1. THE EFFECT OF A LEARNING ANALYTICS
DASHBOARD ON THE PASSING RATE OF A
PROGRAMMING COURSE.
A RANDOMIZED CONTROLLED EXPERIMENT.
Jan Hellings
J.F.HELLINGS@HVA.NL
Amsterdam University of applied science
COMPUTERSCIENCE DEPARTMENT
HBO-ICT
1
7. JAVA PROGRAMMING COURSE
• Introduction course of Java programming of 3 ECTS
• Freshman of computer science education University of applied
science
• 7 weeks 2 lessons of two hours a week
• Flip the classroom
• Two LMS
• Moodle quiz + practical assignment
• MyProgrammingLab of Pearson e-text tutorial
• Book: Introduction to Java Programming, Comprehensive, 10th Ed.
by Daniel Liang.
• Exam: writing a small Java program.
7
8. RESEARCH QUESTIONS
• Will the learning analytics dashboard
improve the passing rate and the grades
of the students participating in the Java
programming course?
• Will the learning analytics dashboard
increase the online activities of students?
8
9. RESEARCH METHOD RCT
9
Freshman
computer
science
n=558
Objection against
participation
Takes part in
theexperiment
n=556
Doesn t take
part in the
experiment
Conditional
randomize
Part of treatment
group
n=276
Part ofcontrol group
n=280
No
Yes
Result
programming
Treatment
group
Result
programming
control group
10. GENERATING DASHBOARDS
PREDICTION MODELS
10
Week 0 Week1
Dashboard
1
Demographic
data
Myproglab
#Attemps
Mastery
Moodle
Quiz
result
Pracitcal
Assignme
nt
Predic
tion
Model
1
week2
Dashboard
2
Myproglab
#Attemps
Mastery
Moodle
Quiz
result
Pracitcal
Assignme
nt
Predic
tion
Model
2
week3
Dashboard
3
Myproglab
#Attemps
Mastery
Moodle
Quiz
result
Pracitcal
Assignme
nt
Predic
tion
Model
3
week4
Dashboard
4
Myproglab
#Attemps
Mastery
Moodle
Quiz
result
Pracitcal
Assignme
nt
Predic
tion
Model
4
week5
Dashboard
5
Myproglab
#Attemps
Mastery
Moodle
Quiz
result
Pracitcal
Assignme
nt
Predic
tion
Model
5
week6
Dashboard
6
Myproglab
#Attemps
Mastery
Moodle
Quiz
result
Pracitcal
Assignme
nt
Predic
tion
Model
6
week7
Dashboard
7
Myproglab
#Attemps
Mastery
Moodle
Quiz
result
Pracitcal
Assignme
nt
Predic
tion
Model
6
week8
Dashboard
8
Myproglab
#Attemps
Mastery
Moodle
Quiz
result
Pracitcal
Assignme
nt
Predic
tion
Model
8
11. RESULTS PASSED
11
n=556. Missing =100. Passed: 332 Failed= 124
figure The mean of passing the exams of the treatment- and control group X
2
(1) ,5,
(1) ,28.
12. RESULTS GRADES
12
Figure the means and standard deviations of the grades for programming of the
treatment – and control group.
13. ONLINE USAGE
MYPROGRAMMINGLAB
13
Figure the mean and standard deviation of the Mastery Level (MMLmastery)
scores of the control – and treatment group extracted from MPL
*p< .10
Mastery level MMLmastery score MyProgrammingLab
per week
Week Mastery Mcontrol Mdashboard p
Week 4 34.928* 42.017* .074
Week 5 27.470* 34.118* .094
Week 6 16.546* 22.508* .051
14. QUIZ RESULTS OF MOODLE
14• Figure means of quiz results of week 1 till 6 extracted from Moodle. No
significant differences.
15. ACCESSING THE DASHBOARD AND
ONLINE USAGE
15
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
1 2 3 4 5 6 7 8
Shareofpopulationoronlie(n=276)
Week
use of dashboard
0 1 >1 online
Online exercises completed
Figure share of population opening dashboard and share online exercises done
16. PREDICTION MODEL PASSING
ACCURACY
16
Table compare precision accuracy (PA), type I error (T1) and type II error (TII) of
passing prediction model for week 1 till 8 of constructed model vs real.
0,000
0,100
0,200
0,300
0,400
0,500
0,600
0,700
0,800
0,900
1,000
1 2 3 4 5 6 7 8
Accuracy/TI/TII
Week
Accuracy of passing prediction model per week
PA PA-Model TI TI-Model TII TII-Model
17. PREDICTION MODEL GRADE (GPM)
Figure R2 for Grade Prediction Model (GPM) per week
1 Tempelaar, D. T., Rienties, B.,& Giesbers, B. (2014).
In search for the most informative data for feedback generation: Learning analytics in a data-rich context.
Computers in Human Behavior. http://doi.org/10.1016/j.chb.2014.05.038
0,000
0,100
0,200
0,300
0,400
0,500
0,600
0,700
0,800
1 2 3 4 5 6 7 8
R2
Week
Coefficient of determination R2 of grade prediction model compared to
literature1 and generated model
GPM Lit model
18. COHORT 2014 VS 2015 MPL
18
Figure MyporgrammingLab mastery level scores (0-100) per week of cohort
2014 and 2015 week 1 (W1_MMLMastery) till week 6 (W6_MMLMastery). Only
W1_MMLMastery p>.000
19. QUIZ RESULTS OF MOODLE
19
Table Moodle quiz results (0-10) of
cohort 2014 vs 2015. W3_quiz and
W6_quiz p >.05
22. CONCLUSIONS RESEARCH QUESTIONS
• Will the learning analytics dashboard improve the
passing rate and the grades of the students participating
in the Java programming course?
• Dashboard had no effect on the passing rate and grade
• Will the learning analytics dashboard increase the online
activities of students?
• Dashboard had a small effect on the online behaviour
about 5% more on p<.1
22
23. CONCLUSIONS
• RCT: differences outcomes between treatment and control
group attributed solely to intervention.
• Prediction models
• Prediction accuracy model passing .528 till .681 Lit :
.75-.863 or .972-.9792. Type II error real model >>
generated model.
• Coefficient of determination R2 of grade model: .152 till
.331. Lit: .4 -.71 Until week 6 better generated model.
• Online usage of cohort 2014 >> cohort 2015
• Flip the classroom make pre-class preparation compulsory or
summative.
23
2 Hu, Y.-H., Lo, C.-L., & Shih, S.-P. (2014). Developing early warning systems to predict students’ online learning performance. Computers in
Human Behavior, 36(0), 469–478. http://doi.org/http://dx.doi.org/10.1016/j.chb.2014.04.002
3 Lauría, E. J. M., Moody, E. W., Jayaprakash, S. M., Jonnalagadda, N., & Baron, J. D. (2013). Open Academic Analytics Initiative: Initial
Research Findings. In Proceedings of the Third International Conference on Learning Analytics and Knowledge (pp. 150–154). New York,
NY, USA: ACM. http://doi.org/10.1145/2460296.2460325
1 Tempelaar, D. T., Rienties, B.,& Giesbers, B. (2014).
In search for the most informative data for feedback generation: Learning analytics in a data-rich context.
Computers in Human Behavior. http://doi.org/10.1016/j.chb.2014.05.038