How Effective is Your Facilitation? Group-Level Analytics
of MOOC Forums
University of South Australia
Delft University of Technology
Teaching Innovation Unit
University of South Australia
Institute for Intelligent Systems
University of Memphis
The facilitation of interpersonal relationships within a respectful
learning climate is an important aspect of teaching practice.
However, in large-scale online contexts, such as MOOCs, the
number of learners and highly asynchronous nature militates
against the development of a sense of belonging and dyadic trust.
Given these challenges, instead of conventional instruments that
reflect learners’ affective perceptions, we suggest a set of
indicators that can be used to evaluate social activity in relation to
the participation structure. These group-level indicators can then
help teachers to gain insights into the evolution of social activity
shaped by their facilitation choices. For this study, group-level
indicators were derived from measuring information exchange
activity between the returning MOOC posters. By conceptualizing
this group as an identity-based community, we can apply
exponential random graph modelling to explain the network’s
structure through the configurations of direct reciprocity, triadic-
level exchange, and the effect of participants demonstrating
super-posting behavior. The findings provide novel insights into
network amplification, and highlight the differences between the
courses with different facilitation strategies. Direct reciprocation
was characteristic of non-facilitated groups. Exchange at the level
of triads was more prominent in highly facilitated online
communities with instructor’s involvement. Super-posting activity
was less pronounced in networks with higher triadic exchange,
and more pronounced in networks with higher direct reciprocity.
MOOCs; forum; facilitation; indicators of social activity; ERGM.
With the massification and commercialization of higher
education, universities place more demands on educators to
deliver quality teaching to more students at lower costs. As a
result, educators need to adapt and modify their pedagogical
approaches to better accommodate a significantly larger and more
diverse group of students. Well-designed assessment to some
extent can help teachers determine if their efforts to help students
learn were effective. However, student performance data alone,
does not provide sufficient information about other aspects of the
learning and teaching context. An alternate approach is required
to explore if a teacher’s implemented facilitation processes were
successful in promoting interpersonal relationships and peer-to-
Various instruments have been designed to capture the evolution
of in-class social relations related to the positive impact on
learning. Examples include the community of inquiry
questionnaire , surveys of social presence  and sense of
belonging . These instruments evaluate the state of trust and
community often at the group-level as the ultimate outcome of
interpersonal relationship formation.
The importance of trust-based relationships has been associated
with small online and face-to-face classes where the group
boundaries are fixed. However, there remains a significant
challenge in leveraging the benefits of such interpersonal
relationships in large learner cohorts particularly in the non-
formal contexts. In massive open online courses (MOOCs) learner
participation in social activities is intermittent , and early
interactions can amass to a level of chaos  that impedes an
individual’s propensity for developing relationships. For instance,
according to Gillani, MOOC forums ‘assemble and disperse as
crowds’ . Additionally, as stated by Poquet et al.  the
relatively short time frames associated with MOOC offerings
further diminish the opportunities for learners to develop
interpersonal trust. Simply put, large course size, short duration of
teaching and the non-formal nature of MOOCs militate against
the development of community. In such instances the use of
established instruments evaluating if learner-learner bonds have
been forged becomes irrelevant.
The amplification of communication between participants could
serve as an effective and proactive indicator for group processes.
The promotion of connections between learners, concepts and
artefacts is favoured by networked learning, connectivism and
socio-material approaches [1, 28, 48]. For example, a connectivist
approach to teaching in networked systems includes controlling
the network of learners through a facilitating role – by amplifying,
curating, way-finding and socially-driven sense-making,
aggregating, filtering, modelling and being persistently present
. In this context, the role of the teacher becomes strongly
aligned with promotion and facilitation of network’s
interconnectedness. Thus, an amplified network of learner
communications may indicate if a structure and a climate
conducive to egalitarian participation has evolved.
This present study uses network indicators to explore social
activity in an open online course. Through these indicators the
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. Copyrights
for components of this work owned by others than ACM must be
honored. Abstracting with credit is permitted. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior
specific permission and/or a fee. Request permissions
LAK '17, March 13-17, 2017, Vancouver, BC, Canada
© 2017 ACM. ISBN 978-1-4503-4870-6/17/03…$15.00
study examined amplification of communication within the
network of learners in ten MOOCs. Exponential random graph
modelling (ERGMs) was applied to understand the extent of
direct and triadic-level reciprocity and the effects of participants
posting activity within each network. ERGMs required a
theoretically-driven hypothesis to be interpreted meaningfully.
Thus, based on our prior work we conceptualized groups of
regularly posting forum users, i.e. forum cores, as identity-based
online communities. Communication within the communities
were abstracted into network graphs for ERGM analysis. Results
highlighted the differences in propensity for tie formation and the
probability of information exchange across networks of different
size and course design. The examined MOOCs were conducted in
STEM disciplines, and delivered by the Delft University of
Technology via the edX platform in 2013-2015.
2. REVIEW OF LITERATURE
2.1 Analytics for Group-Level Social Activity
The social fabric of a learning group evolves from the structures
and processes created as the collective completes the assigned
work. The development of effective group strategies requires
teachers to adopt strong facilitation skills and techniques. In this
context, facilitation skills refer to the ability to design and manage
learning group structures and processes, and employ appropriate
and timely strategies to minimize the common problems that arise
when people work together . Facilitation skills, learning
design and support of knowledge construction are all required to
establish an effective online teaching program . However,
facilitation can be seen as distinct from the other two. Learning
design is geared towards teachers decisions about learning tasks,
course structure and assessment . The support of knowledge
construction, or ‘content facilitation’ , refers to a teacher’s
ability to promote students’ epistemic fluency. Facilitation is
content-neutral in that it promotes a structure and a culture of
participation built on individual dispositions to the group and one
another. Measuring learning outcomes that result from effective
facilitation is particularly challenging because they fall within the
Group-focused analyses can provide novel insights into how
social activity unfolds. For example, Huang et al  examined
the health of forums at the group level in 44 Coursera MOOCs in
relation to super-posting behaviours of its learners. The authors
found that overall higher activity from super-posters is positively
correlated with higher activity from other forum users, concluding
that the super-posters do not ‘drown out the silent majority’
(p.118). It should be noted, however, that such an evaluation of
the health of the forum was limited to the correlation of individual
counts of content-related messages between different types of
learners. The study is a reflective of how social activity on forums
is approached in MOOC research to date.
In current research of platform-based MOOCs, most indicators of
social activity are not well suited to evaluate the state of social
activity in a group. Analyses have been predominantly conducted
at the level of the individual, often as a complement to academic
performance (grades). Numerous studies have analysed the
association of a number of forum posts in relation to an
individual’s performance . For example, Kizilcec et al. 
demonstrated that completing learners who attempted a majority
of the assessments were also more active on the forum. Vu et al.
, however, found that the relationship between learning
performance and social interactions is not bi-directional. Although
high learning performance was positively correlated with active
social interactions, active social interactions were not necessarily
associated with performance. The importance of these analyses
lies in grasping the value of social activity for individual student
certification, however, they offer little insight into the group-level
Studies measuring individual engagement with others in the form
of social capital offer an alternative to evaluating the role of social
activity through post counts. Such an approach is based on
quantifying an individual learner’s position within the network of
posters [8, 25, 26]. However, such an approach does little to
reveal how the network emerges from the collective contributions.
To explain, although individual learning outcomes such as social
capital do provide information about social activity, they are not
sound indicators of group-level activities.
Where social activity in MOOCs has been evaluated at the group-
level, the evaluations have tended to privilege the content-
dimension of interaction. For instance, Hecking et al. 
encapsulated social interaction within content-related posts of the
network into blockmodels describing the overall network pattern.
Or, Kellogg et al.  reported knowledge construction within the
network of learners through qualitative analysis of learner posts.
Similarly, Wang et al.  identified and evaluated groups in the
course based on higher order thinking behaviours reflected
through posted text. These examples demonstrate that the
socioemotional dimension of learning is often taken for granted.
As argued by Kreijns et al. , when a need for the
socioemotional is recognized, the dimension is examined in the
context of a content-loaded learning task or as targeting cognitive
2.2 Social Network Analysis in MOOCs
Social network analysis (SNA) is widely used for evaluating
group-level social activity. SNA as a methodology combines
learner agency with the context where it operates. Thus, it offers
group-level insights into social activity in MOOCs. Early on
Ferguson & Shum  suggested the use of SNA for investigating
interpersonal relations mediated by social platforms. They
proposed to examine social relations through the prism of strong
and weak ties – theoretically loaded terms borrowed from social
science theory. This proposition aligns with the suggested focus
on inquiries into learning ties by Haythornthwaite & de Laat .
However, any inferences about group-level activity in MOOCs
drawn from current SNA research in learning analytics needs to
be enacted with caution.
On the one hand, SNA requires its methodological decisions to be
theoretically driven. Network research is relational, that is, its
main unit of analysis is a theoretically defined relation between
two actors. The meaning assigned to this unit of analysis directly
impacts the interpretations of the results. In short, a network is a
representation of a phenomenon, and consequently, this
phenomenon needs to be conceptually represented in the data .
In practical terms this is done by providing justification for
inclusion and exclusion of ties and edges, as well as drawing
network boundaries. According to Laumann, Marsden and
Prensky , errors in defining these boundaries reach beyond the
consequences of slightly biased estimates of population means,
proportions, or inefficiency in statistical estimation. Flawed
boundary specification may lead to “the fundamental
misrepresentation of the process under study, since …errors of
omitting one actor may distort the overall configuration of actors
in the system and render the entire analysis meaningless” (p.19).
On the other hand, there is a lack of theorising social processes in
open online courses. As a result, there is a divergence in the
research resulting from the differing conceptualizations of
network ties. This largely stems from how post-reply structures
are interpreted. For instance, researchers examining Twitter
networks in MOOCs analyzed communication represented
through person-to-person information flow [15, 50]. Twitter
communication interactions are possibly the least ambiguous to
interpret as they are by design identical with the name networks
where there is little doubt as to ‘who spoke to whom’ . Post-
reply communication structures appear to be replicated in studies
of platform-based MOOCs delivered via edX and Coursera.
However, the approaches adopted to analyse the forum data
differ. For instance, Joksimovic, et al.,  used a conventional
directed post-reply structure commonly used in online education
forums: “if author A2 replied to a message posted by author A1,
we would add a directed edge A2->A1. Further, if A3 posted a
comment on A2’s post, we would include A3->A2 edge as well”.
In contrast, Brown and colleagues ,  used a post-reply
network structure based on communication going from one person
to many prior posters. Gillani, et al.,  adopted another
alternate approach in a Coursera MOOC. In this instance, the
authors represented communication through an edge connecting
“two learners simply if they co-posted in at least one discussion
thread”. The diversity in how researchers are conceptualising
information/ communication flow as network structures impedes
our capacity to make broader inferences about group-level
learning activity. Hence, to draw meaningful interpretations from
graph analytical techniques employed in SNA of MOOCs, it is
imperative to theorise and conceptualise the phenomenon that is
3. THE CURRENT STUDY
3.1. Our Approach
The aim of the current study is to develop indicators to evaluate
social activity in MOOCs, particularly in relation to forum
facilitation. Given that interpersonal relationships at scale may be
unattainable, we propose to examine the amplification of
communication within the networks of MOOC learners.
In our earlier research we argued that teaching online approaches
are derived from formal educational contexts, and there is a strong
need for new analytical frameworks that are distinct for MOOC
settings . While open and formal online learning appear
similar, processes in formal online modes are streamlined by the
strict starting dates and learners’ motivation to complete. The
formal boundaries and learner motivations lie in stark contrast to
the more informal and open MOOC contexts. Yet, both research
and practice use theoretical and methodological frameworks
inherited from formal settings to analyse open online courses. To
overcome the challenge of compatibility of the tools with the
object of analysis, social processes emerging on MOOC forums
first need to be examined for forum sub-groups comparable in
dynamics to formal education settings. In contrast to many
MOOC learners who drop into the forums and leave, persistent
forum contributors experience a greater sense of continuity and
established history in their relationships typical for social groups
. Thus, a sub-set of forum participants, termed as the forum
core, was established.
The forum core are posters who contributed to the forum in at
least any three weeks of the course and received replies to their
posts within the same course week. The underlying characteristic
of this sub-group is repeated presence and timeliness of the
reciprocated communication exchanges. The number of posts
does not distinguish this group from other posters. Some forum
core members could have as few as four to five posts, and some
non-forum core posters could have over a hundred.
To meaningfully capture indicators of participation patterns we
conceptualized forum core, i.e. a group of returning MOOC
posters, as identity-based community. Then, building on prior
research examining the properties of network formation in social
exchange networks, we inquired about the effects different
facilitation strategies may have on network configurations. The
underlying theoretical assumptions were evaluated by statistical
network analyses applied to ten cases of MOOCs. The following
section offers an overview of the prior work and relevant
theoretical and empirical literature that shaped the research
3.2 Identity-Based Communities
Education research has tended to define community development
within two broad categories – community formed through a
common-identity or formed via a common-bond. The differences
between these categories lie within the origin of the group-
attachment. One-to-one attachment among members is a
consequence of people liking each other. When approaching
attachment from a bond-based perspective, the focus is therefore
on interpersonal, dyadic relationships of trust. In contrast,
common-identity attachment is a result of self-identification with
group’s purpose as in topic-based groups, such as sports team or a
school newspaper. These differences between common-identity
and common-bond communities are best captured by what
happens when a member leaves. When attachment is motivated by
identity, other members are perceived as interchangeable, and
turnover of membership does not impact individual behaviour as
much as in common-bond communities where members will leave
following their friends.
A common-identity approach has previously been applied to
informal online groups [41, 47, 51]. Postmes, Spears & Lea 
argued that the power of anonymity in groups where individuals
do not have cognitive representations of other individuals could
enhance the salience of a group’s identity. The authors suggested
that ‘paradigms in which group members do not meet face-to-face
provide precisely those conditions predicted to maximize social
influence exerted by social norms and social identities’ (p. 7).
Furthermore, in a series of experiments, Ren et al. 
demonstrated that facilitation of identity-based attachment had
stronger effects on the frequency of engagement than conditions
targeting interpersonal attachment development. Therefore,
defining a community through identity-based attachment of one-
to-many is of high relevance in educational settings such as
MOOCs where the scale and non-privacy of the environment
aggravate the development of bond-based relationships.
In light of identity-based community theory, this present study
approached forum core as a social entity where attachment is
dictated by self-identification with a group without necessarily
investing in specific person-to-person relationships. Examination
of the discourse produced by forum core indicated traces of social
processes that forum core posters engaged in . More
specifically, through qualitative analysis of forum content, Poquet
 identified the presence of discourse negotiating situational
norms, shared domain of practice, i.e. MOOC forum, and shared
experience, and, MOOC learning and assessment. Furthermore, in
a complementary study of the perceptions among those
contributing regularly Poquet et al.  found that forum core
participants reported the lack of dyadic trust and dyadic
familiarity in interpersonal perceptions but presence of higher
group cohesion perceptions. In short, forum cores in some courses
engaged in collective social processes, and developed perceptions
of group cohesion without underpinning interpersonal trust. This
empirical evidence allows to conceptualize the forum core as an
identity-based community based on information exchange.
3.4 Network Exchange Patterns in Online
Insights into the baseline properties of similar networks are
required to transition from theoretical analysis of forum core as
identity-based communities to applied analysis of the
amplification of its communication exchange. Studies of
electronic networks of practice offer information about network
topography in identity-based communities that help modelling
forum core participation structure.
Electronic networks of practice are conceptually aligned with the
major premises of identity-based communities. They are informal
groups that share knowledge about specific subject of interest
from photography groups on flickr to Wikipedia contributors to
software developers blogging community. According to Wasko et
al., , participants of public electronic networks reported trust
in the quality of information from active members (i.e. attachment
to the group), and that those actively participating were perceived
as trustworthy (i.e. belief in the group), thereby eliciting a
stronger motivation to continue participation (i.e. commitment to
participating in group’s activity). Faraj & Johnson theorised such
online communities as social exchanges between participants
situated in a network context. That is, regardless of the content of
exchange, the interactions are more than just information queries,
but also have social nature and social purpose. In other words, in
their activities posters are conscious of the potential use by
readers, as well as current and future contributors through what
becomes a part of shared history. Simply put, electronic networks
of practice appear to represent a particular kind of an identity-
The work of Faraj & Johnson served to empirically validate the
consistent and predictable network exchange patterns that emerge
from divergent motivations. They proposed three network patterns
that characterize the formation of the network of practice on the
micro-level: direct reciprocity (A replies to B because B helped A
in the past), indirect reciprocity (B helped to A, and thus A will
help C), and preferential attachment where new actors choose to
interact with already well-connected actors. Through a large-
scale longitudinal research of Yahoo! Bulletin boards, they found
a significant positive high propensity of direct reciprocation (A to
B to A), significant positive low propensity for indirect exchange
happening (A to B to C), and a negative propensity for
preferential attachment. The negative propensity for preferential
attachment has also been supported in studies of MOOC forum
networks [27, 30]. In summary, network configurations of
reciprocity lend themselves nicely to modelling baseline network
properties in forum core networks for further examination of the
differences between them in different facilitation contexts.
3.4. Focus of the Study
Building on the discussed theoretical and empirical literature,
MOOC forum cores are assumed to reflect natural patterns
occurring in identity-based communities characterized by network
exchange. Accordingly, forum cores that emerge and evolve in
courses with pedagogical interventions constituted by moderation
strategies would demonstrate different dynamics, varying along
with the intensity of facilitation. These differences in network
structure in courses with moderate and higher moderation would
then indicate the effects of intended forum strategies. In line with
this argument, the following research question was posed:
Are there differences in patterns of direct and triadic-level
reciprocity in forum core networks under different facilitation
To address this research question forum core communication
networks were analysed using models of social selection based on
the p*class models .
4.1. Network Construction
Forum core networks were constructed based on the premise that
they represented identity-based communities. That is, the analyses
were constructed from the forum core posters that were
committed to participating in forum’s activity, and driven by the
mix of altruistic and selfish reasons to engage in information
exchange. Their commitment was captured by the extended
participation criterion that distinguished forum core posters from
Network ties were non-valued and directed representing the
direction of the service offered. The ties were directed to all the
posters in the same thread who preceded the actor based on
timestamps. If A posted, B replied, and C commented, and D
replied, each of the subsequent actors would have a directed tie to
everybody else prior to them. That is, B -> A, C-> B, C->A, D-
>C, D->B, D->A. Such tie inclusion was based on the principle of
collective reciprocity where an individual contributed information
to the group, rather than provided a service to an identified
The network boundaries were set around the course delivery time:
between the week the first video lecture was released, and the
week when the last lecture in the course was released. As
communication between actors can potentially continue well past
the course lectures due to exams or assignments, any established
closing date for the forum analysis would be arbitrary. The
discrete temporal limits applied in this study made the comparison
of courses more feasible.
4.2 Data Description
This present study analysed the forum interactions evolving from
ten (10) STEM courses delivered by the Delft University of
technology via edX platform in 2013-2015. Due to the evaluative
nature of the study, the courses were de-identified. Table 1 offers
an overview of the disciplines, sizes of posting cohort against the
forum core, course duration and qualitative description of the
forum facilitation strategies.
The dataset included five large MOOC with forum core higher
than one hundred people, and five smaller courses with forum
cores ranging from 23–70 people. Three courses were highly
facilitated, i.e. instructor was among active posters, along with
designated staff members and TAs. Five courses had moderate
facilitation, i.e. although staff and TAs helped the information
flow, instructor was not involved with the forum. Finally, two
courses had low facilitation, i.e. nobody moderated the forums
except one or two staff announcement were posted.
Table 1. Course overview with forum facilitation strategies.
4.3 Exponential Random Graph Modelling
Exponential random graph modelling (ERGM) was used to
analyse the network properties describing the structure of forum
core networks. ERGM is a methodology for the analysis of social
selection models. These models assume that characteristics of
actors influence them to select or be selected by others as social
partners . These micro-level interactions resulting from social
selection aggregate into group-level patterns that describe the
network. Additionally, p*/ERGM presumes that multiple
processes can take place simultaneously, and social networks are
both structured and stochastic [36 p.10].
ERGM is a probability model for network analysis [17, 19, 36,
37]. In modelling ERGM estimates the likelihood of a parameter
for a theoretically justified network configuration (form) to occur
beyond chance. This is implemented by examining the likelihood
of a studied configuration (network form) to occur in a generated
distribution of random networks modelled on the network of
interest. Additionally, multi-level analysis within ERGM controls
for the tendency of studied parameters against one another, due to
their theoretical dependency. The output includes a parameter
estimate where zero indicates that the modelled effect is not
different from random. Estimated parameters are considered of
value after the goodness of fit is conducted upon the best fitting
4.4 Modelled Configurations and Effects
This study investigated three configurations within forum core
networks: direct reciprocity characterized by mutuality statistic,
triadic-level exchange characterized by simmelianties statistic,
and the effect of learner super-posting activity on the propensity
of sending and receiving ties. Figure 1 depicts three network
configurations. The first two represent configurations of direct
and indirect reciprocity that, according to Faraj & Johnson are
baseline properties of network formation in identity-based
communities. We further hypothesised that forum core members
who overtime co-occur within direct (A to B to A, Figure 1a) and
indirect reciprocity (A to B to C, Figure 1b) will overtime form
triadic network configurations representing mutual and triadic
exchange that eventually could extend to a cycle if interaction
should continue. Joksimovic, et al.,  previously fitted similar
network configuration (Figure 1c) using simmelianties statistic in
ERGM for MOOC forums. The authors interpreted it as
Simmelian ties . From the analytical perspective, we take
inspiration and build on their approach. Yet, we interpret this
network configuration as amplification of parts of the network
due to generalized information exchange between the actors. Only
direct reciprocation and triadic-level exchange configurations
were therefore modelled.
a) Direct Reciprocity
b) Indirect Reciprocity
c) Triadic-level exchange
Figure 1. Network Features Modelled in ERGMs. Adapted from
Faraj & Johnson (2010).
Markov network configuration were not used in the study to
model triadic exchange. Instead we use simmelainties statistic
available in R statnet package . Morris, Hunter & Handcock
 explain that simmelian triads (Figure 1c) can overlap in terms
of nodes and ties, thus simmelianties is rather a ‘measure of the
clustering of Simmelians (given the number of Simmelians).’
Although social circuit parameters (gwesp) were fitted on some of
the networks, they were not characteristic of all networks, and
thus not used to maintain comparability in the modelled outputs.
Table 2 presents the counts of modelled network configurations
for each of the networks.
Table 2. Counts of network configuration in analyzed
Ties embedded in
Note: *equivalent statnet name for the parameters
Additionally, in ERGMS we controlled for the activity level of
each learner within the network using the nodefactor parameter.
To establish the overall level of activity, we applied in-reach and
out-reach measures of each individual activity as suggested by
Hecking, Hoppe & Harrer . The authors developed measures
of entropy to calculate the diversity of outgoing and ingoing
relations for a node, and account for the weight of an edge, i.e.
frequency of dyadic ties between two given actors. We have
replicated their measure1, and applied it to the entire forum
network. We then applied k-means clustering to divide all forum
posters into three groups: with highest, moderate and low forum
posting activity in the entire network. These attributes of posting
activity were used to control for tie propensity formation in forum
core ERGMs. Table 3 presents the number of actors within each
cluster. Table 3 demonstrates the number of people with different
posting activity in each of the networks. Cluster 1 refers to the
learners with low posting activity. Cluster 2 refers to the posters
with moderate posting activity, and Cluster 3 represents
hyperposters. These interpretations do not fit to describe Course
C, where Cluster 3 and Cluster 4, in fact have similar activity
levels, but Cluster 2 has higher in-reach measures, and Cluster 3
has higher out-reach measure.
Table 3. Number of people in the cluster representing
different posting activity after k-means clustering of in-reach
and out-reach measures of posting activity in the entire
Labelling learners as moderate, or of high and low activity was
relative to one another within the course. That is, a hyperposter in
one course may have similar activity numerically as the moderate
poster in another course. Appendix 1 presents the measures for
This study examined the extent of communication exchange
amplification in the forum core networks of multiple MOOC
courses. Using an ERGM approach, we modelled the baseline
density of networks, direct reciprocity of ties, and triadic-level
exchange as the propensity of simmelianties network
configuration to cluster. We have also controlled for the level of
activity distinguishing between individuals with low, moderate
and high posting behaviour. The research question inquired if any
differences were observed between the measures of direct and
triadic-level exchange in networks where staff implemented
different facilitation strategies.
All fitted ERGMs have converged, were not degenerate, and
showed acceptable goodness of fit. All models were fitted using
the described network configurations. The two exceptions were
course G where models did not converge with the effects of
posting activity, and Course D where no super-posters were a part
of the forum core. Table 4, 5 and 6 outline the results grouped
together for courses with similar facilitation strategies.
First and foremost, our hypothesis that courses with no or low
facilitation reflect the network structures typical for online public
electronic groups was supported. In courses with low facilitation
(Table 4) we observed higher propensity for direct exchange, and
low or no propensity for exchange at the triadic-level. Course B,
as a large course, had posters with high level of activity, and they
were somewhat more likely to form ties than the posters with
moderate participation, though the difference was not stark.
Course G was very small, had no triadic-level exchange features,
and the null model was only slightly improved by adding the
direct reciprocity configuration.
In courses with moderate facilitation (Table 5), where teaching
assistants, staff or community assistants were advancing the
information flow, the dynamics had some similarities and
differences with unmoderated settings. For the similarities, theta
estimates for reciprocity in Courses D and F were higher than
those for triadic-level exchange mirroring the dynamics of
unmoderated courses. In three small courses (A, F and H) triadic-
level exchange parameters were not significant, or even negative.
Interestingly, the propensity of super-posters to form ties as
compared to those posting moderately was more pronounced in
courses with lower or no triadic exchange, and heightened
Table 4. ERGMs outputs for courses with low facilitation
In contrast, course F with highest propensity for triadic-level
exchange features to cluster, indicating higher network
amplification, has least difference between the propensity for tie
formation between those with moderate and high activity.
Table 5. ERGM outputs for courses with moderate facilitation
Main Effects on Ties Formation
These differences between a moderated Course A and
unmoderated Course B can be interpreted as follows. In an
unmoderated network information exchanges are random,
distributed and decentralised. In moderated networks, moderators
have much higher posting activity, sometimes ‘dominating’ in
offering information to other learners. When this occurs, the
triadic-level exchange features seem to be either low, negative or
non-significant. By translating the log odds, we observe that the
high participation posters in course A were 22 times more likely to
form a tie as compared to low posters, while moderate posters
were three times more likely. At the same time, in an unmoderated
course B: posters with higher activity were six times as likely to
make a post than those with low posting behavior, and moderate
posters were three times more likely. We can speculate that in
cases where super-posters do not over-dominate, activity of those
with moderate posting behavior overtime grows, and these actors
engage in more conversations with one another as well as
interacting with super-posters, thereby resulting in an
amplification of communication exchange.
Finally, network structures in courses with high facilitation
differed from both moderated and unmoderated courses without
instructor participation (Table 6). Overall, it appears that in
courses with high facilitation, triadic exchange configuration has
a higher likelihood to occur than direct reciprocity. Such
dynamics are demonstrated in courses E and J, where clustering
of reciprocal triads are more characteristic of the network than
person-to-person reciprocations. In a course I, however, direct
reciprocity is still more characteristic of the network, which we
interpret as the lack of amplification within the information
exchange. Again, similar to the pattern observed with the
moderately facilitated courses, in course I, high posting
individuals are much more likely to form ties. In fact, by
converting log odds, we found that a super poster in course I was
eleven times more likely to form network ties than those with a
low level of forum activity. In courses E and J the likelihood for
super-posters to form ties was three times more than participants
with a low posting behavior. To note, the individual number of
posts by super-posters in course I was actually lower by count
than that of super-posters in the other two courses.
Table 6. ERGM outputs for courses with high facilitation
To extrapolate, the network dynamics is opposite between courses
with high facilitation and low or no facilitation. Direct reciprocity
of knowledge exchange seems to be inherent in identity-based
communities, such as forum core. Overtime and with facilitative
efforts network structure starts being defined by core and
periphery: with both random reciprocation and low clustering of
more generalized, triadic, exchange, likely at the core of the
group. This clustering may also be interpreted as clique
formation, or polarization of power and access that takes place as
the network shifts from distributed to amplified. Facilitated forum
core networks appear to be characterized by higher degree of
clustering of reciprocated triads, and lower level of random
reciprocation. This could mean that random direct exchanges
within the group decrease, as information flow gets to amplify
across more and more members, shifting from cliquish core and
distributed periphery to an amplified interconnected cluster. We
also observed that super-posting activity does seem to be
associated with lower level of exchange between triads. In other
words, a moderator may be taking over what somebody else could
address by offering her services too much or too fast, therefore
not allowing other members to indirectly reciprocate to the group.
To conclude, according to ERGM results, network features
modelled to gain insight into network amplification were useful in
highlighting differences between the courses with different
facilitation strategies. More specifically, direct reciprocation was
characteristic of non-facilitated groups, while triadic-level
exchange was more prominent in highly facilitated online
communities with instructor’s involvement. Finally, super posting
activity was less pronounced in networks with higher triadic-level
exchange, and more pronounced in networks with higher direct
The aim of the current study was to establish potential indicators
for evaluating social activity in MOOCs, particularly in relation to
forum facilitation. Thus, in lieu of measuring affective individual
perceptions as indicators of socio-emotional processes, the extent
of amplification in communication exchanges between MOOC
learners was examined. A sub-group of MOOC forum posters was
conceptualized as an identity-based community. Building on the
prior research on the properties of network formation in social
exchange networks, we modelled ten forum core networks using
network configurations of direct reciprocity and triadic-level
exchange, while controlling for the overall level of individual
Results support our hypotheses about the nature of the forum core
in MOOCs, and suggest that the chosen indicators of forum core
networks may be useful in evaluating the effects of facilitation.
That is, courses with varying levels of moderation differ in the
likelihood of direct and triadic exchanges that take place. In non-
facilitated courses, the dynamics of the networks was similar to
electronic networks of practice, and largely characterized by
direct reciprocity. In highly moderated networks, triadic-level
exchange configurations were more prominent than direct
reciprocity. That is, these networks were amplified by the
propensity of reciprocated ties in triads to cluster. Furthermore, a
higher degree of triadic-level exchange seemed to take place in
networks where activity of super-posting participants was less
pronounced as compared to those posting moderately or at low
The main contribution of the study is that its indicators of group-
level social activities differentiated between the network
structures in forums with different facilitation strategies. Our
results demonstrated that moderating the forum per se is
insufficient for the effective evolution of participation. Courses
with teaching assistants and staff demonstrate different patterns
than those with instructor involvement. In courses with moderate
facilitation interventions reciprocal triads have lower likelihood to
occur than in highly facilitated forums with instructor’s
involvement. These findings re-iterate the importance of teacher’s
social presence and its impact on the level of engagement in open
It is not surprising to learn that an instructor’s presence motivates
learners to participate - even in large scale courses. This finding
has been widely discussed and supported in formal online
education research. This study offers additional insights into the
roles, played by the staff, teaching and community assistants.
MOOC research claimed their importance in supporting the spirit
of the forums. Our findings further demonstrate that moderated
forums appear to transition from their baseline properties, as the
structure shifts from distributed to more hierarchical network.
The results of the present study also support Huang et al.’s
conclusion that super-posters do not drown out the silent majority.
The observed interplay between the prominence of moderator
involvement and the development of exchange at the triadic-level
features suggest that the health of the forum may still be affected.
In courses where super-posting activity is more distinct than that
of moderate posters, features of triadic-level exchange are not
more pronounced than direct reciprocity features. Such dynamics
may indicate that super-posters are limiting engagement
opportunities for learners who are moderately active. In other
words, if there is a moderator who always replies, and no teacher
to stimulate the dialogue, then the community has less reason to
engage with one another. It seems that super-posters, who are
staff and teaching assistants at early stages, help the network to
develop, but they are not always skillful at decreasing their
activity at an optimal time.
This study’s analysis yielded many promising results. However,
further work is still required. Much of the interpretation of this
work and the associated implications are at this point speculative.
Future work should seek to account for the temporal aspect of
how the network unfolds and interacts, as well as include forum
core interactions with the rest of the forum posters in MOOCs. An
extrapolation of the findings relates to the time based dynamics of
network formation. We suggest that facilitation shifts the forum
core network structure from a distributed to cliquish to egalitarian
structure. Here an egalitarian structure is the product of combined
direct and indirect reciprocity. Forum core is in its turn situated
within the network exchanges of intermittent posters, and the
relationship of forum core to those posters is based on the features
of indirect reciprocity. That is, if a forum core member gets
reciprocated by a staff member at an early stage, she would be
more likely to offer answer to an intermittent poster later on. In
other words, measuring the extent of network amplification over
time offers partial explanation of the nature of social exchanges,
while indirect reciprocation between the forum core and the other
poster explains the remaining dynamics. These speculations form
the basis of our future research.
Several direct implications stem from this study’s analyses. First,
empirical validation of forum core networks as identity-based
communities is valuable as MOOC forum research can build on
the established research agenda both in social science and in
studies of human-computer interaction. For instance, Ren et al.
 developed a set of design principles for the facilitation of
identity-based communities, and these could be applied in
MOOCs. Yet, as the dataset of courses was limited, more diverse
sets of courses need to be analyzed to see if the insights
demonstrate consistent patterns across disciplines, student cohorts
and class sizes.
While this study did not include text mining and analysis of
learner posts, identity-based online communities have been
researched using text analysis along with SNA. Particularly 
suggested that measures of entropy capture the lack of topical
diversity in identity-based communities, what they refer to as
‘topical groups’, and the divergence of discussion themes in what
they refer to as ‘social groups’. Furthermore, granular text mining
for the concepts related to power and authoritativeness [e.g. 23]
may offer insights as to where the shifts in network structure are
reflected in the text features of the various learner posts. On a
final note, the indicators proposed from this work should be
validated in formal online education environments, to assess for
comparability between formal and open online settings.
Our thanks to TU DELFT and Open Online Learning at DelftX.
 Bell, F. (2011). Connectivism: Its place in theory-informed
research and innovation in technology-enabled learning. The
International Review of Research in Open and Distance
 Brandes, U., Robins, G., McCranie, A. and Wasserman, S.
(2013). What is network science? Network Science, 1(1), 1–
 Brinton, C.G., Chiang, M., Jain, S., Lam, H., Liu, Z. and
Wong, F.M.F. (2013). Learning about social learning in
MOOCs: From statistical analysis to generative model. IEEE
Transactions on Learning Technologies, 7(4).
 Brown, R., Lynch, C.F., Eagle, M., Albert, J., Barnes, T.,
Baker, R., Bergner, Y. and McNamara, D. (2015). Good
communities and bad communities: Does membership affect
performance? Proceedings of the 8th International
Conference on Educational Data Mining, 612–614.
 Brown, R., Lynch, C., Wang, Y., Eagle, M., Albert, J.,
Barnes, T., ... McNamara, D. (2015). Communities of
performance & communities of preference. In CEUR
Workshop Proceedings. (Vol. 1446). CEUR-WS.
 Conole, G. (2013). Designing for learning in an open world.
New York: Springer.
 De Laat, M., Lally, V., Lipponen, L. and Simons, R.-J.
(2007). Online teaching in networked learning communities:
A multi-method approach to studying the role of the teacher.
Instructional Science, 35(3), 257–286.
 Dowell, N. M., Skrypnyk, O., Joksimović, S., Graesser, A.
C., Dawson, S., Gašević, S., … Kovanović, V. (2015).
Modeling learners’ social centrality and performance through
language and discourse. In C. Romero & M. Pechenizkiy
(Eds.), Proceedings of the 8th International Conference on
Educational Data Mining (pp. 250–257). International
Educational Data Mining Society.
 Ferguson, R., Shum, S.B. (2012). Social learning analytics:
five approaches. Proceedings of the 2nd international
conference on learning analytics and knowledge, 23–33.
 Gillani, N., & Eynon, R. (2014). Communication patterns in
massively open online courses. The Internet and Higher
Education, 23, 18-26.
 Gillani, N., Yasseri, T., Eynon, R., & Hjorth, I. (2014).
Structural limitations of learning in a crowd: communication
vulnerability and information diffusion in MOOCs. arXiv
preprint arXiv:1411.3662 [physics.soc-ph].
 Goldberg, L. R., Bell, E., King, C., O’Mara, C., McInerney,
F., Robinson, A., & Vickers, J. (2015). Relationship between
participants’ level of education and engagement in their
completion of the Understanding Dementia Massive Open
Online Course. BMC medical education, 15(1), 1.
 Goodyear, P., Salmon, G., Spector, J. M., Steeples, C., &
Tickner, S. (2001). Competences for online teaching: A
special report. Educational Technology Research and
Development, 49(1), 65-72. http://dx.doi.org/
 Grabowicz, P. A., Aiello, L. M., Eguíluz, V. M., & Jaimes,
A. (2013). Distinguishing topical and social groups based on
common identity and bond theory. In Proceedings of the
sixth ACM international conference on Web search and data
mining, 627-636. http://dx.doi.org/
 Gruzd, A., & Haythornthwaite, C. (2013). Enabling
community through social media. Journal of medical
Internet research, 15(10), e248.
 Gruzd, A. & Haythornthwaite, C. (2008). Automated
discovery and analysis of social networks from threaded
discussions. Paper presented at the International Network of
Social Network Analysis, St. Pete Beach, FL, USA.
 Handcock, M., Hunter, D., Butts, C., Goodreau, S.,
Krivistky, P. and Morris, M. 2015. ergm: Fit, Simulate and
Diagnose Exponential-Family Models for Networks. The
Statnet Project. http://www.statnet.org.
 Handcock, M. S., Hunter, D. R., Butts, C. T., Goodreau, S.
M., & Morris, M. (2003). statnet: Software tools for the
Statistical Modeling of Network Data. Seattle, WA. Version,
 Harris, J. K. (2013). An introduction to exponential random
graph modeling (Vol. 173). Sage Publications.
 Haythornthwaite, C. and De Laat, M. (2012). Social network
informed design for learning with educational technology. In
A.D. Olofsson, J.O. Lindberg, K. Klinger, C. Shearer (Eds.),
Informed Design of Educational Technologies in Higher
Education: Enhanced Learning and Teaching (pp.352–374).
 Hecking, T., Chounta, I.-A. and Hoppe, H.U. (2016).
Investigating social and semantic user roles in MOOC
discussion forums. Proceedings of the Sixth International
Conference on Learning Analytics & Knowledge, 198–207.
 Hecking, T., Hoppe, H.U. and Harrer, A. (2015). Uncovering
the Structure of Knowledge Exchange in a MOOC
Discussion Forum. Proceedings of the 2015 IEEE/ACM
International Conference on Advances in Social Networks
Analysis and Mining 2015, 1614–1615. http://dx.doi.org/
 Howley, I., Mayfield, E. and Rosé, C.P. (2011). Missing
something? Authority in collaborative learning. In A.
Dimitracopolou, C. O’Malley, D- Suthers, P-Reimann (Eds.),
Proceedings of the 9th Computer Supported Collaborative
Learning Conference (pp.336373).
 Huang, J., Dasgupta, A., Ghosh, A., Manning, J. and
Sanders, M. (2014). Superposter behavior in MOOC forums.
Proceedings of the first ACM conference on
Learning @ Scale, 117–126.
 Jiang, S., Warschauer, M., Williams, A.E., O’Dowd, D. and
Schenke, K. (2014). Predicting MOOC performance with
week 1 behavior. Proceedings of the 7th International
Conference on Educational Data Mining, 273–275.
 Joksimović, S., Dowell, N., Skrypnyk, O., Kovanović, V.,
Gašević, D., Dawson, S. and Graesser, A.C. (2015). How do
you connect? Analysis of social capital accumulation in
connectivist MOOCs. Proceedings of the Fifth International
Conference on Learning Analytics and Knowledge, 64–68.
 Joksimović, S., Manataki, A., Gašević, D., Dawson, S.,
Kovanović, V. and de Kereki, I.F. (2016). Translating
network position into performance: importance of centrality
in different network configurations. Proceedings of the Sixth
International Conference on Learning Analytics &
 Jones, C. (2004). Networks and learning: communities,
practices and the metaphor of networks. Research in
Learning Technology, 12(1).
 J Justice, T., & Jamieson, D. W. (2012). The facilitator's
fieldbook. AMACOM Div American Mgmt Assn.
 Kellogg, S., Booth, S. and Oliver, K. (2014). A social
network perspective on peer supported learning in MOOCs
for educators. The International Review of Research in Open
and Distributed Learning, 15(5).
 Kizilcec, R.F., Piech, C. and Schneider, E. (2013).
Deconstructing disengagement: Analyzing Learner
Subpopulations in Massive Open Online Courses.
Proceedings of the Third International Conference on
Learning Analytics and Knowledge - LAK ’13.
 Krackhardt, D. (1998). Simmelian ties: Super strong and
sticky. In R. Kramer, M. Neale (Eds.), Power and influence
in organizations (pp. 2138). SAGE Publications: USA.
 Kreijns, K., Kirschner, P. A., & Jochems, W. (2003).
Identifying the pitfalls for social interaction in computer-
supported collaborative learning environments: a review of
the research. Computers in human behavior, 19(3), 335-353.
 Kreijns, K., Kirschner, P. A., Jochems, W., & Van Buuren,
H. (2011). Measuring perceived social presence in
distributed learning groups. Education and Information
Technologies, 16(4), 365-381.
 Laumann, E., Marsden, P. and Prensky, D. (1983). The
boundary specification problem in network analysis. In
L.Freeman, D. White (Eds.), Research methods in social
network analysis (pp.6186). Transaction Publishers: London,
 Lusher, D., Koskinen, J. and Robins, G. (2012). Exponential
random graph models for social networks: Theory, methods,
and applications. Cambridge University Press: USA.
 Morris, M., Handcock, M. S., & Hunter, D. R. (2008).
Specification of exponential-family random graph models:
terms and computational aspects. Journal of statistical
software, 24(4), 1548 7660.
 Poquet, O. (2016, September). Needle in a haystack:
Analysis of social processes in MOOCs. Presentation at the
EASS HDR Forum, School of Education, University of
 Poquet, O., Kovanovic, V., de Vries, P., Hennis, T.,
Joksimovic, S., Gasevic, D. and Dawson, S. (n.d.) Social
presence in Massive Open Online Courses. Manuscript
submitted for review.
 Poquet, O. and Dawson, S. (2016). Untangling MOOC
learner networks. Proceedings of the Sixth International
Conference on Learning Analytics & Knowledge, 208–212.
 Postmes, T., Spears, R., & Lea, M. (1999). Social identity,
normative content, and "deindividuation" in computer-
mediated groups. In R. Spears, B. Doosjie & N. Ellemans
(Eds.), Social identity: Context, commitment, content (pp.
164183). Oxford: Blackwell.
 Ren, Y., Harper, F. M., Drenner, S., Terveen, L. G., Kiesler,
S. B., Riedl, J., & Kraut, R. E. (2012). Building Member
Attachment in Online Communities: Applying Theories of
Group Identity and Interpersonal Bonds. Mis Quarterly,
 Ren, Y., Kraut, R., & Kiesler, S. (2007). Applying common
identity and bond theory to design of online communities.
Organization studies, 28(3), 377-408.
 Robins, G., Pattison, P., & Elliott, P. (2001). Network
models for social selection processes. Psychometrika,
 Rourke, L., Anderson, T., Garrison, D. R., & Archer, W.
(2007). Assessing social presence in asynchronous text-
based computer conferencing. International Journal of E-
Learning & Distance Education, 14(2), 50-71.
 Rovai, A. P. (2002). Building sense of community at a
distance. The International Review of Research in Open and
Distributed Learning, 3(1).
 Sassenberg, K. (2002). Common bond and common identity
groups on the Internet: Attachment and normative behavior
in on-topic and off-topic chats. Group Dynamics: Theory,
Research, and Practice, 6(1), 27.
 Siemens, G. (2004, December). Connectivism: A learning
theory for the digital age. [Web log post]. Retrieved from
 Siemens, G. (2008). Learning and Knowing in Networks:
Changing Roles for Educators and Designers. Presented at
 Skrypnyk, O., Joksimovic, S., Kovanovic, V., Gasevic, D.
and Dawson, S. 2014. Roles of course facilitators, learners,
and technology in the flow of information of a cMOOC. The
International Review of Research in Open and Distributed
 Utz, S. (2003). Social identification and interpersonal
attraction in MUDs. Swiss Journal of
Psychology/Schweizerische Zeitschrift für
Psychologie/Revue Suisse de Psychologie, 62(2), 91.
 Vu, D., Pattison, P. and Robins, G. (2015). Relational event
models for social learning in MOOCs. Social Networks, 43,
 Wang, X., Wen, M. and Rosé, C.P. (2016). Towards
triggering higher-order thinking behaviors in MOOCs.
Proceedings of the Sixth International Conference on
Learning Analytics & Knowledge, 398–407.
 Wasko, M., Teigland, R. and Faraj, S. (2009). The provision
of online public goods: Examining social structure in an
electronic network of practice. Decision Support Systems,
47(3), 254–265. http://dx.doi.org/10.1016/j.dss.2009.02.012
 Yang, D., Sinha, T., Adamson, D. and Rose, C. (2013).
“Turn on, Tune in, Drop out”: Anticipating student dropouts
in Massive Open Online Courses. Presented at the NIPS
Workshop on Data Driven Education, Lake Tahoe, Nevada,
Table A1. Results of k-means clustering of in-reach and out-reach measures derived from the entire MOOC forum network in each
of the courses. Attributes for clusters were used to control for activity in forum core ERGM. In most courses clusters were
interpreted as posters with high, moderate and low posting activity (except for C)
Cluster 3 - High Posting Activity
Cluster 2 - Moderate Posting Activity
Cluster 1 - Low Posting Activity