Stimulating collaboration by providing insights in the participation of students in a CSCL environment

(1)

Stimulating collaboration by providing

insights in the participation of students in a

CSCL environment

July 2, 2017

Student: Hannah Lim 10588973 Supervisor: Bert Bredeweg Course: Thesis Project AI Course code: 5082ABK18Y Abstract

For complex domains of science conceptual modeling and collaborative learning are beneficial for learning results. In this research an attempt is made to combine conceptual modeling and collaborative learning. Previous research has shown that collaborative learning does not necessarily occur when students are set in a collaborative environment. If the environment does not satisfy the requirements for collaboration ”free-riding” may occur. Free-riding means that a student is leaning on a collective effort without or by less contributing then others. Therefore in this research a collaborative modeling environment is created, to which a feedback tool using learning analytics is added. The feedback tool provides insights in the distribution of the work of the students. This environment is created in DynaLearn, which is an online learning environment that makes use of conceptual modeling. The results seem to indicate that the feedback tool influences collaboration positively. Furthermore, with the results obtained in this research it becomes possible to detect free-riders.

(2)

Thesis Project AI

1 Introduction

For effective learning in cognitive complex domains of science it is beneficial for students to engage in modeling the subject matter [Schwarz and White, 2005]. According to Schwarz and White modeling is a core component of science as well as for human cognition. Therefore, involving students in a process of creating, testing and revising models, will help them develop their thinking and positively influences learning results and the development of subject matter expertise. Further it can improve the learning process by making the subject matter more easily accessible and interesting. A more specific form of modeling is conceptual modeling, where students are modeling theoretical concepts and perform concept predictions and explanations [Zitek et al., 2013]. Conceptual modeling requires more complex relations then merely modeling and therefore teaching students with conceptual modeling could mean an impor-tant improvement for science education.

In addition, collaborative learning in conceptual complex domains as in science education benefits learn-ers in comparison to working individually [Johnson and Johnson, 1999, Kirschner et al., 2011] because students are able to verbalize their thoughts in ways that clearly articulate and explain their opinions and interpretations [Andrews and Rapp, 2015]. However, collaborative learning does not necessarily occur when learners are assigned to work collaboratively [Andrews and Rapp, 2015]. Therefore Johnson and Johnson have conceived five basic elements for collaboration: (1) positive interdependence, (2) in-dividual accountability, (3) face-to-face promotive interaction, (4) social skills and (5) group processing (see Section 2.2). If these basic elements are not included in the cooperation, collaborative learning will probably not occur and additional the chance of “free-riding” increases. Free-riding means that a student is leaning on a collective effort without or by less contributing then others [Belgiorno et al., 2010]. When free-riding occurs, other students tend to be less motivated and the group performance could decrease.

Therefore, for science education it could be desirable to create a collaborative modeling environment. With the benefits of creating, testing and revising models by modeling on the one hand and the ben-efits of verbalizing thoughts of collaborative learning on the other hand, better learning results could be achieved. Therefore, this study attempts to create a collaborative modeling environment. This envi-ronment is created within DynaLearn, which is an intelligent online modeling envienvi-ronment that allows learners to acquire conceptual knowledge by constructing and simulating models of how systems behave [Bredeweg et al., 2013]. Collaboration within DynaLearn implies a form of computer-supported collab-orative learning (CSCL). In general, the interaction between the learners is shaped using technologies [Dillenbourg et al., 2009]. Based on the five basic elements of collaboration [Johnson and Johnson, 1999] in this study only face-to-face interaction between learners is maintained to ensure good communication conditions.

To create an effective CSCL conceptual modeling environment this study attempts to stimulate collabo-ration and avoid the chance of free-riding by providing insights in the participation of students. To find the best method to stimulate collaboration, the following research question was posed:

How to enable effective learning in a CSCL modeling environment?

This question is narrowed down in the following sub questions: 1. What is and influences collaborative learning?

2. How can learning analytics be used to provide feedback that improves collaborative learning? 3. How can free-riding be prevented by providing feedback based on learning analytics?

4. How can the level of collaboration be measured?

The outline of this paper is as follows: In Chapter 2 a more extended explanation of DynaLearn will be discussed. In the same chapter an outline of relevant theories concerning collaboration and methods to measure the occurrence of collaboration will be presented. In Chapter 3 the methodology of the

(4)

Thesis Project AI

implementation and the evaluation of the tool will be explained. Thereafter in Chapter 4 the results of the evaluation of the tool will be presented. In Chapter 5 the conclusions will be made on basis of the results and lastly, further suggestions and work for this research will be discussed in Chapter 6.

2 Theoretical Framework

The objective of this research is to provide a CSCL environment with DynaLearn. Research has shown that effective collaboration does not occur spontaneously [Andrews and Rapp, 2015]. To transform DynaLearn into a CSCL environment, features that positively influence collaboration need to be imple-mented into the DynaLearn software. In order to do this, ideas need to be formed about how collaboration can be stimulated within DynaLearn. Therefore it is necessary to obtain knowledge about collaboration and how to influence it, and project this on the environment we can provide with DynaLearn.

This literature review is divided into three parts. First, the working of the DynaLearn program is described in combination with the learning advantages of conceptual modeling. Second, the definition and meaning of collaboration will be established with a focus on the chance and danger of free-riding. Third, related work for observing and measuring the occurrence of collaboration will be discussed.

2.1 DynaLearn

DynaLearn is an intelligent online learning environment that allows learners to acquire conceptual knowl-edge by constructing and simulating models of how systems behave [Bredeweg et al., 2013]. Conceptual modeling helps learners express and externalize their thinking, and visualize and test components of their theories. Because of this, their understanding of systems, and of the systems behaviour will be improved. In addition, the activity of modeling makes the material more interesting for learners. The program makes use of logic-based representations for expressing conceptual knowledge and qualita-tive vocabulary for users to construct their explanation about systems. DynaLearn has a workspace for knowledge construction that accommodates the true nature of conceptual knowledge. Because modeling is science related, DynaLearn is mainly useful for science education.

DynaLearn enables learners to construct a model of a system and its behaviour in the workspace of the program. These models are dynamical, which implies that they change over time. An example of such a system could be an ecosystem, like shown in figure 1. A model consists of entities, which are physical objects of the system that are connected with each other. These entities have changeable and measure-able features that explain the behaviour of the entities, which are called quantities. Every quantity has a value-range, which defines the values the quantity can have. The quantities can influence each other, causing the values of the quantities to change which makes the system dynamic. So cause-effect is of big importance with constructing systems. In DynaLearn different kind of influences can be used; direct influences and indirect influences. Direct influences initiate change while indirect influences propagate change.

In DynaLearn the learner can simulate the model he or she created. By letting an attribute decrease, remain the same or increase, the learner can initialize the model by setting a starting value for one or more attributes. After the initialization the learner can simulate the model and in this way dynamically test his theories about the system modelled. As a result the learner is expected to gain further insights in his own thinking about the system and his conceptual knowledge is expected to increase.

(5)

Thesis Project AI

(a) The initial state of the model (b) The simulation of the model

Figure 1: An example of a model in DynaLearn

2.2 Collaboration and collaborative learning

Collaboration occurs when two or more learners are working together effectively. Johnson and John-soncame up with five basic elements required for collaboration, namely: (1) positive interdependence, which contains that learners are not able to succeed unless their group mates succeed. To realize positive interdependence different influencing features could be presented, for example joint rewards, divided resources and complementary roles. (2) Individual accountability is required for a good form of col-laboration to ensure every learner will have the responsibility to participate. Examples of structuring individual accountability are taking an individual test after the collaboration, randomly selecting a stu-dent to represent the entire group and having each learner explain what they have learned to a group member. (3) Face-to-face promotive interaction makes helping, assisting, supporting, encouraging, and praising each other’s work possible to improve the collaboration. (4) Social skills are necessary for all group members as it is an important precondition for working together. (5) Group processing occurs when group members discuss their achievements and failures within the group and stimulate group co-herence. Without the occurrence of these five elements, a good form of collaboration could not occur.

Collaborative learning benefits learning outcomes in comparison to working individually [Andrews and Rapp, 2015]. This is mainly because of the active role students have in their own learning process instead of passively ‘receiving’ knowledge [Ruël et al., 2003]. Furthermore Andrews and Rapp states that collaboration leads to verbalizing thoughts, which will strengthen existing knowledge structures and may help individuals acquire new knowledge because the perspectives of others allow them to test their theories. To be able to obtain these benefits, the preconditions for establishing collaboration have to be taken into account to [Andrews and Rapp, 2015]. First, clear instructions and structure for the collaborative work must be provided. Without this clarity about what to do, there is a great possibility that collaboration will not occur. Second, the complexity of the task is of importance. A high-complexity task will improve the collaboration and the learning outcomes. If the task is to easy for the students, the learning outcomes will be better when performing the task individually. Third, the division of the ability of the students is of influence on the collaboration. Low-ability students in a homogenous group will be less likely to benefit compared to a heterogeneous group. Average-ability students benefit most from homogenous groups and high-ability students benefit from both homogenous and heterogeneous groups. Last, an equal contribution to the collaboration of all group members is necessary to assure that all group members benefit. Without this, social loafing and free-riding might occur meaning that a student decreases his effort and leans on a collective effort without or by less contributing then others [Belgiorno et al., 2010]. The free-riding student tends to score lower on an after test and the group members are more dissatisfied with their group [Ruël et al., 2003]. Free-riding in student groups usually has a negative effect on team performance. This is not necessarily the case [Ruël et al., 2003], only if the free-riding behaviour in collaboration demotivates the other group members and therefore will decrease

(6)

Thesis Project AI

the overall group performance. However free-riding is never contributing and to be able to avoid social loafing from happening it is important to know when it does or does not occur. There are several conditions, which influence the chance of free-riding, namely: (1) the type of task to be performed, (2) the number of students within a team, (3) the type of performance and reward (on an individual or a group basis), (4) the identifiability of the individual contribution and (5) certain group characteristics (like trust). These correspond with the five elements and the benefits of collaboration. If the number of students within a team is smaller, the chance of free-riding is less likely to occur. Collaboration in pairs decreases the chance of free-riding [Belgiorno et al., 2010]. Also Belgiorno states that if the sizes of the collaborating groups are big, free-riding could still be avoided by letting the students have insights in what the individual input to the collaboration is of every student.

2.3 Observing and measuring collaboration in CSCL

The research in CSCL about observing and measuring collaboration is restricted and focuses mostly on results. The passed 10 years interest in the process of collaboration is emerging, but up till now current research does not seem to offer clear handles for observing and measuring. The context of most research is distance education. Further as both Dillenbourg et al. [2009] and more recently Andrews and Rapp [2015] point out in their overviews, the collaboration process is complex and a lot of angels can be taken to improve and measure the level of collaboration. Dillenbourg et al. elaborate on the importance of research on motivation.Andrews and Rapp give a comprehensive overview on the factors that benefit and challenge collaboration and stress the importance of equal participation for effective group learn-ing. However, the overall tendency is that current research does not offer clear handles on measuring collaboration and further research is still required. Below first a short overview is given of approaches and methods for observing and measuring the collaboration process in a learning environment. Second an example is given of the application of machine learning to analyse student interactions and improve the collaboration process.

The research of Daradoumis et al. [2006] is in the context of distance education. Two measurement levels relevant for measuring collaboration are distinguished: the process (group functioning) and the product (task performance). Most of the previous research has analysed collaboration in CSCL mainly on the first measurement level focusing on communication between the learners [Daradoumis et al., 2006] [Anaya and Boticario, 2011] while most of the time the context of CSCL research is distant learning for which good communication is crucial for collaboration and can be monitored relatively easy by inspecting digital data gathered by the system.

For analysis Daradoumis et al. used high degree and low-mid/level degree as indicators. Task perfor-mance, group functioning, social support and help services are high-level interactions indicators. They found that further specification in lower degree levels was required to make an in-depth and effective evaluation of the collaboration process possible. In their study they refined only task performance and group functioning.

The research of Collazos et al. [2007] is in context of distant as well as direct interaction between team-mates. Based on the concepts of collaboration defined by Johnson and Johnson [1999] four indicators were developed that measure activities relevant for the collaboration process: use of strategies, intra-group cooperation, reviewing success criteria, monitoring and performance on the group. These indicators are measured directly from the data collected by the games and tools developed especially for this research. The activities measured are largely specific for the task and environment in which the team collaborates.

Besides the evaluation of the collaboration process with digital data generated by the system, Collazos et al. included in the evaluation attitude indicators reflecting the users’ perception about the collab-oration using aspects related with community psychology and social and educational psychology. To measure these attitude queries were used. Psychologists believe that a positive attitude to collaborative

(7)

Thesis Project AI

work can result from effective participation in collaborative learning processes and therefore can be seen as an indicator for good collaboration.

Further because students seem to need information (in a literal or visual form) on their own actions to support awareness, meta-cognition and thereby self-regulation of their learning process Daradoumis et al. [Collazos et al., 2007] both tried to present feedback on the collaboration and the task, by letting the learners evaluate themselves. Also Daradoumis et al. developed a set of indexes for a social network analysis to be able, amongst others, to identify the more and less active students, relevant because ac-cording to the research active participation is used as an indicator for good collaboration.

For the context of a working environment where employees with different expertise need to work together Baeza-Yates and Pino [2006] created a formal evaluation method for computer supported collaborative work (CSCW), in which the focus is on the final result and the efficiency of the collaboration process. The evaluation considers three aspects: Quality, Time and Work. (1) Quality is mainly dependent on the results of the task performed by the group. (2) Time describes the time used to complete the task. (3) Work signifies the amount of work done to complete the task. Collazos et al. indicates that in the CSCL context as for example created in their research where the information of the task was divided over the students, such that to complete the task they had to collaborate, this research can be of value for measuring collaboration.

Finally machine learning (ML) is a promising technique for getting a handle on the analysis of the collaboration process to be able to better support and stimulate collaboration Anaya and Boticario. With ML it is possible to analyse student collaboration regularly and frequently without interfering in the process. The ambition of Anaya and Boticario is to implement a domain independent tool for analysis of computer supported environments for distant collaborative learning. The analysis is performed through a combination of unsupervised learning, clustering and classifying students based on their interactions, and supervised learning with decisions trees, classifying student interaction according to a collaboration level to construct a set of metrics to measure collaboration. Domain independent indicators of student interactions are input for the datasets that are used. The analysis methods were validated the first time with help of an expert. The results are promising where in this context the researchers found the supervised classifying metric approach appeared more appropriate.

3 Methodology

To find out how collaboration using DynaLearn can be improved by implementations in the DynaLearn software the method was broken up in three parts corresponding with the sub-questions posed in the introduction; collecting information of influencing collaboration, implementing the tool in the DynaLearn program and lastly the evaluation of the level of collaboration of the tool implemented in an experiment.

3.1 Collecting information of influencing collaboration

To find out what influences collaboration, literature research has been done. In the Theoretical Frame-work most of this part of the question has been answered. The two findings most relevant for the implementation of the tool are the prevention of free-riding and the importance of self-regulation of learning activity. If free-riding occurs, the learning results of the ”free-rider” will decrease and the rest of the group will be less motivated. If students were provided with information on their own actions, awareness was supported and thereby self-regulation of their learning activity. If learners believed that their teammates were providing equally effort as themselves they were motivated more and were likely to work harder.

The elements for collaboration (positive independence, individual accountability, face-to-face promotive interaction, social skills and group processing) are requirements for collaboration. To ensure that these

(8)

Thesis Project AI

elements will occur, interventions would be added in the class and the task.

3.2 Implementing the tool in the DynaLearn program

DynaLearn was only available to work in individually. For this study an environment is created to provide the possibility of working collaboratively. The workplace looks the same as it did when working individually, except that the students can share their models with other students and work simultaneously within the same model. By clicking on the ‘shaking-hands’ button shown in figure 2, the students can share their model with another student.

Figure 2: In red surrounded the added button for collaboration within DynaLearn

To create a collaborative conceptual learning environment in DynaLearn that prevents free-riding, the visual in figure 3 was implemented by providing the students with insights of their activity and that of their teammate in creating the model. The visual makes use of learning analytics and the percentages presented in the tool are calculated with the following formula:

Amountof Actionsstudent

Amountof Actionsstudent+ Amountof Actionsteammate

· 100

In the horizontal bar the percentages of the total amount of actions is presented, which concern all the actions i.e. create, modify and delete towards all the concepts of the model i.e. entity, quantity, configuration and influences. In addition, the percentages of the actions towards a specific concept are presented in the vertical bars, which are calculated with the same formula. The first vertical bar con-cerns the percentages of the total amount of actions towards the concept entity. The second vertical bar concerns the percentages of the total amount of actions towards the concept quantity. In the third vertical bar the percentages are shown of the total amount of actions concerning the total amount of actions towards all concepts omitted entity and quantity.

Figure 3: The feedback tool implemented within DynaLearn

The learning activity of the students to calculate the percentages is derived from DynaLearn. The pro-gram saves every action students make with the following features: username, modelID, model ,moment, action, target, targettype, arguments. In table 1 these features are explained more clearly.

(9)

Thesis Project AI

Feature Meaning

Username The username of the user who executed the action ModelID The ID of the model where the action was executed Model The name of the model where the action was

exe-cuted

Moment The time and date when the action was executed Action The type of action i.e. create, delete, modify, undo,

redo, newmodel, copymodel

Target The specific target the action is run on

Targettype The type of target the action is run on i.e. en-tity, quanen-tity, configuration, proportionality posi-tive, proportionality negaposi-tive, derivative value Arguments Arguments that belong to the type of action

Table 1: Features of an action in DynaLearn

The implementation of the tool into the program is done in Javascript.

3.3 Evaluation

The feedback tool is evaluated by doing an experiment with students. In a systematic variation experi-ment [Field, 2009] a treatexperi-ment group and a control group both made use of the collaborative modeling environment DynaLearn, while the feedback tool was only presented to the treatment group. The data from the experiment consists of high-level and low-level indicators of collaboration. This data is statis-tically analyzed using SPSS and machine learning with Python. The analysis shows the extent of the feedback tool to the support of the collaboration and therefore the learning results of students.

3.3.1 Experiment

The experiment is done with high school students as part of their ongoing biology lessons.

Participants

Participants were 55 students from the forth grade from three different classes (class 1: 20 students, class 2: 20 students, class 3: 15 students) between the age of 15 and 17, from which 55 % of them are girls. All the participants studied a physics profile.

Procedure

The lesson was given to three different classes. The classes were divided over the treatment group and the control group. The control group consisted of class 1 (n=20) and the treatment group consisted of class 2 and class 3 (n=35). Because class 3 consisted of an uneven number of students, one student did the task alone instead of in a group of three to prevent the loss of further subjects.

Both groups did a task in pairs of two using the CSCL modeling environment DynaLearn with collabora-tion directed instruccollabora-tions, but the feedback tool was only available to the treatment group. To create the CSCL environment in the experiment, the lessons took place in a computer classroom. Each student was sitting in front of a computer and next to their teammate so they could exchange thoughts face-to-face, because face-to-face interaction is one of the requirements of collaboration [Johnson and Johnson, 1999]. Between the pairs a computer was empty to create a free space.

The subject matter of the lesson was ecosystems. An experimenter was present during the treatment so the children could ask questions about the task and the software.

(10)

Thesis Project AI

The lesson began with a short explanation of what they had to do. Also it was explained how the lesson was graded, namely for the task, the level of collaboration, which will be explained later on, and the individual final test. Then the lesson began:

First, each student had to do a test individually to check their conceptual knowledge before having worked with DynaLearn. This will be referred to as the (1) pretest. Second, (2) the task was done in pairs of two. After the task the students received an individual test to see how much knowledge they had gained, which will be referred to as the (3) posttest. Last, the students had to fill in a (4) questionnaire of the lesson.

3.3.2 Tasks in experiment

In the Appendix A the material used for this experiment is presented. The task, part 3 of the posttest and the questionnaire are adjustments of previous research [Schlatter, 2017].

1. Pretest

To measure the pre-existing knowledge of conceptual models of the students taking a test is necessary before the treatment starts. This test was not dependent on the subject matter of the class, namely ecosystems but dependent on their conceptual knowledge. In the pretest the students were asked to draw a model of an ecosystem, which is described widely.

2. Task

In the task the students learned how to work with DynaLearn and they are asked to construct a model of an ecosystem. There were two versions of the task, A and B, with both different parts of the information necessary to complete the task. In this way the positive interdependence was stimulated and forced the students to cooperate and discuss to complete the task [Collazos et al., 2007].

Because the students were beginners with modeling in DynaLearn the students only made use of primary causes to prevent the difficulty of the program influencing the collaboration.

3. Posttest

To influence the individual accountability of the students, an individual test after the collaboration was taken. This was told in the explanation in the beginning of the lessons and motivated the students to participate in the collaboration [Johnson and Johnson, 1999]. Only the saying that an individual test will be taken would influence the collaboration. Thus, the results of the test were for the individual accountability irrelevant, but for the data analysis the results of the test were relevant to measure the conceptual knowledge.

For the developing of the Posttest Bloom’s Taxonomy of Educational Objectives for Knowledge-Based Goals is used [Omar et al., 2012]. This is a model for classifying learning objectives within education into six levels of complexity. In the table below the questions with its corresponding knowledge level will be presented.

(11)

Thesis Project AI

Level Meaning Part of Posttest

1. Knowledge level Having the ability to recall terms and ideas from previous lessons

Question 1

2. Comprehension level

Having the ability to interpret, translate, extrapolate, classify and explain terms and ideas from the subject

Question 2

3. Application level Having the ability to understand the concept and apply it to a cer-tain situation

Question 3: if all the correct con-nections have been made i.e. correct direct and value 4. Analysis level Having the ability to analyze the

knowledge they have learned and to distinguish between fact and opinion.

Question 3: if no mistakes are made in the model or if no useless items are added

5. Synthesis level Having the ability to integrate and combine ideas or concepts by rearranging components into a whole

Question 3: if configurations are added to the model

6. Evaluation level Having the ability to judge, crit-icize, support and defend a term or idea referring to the subject

Question 4

Table 2: Levels Blooms Taxonamy assigned to the questions of the Posttest

4. Questionnaire

To measure the attitude of the students towards collaboration and about the lesson a questionnaire is taken after the lesson. The questionnaire consisted of eight statements about the lesson with answers on a five points scale with 1 being ”complete disagreement” and 5 being ”complete agreement”. In table 3 the measured attitudes are presented with its referring questions.

Measured attitude towards Questions

The lesson Question 1

The task Question 2, Question 3, Question 4, Question 7

The collaboration Question 6

DynaLearn Question 5

The tool Question 8

Table 3: Measured attitudes and its reference to the questions

It was expected if the students enjoyed working together, the students experienced a good form of col-laboration.

3.3.3 Data

In order to analyze collaboration it is necessary to collect data. The data used in this research consists of data provided by DynaLearn, scores of the pretest, task and posttest of the students scored by two correctors, and at last the opinions of the student concerning the lesson.

(12)

Thesis Project AI

In this research different kind of indicators for collaboration were used. Like Daradoumis et al. [2006] in his research did, high-level indicators are used, shown in table 4 and low-level indicators are shown in figure 4 and figure 5.

Indicator Meaning Conceived

Difference in distribution of actions

The distribution of the actions done by the students in per-centage

Absolute(∆(P ercentageStudent1−

P ercentageStudent2))

Score task The score of the model build in pairs

Scored by one corrector and the score is normalized.

Score posttest

The score on the in-dividual test after the lesson

Scored by two correctors, from which the mean of the two scores is used as posttest score. The score is normalized.

Progress The progress the students made

ScoreP osttest − ScoreP retest.

Both tests are scored by two correctors. The scores are normalized.

Amount of actions

Amount of actions DynaLearn saves every action done by the students. These ac-tions are summed up

Time Time used to

com-plete the task in seconds

Calculated by contracting the time of the first action of the time of the last action.

Table 4: High-level indicators

(a) Low-level indicators of the indicator Amount of actions

(b) Low-level indicators of the indicator Distribution of actions

Figure 4: High-level indicators: Amount of actions and Distribution of actions with their low-level indicators

(13)

Thesis Project AI

Figure 5: Low-level indicators of the indicator Score Posttest

The scores of the pretest are not used as indicator of collaboration, but to measure the differences between the groups. These test are scored by two correctors. Both scores are normalized. The total score on the pretest is the mean of both scores.

3.3.4 Statistical analysis

To analysis the data two forms of statistical analysis are used; (1) statistical analysis using machine learn-ing in Python and (2) statistical analysis in SPSS. Because no general method to analyze collaboration is found a few assumptions have been made:

1. If the amount of actions from the students in collaboration are approximately equal, the collabo-ration is strong.

2. If the amount of actions of the students in collaboration are out of balance, there is a great possibility free-riding occurs.

With machine learning the collaboration of the students depending on the distribution of actions are classified in three categories shown in table 5. In addition, the students are classified individually based on their contribution to the model as ’free-riders’ or ’non free-riders’ (see table 6). To evaluate the effectiveness of the feedback tool statistical analysis is used to detect (in)significant differences between the treatment group and the control group. To notify possible differences between the three classes, statistical analysis is used. In addition, a statistical analysis is used to inspect the effects of the form of collaboration on the task, the Posttest and the progress of the students. This concerns the classifications of collaboration and the classifications of the individual. Last, a statistical analysis is used to notify differences in answers on the questionnaire between the treatment group and the control group, and between the good, moderate and bad collaboration.

Classification Meaning

Good form of collaboration Action difference below 20%

Moderate form of collaboration Action difference above 20% and below 50 Bad form of collaboration Action difference above 50%

(14)

Thesis Project AI

Classification Meaning

Free-rider Actions below 25% Non free-rider Actions above 25%

Table 6: Classifications for individuals based on their contribution to the model

4 Results

4.1 Compare on basis of pretest

In table 7 the differences of the scores on the pretest of the control group and the treatment group are presented. The control group scored on average 0.61248 higher on the pretest then the treatment group. The Levene’s test suggests that the two groups differ in variance, but not significantly (p = 0.054).

In table 8 the differences of the scores on the pretest of the different classes are presented. Class 3 scores on average 0.83211 lower on the pretest then class 1 and class 2. This difference is insignificant, based on the one-way ANOVA (p=0.111).

Group Mean SD Range

Treatment group 5.251 1.0557 (3.208, 6.695) Control group 5.863 1.626 (2.660, 8.482)

Table 7: The scores of treatment group and the control group on the pretest

Class Mean SD Range

Class 1 5.863 1.626 (2.660, 8.482) Class 2 5.602 1.055 (3.351, 6.695) Class 3 4.900 1.056 (3.208, 6.695)

Table 8: The scores of Class 1, Class 2 and Class 3 on the pretest

4.2 Effects feedback tool and collaboration

4.2.1 Effectiveness feedback tool on collaboration

The treatment group seemed to collaborate better (mean of distribution of actions: 35.5%) then the control group (mean of distribition of ations: 42.4%) (see Table 9), but the t-test showed no significant effect was found (p = 0.253).

The classification of the forms of collaboration (see Table 5) indicate that there seems to be a difference between the treatment group and the control group. 41% of the treatment group 11% of the control group were classified as ’Good Collaboration’. 24% of the treatment group and 33% of the control group were classied as ’Bad Collaboration’, with the possibility of free-riding to occur. However, this difference was not statistically significant (p = 0.285).

Group Mean SD Range

Treatment group 35.551 28.920 (2.804, 92.982 ) Control group 42.377 22.210 (5.556, 8.1667)

(15)

Thesis Project AI

Classification Treatment Control

Good Collaboration 41% 11%

Moderate Collaboration 35% 56%

Bad Collaboration 24% 33%

Table 10: The percentages of the treatment and the control group on the distribution of the classifiers good, moderate and bad collaboration.

4.2.2 Influence collaboration on scores

The distribution of actions between the students collaborating on the score of the task, the score of the posttest and the progress were statistically insignificant. Correlation between distribution of actions with the collaborative task is 0.185 with a p-value of 0.366. Correlation between distribution of actions with the score on the posttest is -0.111 with a p-value of 0.588. Correlation between distribution of actions with the progress is -0.157 with a p-value of 0.443.

In the scatter plots of figure 6 the distributions of the pairs are shown in relation with their score on the task, the posttest and progress.

Figure 6: Scatter plots with on the x-as the distribution of actions and on the y-as te scores of (a) the task, (b) the posttest and (c) the progress

Table 11 shows that the pairs classified with ’Bad Collaboration’ scored on average 0.512 higher on the task then the pairs classified with ’Good Collaboration’ and ’Bad Collaboration’. The one-way ANOVA tests these differences to be insignificant (p = 0.682). Table 12 shows that the pairs classified with ’Bad Collaboration’ scored on average 0.5875 lower on the posttest then the others. The one-way ANOVA tests these differences to be insignificant (p = 0.615). Table 13 shows that the pairs classified with ’Bad Collaboration’ had on average an 1.775 lower progress then the other. The one-way ANOVA tests these differences to be insignificant (p = 0.724).

Group n Mean SD Range

Good collaboration 8 7.265 2.103 (3.541,9.357) Moderate collaboration 11 7.841 1.834 (5.000, 10.000) Bad collaboration 7 8.065 1.515 (6.250, 9.791)

Table 11: The score of the classifications good, moderate and bad collaboration on the collaborative task

(16)

Thesis Project AI

Good collaboration 8 6.248 1.339 (4.130, 10.250) Moderate collaboration 11 6.691 2.218 (2.259,8.861) Bad collaboration 7 5.882 0.916 (4.750,7.519)

Table 12: The score of the classifications good, moderate and bad collaboration on the posttest

Good collaboration 8 1.760 2.614 (-2.637, 9.056) Moderate collaboration 11 2.433 4.534 (-4.129, 10.298) Bad collaboration 7 0.919 3.912 (-3.183, 7.860)

Table 13: The score of the classifications good, moderate and bad collaboration on the progress

4.2.3 Influence free-riding

The ”free-riders” (students with the amount of actions < 25%) scored significant lower on the posttest (p=0.029) and lower on the progress level (p=0.019) results form the t-test. Table 14 shows that the ”free-riders” scored on average 4.839 on the posttest and the ”non free-riders” 6.473 (difference = 1.634). Table 15 shows that the ”riders” scored on average -0.913 on the progress level and the ”non free-riders” 1.060 (difference = 1.973). In the figure 7 the free-riders are clearly presented in the scatter plot. As can be seen in the scatter plot of figure 7 that the free-riders score beneath the mean all the students.

Student n Mean SD Range

Free-rider 7 4.839 0.727 (3.852, 6.019 ) Non free-rider 45 6.473 1.886 (2.093, 10.352)

Table 14: The score of the classifications free-rider and non free-rider on the posttest

Free-rider 7 -0.913 0.620 (-1.731, 0.024) Non free-rider 45 1.060 2.106 (-3.895, 5.291)

(17)

Thesis Project AI

Figure 7: Scatter plot of the students with on the x-as the indicator Actions in comparison to teammate and on the y-as the indicator Score posttest The students are clustered on the basis of the classification of free-riding or non free-riding. The horizontal line is set at the mean (≈ 6.253) of the

score of the posttest of all the students.

Free-rider 7 5.622 1.012 (4.050, 6.522) Non free-rider 45 5.473 1.340 (2.660, 8.482)

Table 16: The score on the pretest of the classifications free-rider and non free-rider. T-test results in p-value = 0.780

4.3 Further results

4.3.1 Collaboration and the indicators Amount of actions and Time

In table 17 the further indicators ’Amount of actions’ and ’Time’ are presented on the classifications of collaboration. There was no significant difference in the amount of actions used to complete the task between the classifications of collaboration (p=0.224, p>0.05). The ’Good Collaborators’ used significantly less time to complete the task then the rest (p=0.031, p<0.05). The shortest duration of all the pairs to complete the task is 1457 seconds (≈ 24 minutes) and the longest duration of all the pairs to complete the task is 5723 seconds (≈ 95 minutes). In figure 8 the distribution of the pairs are shown on the basis of time used to complete the task and the score of the task. 76% of the pairs finished the test within 4000 seconds (≈ 67 minutes) and 61% of the pairs finished the test within 3286.88 (≈ 54 minutes) which is the mean of the time used to complete the task.

(18)

Thesis Project AI

Indicator Collaboration group Mean SD Range P-value

Amount of actions ’Good Collaboration’ 131.125 45.840 (92,234)

’Moderate Collaboration’ 115.727 23.290 (85,146) 0.224

’Bad Collaboration’ 103/286 12.829 (86,120)

Time (in seconds) ’Good Collaboration’ 2640.500 322.721 (2079,3253)

’Moderate Collaboration’ 3914.8182 1028.411 (2189, 5723) 0.031 ’Bad Collaboration’ 3038.8571 1389.971 (1457,5288)

Table 17: The classifications good, moderate and bad collaboration on the indicator amount of actions and the indicator time

Figure 8: Scatter plot of the pair collaborating with on the x-as the indicator Time and on the y-as the indicator Score Task. The pairs are clustered on the basis of the classification of collaboration. The

vertical line is set at the mean (≈ 3286.88) of the time used in seconds to complete the task.

4.3.2 Progress treatment group and control group

The differences between the treatment group and the control group on the posttest (p=0.925) and the progress (p=0.389) seem to be insignificant resulted from the t-test. In table 18 the means of the treatment group and the control group on the Posttest and the progress are presented.

Group Mean posttest Mean progress

Treatment group 6.360 1.146

Control group 6.293 0.462

Table 18: The score on the posttest and the progress of the treatment group and the control group

4.3.3 Progress classes

Table 19 shows that class 3 has a progress 1.7015 higher then the other classes. This difference in progress is insignificant (p=0.122) results from a one-way ANOVA.

(19)

Thesis Project AI

Class Mean SD Range

Class 1 0.462 1.958 (-2.065, 3.930) Class 2 0.441 1.802 (-1.319, 3.619) Class 3 2.153 1.958 (-0.069, 5.149) Table 19: Class 1, Class 2 and Class 3 on progress

4.3.4 Results attitudes of students

The results of the attitude of the students towards the lesson from the treatment group and the control group can be seen in Appendix B.1. A t-test resulted in an insignificant effect (0.425) of the tool on the insights in the distribution of contribution. The control group filled in to agree 0.3987 more with question 8. Further, the differences between the answers of the groups were not statistically significant (p<0.05).

There was no significant effect found of the classification of collaborations and the questions. In table 20 it is shown that pairs classified with ’Bad Collaboration’ filled in to agree 1.172 more with question 8 then the rest. This difference was insignificant (p=0.061).

Question Collaboration group Mean SD Range P-value

1 Good 8.143 1.069 (7,9) Moderate 8.273 1.191 (6,10) 0.723 Bad 7.875 0.835 (7,9) 2 Good 7.286 1.704 (5,9) Moderate 7.273 1.794 (3,9) 0.757 Bad 6.750 1.389 (5,9) 3 Good 6.429 1.134 (5,8) Moderate 7.091 1.221 (4,8) 0.166 Bad 6.125 0.835 (5,7) 4 Good 6.429 1.134 (6,7) Moderate 7.727 1.348 (6,10) 0.155 Bad 6.875 1.642 (4,9) 5 Good 6.429 1.902 (4,9) Moderate 6.727 1.737 (4,9) 0.938 Bad 6.750 2.252 (3,10) 6 Good 9.143 1.574 (6,10) Moderate 9.273 0.905 (7,10) 0.644 Bad 9.625 0.518 (9,10) 7 Good 7.286 1.380 (6,10) Moderate 7.909 1.136 (6,9) 0.529 Bad 7.750 0.886 (6,9) 8 Good 8.429 1.272 (7,10) Moderate 8.727 1.272 (6,10) 0.061 Bad 9.750 0.463 (9,10)

Table 20: The classifications good, moderate and bad collaboration on the questionnaire. In the last column the p-value of the one-way ANOVA between the tree classifications is shown

(20)

Thesis Project AI

5 Conclusion

In this research a first form of collaboration in the DynaLearn program is tested, based on a vast amount of research that proves collaborative learning to be effective especially in conceptual complex domains [Andrews and Rapp, 2015]. For this project a joined workspace is created. As just providing an envi-ronment for collaboration, does not ensure collaboration occurring [Andrews and Rapp, 2015], a tool that aims at stimulating collaboration by giving feedback on the participation of team members is im-plemented in the workspace. The tool is using learning analytics and provides feedback to the students about input and amount of actions each of them contributes to the model they are creating together. The feedback tool is evaluated with high school students as part of their ongoing lessons. In a systematic variation experiment an treatment and a control group both had the joined workspace at their disposal, while the feedback tool was only available for the treatment group. The instructions given to both groups were the same and aimed at encouraging students for collaborate effectively.

First of all, the results seem to indicate that the feedback tool has an effect on the collaboration, as in the treatment group a higher percentage of pairs can be classified as good collaboration then in the control group. However, care should be taken as the results is not statistically significant. Further, this research confirms that free-riders learn less then non free-riders. A bad form of collaboration is considered as the occurrence of free-riding. The results show that also in the DynaLearn environment free-riding is a risk and has a significant negative effect on learning. Free-riders show no progress or even show declining progress. Moreover, no significant relationship was found between attitudes of students and the level of collaboration, though compared to moderate and good collaborating pairs, the pairs that were classified, as bad collaborators believed more that they have insights in what they and their teammate were contributing to the model. Finally, there was a great difference in time that pairs used to complete the task. However this difference showed no correlation with the score on the task.

Overall, there are indications that the feedback tool has some effect in terms of stimulating collaboration. Based on the results the research can be seen as a small start towards a successful implementation of collaborative learning with DynaLearn. In the next chapter possible ideas for future research will be discussed.

6 Discussion and further research

This research is a first step in creating a CSCL modeling environment. Small, non significant, effects have been found that indicate that the feedback tool does influence collaboration in a positive way. Also, based on this research it is possible to detect free-riders. Further insights about influencing collaboration and free-riding could be interesting for future research. Therefore, in this chapter the approach of this research and recommendations for future work will be discussed.

The first part of discussion is about the complexity of the task used in the experiment. Given the fact that collaboration especially pays of in complex tasks [Johnson and Johnson, 1999, Kirschner et al., 2011, Andrews and Rapp, 2015] and the average progress of the students was limited, it might be possible that the task was not complex enough. Therefore, in future research it is important to carefully select the task to be a good match with the level of competence of the students in the experiment. In addition, DynaLearn has a lot of features that could have been used. In the task used in this approach, only a few of them have been available for the students. By using more features of DynaLearn the complexity of the task could increase.

A second point of discussion is the amount of time given to complete the task. In the experiment there was a great difference in the time it took pairs to complete the task. The pairs had approximately 90 minutes to complete the task, but most of the pairs had finished within 50 minutes. Possibly pairs that found the task more difficult, used more time. Because there was no time limit, those pairs were also

(21)

Thesis Project AI

able to finish the task.With a shorter and stricter time limit to complete the task, this might result in a greater difference in scores between the pairs. In future research it is important to guarantee comparable learning time brackets by assuring the same amount of modeling time.

Another point of discussion, is the questionnaire as research instrument since the questionnaire used in this research was not anonymous. This may have led for students attempting to answer with socially desirable answers. The choice for a non-anonymous evaluation was made because of the interest of corre-lations between student’s attitude to perception of collaboration and how well they actually collaborated. Perhaps in future research another form of evaluation needs to be designed.

Last, an important finding is that free-riding is a realistic risk in DynaLearn, that indeed leads to significantly less progress and can be detected with help of the feedback tool. In addition, it seems that free-riding pairs were more aware of the contributions to the work of team members. It is possible that when participation is far out of balance it is obvious for the students what the distribution of the contribution is. This indicates a beginning of awareness of free-riding. Therefore, the possibility of detecting free-riders on the basis of the feed-back tool and the awareness of the students, might be desirable to intervene and support students during the process to improve their collaboration.

References

Antonio R Anaya and Jes´us G Boticario. Application of machine learning techniques to analyse student interactions and improve the collaboration process. Expert Systems with Applications, 38(2):1171–1181, 2011.

Jessica J Andrews and David N Rapp. Benefits, costs, and challenges of collaboration for learning and memory. Translational Issues in Psychological Science, 1(2):182, 2015.

Ricardo Baeza-Yates and Jos´e A Pino. Towards formal evaluation of collaborative work. Information Research: An International Electronic Journal, 11(4):n4, 2006.

Furio Belgiorno, Ilaria Manno, Giuseppina Palmieri, and Vittorio Scarano. Free-riding in collaborative diagrams drawing. Sustaining TEL: From Innovation to Learning and Practice, pages 457–463, 2010. Bert Bredeweg, Jochem Liem, Wouter Beek, Floris Linnebank, Jorge Gracia, Esther Lozano, Michael Wißner, Ren´e B¨uhling, Paulo Salles, Richard Noble, et al. Dynalearn–an intelligent learning environ-ment for learning conceptual knowledge. AI Magazine, 34(4):46–65, 2013.

C´esar A Collazos, Luis A Guerrero, Jos´e A Pino, Stefano Renzi, Jane Klobas, Manuel Ortega, Miguel A Redondo, and Crescencio Bravo. Evaluating collaborative learning processes using system-based mea-surement. Journal of Educational Technology & Society, 10(3), 2007.

Thanasis Daradoumis, Alejandra Mart´ınez-Mon´es, and Fatos Xhafa. A layered framework for evaluating on-line collaborative learning interactions. International Journal of Human-Computer Studies, 64(7): 622–635, 2006.

Pierre Dillenbourg, Sanna J¨arvel¨a, and Frank Fischer. The evolution of research on computer-supported collaborative learning. Technology-enhanced learning, pages 3–19, 2009.

Andy Field. Discovering statistics using SPSS. Sage publications, 2009.

David W Johnson and Roger T Johnson. Making cooperative learning work. Theory into practice, 38 (2):67–73, 1999.

Femke Kirschner, Fred Paas, and Paul A Kirschner. Task complexity as a driver for collaborative learning efficiency: The collective working-memory effect. Applied Cognitive Psychology, 25(4):615–624, 2011.

(22)

Thesis Project AI

Nazlia Omar, Syahidah Sufi Haris, Rosilah Hassan, Haslina Arshad, Masura Rahmat, Noor Fari-datul Ainun Zainal, and Rozli Zulkifli. Automated analysis of exam questions according to bloom’s taxonomy. Procedia-Social and Behavioral Sciences, 59:297–303, 2012.

Gwendolynn Charmaine Ru¨el, Nienke Bastiaans, and Aukje Nauta. Free-riding and team performance in project education. University of Groningen, 2003.

Erika Schlatter. The effect of modelling on learning subject-specific content: an experimental approach. MSc thesis Artificial Intelligence, University of Amsterdam, 2017.

Christina V Schwarz and Barbara Y White. Metamodeling knowledge: Developing students’ understand-ing of scientific modelunderstand-ing. Cognition and instruction, 23(2):165–205, 2005.

Andreas Zitek, Michaela Poppe, Michael Stelzhammer, Susanne Muhar, and Bert Bredeweg. Learning by conceptual modeling–changes in knowledge structure and content. IEEE Transactions on Learning Technologies, 6(3):217–227, 2013.

(23)

Thesis Project AI

A

Material used for the experiment

The website mentioned in the task differed between the treatment group and the control group. The treatment group could use the collaborative modeling environment on the website dev.samen.dynalearn.nl and the control group on the website dev1.dynalearn.nl.

(24)

Naam Klas

De Loosdrechtse plassen

In het ecosysteem van de Loosdrechtse plassen zijn verschillende organismen te vinden. Eén daarvan zijn algen. Algen produceren 73% tot 87% van het zuurstof die voor de landorganismen ter beschikking staat. Tegenwoordig plaatsen bewoners grote vlotten aan de plas waardoor er minder zonlicht in het water terecht komt. Algen maken gebruik van fotosynthese en hierdoor is voor de groei van algen zon erg belangrijk. Wanneer er minder zonlicht in het water aanwezig is, kunnen de algen minder goed groeien. Minder algen zorgen dus voor minder zuurstof in de buitenlucht. Daarnaast worden algen gegeten door kreeftjes, dus verdwijning van de algen kan zorgen voor het terugdringen van de populatie kreeftjes. De kreeftjes worden op zijn beurt weer gegeten door de forel. Wanneer de kreeftjes afsterven zullen ze gegeten worden door de reducenten in het water.

Maak aan de hand van deze bovenstaande informatie een schema die het gedrag van het ecosysteem weergeeft. Doe dit door verbanden tussen eigenschappen van het ecosysteem en de organismen te laten zien. Zijn deze verbanden positief of negatief?

Thesis Project AI

(25)

Versie A Jouw naam: Naam teamgenoot: Klas Let op! Deze les is onderdeel van een onderzoek. Daarom is het belangrijk dat jullie als zelfstandig team aan de opdrachten werken. Als jullie een probleem hebben of ergens niet uitkomen, probeer er dan samen nog eens rustig over te praten om er toch uit te komen. Als jullie er dan toch niet uitkomen, kunnen jullie je hand op steken en zal de onderzoeker jullie helpen.

Ecosysteem Yellowstone

Een krantenkop luidt: Noorse regering wil 2/3 van de wolvenpopulatie afschieten. Hierover is veel ophef. De Noorse regering voert als argument aan dat andere diersoorten te lijden hebben onder de wolven populatie. In veel Europese landen zijn geen of nog maar weinig wolven. En als er wolven zijn, zijn ze niet altijd gewenst, bijvoorbeeld omdat ze kleinere dieren eten. Maar is het als je naar het heel ecosysteem kijkt, wel gewenst om zoveel wolven af te schieten? In deze les gaan jullie op zoek naar argumenten om wolven wel of niet te beschermen. Om hierachter te komen zullen jullie eerst kijken naar een andere casus over wolven: het ecosysteem in Yellowstone, het oudste National Park in de Verenigde Staten. Door de invloed van de wolven op het ecosysteem in Yellowstone te bestuderen krijgen jullie informatie om een oordeel te vellen. Om meer te leren over dat ecosysteem maken jullie gezamenlijk een model. Een model is een vereenvoudigde voorstelling van de werkelijkheid, die je kan gebruiken om ergens over te leren, je ideeën ergens over te ontwikkelen of je ideeën te testen. Wij gaan dit model maken in DynaLearn.

1.

De les gaat over de invloed van de aanwezigheid van wolven op dieren als beren, bevers en verschillende vogelsoorten. Bespreek eerst wat jullie verwachten dat er gebeurt met de populatie van elk dier, wanneer de populatie wolven toe neemt. Vul hierna voor elke diersoort in of jullie denken dat de populatie toeneemt, afneemt of gelijk blijft als de populatie wolven toeneemt. 1. De populatie wolven groeit, dus de populatie beren neemt af / blijft gelijk /neemt toe. 2. De populatie wolven groeit, dus de populatie bevers neemt af / blijft gelijk /neemt toe. 3. De populatie wolven groeit, dus de populatie vogels neemt af / blijft gelijk /neemt toe.

2.

Ga naar devsamen.dynalearn.nl en log in met de inlognaam en het wachtwoord dat je gekregen hebt. Sla het nog lege model op door op te klikken. Geef het de naam ‘Ecosysteem Yellowstone’. Om te gaan Thesis Project AI

(26)

Versie A samenwerken moet je op klikken en vervolgens op . Hiermee krijg je een code die je teamgenoot moet invullen. Jullie gaan eerst een klein model maken, met daarin alleen een specifieke groep producenten: struiken. De basis van een DynaLearn-model bestaat uit entiteiten. Een entiteit is een onderdeel van het model. Je maakt een entiteit aan door in het menu links op te klikken, en daarna op het scherm te klikken waar deze entiteit moet komen te staan.

3.

Maak een entiteit aan en geef het de naam ‘ecosysteem Yellowstone’.

4.

Maak ook de entiteit ‘struiken’ en koppel deze aan de entiteit ecosysteem Yellowstone. Als je op struiken klikt verschijnt er een menu. Kies voor en klik dan op ecosysteem Yellowstone. Vul bij ‘configuratie’ in ‘staat in’. Dit is het verband tussen struiken en het ecosysteem. De twee entiteiten geven al een eerste indruk van het ecosysteem. Maar entiteiten kunnen ook eigenschappen hebben, die elk het systeem op hun eigen manier beïnvloeden. Zo een eigenschap noem je een grootheid. Een voorbeeld van een grootheid die bij struiken kan horen, is de hoeveelheid.

5.

Maak deze grootheid aan. Klik op de entiteit struiken en kies voor . Geef de grootheid de juiste naam. De struiken leven natuurlijk niet alleen in Yellowstone. Je teamgenoot beschikt over de informatie over de consumenten die leven in Yellowstone. Bespreek dit kort.

6.

Welke drie organismen kunnen jullie uit de tekst over consumenten in Yellowstone vinden? 1. 2. 3.

7.

Maak de organismen als entiteiten aan, elk met de grootheid ‘aantal’. Maak hierbij ook het verband aan tussen de nieuwe entiteiten en de entiteit ‘ecosysteem Yellowstone’.

8.

Welke grootheid van de entiteit ‘struiken’ hebben jullie nog meer kunnen vinden in de tekst en maak deze aan. Grootheden kunnen elkaar beïnvloeden. Als dat zo is, kun je in DynaLearn een verband aanmaken. Door zulke verbanden kun je uiteindelijk een model simuleren (daarover

(27)

straks meer) Je maakt een verband aan door op de grootheid te klikken en in het menu te kiezen voor . Er verschijnt dan een nieuw menu: • Bij een positief verband is het zo dat als de ene grootheid toeneemt, de andere grootheid ook toeneemt. Of andersom: de ene neemt af, waardoor de andere ook afneemt. • Bij een negatief verband is de relatie tegenovergesteld: als de ene grootheid toeneemt, neemt de andere af. Of andersom: de ene grootheid neemt af, waardoor de andere toeneemt.

9.

De grootheden van de drie nieuwe entiteiten (uit vraag 6) zijn elk verbonden met één of meer grootheden van de entiteit struiken. Overleg hoe deze nieuwe grootheden verbonden zijn met de grootheden van de entiteit struiken. En hoe beïnvloeden de grootheden van de entiteit struiken elkaar? Maak ook dit verband aan in jullie model. De volgende stap is het simuleren van jullie model. Bepaal samen een beginsituatie waarin jullie van de grootheden van de struiken aangeven of ze afnemen, gelijk blijven of toenemen. Of een grootheid afneemt of toeneemt zie je aan 𝛿. Door op het driehoekje van met de punt naar beneden te klikken en vervolgens op , neemt de grootheid af. Door op ∅ te klikken, blijft de grootheid gelijk. En door op het driehoekje met de punt naar boven te klikken, neemt de grootheid toe. Niet elk waarde bereik hoeft een beginsituatie te hebben. Alleen zij die aan het begin van een causale keten staan.

10.

Simuleer nu het model door op te klikken en dan op die rechts in beeld verschijnt. Wat gebeurt er met de grootheden van de drie consumenten? Bespreek of dit is wat jullie verwacht hadden? Waarom wel/niet? Jij bent de enige die het model kan simuleren, dus laat jouw teamgenoot meekijken op jouw scherm tijdens de simulatie. Als niet in beeld komt, dan werkt de simulatie niet volgens verwachting en moeten jullie het model waarschijnlijk aanpassen (zie 11).

11.

Als de simulatie verlopen is zoals jullie hadden verwacht, kunnen jullie verdergaan en deze vraag overslaan. Anders moeten jullie het model eerst aanpassen. Denk hierbij aan: • De richting van het verband (van x naar y of andersom?) • Het type verband (positief of negatief) De hoeveelheid planten heeft niet alleen invloed op hoeveel dieren er kunnen leven. Omdat dieren planten eten, kunnen zij ook de hoeveelheid planten beïnvloeden. Over één van de dieren die de hoeveelheid planten op bepaalde plekken in Yellowstone sterk beïnvloedt gaat de tekst hieronder. Deze keer beschik jij over de informatie in plaats van jouw teamgenoot. Deel deze informatie ook met hem/haar door het kort te bespreken.

(28)

Versie A De wapiti en de wolven in Yellowstone Wapiti’s horen bij de familie van de hertachtigen. Ze verplaatsen zich weinig en vormen grote kuddes. Hierdoor leggen zij een grote druk op de begroeiing: struiken en bossen verdwijnen. Hierdoor groeien er ook minder bessen aan de struiken. Als er echter wolven in de omgeving zijn verplaatsen wapiti’s zich vaker. Ze leven dan in kleinere kuddes en kunnen zich daardoor gemakkelijker verplaatsen. Als de wapiti’s zich meer verplaatsen, kunnen de struiken weer beter groeien.

12.

Bij de eerdere dieren die jullie als entiteiten aan het model hebben toegevoegd, hebben jullie enkel de grootheid ‘aantal’ gebruikt. In de tekst hierboven kunnen jullie meer grootheden vinden. Bespreek welke grootheden nog meer bij de entiteit ‘Wapiti’ passen? Maak de entiteit ‘Wapiti’ aan met drie grootheden die jullie denken dat hierbij horen, dit mag ook de grootheid ‘aantal’ zijn. Verbind de entiteit ‘Wapiti’ ook aan het ‘Ecosysteem Yellowstone’.

13.

Overleg welke grootheden van de entiteit ‘Wapiti’ in verband staan met de grootheden ‘hoeveelheid’ en ‘aantal bessen’ van de entiteit ‘struiken’? En wat voor een verband is het (positief of negatief)? Maak deze verbanden aan in jullie model.

14.

Maak een entiteit ‘wolven’ met de grootheid ‘aantal’ en verbindt ook deze entiteit met de entiteit ‘Ecosysteem Yellowstone’. Gebruik de tekst over de wapiti en wolven om samen te bedenken welke verbanden (positief of negatief) er zijn.

15.

Maak de verbanden en simuleer wat er gebeurt als het aantal wolven toeneemt. Kijk hierbij vooral naar de aantallen van de vogels, beren en bevers. Bespreek kort of het simulatieresultaat overeenkomt met wat jullie hadden verwacht? Kijk daarbij ook weer even naar opdracht 1. Als het resultaat klopt met jullie verwachting, zijn jullie klaar met de opdracht.

(29)

Versie B Jouw naam: Naam teamgenoot: Klas Let op! Deze les is onderdeel van een onderzoek. Daarom is het belangrijk dat jullie als zelfstandig team aan de opdrachten werken. Als jullie een probleem hebben of ergens niet uitkomen, probeer er dan samen nog eens rustig over te praten om er toch uit te komen. Als jullie er dan toch niet uitkomen, kunnen jullie je hand op steken en zal de onderzoeker jullie helpen.

Ecosysteem Yellowstone

Een krantenkop luidt: Noorse regering wil 2/3 van de wolvenpopulatie afschieten. Hierover is veel ophef. De Noorse regering voert als argument aan dat andere diersoorten te lijden hebben onder de wolven populatie. In veel Europese landen zijn geen of nog maar weinig wolven. En als er wolven zijn, zijn ze niet altijd gewenst, bijvoorbeeld omdat ze kleinere dieren eten. Maar is het als je naar het heel ecosysteem kijkt, wel gewenst om zoveel wolven af te schieten? In deze les gaan jullie op zoek naar argumenten om wolven wel of niet te beschermen. Om hierachter te komen zullen jullie eerst kijken naar een andere casus over wolven: het ecosysteem in Yellowstone, het oudste National Park in de Verenigde Staten. Door de invloed van de wolven op het ecosysteem in Yellowstone te bestuderen krijgen jullie informatie om een oordeel te vellen. Om meer te leren over dat ecosysteem maken jullie gezamenlijk een model. Een model is een vereenvoudigde voorstelling van de werkelijkheid, die je kan gebruiken om ergens over te leren, je ideeën ergens over te ontwikkelen of je ideeën te testen. Wij gaan dit model maken in DynaLearn.

1.

De les gaat over de invloed van de aanwezigheid van wolven op dieren als beren, bevers en verschillende vogelsoorten. Bespreek eerst wat jullie verwachten dat er gebeurt met de populatie van elk dier, wanneer de populatie wolven toe neemt. Vul hierna voor elke diersoort in of jullie denken dat de populatie toeneemt, afneemt of gelijk blijft als de populatie wolven toeneemt. 1. De populatie wolven groeit, dus de populatie beren neemt af / blijft gelijk /neemt toe. 2. De populatie wolven groeit, dus de populatie bevers neemt af / blijft gelijk /neemt toe. 3. De populatie wolven groeit, dus de populatie vogels neemt af / blijft gelijk /neemt toe.

2.

Ga naar devsamen.dynalearn.nl en log in met de inlognaam en het wachtwoord dat je gekregen hebt. Om te gaan samenwerken klik je op . Jouw teamgenoot heeft een code gekregen die jij kan invullen om de samenwerking te starten. Lukt het niet, steek dan je hand op. Thesis Project AI

(30)

Jullie gaan eerst een klein model maken, met daarin alleen een specifieke groep producenten: struiken. De basis van een DynaLearn-model bestaat uit entiteiten. Een entiteit is een onderdeel van het model. Je maakt een entiteit aan door in het menu links op te klikken, en daarna op het scherm te klikken waar deze entiteit moet komen te staan.

3.

Maak een entiteit aan en geef het de naam ‘ecosysteem Yellowstone’.

4.

Maak ook de entiteit ‘struiken’ en koppel deze aan de entiteit ecosysteem Yellowstone. Als je op struiken klikt verschijnt er een menu. Kies voor en klik dan op ecosysteem Yellowstone. Vul bij ‘configuratie’ in ‘staat in’. Dit is het verband tussen struiken en het ecosysteem. De twee entiteiten geven al een eerste indruk van het ecosysteem. Maar entiteiten kunnen ook eigenschappen hebben, die elk het systeem op hun eigen manier beïnvloeden. Zo een eigenschap noem je een grootheid. Een voorbeeld van een grootheid die bij struiken kan horen, is de hoeveelheid.

5.

Maak deze grootheid aan. Klik op de entiteit struiken en kies voor . Geef de grootheid de juiste naam. De struiken leven natuurlijk niet alleen in Yellowstone. In de tekst hieronder leer je meer over een paar andere bewoners. Jouw teamgenoot beschikt niet over deze informatie, dus deel deze informatie met hem/haar door het kort te bespreken. Consumenten in Yellowstone In Yellowstone leven veel verschillende soorten dieren. Eén van de diersoorten die je tegen kan komen is de bever. Bevers gebruiken takken van bomen en struiken voor het maken van hun dammen. Het vinden van takken die ze kunnen afknagen is dus noodzakelijk voor hun voortbestaan. Ook vogels zijn erg afhankelijk van bomen en struiken. Ze bieden een veilige plek, hoog boven de grond. Maar vogels gebruiken ook de bessen die groeien in struiken om van te eten. Zonder bomen en struiken hebben vogels dus geen eten en geen slaapplek. In Yellowstone komen ook veel roofdieren voor, zoals beren. Beren eten niet alleen kleinere dieren, maar ook bessen die aan struiken en bomen groeien.

6.

Welke drie organismen kunnen jullie in de tekst hierboven vinden? 1. 2. 3.

7.

Maak de organismen als entiteiten aan, elk met de grootheid ‘aantal’. Maak hierbij ook het verband aan tussen de nieuwe entiteiten en de entiteit ‘ecosysteem Yellowstone’.

Stimulating collaboration by providing insights in the participation of students in a CSCL environment