Documente Academic
Documente Profesional
Documente Cultură
1. PhD Candidate, Information Technology in Studies, Faculty of Education, The University of Hong
Kong
2. PhD Candidate, Electrical and Electronic Engineering, Faculty of Engineering, The University of
Hong Kong
4. Clinical Associate Professor, Implant Dentistry, Prosthodontics, Faculty of Dentistry, The University
of Hong Kong
*Corresponding author:
Nikos Mattheos
This article has been accepted for publication and undergone full peer review but has not
been through the copyediting, typesetting, pagination and proofreading process, which may
lead to differences between this version and the Version of Record. Please cite this article as
doi: 10.1111/eje.12428
Abstract
Background: Massive Open Online Courses (MOOCs) have been often described as a major
innovation Higher Education, but their application in the teaching of clinical disciplines is still very
limited, while there is a lack of scientific evaluations in this domain. The aim of this study was to
investigate learners' behaviors and correlate patterns of Self-Regulated Learning (SRL) with
Methods: The clickstream data of learners from the first run of the MOOC Implant dentistry by The
University of Hong Kong, was modelled and quantified based on Zimmerman’s SRL model. The data
was quantitatively analysed by means of k-means clustering for evidence of five SRL behavioural
indicators of student activity. The clusters identified were then correlated with student engagement
Results 7608 individuals enrolled, 5014 engaged (active learners 65.90 %), 1277 of them (25.47%)
completed the course and 1232 purchased a certificate. Two major groups of learners emerged:
Attentive (n=1433) who were more likely to follow the prescribed pathway in the MOOC and
Auditors (n=3581) who accessed content selectively. There was significantly higher engagement,
achievement and completion rates among Attentive than Auditors. Both groups included
subcategories (Browser, Digger, Test-driven, Sampler, Persistent) which might reflect different SRL
strategies.
compared to current benchmarks. There appears to be a wide diversity of learning behaviors among
Accepted Article
learners, with two however dominant patterns. Learners with a linear learning pathway achieved
significantly higher grades and completion rates than those who accessed content irregularly and
selectively. Such differences however might be influenced by the demographic and professional
background of the learner, as well as their motivation to attending the MOOC. Certain learning
behaviors, in particular how learners access content in relation to assessments might be closer related
to SRL.
Introduction
A Massive Open Online Course (MOOC) is an open online learning environment which can
simultaneously cater for an unlimited number of learners. The term is reported to have been
introduced in 2008 (1) and quickly gained traction, describing a learning instrument that offers
learners and tutors an open and continual form of networking beyond the limits of conventional online
courses (2,3). A large number of MOOCs are available on an ever-growing array of subjects, with the
most established online platforms reported to reach more than 101 million learners worldwide (2).
The recent development of MOOCs has been hailed by many educators as a significant innovation
within teaching and learning, with the potential to revolutionise the landscape of university education
worldwide (1-3).
Despite the rapid growth of MOOCs in most university disciplines, expansion in clinical sciences has
been slow. The role of open and online learning in the teaching of clinical disciplines, such as
Medicine and Dentistry, has been debated, with educators being divided when assessing the potential
of MOOCs (4,5). It is clear that MOOCs are not to replace dominant models of clinical education
such as residencies and face-to-face teaching environments. However, some educators believe
MOOCs could become the transformative pedagogy to help dental schools overcome many of the
and associated faculty shortages, as well as the high cost of clinic operations (6).
Accepted Article
Implant Dentistry is one of the most recent and rapidly developing clinical subjects within Oral
Health Sciences, evolving around the replacement of missing teeth with surgically placed endosseous
titanium implants. The teaching of Implant Dentistry in the undergraduate curricula is limited, and it
varies significantly between regions and institutions, with universities having many difficulties to
adequately implement it (7,8). As a result, the majority of dental graduates worldwide start their
careers with a very limited understanding of dental implants as part of comprehensive care. Even
long-practicing dentists who were never taught implant dentistry in their curricula have little
opportunity to reach flexible, quality-assured and unbiased education in this discipline. The lack of
human and material resources and expertise is repeatedly cited as one of the obstacles for the
implementation of Implant Dentistry to the undergraduate curriculum to the required extend (9,10). In
that sense, sharing of teaching resources and content at a global level through open and online
learning has been often proposed as a possible remedy to address such deficiencies (11). As of the
time this project was launched (2015), there was no MOOC attempted for the teaching of clinical
disciplines in Dentistry, while few MOOCs had been launched within disciplines of clinical medicine
(12-13). The methodological challenge therefore was evident, to design a robust pedagogical
framework but also to produce evidence of the learning impact for the teaching of clinical sciences
with this medium. Under such a massive and autonomous learning environment, students’ self-
learning ability. According to Zimmerman (14), the degree to which students are metacognitively,
motivationally, and behaviorally active in steering their own learning is described as self-regulated
learning (SRL). As a MOOC caters for large numbers of learners with diverse motivation,
professional background and learning needs, the SRL ability greatly varies among learners, some of
who are students, while others are practicing clinicians under different settings, expertise and
competence.
The aim of this study was to evaluate learners’ dynamics and learning behaviours during the first run
Accepted Article
of a MOOC in Implant Dentistry. In particular, the study aimed to investigate patterns of SRL
behaviours through data mining and investigate correlations of such patterns with the learner’s
This study was approved by the HKU/HA West Cluster Institutional Review Board with reference
number UW 16-1005.
The authors used Zimmerman’s SRL model (14), in order to develop the pedagogical framework of
the MOOC. Zimmerman approached SRL from a cognitive point, where SRL is characterized by
three phases:
Using the three phases of the SRL theory as scaffold, a multidisciplinary learning environment was
designed which could foster the growth of effective learning in three learning “pathways”, developing
(A) Theoretical fundamentals and foundation knowledge, mainly presented through short seminars,
supported by discussion boards suggested readings and assessed by Multiple Choice Quizzes.
and case studies, supported by discussion boards suggested readings and assessed by Multiple Choice
Accepted Article
Quizzes.
(C) Application of knowledge, competences, decision making and guided reflection, by means of
three tailor-made “virtual” patients seeking help. This was also supported by discussions and
assessment was done through non-graded peer and self-assessment. Each patient is presented initially
at the start of the course and the problems are unfolding gradually as the course advances, while
certain tasks are requested by the students at each stage. The model was adopted from the “Interactive
Examination” concept (15), during which the students engage in a guided process with an actual
patient problem, define the parameters of the problem and propose solutions based on the theoretical
and experiential content. A peer solution is then utilized to kickstart a guided reflection, aiming to
help the students identify strengths and weaknesses in their own thinking and define new learning
objectives (15).
b) Learning content
The content and learning objectives of the MOOC were derived from the competencies defined by the
European Consensus Workshop in the university teaching of Implant Dentistry (16) for the graduating
dentist. Fifty-two distinct competencies (11 major and 41 supporting) were encoded into 28 learning
objectives, which were consequently mapped to the three learning pathways and organized in 5
An evident but non-binding “prescribed” learning pathway was established, which encouraged the
learners to follow a sequential access to the different components of the course completing one
module per week. Gradual advancement on all three pathways through the 5 weeks of the course was
engineered through regular “checkpoints” and mentoring by experienced tutors and peers. In addition,
for each one e.g. in video backgrounds, notes etc. Learners remained however free at all times to
Accepted Article
access any content in their own preferred sequence and timing.
Twenty internationally acknowledged experts in respective fields of Implant Dentistry were invited to
contribute according to their expertise by developing content as directed by the map of the identified
competences and learning objectives. The MOOC ran for four rounds on Coursera
(www.coursera.org) during 2016-2017. The analyses in this study are based on the sample form the
first run which took place from 25th October to 5th December 2016.
c) Data collection
The numbers of learners enrolled in the course, as well as the number of course completers and
certificate holders were recorded. The clickstream data of all enrolled learners were collected,
representing user interactions with the content within the learning platform. Such data are matched
with the unique user ID and include any course item clicked or accessed by the user, in the sequence
d) Data analysis
The course was organised in 5 modules, each of which has a number of lessons. Each lesson was
composed by a variety of resources such as lectures, clinical procedures videos, discussion boards and
a graded assessment. The way the students navigated the content was recorded in their individual
clickstream. The clickstream data was modelled and quantified based on the Zimmerman’s SRL
model The data was analysed for evidence of five SRL indicators of student activity performance,
pointing to student learning patterns: a) Time management (TM), b) Task strategy (comprehension)
and e) Application (AP). Definitions of the SRL indicators used can be seen in Table 1. Engagement
Accepted Article
(EN) was defined as the number of completed lessons out of the total number of lessons accessed and
was expressed in percentage from 0-100%. Scores on the basis of the completed graded course
assessments were calculated for each learner and expressed as a final grade (percentage passing grade:
75% - maximum possible 100%). A Python script was programmed in Jupyter web-based platform
(Project Jupyter, jupyter.org, version 5.4.0). Quantitatively, the toolkit of pandas, counter, sklearn
(e.g. k-mean clustering), and n-gram were used to identify clusters of learners based on the indicators.
The identified clusters were qualitatively classified using the constant-comparative method, a
technique that is typically recommended for systematic review pursuits (17). Students behavior for
each indicator was defined as “high level” if the clickstream suggested a score above the average and
“low level” if below the average, based on the defined metrics for each indicator. Pearson correlation
was used to investigate correlations between learners’ behaviors and performance, (grades,
engagement completion rates). Mann-Whitney U test and Chi-square test for independent samples
run on SPSS (IBM, version 24) were used to investigate significant differences in prevalence of
Results
During the first run, 7608 individuals enrolled in the MOOC. Out of them, 5014 learners (65.90 %)
executed at least one action interacting with the content thus becoming “active learners”, and 2957 of
them (58.97 %) completed at least one assessment. 1277 learners completed the course, while 1232 of
them went further to purchase a certificate. Completion rate was 25.47% of active learners or 16.78 %
of the enrolled learners. A certificate was purchased by 96.48 % of the learners who qualified
(completers). The first certificate was acquired already on the first day after the start of the course,
majority of completions and certificates were achieved before the 5th and last week of the course.
Accepted Article
The results of the correlation of the indicators can be seen in table 2. Time management (TM) was
moderately and negatively correlated with engagement (EN) (r = -.303, p < 0.01) and also to the final
grade (r = -.300, p < 0.01). The self-instruction of systematic learning was highly and positively
correlated with the final grade (r = .841, p < 0.01), completion (r = .792, p < 0.01), earning certificate
K-means clustering resulted in the identification of eight clusters among participants in the MOOC
(Table 2). The learners' types were further organised in two major patterns in terms of their learning
engagement on activities, namely “Auditors” (n= 3581, 71.42 %) and “Attentive” (n=1433, 28.58 %)
. Auditor describes participants who demonstrated interest in specific parts of the content and
typically did not follow the prescribed sequence of instructional design (clusters 2, 3, 4, 5, 7, 8 ).
Attentive describes participants who appeared to follow the prescribed pathway, typically including
passing the assessments in sequence and completion of the course. Based on the demonstrated
behaviors, Auditors included a) Browser, b) Test-driven, and c) Digger, while Attentive included e)
Browser refers to participants who went through a few activities mainly at the initial modules of the
course, which they did not revisit such as the participants of clusters 2, 4 and 8. Test driven describes
participants who appeared to revisit activities especially after taking graded assessments, such as the
participants of cluster 7. Digger characterizes participants who were focused in few specific materials
such as video lectures, which they however accessed repeatedly multiple times, such as the
revisit already accessed content, even after taking assessments as in cluster 1. Persistent were learners
Accepted Article
who were more frequently revisiting content, in particular revisiting lectures after taking assessments,
such as cluster 6. In terms of achievement, significant differences were found in final grades,
completion rates as well as acquisition of certificates for learners of the different clusters (Table 3).
Discussion
MOOCs are no longer a novelty in the Higher Education landscape, with the most popular platforms
collectively reaching more than 81 million students already. Nevertheless, learners' behavior and
dynamics within these new learning environments remains poorly understood, as research remains
still scarce. With regards to healthcare, although there is at present a growing number of MOOCs,
published studies are very few and merely reporting expert opinions (4) or descriptive case studies
(18). Although peer-reviewed publications based on MOOCs in healthcare disciplines are emerging,
published studies so far discuss mainly non-clinical MOOCs in domains such as Pharmacology and
Allied Health (18), Medicine as Business (19, Dentistry for non-dentists (20), while there is still a
approach such learning environments in a scientific and systematic manner, in order to determine the
potential role of such learning platforms in the teaching framework of clinical disciplines.
It is difficult to benchmark enrolment and completion rates in a MOOC, as the figures vary
significantly among courses and disciplines and the drop out rated is typically very high. Completion
rates among MOOCs typically vary between 2-11%. In healthcare, A recent report on 4 non-clinical
MOOCs reported higher than the average completion rates, placed between 4.3-11 % (18).
Conclusively, one can note that in comparison to the current benchmarks, Implant Dentistry was well
attended and achieved exceptionally high completion rate, pointing to a favorable potential of this
implies not only quantitative but also qualitative characteristics. In some reports engagement is
Accepted Article
defined as the percentage of learners that performed an "action" among the complete number of
students enrolled (18). As this action could however imply just a single click in a video lecture, what
this indicator mainly shows is rather how many students enrol without ever bothering to just click
once anything in the course. It is apparent that we need much better indicators in order to understand
and assess the actual engagement of students with the learning experience. At the same time there is a
limitation in what clickstream data can offer, without being supplemented with qualitative data from
each user. Unfortunately, in Coursera it proved to be impossible to collect background data from the
learners which could be analysed in parallel with the clickstream and help us pinpoint patterns to
One of the major challenges today in evaluating MOOCs is to find the appropriate instruments,
metrics and indicators to assess the success of a such a course. Traditional metrics and assessments
based on completion and student grades do not completely approach such massive, student-driven
pursuing their own specific learning objectives in attending MOOCs, which are not necessarily met
percentage of resources accessed, the authors defined engagement as the amount of lessons completed
out of all lessons accessed. This was an attempt to capture a more qualitative aspect of engagement, as
each lesson included a variety of resources, videos, discussions boards and assessment. Attending
resources, discussions and assessment even if in one focused area of the course, might define a
different level of engagement than accessing multiple lessons and modules but without utilising all
possibilities of the learning environment, especially the more interactive ones such as discussions and
assessments.
It was obvious that learners approached the course in very different ways and there was a wide
diversity in all indicators that define learner's behavior. The analysis of the clickstream by clustering
certain indicators allowed the identification of some major patterns of navigation and interaction with
achievement in terms of grades and completion. At the same time, as the demographic and
Accepted Article
professional data of the students were not available, it was not possible to identify the types of
learners populating each cluster and investigate their individual objectives and motivation attending
the MOOC. There was a clear division between two patterns: a sequential navigation of the content as
directed by the instructional design (Attentive) and a more irregular and selective access of specific
areas (Auditors). Students with attentive patterns achieved significantly higher grades and were the
majority of the completers. In some studies, such distinct patterns have been related to the SRL
ability of the learner. Littlejohn, et al (21) interviewed 32 MOOC participants who were classified as
higher and lower self-regulated learner (SRLer) based on an SRL survey. They found that higher SR
Learners were less likely to follow a linear progression through the MOOC, while lower SR Learners
were more likely to follow the course in a structured way. Nevertheless, in the case of Implant
Dentistry Auditors and Attentive behaviors are also likely to reflect differences in the professional
background and experience of the learners, rather than SRL ability. As was obvious from the non-
identifiable demographics available, the course was attended by a diverse group of learners ranging
from undergraduate students, fresh graduates and junior clinicians to senior and experienced
practitioners. The first, being at a novice level and less familiar with the entire content were more
likely to follow the prescribed pathway, while the latter, being experienced were more likely to
selectively attend or interact in their areas of interest. For experienced clinicians such areas could be
topics related to their focus of practice, content of certain type (e.g. clinical procedure videos) or input
from certain lecturers. The motivation for completion and certification might have also been different
among learners with different professional background, as purchasing a verified MOOC certificate
might have been of much higher value to students and novice learners than experienced and already
established professionals. This might be one of the reasons why so many of the completers actually
completed the course and acquired the certificates much earlier than the prescribed time. Looking
closer at 31 learners who completed all assessments withon one day, we could find that 26 were
attentive and only 5 auditors. The percentage of certificate acquisitions was exceptionally high for a
MOOC, regardless of the background of those who purchased the certificates. This might point a
need especially within clinical disciplines. Based on the available demographics offered from
Accepted Article
Coursera in August 2018, 64% of enrolled learners were male, while 51% of all learners were in the
age bracket between 25-34. Furthermore, 35% of them were full time students, while 40 % were
employed full time. In terms of education, 20 % already had a Masters degree and 6 % a Doctorate.
Unfortunately, this data is offered as a sample based on the more than 30,000 students enrolled up to
August 2018 and cannot be in anyway correlated to the actual sample used in this study, as
Even if Auditors and Attentive might reflect different professional backgrounds, the sub categories
might be closer to demonstrating actual SRL, as they describe more specific patterns of approaching
the content and assessments. A study based on an SRL survey and clickstream data from 4831 MOOC
participants showed that higher SR Learners were more likely to revisit previously studied course
There is at present little knowledge as to the selection of the ideal indicators in order to study
engagement and learners' behaviors through clickstream data. The 5 indicators modeled for this study
were chosen among many possible combinations, in an attempt to identify certain behaviors based on
the performance phase in the Zimmerman’s model. Although there is a certain risk of bias in the
selection of the five indicators, the fact that some significant differences were finally identified among
the achievement and performance in each cluster suggests that the modeling was relevant to the aims
of the study. In future research, these indicators could be replicated, modified and enriched, hopefully
allowing a deeper understanding of how to best analyse clickstream data. Furthermore, the
availability of anonymised demographic and professional data or learners will empower clickstream
data to offer a much deeper understanding of learners' behaviors in a MOOC and platforms that offer
MOOCs need to seriously consider how to make such data available for research purposes.
learning environment for the teaching of clinical disciplines and can attract a wide diversity of
Accepted Article
learners, achieving much higher completion and certification rates than typically reported. There
appear two fundamental approaches in the navigation and interaction with the content, one of a
sequential engagement and one of more targeted and selective access of resources. The first learning
pattern is significantly correlated with higher engagement, performance and completion rates than the
latter in the MOOC context. These patterns might be influenced by the demographic and professional
background of the learners, as well as their motivation for attending the MOOC. Certain subcategories
however, in particular how learners access content in relation to assessments might be closer related
to SRL. In the future, research based on data from clickstreams combined with an analysis of the
demographic, professional and motivational background of the learners could help us better
understand learners' behaviors and the potential of MOOCs in the teaching of clinical disciplines in
particular.
References
1. Skiba DJ. Disruption in higher education: Massively open online courses (MOOCs). Nursing
2. Shah D. By The Numbers: MOOCs in 2018: Class Central. 2018 [Available at: https://www.class-
3. Kaplan AM, Haenlein M. Higher education and the digital revolution: About MOOCs, SPOCs,
social media, and the Cookie Monster. Business Horizons. 2016 : 59(4):441-450.
4. Kearney RC, Premaraj S, Smith BM, Olson GW, Williamson AE, Romanos G. Massive open
online courses in dental education: two viewpoints: viewpoint 1: massive open online courses offer
transformative technology for dental education and viewpoint 2: massive open online courses are not
Journal of Medical Internet Research. 2014 :16 (8): e197. doi: 10.2196/jmir.3798.
Accepted Article
6. Walker MP, Duley SI, Beach MM, Deem L, Pileggi R, Samet N, et al. Dental education
dentistry education in Europe: the continuum from undergraduate to postgraduate education and
continuing professional development. Eur J Dent Educ 2014 :18 Suppl 1:3-10.
10. Koole S, Vandeweghe S, Mattheos N, De Bruyn H. Implant dentistry in Europe: 5 years after the
ADEE consensus report. Eur J Dent Educ 2014: 18 (Suppl. 1): 43–51.
information technology in dental education. Eur J Dent Educ 2008 : 12 Suppl 1:85-92.
https://www.coursera.org/learn/cataract-surgery
17. Hew, K.F., & Cheung, W.S. Students’ and instructors’ use of massive open online courses
18. Maxwell WD, Fabel PH, Diaz V, Walkow JC, Kwiek NC, Kanchanaraksa S, Wamsley M, et al.
Massive open online courses in U.S. healthcare education: Practical T considerations and lessons
learned from implementation. Currents in Pharmacy Teaching and Learning 2018: 10: 736–743.
19. Robinson R. Delivering a medical school elective with massive open online course (MOOC)
20. Stokes CW, Towers AC, Jinks PV, Symington A. Discover Dentistry: encouraging wider
participation in dentistry using a massive open online course (MOOC). British Dental Journal. 2015
Jul;219(2):81.
21. Littlejohn A, Hood N, Milligan C, Mustain P. Learning in MOOCs: Motivations and self-
regulated learning in MOOCs. The Internet and Higher Education 2016 : 29: 40-8.
22. Kizilcec RF, Pérez-Sanagustín M, Maldonado JJ. Self-regulated learning strategies predict learner
behavior and goal attainment in Massive Open Online Courses. Computers & education. 2017:
104:18-33.
Time management The number of modules The number of modules The higher the indicator,
(TM) the participant accessed accessed (1-5) / the number the more likely is the
per number of days logged of the days participant student to have spread
in logged in (1-35) attendance over more
moduleson fewer days
logged in
Task strategy for The ratio of replays of The times of video lectures The higher the indicator,
comprehension (TS) individual lectures from replayed / the total number the more likely the
the total lectures the of the video lectures participant was to revisit
participant accessed accessed and replay video lectures
Self-instruction The ratio of replays of the The number of replays of The higher the indicator,
(assessment driven) video lectures after a quiz video lectures immediately the more likely the learner
(SI-ad) out of in the total number after an assessment / the to demonstrate
of the assessments the number of the assessments assessment driven learning
participant took the participant took behaviours
Self-instruction of The ratio of the number of The number of lessons The higher the indicator,
systematic learning lessons followed in the accessed in the prescribed the more likely the
(SI-sl) prescribed sequence out sequence / the total participant to have
of total lessons the number of lessons the followed the prescribed
participant accessed participants accessed sequence of the
instructional design
Application (AP) The ratio of the The number of discussion The higher the indicator,
discussions accessed threads attended related to the more likely the student
related to the clinical the clinical problem/ the to be engaged in the
problems set, out of number of clinical problems clinical problems and
number of clinical attended related discussion
problems attended
3. Test Driven 261 238 (91.19%) 37(14.2%) 36 (13.8%) 25.4% 0.26 0–1 ** ** -
4. Sampler 973 940 (96.61%) 791 (81.3%) 766 (78.7%) 69.2% 0.87 0.40 – 1 ** ** ** -
Attentive 5. Persistent 460 451 (98.04%) 402 (87.4%) 384 (83.5%) 69.4% 0.88 0.33 – 1 ** ** ** .077 -
Total 1433 1391 (97.07%) 1193 (83.3%) 1150 (80.3%) 69.2% 0.88 0.33 – 1 - - - - **
EG = Engagement (%)
** means difference of final grade between types at the significant level (p < .01)