of the total effect was considered in the moderated mediation models examined. Unstandardized regression weights are reported. All continuous predictors were centered prior to the analyses

prior knowledge learners’ deep text comprehension to the degree it induced them to perform more constructive activities.

Interactive discussion activities We found a significant moderated mediation effect: Mediated via interactive activities the script guidance level had a significantly positive effect on deep text comprehension, but only for participants with prior knowledge at one standard deviation below the mean and at the mean (Fig. 4, indirect path). Consulting the a- and b-path analyses (Table 6) shows that although the high guidance script induced more interactive activities than the low guidance script for all learners irrespective of prior knowledge (Fig. 4, a-path), the interactive activities were, in turn, only conducive to low and average prior knowledge learners’ deep text comprehension (Fig. 4, b-path). In sum, compared to the low guidance script, the high guidance script benefitted low and average prior knowledge learners’ deep text comprehension to the extent it induced more interactive activities.

Learning time The results reveal a significant moderated mediation effect: Mediated through the learning time the script guidance level had a significantly negative effect on deep text comprehension, but only for learners with prior knowledge at the mean and one standard deviation above (Fig. 4, indirect path). Consulting the a- and b-path analyses (Table 7) shows that although the high guidance script increased learning time compared to the low guidance

123

Who benefits from which CSCL script and why?

459

script for all learners irrespective of prior knowledge (Fig. 4, a-path), this learning time increase was detrimental only to average and high prior knowledge learners’ deep text comprehension (Fig. 4, b-path). In sum, the high guidance script (as compared to the low guidance script) was detrimental to average and high prior knowledge learners’ deep comprehension to the extent it led to a higher learning time.

Discussion The present study examined the question of how a low versus a high guidance level CSCL script targeted at fostering interactive activities may interact with learners’ prior knowledge in affecting deep text comprehension outcomes. We found prior knowledge to be a significant moderator for the effects of the script guidance level on deep text comprehension (research question 1). That is, learning under the high guidance READ script has been slightly more helpful for low prior knowledge learners than learning with the low guidance script. By contrast, high prior knowledge learners experienced disadvantages from the high guidance script. The findings of the moderated mediation analyses (research question 2) provide possible explanations for this pattern of results. We will discuss the following questions: why did low prior knowledge learners benefit from the high guidance script? Why were high prior knowledge learners impeded by the high guidance script?

Why did low prior knowledge learners benefit from the high guidance script? The high guidance script successfully induced low prior knowledge learners to perform the targeted interactive activities which were, in turn, conducive to their deep text comprehension. With lower guidance, low prior knowledge learners relied primarily on active activities and performed an equally low number of constructive activities and fewer interactive activities than when provided with the higher guidance of the READ script. Hence, low prior knowledge learners were under-scripted in the low guidance condition (cf. Stegmann et al. 2011). In line with prior CSCL and individual learning research, this suggests that low prior knowledge learners often need additional guidance to go beyond shallow learning activities (e.g., Hmelo et al. 2000; Kirschner et al. 2006). Active activities were negatively associated with low but not high prior knowledge learners’ deep comprehension. This suggests that while high prior knowledge learners may be able to activate their existing deep knowledge structures through active activities (cf. Chi and Wylie 2014) there might be a greater need for low prior knowledge learners to build deep knowledge structures by constructive or interactive activities. Finally, interactive activities were conducive to low but not high prior knowledge learners’ deep comprehension. This indicates that low prior knowledge learners profited from using the co-learner as an additional learning resource to draw inferences in the service of deep text comprehension (Chi and Wylie 2014; Deiglmayr and Schalk 2015; Nokes-Malach et al. 2015).

Why were high prior knowledge learners impeded by the high guidance script? In the low guidance script condition, high prior knowledge learners performed more constructive activities, spent a shorter learning time, and achieved a higher level of

123

460

S. Mende et al.

performance in terms of deep comprehension than in the high guidance condition. Similar to the low prior knowledge learners, the high guidance script guided high prior knowledge learners to perform more interactive activities within a longer learning time. Yet, interactive activities were, in turn, not conducive to high prior knowledge learners’ deep comprehension. By contrast, drawing inferences based primarily on their own knowledge (constructive activities) benefitted their deep comprehension. This pattern of results is in line with the assumptions (a) that high prior knowledge learners may spontaneously draw more inferences than low prior knowledge learners—but more likely by their own rather than by referring to co-learners’ contributions (cf. Janssen et al. 2010; Nokes-Malach et al. 2012) and (b) that doing so is also more beneficial for them in terms of deep comprehension (cf. Kintsch 2004; Nokes-Malach et al. 2015; Wittwer and Renkl 2008). Hence, as the high guidance script prevented high prior knowledge learners from performing a learning activity spontaneously that was more effective for them (constructive activities), over-scripting seems to have occurred. This would explain why the general learning time increase due to the high guidance script was detrimental for high but not low prior knowledge learners: Dealing with a script which provides redundant guidance may have decreased the effective learning time, which was, in turn, detrimental to deep comprehension (cf. Stegmann et al. 2011).

Limitations At least three limitations concerning the generalizability of the present study should be pointed out. First, the dyads were formed randomly due to methodological reasons. That is, the learning dyads could have been homogeneous or heterogeneous regarding their members´ prior knowledge. While maintaining the ecological validity of dyads with different prerequisites, a homogeneous or heterogeneous a priori grouping with respect to prior knowledge would have allowed for examining the effects of script guidance level (and collaborative activities) more systematically (cf. Nokes-Malach et al. 2012). Second, although the high-guidance READ-script led to more interactive activities than the low guidance script, the absolute number of discussion phases (out of 22) in which learners showed interactive activities was still fairly low. The treatment check revealed that the open self-explanation prompt applied in the individual externalization phases (E) of the READ-script was only modest in fostering learners to draw inferences. Using a prompt that more directly guides the inferencing might have been more effective in this concern (cf. Nokes et al. 2011). In addition, the awareness induction phase (A) was not accompanied by specific scaffolds. Implementing tools which support learners to systematically extract information about a co-learner’s knowledge and understanding could have been more effective in stimulating mutual exchange (Engelmann et al. 2009; Janssen and Bodemer 2013). Finally, no scaffolds were provided in the discussion phases (D) which could have fostered interactive activities (Fischer et al. 2013a; Weinberger 2011). Third, with the present experimental design the effects of the READ-script were investigated as a whole. This does not allow disentangling the effects of its single elements (i.e. reading, externalizing, awareness induction, discussion). Hence, it remains unclear, to what extent the different READ-components contributed to the pattern of results.

Future research In the present study ICAP (Chi and Wylie 2014) was used as overarching framework for investigating the effects of varying script-guidance levels in CSCL. Following prior

123

Who benefits from which CSCL script and why?

461

suggestions to extent the majority of the previous ICAP research (Deiglmayr et al. 2015) we did not only measure deep comprehension effects of the scripts and subsequently draw conclusions about whether the scripts which were intended to induce interactive learning activities did so. We rather also investigated which observable activities the learners actually executed. In addition, we considered the role of prior knowledge as potential moderator. The present findings offer two interesting points for the extension of ICAP research. First, the ICAP framework provides comprehensive information on which tasks or instructions can principally induce a certain ICAP-activity. However, less is known about how specific task characteristics (e.g., the amount of supportive information or guidance, cf. Proske et al. 2012) or learner characteristics (e.g., prior knowledge) may affect learners’ ability to really carry out the requested activity. For example, a learner might understand that the task requires him to draw inferences (a constructive activity) but he may hardly be able to do so due to a lack of necessary prior knowledge. Altogether, our results indicate that there can be a ‘‘too little’’ and also a ‘‘too much’’ of guidance with respect to learners’ prior knowledge. Considering the interplay of these variables in future research could add valuable information for how to design appropriate forms of guidance by taking into account learners’ prerequisites in order to reliably induce a desired learning activity. Second, the ICAP framework provides well supported assumptions concerning the effects of the distinguished learning activity classes on learning outcomes. However, it is largely unknown if and how moderating conditions affect the relationships between learners’ executed ICAP activities and their learning outcomes. In the present study, the proposed superiority of interactive over constructive activities was affected by learners’ prior knowledge. This indicates that the demands of a learning situation must be high enough in relation to learners’ prior knowledge so that learners can profit from collaboration and interactive activities in terms of deep comprehension outcomes (cf. Fischer et al. 2013b; Nokes-Malach et al. 2015). Hence, future studies in the field of CSCL should consider the interplay between executed (ICAP) learning activities, prior knowledge and the demands of the learning situation in more detail.

Conclusion Our study indicates that prior knowledge is an important moderator concerning the effects of the CSCL script guidance level on deep comprehension that should be considered in further research. Additionally, considering the activities executed by the learners in the CSCL process has provided valuable insights into the dynamics of CSCL which would have been overlooked otherwise. Specifically, our analyses revealed that the ICAP assumption of interactive activities being superior to constructive activities may have to be qualified. Further research is needed to examine the role of prior knowledge and further moderators for the effects of the ICAP activities on deep comprehension. In sum, the present study suggests that CSCL script designers shouldn’t base their work on the idea of ‘‘one size fits all’’ but rather should tailor their design decisions to learners’ prior knowledge. Acknowledgements We would like to thank Professor Cindy Hmelo-Silver and three anonymous reviewers for their very valuable critiques and suggestions.

123

462

S. Mende et al.

Appendix 1 Self-explanation prompt adapted from Chi et al. (1994, p. 477) Please explain in this text box what the text paragraph means to you, that is • What new information does it provide to you? • How does it relate to what you have already read? • Does it give you a new insight into your understanding how the circulatory system works? • Does it raise questions in your mind? Tell us whatever is going through your mind, even if it seems unimportant!

Appendix 2 Examples of posttest questions for the assessment of deep text comprehension adapted from Chi et al. (2001, p. 523, 525). What results at the cellular level from having a hole in the septum? (a) (b) (c) (d)

This can result in the carbon dioxide concentration being increased in the body cells X This can result in the body cells no longer receiving any oxygen This can result in the blood of the pulmonary veins being less oxygenated This can result in the lungs having to take up more carbon dioxide

Where is the failure located, in most instances if the heart stops functioning properly? (a) (b) (c) (d)

Left atrium Left ventricle X Right atrium Right ventricle

To identify, for instance, the correct answer to the question of where the failure is located in the most cases when the heart stops working properly, one has to make the following line of reasoning: based on the text information that the atria pumps blood to the ventricles, the right ventricle pumps blood to the lungs and the left ventricle pumps blood to the whole body, one has to insert the information that pumping blood in the whole body (a great area with long distances) is a task requiring more physical effort than the tasks of transporting blood in the ventricles or lungs from prior knowledge. Based on the knowledge that more effort causes a greater physical strain, this allows the conclusion that, in the majority of cases, the left ventricle is the cause of heart problems (cf. Chi et al. 2001).

Appendix 3 See Table 8.

123

2.65

36.91

Interactive (M3)

Learning time (M4)

2.16

-3.48

Interactive (M3)

Learning time (M4)

2.74

-2.47

Constructive (M2)

Active (M1)

a*b-paths (MeX*YMe.X)

-2.81

-.09

Learning time (M4)

c-path (YX)

.84

.82

Constructive (M2)

Interactive (M3)

Active (M1)

-.28

-2.94

Constructive (M2)

b-paths (YMe.X)

-9.92

2.37

1.27

1.85

2.79

2.65

.07

.44

.64

.30

3.37

.73

.51

1.01

[-8.07, 1.47]

[.12, 5.15]

[-6.46, .76]

[-2.66, 8.26]

[-7.90, 2.51]

[-.23, .03]

[-.13, 1.67]

[-.39. 2.31]

[-.93, .28]

[30.77, 43.52]

[1.20, 4.10]

[-4.00, -1.80]

[-11.88, -7.90]

.78

5.78

.78

8.17

.02

1.73

-.61

-.79

–

–

-1.27

–

2.82

2.91

1.32

3.74

.07

.65

.85

.33

–

–

.60

–

Boot SE

[-4.19, 7.02]

[1.36, 12.98]

[-.91, 4.97]

[2.37, 17.59]

[-.12, .19]

[.45, 3.06]

[-2.31, 1.56]

[-1.47, -.20]

–

–

[-2.52, .03]

–

BCa CI95%

B

BCa CI95%

B

Boot SE

Prior knowledge at -1SD

Overall effect

Active (M1)

a-paths (MeX)

Path

Table 8 Summary of the results of mediation and moderated mediation analyses

-5.96

2.29

-1.13

2.66

-.16

.87

.38

-.27

–

–

-2.93

–

B

2.54

1.38

1.91

2.97

.07

.41

.68

.31

–

–

.49

–

Boot SE

[-11.24, -.98]

[.04, 5.89]

[-5.07, 2.87]

[-2.62, 9.61]

[-.30, -.04]

[.09, 1.70]

[-.94, 1.98]

[-.94, .34]

–

–

[-3.91, -1.94]

–

BCa CI95%

Prior knowledge at M

-.34

-.01

1.38

.26

-10.73

-.01

-6.35

-2.44

–

–

-4.60

–

B

3.31

1.38

2.96

3.22

.11

.62

.66

.35

–

–

.69

–

Boot SE

[-18.51, -5.06]

[-3.26, 2.34]

[-12.11, -.51]

[-9.37, 3.67]

[-.56, -.15]

[-1.07. 1.10]

[.22, 2.54]

[-.44, .93]

–

–

[-5.93, -3.29]

–

BCa CI95%

Prior knowledge at ? 1SD

Who benefits from which CSCL script and why? 463

123

123

YX.Me

3.79

2.77

3.14

4.16

[-8.09, 9.00]

[-10.24, .60]

[-7.87, 6.41]

[-15.66, 3.24]

BCa CI95%

Boot SE

2.14

-5.66

-.05

-5.66

3.74

2.85

3.22

4.22

YX.Me*Mo

B

[-7.35, 11.21]

[-11.34, .36]

[-7.67, 6.36]

[-16.22, 3.46]

BCa CI95%

Prior knowledge at M B

Boot SE

BCa CI95%

Prior knowledge at ? 1SD

X script guidance level (0 = low guidance, 1 = high guidance), Y deep text comprehension, Mo prior knowledge, Mei mediator, YX effect of the script guidance level on deep text comprehension, MeX effect of the script guidance level on a mediator, YMe.X effect of a mediator on deep text comprehension while controlling for the script guidance level, MeX*YMe.X product of MeX and YMe.X, YX.Me effect of condition on deep text comprehension while controlling for a mediator, YX.Me*Mo effect of condition on deep text comprehension while controlling for a mediator, prior knowledge and their interaction. All continuous predictors were centered prior to analyses. Unstandardized regression coefficients are reported. Accelerated and bias-corrected bootstrap confidence intervals indicating significant regression weights are written in bold. Simple slopes for three values of prior knowledge (one SD below the mean, at the mean and one SD above the mean) are only presented in cases where prior knowledge moderated a path significantly

.68

-4.97

Interactive (M3)

Learning time (M4)

-.34

-5.55

Constructive (M2)

Active (M1)

Boot SE

B

BCa CI95%

B

Boot SE

Prior knowledge at -1SD

Overall effect

c0 -paths (YX.Me and YX.Me*Mo)

Path

Table 8 continued

464 S. Mende et al.

Who benefits from which CSCL script and why?

465

References Afifi, A. A., Kotlerman, J. B., Ettner, S. L., & Cowan, M. (2007). Methods for improving regression analysis for skewed continuous or counted responses. Annual Review of Public Health, 28, 95–111. doi:10. 1146/annurev.publhealth.28.082206.094100. Berkowitz, M. W., Althof, W., Turner, V. D., & Bloch, D. (2008). Discourse, development, and education. In F. K. Oser & W. Veugelers (Eds.), Getting involved: Global citizenship development and sources of moral values (pp. 189–201). Rotterdam: Sense Publishers. doi:10.1080/03057240903101705. Berkowitz, M. W., & Gibbs, J. C. (1983). Measuring the developmental features of moral discussion. Merrill-Palmer Quarterly, 29(4), 399–410. Best, R. M., Rowe, M., Ozuru, Y., & McNamara, D. S. (2005). Deep-level comprehension of science texts: The role of the reader and the text. Topics in Language Disorders, 25(1), 65–83. doi:10.1097/ 00011363-200501000-00007. Bromme, R., Hesse, F. W., & Spada, H. (2005). Barriers and biases in computer-mediated knowledge communication. And how they may be overcome. New York: Springer. doi:10.1007/b105100. Chan, C. K., Burtis, P. J., Scardamalia, M., & Bereiter, C. (1992). Constructive activity in learning from text. American Educational Research Journal, 29(1), 97–118. doi:10.2307/1162903. Chi, M. T. H. (2000). Self-explaining: The dual processes of generating inference and repairing mental models. In R. Glaser (Ed.), Advances in instructional psychology (pp. 161–238). Mahwah, NJ: Lawrence Erlbaum Associates Publishers. Chi, M. T. H. (2009). Active-constructive-interactive: A conceptual framework for differentiating learning activities. Topics in Cognitive Science, 1(1), 73–105. doi:10.1111/j.1756-8765.2008.01005.x. Chi, M. T. H., de Leeuw, N., Chiu, M. H., & LaVancher, C. (1994). Eliciting self-explanations improves understanding. Cognitive Science, 18(3), 439–477. Chi, M. T. H., & Menekse, M. (2015). Dialogue patterns in peer collaboration that promote learning. In L. B. Resnick, C. Asterhan, & S. Clarke (Eds.), Socializing intelligence through academic talk and dialogue (pp. 263–274). Washington, DC: American Educational Research Association. Chi, M. T. H., Siler, S. A., Jeong, H., Yamauchi, T., & Hausmann, R. G. (2001). Learning from human tutoring. Cognitive Science, 25(4), 471–533. doi:10.1016/S0364-0213(01)00044-1. Chi, M. T. H., & Wylie, R. (2014). The ICAP framework: Linking cognitive engagement to active learning outcomes. Educational Psychologist, 49(4), 219–243. doi:10.1080/00461520.2014.965823. Chiu, J. L., & Chi, M. T. H. (2014). Supporting self-explanation in the classroom. In V. A. Benassi, C. E. Overson, & C. M. Hakala (Eds.), Applying science of learning in education: Infusing psychological science into the curriculum (pp. 91–103). Washington, DC: Society for the Teaching of Psychology. Cohen, E. G. (1994). Restructuring the classroom: Conditions for productive small groups. Review of Educational Research, 64(1), 1–35. doi:10.2307/1170744. Cote´, N., Goldman, S. R., & Saul, E. U. (1998). Students making sense of informational text: Relations between processing and representation. Discourse Processes, 25(1), 1–53. doi:10.1080/ 01638539809545019. Cronbach, L. J., & Snow, R. E. (1977). Aptitudes and instructional methods: A handbook for research on interactions. Oxford: Irvington. De Backer, L., Van Keer, H., & Valcke, M. (2014). Socially shared metacognitive regulation during reciprocal peer tutoring: Identifying its relationship with students’ content processing and transactive discussions. Instructional Science, 43(3), 323–344. doi:10.1007/s11251-014-9335-4. Deiglmayr, A., Rummel, N., & Loibl, K. (2015). The mediating role of interactive learning activities in CSCL: An INPUT-PROCESS-OUTCOME model. The 11th Conference on Computer Supported Collaborative Learning (CSCL), Gothenburg, Sweden. Deiglmayr, A., & Schalk, L. (2015). Weak versus strong knowledge interdependence: A comparison of two rationales for distributing information among learners in collaborative learning settings. Learning and Instruction, 40, 69–78. doi:10.1016/j.learninstruc.2015.08.003. Diehl, M., & Stroebe, W. (1987). Productivity loss in brainstorming groups: Toward the solution of a riddle. Journal of Personality and Social Psychology, 53(3), 497–509. doi:10.1037/0022-3514.53.3.497. Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving students’ learning with effective learning techniques: Promising directions from cognitive and educational psychology. Psychological Science in the Public Interest, 14(1), 4–58. doi:10.1177/ 1529100612453266. Engelmann, T., Dehler, J., Bodemer, D., & Buder, J. (2009). Knowledge awareness in CSCL: A psychological perspective. Computers in Human Behavior, 25(4), 949–960. doi:10.1016/j.chb.2009.04.004. Ertl, B., Kopp, B., & Mandl, H. (2004). Effects of individual prior knowledge on collaborative knowledge construction and individual learning outcome in videoconferencing. (Research Report No. 171).

123

466

S. Mende et al.

Munich: Ludwig-Maximilians-University, Department of Psychology, Institute for Educational Psychology. Fiorella, L., & Mayer, R. E. (2016). Eight ways to promote generative learning. Educational Psychology Review, 28(4), 717–741. Fischer, F., Kollar, I., Stegmann, K., & Wecker, C. (2013a). Toward a script theory of guidance in computersupported collaborative learning. Educational Psychologist, 48(1), 56–66. doi:10.1080/00461520. 2012.748005. Fischer, F., Kollar, I., Stegmann, K., Wecker, C., Zottmann, J., & Weinberger, A. (2013b). Collaboration scripts in computer-supported collaborative learning. In C. E. Hmelo-Silver, C. A. Chinn, C. K. K. ´ Donnell (Eds.), The international handbook of collaborative learning (pp. 403–419). Chan, & A. O New York: Routledge. Fonseca, B. A., & Chi, M. T. H. (2011). Instruction based on self-explanation. In R. E. Mayer & P. A. Alexander (Eds.), The handbook of research on learning and instruction (pp. 296–321). New York: Routledge. Gadgil, S., & Nokes-Malach, T. (2012). Overcoming collaborative inhibition through error correction: A classroom experiment. Applied Cognitive Psychology, 26(3), 410–420. doi:10.1002/acp.1843. Gijlers, H., & de Jong, T. (2009). Sharing and confronting propositions in collaborative inquiry learning. Cognition and Instruction, 27(3), 239–268. doi:10.1080/07370000903014352. Ha¨kkinen, P., & Ma¨kitalo-Siegl, K. (2007). Educational perspectives on scripting CSCL. In F. Fischer, I. Kollar, H. Mandl, & J. Haake (Eds.), Scripting computer-supported collaborative learning—Cognitive, computational and educational approaches (pp. 263–271). US: Springer. Hayes, A. F. (2013). Introduction to mediation, moderation, and conditional process analysis: A regressionbased approach. New York: Guilford Press. Hmelo, C. E., Nagarajan, A., & Day, R. S. (2000). Effects of high and low prior knowledge on construction of a joint problem space. Journal of Experimental Education, 69(1), 36–56. doi:10.1080/ 00220970009600648. Janssen, J., & Bodemer, D. (2013). Coordinated computer-supported collaborative learning: Awareness and awareness tools. Educational Psychologist, 48(1), 40–55. doi:10.1080/00461520.2012.749153. Janssen, J., Erkens, G., & Kanselaar, G. (2007). Visualization of agreement and discussion processes during computer-supported collaborative learning. Computers in Human Behavior, 23(3), 1105–1125. doi:10. 1016/j.chb.2006.10.005. Janssen, J., Kirschner, F., Erkens, G., Kirschner, P. A., & Paas, F. (2010). Making the black box of collaborative learning transparent: Combining process-oriented and cognitive load approaches. Educational Psychology Review, 22(2), 139–154. doi:10.1007/s10648-010-9131-x. Jeong, H., & Chi, M. T. H. (2007). Knowledge convergence and collaborative learning. Instructional Science, 35(4), 287–315. doi:10.1007/s11251-006-9008-z. Jeong, H., & Hmelo-Silver, C. (2016). Seven affordances of computer-supported collaborative learning: How to support collaborative learning? How can technologies help? Educational Psychologist, 51(2), 247–265. doi:10.1080/00461520.2016.1158654. Kalyuga, S., Ayres, P., Chandler, P., & Sweller, J. (2003). The expertise reversal effect. Educational Psychologist, 38(1), 23–31. doi:10.1207/S15326985EP3801_4. King, A. (1999). Discourse patterns for mediating peer learning. In A. M. O’Donnell & A. King (Eds.), Cognitive perspectives on peer learning (pp. 87–115). Mahwah, NJ: Lawrence Erlbaum Associates Publishers. Kintsch, W. (1998). Comprehension: A paradigm for cognition. New York: Cambridge University Press. Kintsch, W. (2004). The construction-integration model of text comprehension and its implications for instruction. In R. Ruddell & N. Unrau (Eds.), Theroretiocal models and processes of reading (5th ed., pp. 1270–1328). Newark: International Reading Association. Kirschner, P. A., & Erkens, G. (2013). Toward a framework for CSCL research. Educational Psychologist, 48(1), 1–8. doi:10.1080/00461520.2012.750227. Kirschner, P. A., Sweller, J., & Clark, R. E. (2006). Why minimal guidance during instruction does not work: An analysis of the failure of constructivist, discovery, problem-based, experiential, and inquirybased teaching. Educational Psychologist, 41(2), 75–86. doi:10.1207/s15326985ep4102_1. Kollar, I., Fischer, F., & Hesse, F. (2006). Collaboration scripts—A conceptual analysis. Educational Psychology Review, 18(2), 159–185. doi:10.1007/s10648-006-9007-2. Lam, R. J. (2013). Maximizing the benefits of collaborative learning in the college classroom. Paper presented at the Annual Meeting of the Cognitive Sciences Society, Berlin. McNamara, D. S. (2004). SERT: Self-explanation reading training. Discourse Processes, 38(1), 1–30. doi:10.1207/s15326950dp3801_1.

123

Who benefits from which CSCL script and why?

467

McNamara, D. S., & Magliano, J. P. (2009a). Self-explanation and metacognition: The dynamics of reading. In D. J. Hacker, J. Dunlosky, & A. C. Graesser (Eds.), Handbook of metacognition in education (pp. 60–81). New York, NY: Routledge. McNamara, D. S., & Magliano, J. P. (2009b). Toward a comprehensive model of comprehension. In B. H. Ross (Ed.), The psychology of learning and motivation (pp. 297–384). San Diego, CA: Elsevier Academic Press. doi:10.1016/S0079-7421(09)51009-2. Neal, D. J., & Simons, J. S. (2007). Inference in regression models of heavily skewed alcohol use data: A comparison of ordinary least squares, generalized linear models, and bootstrap resampling. Psychology of Addictive Behaviors, 21(4), 441–452. doi:10.1037/0893-164X.21.4.441. Nokes, T. J., Hausmann, R. G. M., VanLehn, K., & Gershman, S. (2011). Testing the instructional fit hypothesis: The case of self-explanation prompts. Instructional Science, 39(5), 645–666. doi:10.1007/ s11251-010-9151-4. Nokes-Malach, T., Meade, M. L., & Morrow, D. G. (2012). The effect of expertise on collaborative problem solving. Thinking & Reasoning, 18(1), 32–58. doi:10.1080/13546783.2011.642206. Nokes-Malach, T., Richey, J. E., & Gadgil, S. (2015). When is it better to learn together? Insights from research on collaborative learning. Educational Psychology Review, 27(4), 645–656. doi:10.1007/ s10648-015-9312-8. Noroozi, O., Teasley, S. D., Biemans, H. J. A., Weinberger, A., & Mulder, M. (2013). Facilitating learning in multidisciplinary groups with transactive CSCL scripts. International Journal of Computer-Supported Collaborative Learning, 8(2), 189–223. doi:10.1007/s11412-012-9162-z. O’Donnell, A. M. (1999). Structuring dyadic interaction through scripted cooperation. In A. M. O’Donnell & A. King (Eds.), Cognitive perspectives on peer learning (pp. 179–196). Mahwah, NJ: Lawrence Erlbaum Associates Publishers. Pfister, H. R. (2005). How to support synchronous net-based learning discourses: Principles and perspectives. In R. Bromme, F. W. Hesse, & H. Spada (Eds.), Barriers and biases in computer-mediated knowledge communication: And how they may be overcome (pp. 39–57). New York: Springer. Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior Research Methods, 40(3), 879–891. doi:10. 3758/BRM.40.3.879. Proske, A., Ko¨rndle, H., & Narciss, S. (2012). Interactive learning tasks. In N. M. Seel (Ed.), Encyclopedia of the sciences of learning (pp. 1606–1610). New York: Springer. Renkl, A. (1999). Learning mathematics from worked-out examples: Analyzing and fostering self-explanations. European Journal of Psychology of Education, 14(4), 477–488. doi:10.1007/BF03172974. Rietzschel, E. F., Nijstad, B. A., & Stroebe, W. (2006). Productivity is not enough: A comparison of interactive and nominal brainstorming groups on idea generation and selection. Journal of Experimental Social Psychology, 42(2), 244–251. doi:10.1016/j.jesp.2005.04.005. Roscoe, R. D. (2014). Self-monitoring and knowledge-building in learning by teaching. Instructional Science, 42(3), 327–351. doi:10.1007/s11251-013-9283-4. Roscoe, R. D., & Chi, M. T. H. (2007). Understanding tutor learning: Knowledge-building and knowledgetelling in peer tutors’ explanations and questions. Review of Educational Research, 77(4), 534–574. doi:10.3102/0034654307309920. Roscoe, R. D., & Chi, M. T. H. (2008). Tutor learning: The role of explaining and responding to questions. Instructional Science, 36(4), 321–350. doi:10.1007/s11251-007-9034-5. Rummel, N., & Spada, H. (2005). Learning to collaborate: An instructional approach to promoting collaborative problem solving in computer-mediated settings. Journal of the Learning Sciences, 14(2), 201–241. doi:10.1207/s15327809jls1402_2. Stegmann, K., Mu, J., Gehlen-Baum, V., & Fischer, F. (2011). The myth of over-scripting: Can novices be supported too much? In H. Spada, G. Stahl, N. Miyake, & N. Law (Eds.), Connecting computersupported collaborative learning to policy and practice: CSCL2011 conference proceedings volume I—Long papers (pp. 406–413). Hong Kong: International Society of the Learning Sciences. Teasley, S. (1997). Talking about reasoning: How important is the peer in peer collaboration? In L. B. Resnick, R. Sa¨ljo¨, C. Pontecorvo, & B. Burge (Eds.), Discourse, tools and reasoning: Essays on situated cognition (pp. 361–384). Berlin: Springer. Trauzettel-Klosinski, S., & Dietz, K. (2012). Standardized assessment of reading performance: The new international reading speed texts IReST. Investigative Ophthalmology & Visual Science, 53(9), 5452–5461. doi:10.1167/iovs.11-8284. VanLehn, K., Graesser, A. C., Jackson, G. T., Jordan, P., Olney, A., & Rose´, C. P. (2007). When are tutorial dialogues more effective than reading? Cognitive Science, 31(1), 1–60. doi:10.1080/ 03640210709336984.

123

468

S. Mende et al.

Vogel, F., Wecker, C., Kollar, I., & Fischer, F. (2016). Socio-cognitive scaffolding with computer-supported collaboration scripts: A meta-analysis. Educational Psychology Review. doi:10.1007/s10648-0169361-7. Webb, N. M. (1989). Peer interaction and learning in small groups. International Journal of Educational Research, 13(1), 21–39. Weinberger, A. (2011). Principles of transactive computer-supported collaboration scripts. Journal of Digital Literacy, 6(3), 189–202. Weinberger, A., Kollar, I., Dimitriadis, Y., Ma¨kitalo-Siegl, K., & Fischer, F. (2009). Computer-supported collaboration scripts. Perspectives from educational psychology and computer science. In N. Balachef, S. R. Ludvigsen, T. de Jong, S. Barnes, & A. W. Lazonder (Eds.), Technology-enhanced learning. principles and products (pp. 155–173). Berlin: Springer. Wittrock, M. C. (1989). Generative processes of comprehension. Educational Psychologist, 24(4), 345–376. doi:10.1207/s15326985ep2404_2. Wittwer, J., & Renkl, A. (2008). Why instructional explanations often do not work: A framework for understanding the effectiveness of instructional explanations. Educational Psychologist, 43(1), 49–64. doi:10.1080/00461520701756420.

123