=Paper=
{{Paper
|id=Vol-1183/gedm_paper03
|storemode=property
|title=What is the Source of Social Capital? The Association between Social Network Position and Social Presence in Communities of Inquiry
|pdfUrl=https://ceur-ws.org/Vol-1183/gedm_paper03.pdf
|volume=Vol-1183
|dblpUrl=https://dblp.org/rec/conf/edm/KovanovicJGH14
}}
==What is the Source of Social Capital? The Association between Social Network Position and Social Presence in Communities of Inquiry==
https://ceur-ws.org/Vol-1183/gedm_paper03.pdf
What is the Source of Social Capital?
The Association Between Social Network Position and Social Presence in
Communities of Inquiry
∗
Vitomir Kovanovic Srecko Joksimovic Dragan Gasevic
School of Interactive Arts and School of Interactive Arts and School of Computing Science
Technology Technology Athabasca University
Simon Fraser University Simon Fraser University 1 University Drive
250 - 13450 102nd Avenue 250 - 13450 102nd Avenue Athabasca, AB, T9S 3A3
Surrey, BC, V3T0A3 Canada Surrey, BC, V3T0A3 Canada Canada
vitomir_kovanovic@sfu.ca sjoksimo@sfu.ca dgasevic@acm.org
Marek Hatala
School of Interactive Arts and
Technology
Simon Fraser University
250 - 13450 102nd Avenue
Surrey, BC, V3T0A3 Canada
mhatala@sfu.ca
ABSTRACT ing insights into the very complex nature of the learning phenom-
It is widely accepted that the social capital of students – developed ena. Among the different ways of researching students’ social in-
through their participation in learning communities – has a signif- teractions Quantitative Content Analysis (QCA) [38, 19] and Social
icant impact on many aspects of the students’ learning outcomes, Network Analysis (SNA) [52, 46] represent two commonly used
such as academic performance, persistence, retention, program sat- methods.
isfaction and sense of community. However, the underlying social
processes that contribute to the development of social capital are A widely accepted model of distance education which makes a use
not well understood. By using the well-known Community of In- of QCA is the Community of Inquiry (CoI) model [28]. According
quiry (CoI) model of distance and online education, we looked into to Garrison and Arbaugh [30], it is one of the leading models of
the nature of the underlying social processes, and how they relate distance education that describes the key constructs of the overall
to the development of the students’ social capital. The results of educational experience. The CoI model provides the in-depth as-
our study indicate that the affective, cohesive and interactive facets sessment of teaching, cognitive and social dimensions of learning
of social presence significantly predict the network centrality mea- phenomena, and how those three dimensions affect: i) the overall
sures commonly used for measurement of social capital. success of the learning process, and ii) the attainment of learning
objectives [28]. Empirical research showed that the social dimen-
General Terms sion of learning plays an important role in the learning communities
Social Network Analysis, Community of Inquiry, Social Presence by mediating the relationship between the teaching and cognitive
dimensions [31]. Still, the CoI model does not explicitly address
the question of student social networks, their structure, or the ef-
1. INTRODUCTION fects they have on the overall educational experience and learning
Asynchronous online discussions have been frequently used both in outcomes. Given the amount of evidence from the studies of stu-
blended and fully online learning [41]. However, with the broader dent social networks [46], this warrants further investigation.
adoption of social-constructivist pedagogies and the shift towards
the collaborative learning [2], they are viewed as one of the impor- One of the central aspects in the study of social networks is the
tant study tools for the computer-supported collaborative learning idea of the social capital [13, 12]. Generally speaking, social capi-
(CSCL) within the online learning environments. Their use has tal can be defined as a value resulting from occupying a particularly
produced an enormous amount of data about the interactions be- advantageous position within a social network [12]. Over the years,
tween students and instructors [21]. The distance education and the study of social capital has become increasingly popular in the
CSCL research communities have tried to use these data for gain- field of education [14]. The large number of studies in the distance
∗Corresponding Author education field indicated an important connection between the stu-
dents’ social capital and many important aspects of education and
learning including academic performance [33, 15, 7, 49, 43], re-
tention [23], persistence [50], program satisfaction [7], and sense
of community [17]. Still, research of the student social networks
have involved mostly isolated studies that were focused on the un-
derstanding of the relationship between a particular set of con-
structs selected by the researchers and the students’ network po-
sition. Likewise, the underlying mechanisms responsible for the
observed social structure are typically not addressed, which is un-
derstandable given the lack of educational theories that explicitly
take into the consideration student social networks. 2.1.2 Social network analysis in education
While social network analysis has been widely adopted in social
In this paper, we present the results of the study which explored and behavioral sciences, its adoption in the field of education was
the links between the CoI model and the social network analysis initially very limited [14]. According to Carolan [14], the main
of student networks. With the current advancement within the CoI reasons for this are “overemphasis on individual explanations of
research and most recent validations of the model [31], the model educational opportunities and outcomes, a quest for scientific le-
is mature enough and empirically sound to provide this missing gitimacy, and a preference for experimental designs that estimate
theoretical foundation for understanding the structure of students’ the causal effects of ‘educational interventions’ ” [14, 32]. Never-
social networks. Likewise, the understanding of the structure of theless, over the years, the number of studies that indicated the im-
social networks can provide a more comprehensive overview of the portance of social connections on the overall academic experience
social dimension of learning that it is already accounted for in the has grown considerably. A good example is the study of students’
research of the CoI model. overall academic experience from early 1990s by Astin [5] in which
he concluded that: i) the environment made by the instructors and
Given the exploratory nature of this study, we focused on the re- students is crucial, and ii) the single most important environmental
lationship between social capital and social processes which are influence is peer group.
indicative of the student social presence development. The main
question we aim to answer, in this paper is which social processes, In the context of distance education, there have been many studies
and to what extent, are indicative of the development of the social recently that looked at the connection between several important
capital in a communities of inquiry? Given the detailed charac- learning constructs and social capital of students. Likewise, in the
terization of social aspects of learning in the CoI model through fields of educational data mining (EDM) [6] and learning analyt-
the construct of social presence, we explored how this construct ics [40], the interest in SNA has been growing. The recent review
relates to the students’ social capital, as characterized by their po- of the EDM field by Romero and Ventura [44] noted a growing in-
sition in social networks formed around communities of inquiry. terest in SNA; likewise, in the learning analytics community, SNA
As the community of inquiry provides characterization of different was recognized as one of the most important techniques of social
sociological processes that constitute social presence, we looked learning analytics [11, 25].
how each of them contributed to the development of social capital
withing students’ social network. As expected, academic performance was the focus of a large ma-
jority of the studies [33, 50, 15, 7, 49, 43] that have found positive
2. THEORETICAL BACKGROUND effects of student positions in social networks on academic perfor-
mance. Still, academic performance was not the only construct that
2.1 Social network analysis was examined. The study of retention by Eckles and Stradley [23]
2.1.1 Social capital found that for each friend that leaves an academic degree program
The study of social networks has attracted much attention in social makes a student five times more likely to leave as well, while ev-
and behavioral sciences [17, 14]. The focus in social network anal- ery friend who stays makes a student 2.25 times more likely to
ysis is on the study of relationships, also known as ties, between also stay in college. The study of student persistence and integra-
a set of actors, or participants [14]. Through the relationships, tion by Thomas [50] found that students with a broader set of ac-
members of a network engage in sharing, exchange or delivery of quaintances are more likely to persist in the academic program of
various resources including information [36]. Social network anal- a higher education institution, and that students with a higher pro-
ysis draws much of its ideas from the mathematical graph theory portion of ties outside their peer group also perform better academi-
and the sociometric studies of the human relationships [52]. cally. This is aligned with the findings of Dawson [17] who showed
that students’ sense of community membership was positively re-
An important concept in the study of social networks is the idea of lated to their closeness and degree centrality measures. Similarly,
relation strength [34], which is used to make a distinction between in the study of a team-based MBA program by Baldwin et al. [7], it
strong social ties, which require a substantial commitment (e.g., was found that the high embeddedness in the friendship network in-
family, close friends), and weak social ties which do not obligate a creased students’ perception of learning and enjoyment in the pro-
strong commitment (e.g., acquaintances). Likewise, the idea of net- gram; as well, the centrality in the communication networks was
work brokerage builds on the fact that in a large network, the den- found to be positively linked with the student grades.
sity of relationships is not uniform, which indicates the existence of
smaller sub-communities within a large social network [12, 13]. In One important thing to notice is that the majority of the studies
his seminal paper, Granovetter [34] stressed the tremendous impor- did not draw their theoretical foundations of network formation
tance of weak social ties, as they provide access to novel informa- from the established educational theories. As pointed out by Riz-
tion from different parts of a social network and provide pathways zuto et al. [43], there is a lack of “theory of academic performance
of information exchange between sub-communities. An individ- that combines individual characteristics as well as social and in-
ual who possesses a large number of weak ties in many different frastructural factors” (p180). The main exception is the use of
sub-communities is able to take advantage by combining diverse retention theories by Tinto [51] and Bean [8] in the study of stu-
information coming from different sub-communities, and to even dent persistence and retention. The other notable theories that are
control to a certain degree the spread of information from one sub- adopted, such as Feld’s theory of focused choice [24], or Lin’s the-
community to another [12]. This ability to create a value from oc- ory of social resources [39] are general sociological theories that
cupying a particular position in a social network is known as so- do not take into the account the specific of learning processes and
cial capital [13]. To study and assess values of different network educational contexts.
positions, the principles of graph theory are the most commonly
used [52]. The notion of centrality is particularly important. This
notion captures the relative importance of individuals in social net-
2.2 The community of inquiry (CoI) model
works [52]. Given the complexity of measuring actors’ relative 2.2.1 Overview
importance, a large number of centrality measures were proposed The Community of Inquiry (CoI) model is a general model of dis-
over the years out of which degree, closeness and betweenness cen- tance education which explains the constructs that contribute to the
tralities are the most frequently used [26]. overall learning experience. It is rooted in the social constructivist
philosophy, most notably in the work of John Dewey [20], and is 2) Interactivity and open communication: In order to promote
particularly well suited for understanding different aspects of learn- the development of higher-order critical thinking skills, the no-
ing within the learning communities. The main goal of the CoI tion that the other side is listening and attending is crucial [45].
model was to define the constructs that characterize a worthwhile Thus, activities such as praising of the student work, actions, or
educational experience, and a methodology for their assessment. comments contribute to the teacher immediacy, which in turn
The CoI model consists of the three interdependent constructs, also leads to affective, behavioral and cognitive learning [45]. Sim-
known as presences, that together provide a comprehensive cover- ilarly, open communication is defined as “reciprocal and re-
age of the distance learning phenomena: spectful exchanges of messages” [28, p100] and together with
interactivity provide a basis on which productive social learning
1) Cognitive Presence explains different phases of students’ knowl- can be established.
edge construction process through social interactions within a 3) Cohesiveness: The activities that “build and sustain a sense of
learning community [28]. group commitment” [28, p101] define cohesiveness. The goal
2) Teaching Presence describes the instructor’s role in course de- is to create a group where the members possess strong bonds
livery and during course design and preparation [3]. to both i) each other and ii) the group as a whole. This in turn
3) Social Presence explains the social relationships and the social stimulates productive learning and the development of critical
climate within a learning community that have a significant ef- thinking skills.
fect on the success and quality of social learning [45].
Given that there are three different dimensions of social presence,
The CoI model is well-researched and widely accepted within the the coding scheme for social presence (see Table 1) defines a list
distance learning research community as shown by a recent two- of indicators for each dimension. By looking at the content and the
part special issue of The Internet and Higher Education journal [1]. timing of each message, it is possible to see how the social climate
The model defines its own coding schemes that are used to assess unfolded during the course delivery. This provides a way of un-
the levels of the three presences through the QCA in transcripts of derstanding and evaluating the different pedagogical interventions
asynchronous online discussions. More recently, instead of rely- with respect to the development of a productive social climate in a
ing on the QCA, a CoI survey instrument [4] was developed as an learning community which enables for the meaningful social inter-
alternative way of assessing the levels of the three presences. actions [53].
2.3 Research Question: Characterization of
2.2.2 Social presence
Social presence is defined as the “ability of participants in a com-
social capital through social presence
As indicated in the previous sections, there is a strong evidence that
munity of inquiry to project themselves socially and emotionally,
social capital plays an important role in the shaping of the overall
as “real” people (i.e., their full personality), through the medium
learning experience. The main research question that we investi-
of communication being used” [28, p3]. Critical thinking, social
gate in this paper:
construction of knowledge and the development of the cognitive
presence are more easily developed in the cases where the appro- What is the relationship between the students’ social
priate levels of social presence have been established [28]. capital, as captured by social network centrality mea-
sures, and students’ social presence, as defined by the
Given the form of delivery in distance education, face-to-face com- three categories in the Community of Inquiry model?
munication that is typical for more traditional forms of education
delivery is not possible. Hence, establishing and sustaining social The higher the social capital of a learner is, the more capable the
presence is more challenging. Distance education was often criti- learner is in terms of learning opportunities, information exchange,
cized as being inferior to more traditional forms of education, par- or integration within the academic environment. Still, the origins
ticularly because of the inability to create social presence between of social capital are not fully understood. Why certain students
the members of a learning community [2]. However, according occupy advantageous positions in social networks? What are the
to Garrison et al. [28], the form of communication is not the solely social processes that enable them to take advantage of their social
factor determining the development of social presence. A key as- relationships? As for now, not a single theory of learning addresses
pect of establishing social presence in face-to-face settings are vi- the question of social capital directly, even though the impact of
sual cues, while participants in online communities use different social context on learning is widely acknowledged.
techniques – such as emoticons – to convey the affective dimension
of communication that lacks in typical text-based communications. As indicated by the previous study by de Laat et al. [18], content
analysis techniques can be used in combination with SNA to pro-
As described by Rourke et al. [45], the origins of social presence vide a more comprehensive view of the social learning processes.
can be found in the work of Mehrabian [42] and his notion of im- In this paper, we propose the use of the Community of Inquiry
mediacy which is defined as “the extent to which communication model, given its holistic view of educational experience and exten-
behaviors enhance closeness to and nonverbal interaction with an- sive empirical evaluation by the research community [29], with the
other” [42, p203]. This, and the set of follow-up studies by com- aim to characterize the origins of social capital in communities of
munication theorists, defined the theoretical background on which inquiry. The CoI model description of important behavioral indices
the construct of social presence was based [45]. The social pres- that contribute to the development of the positive social climate
ence in the CoI model is defined as consisting of three different could be used to interpret the observed differences among students
dimension of communication: positions in a social network.
1) Affectivity and expression of emotions: Since emotions are Likewise, the synergistic effect of using those two perspectives on
strongly associated with motivation and persistence, they are student interactions provide a value for the CoI model by emphasiz-
indirectly connected to critical thinking and communities of in- ing the effects of the theorized social processes. For example, are
quiry. More formally, emotional expression has been indicated interactivity and open communication important for the develop-
by the “ability and confidence to express feelings related to the ment of social capital? Are the students who show group cohesion
educational experience” [28, p99]. the ones who take brokerage positions? Recently, there have been
Table 1: Social Presence Categories and Indicators as defined by Rourke et al. [45]
Category Code Name Definition
Affective A1 Expression of emotions Conventional expressions of emotion, or unconventional expression of emotion,
includes repetitions punctuation, conspicuous capitalization, emoticons.
A2 Use of humor Teasing, cajoling, irony, understatements, sarcasm.
A3 Self-disclosure Presenting details of life outside of class, or express vulnerability.
Interactive or Open I1 Continuing a thread Using reply feature of software rather than starting a new thread.
Communication I2 Quoting from others’ messages Using software features to quote others entire messages or cutting and pasting
selections of others’ messages.
I3 Referring explicitly to others’ messages Direct references to contents of others’ posts
I4 Asking questions Students ask questions of other students or the moderator.
I5 Complementing, expressing appreciation Complimenting others or contents of others’ messages.
I6 Expressing agreement Expressing agreement with others or content of others’ messages.
Cohesive C1 Vocatives Addressing or referring to participants by name.
C2 Addresses or refers to the group using Addresses the group as we,us, our, group.
inclusive pronouns
C3 Phatics, salutations Communication that serves a purely social function: greetings, closures.
some attempts [47, 48] that make use of SNA in conjunction with
the CoI model to provide insights into particular aspects of learn- Table 2: Course offering statistics
ing, such as self-regulation [9]. Still, the central question of social Student count Message count Graph density
capital is left unexplored and that is the goal in our study. Winter 2008 15 212 0.52
Fall 2008 22 633 0.69
3. METHODS Summer 2009 10 243 0.84
Fall 2009 7 63 0.58
3.1 Dataset Winter 2010 14 359 0.84
For our study, we used the dataset consisting of six offers (Win- Winter 2011 13 237 0.77
ter 2008, Fall 2008, Summer 2009, Fall 2009, Winter 2010, Win-
Average 13 291 0.71
ter 2011) of the masters level software-engineering course offered Total 81 1747
through the fully online instructional condition at a Canadian open
public university. The course is 13 weeks long, research-intensive,
and focuses on understanding of current research trends and chal- Table 3: Descriptive statistics of social network metrics
lenges in the area of software engineering. Students were requested: M ean SD M in M ax
i) to participate in online discussions for which they received 15%
Betweenness 9.04 14.51 0.00 74.20
of their final grade (see details in [32]), and ii) to work on a four In-degree 19.84 8.62 4.00 42.00
tutor marked assignments. Overall, 81 student created the total of Out-degree 19.86 9.37 3.00 44.00
1747 discussion messages which were then used as the main data In-closeness 0.09 0.04 0.04 0.17
source for this study. The total number of students and messages Out-closeness 0.08 0.04 0.03 0.18
for all six course offerings are shown in Table 2.
3.2 Social network measures to calculate, as it takes into account only the direct relationships
between the actors [52].
In order to measure students’ social capital we extracted student so-
3) Closeness centrality represents the distance of an individual
cial network graphs from the interactions on the discussion boards.
participant in the network from all the other network partici-
We extracted directed social graphs, so that whenever a student X1
pants [26]. It is defined as the inverse of the sum of the distances
responded to a message from another student X2, we created a di-
to all other participants [14], and hence takes into account both
rect relationship between the two of them (X1 ⇒ X2). Since
direct and indirect relationships [52]. Much like degree central-
two students can exchange more than one message, we extracted a
ity, given that the student graphs are directed, we calculated the
weighted graph where the weights corresponded to the number of
in-closeness and the out-closeness centrality measures. For a
exchanges between a given pair of students. We created a separate
given actor A, in-closeness centrality measures how many indi-
social graph for each of the course offerings independently and the
rect steps are needed for all other actors to reach the actor A,
graph densities for each offering are shown in Table 2.
while out-closeness measures how many indirect steps the actor
A requires in order to reach all the other actors in the network.
From the constructed social network graphs, we extracted the three
network centrality measures which are most frequently used for the
Table 3 shows the descriptive statistics for all five extracted central-
study of the educational social networks [14]:
ity measures. We can see that on average the students wrote around
20 messages, and also received on average around 20 responses.
1) Betweenness centrality captures brokerage opportunities of ac-
This level of activity was expected, as by the course design the stu-
tors in a network and is the most directly related to the social
dents were expected to spend a significant amount of time on the
capital construct [13, 12]. For a given actor A, it is mathemat-
online discussions. Still, from the descriptive statistics reported in
ically defined as the number of shortest paths between any two
Table 3, we can observe the large differences between the individ-
other actors that “pass through” the actor A [26].
ual students in the case of all five centrality measures.
2) Degree centrality measures the total number of relationships
that each participant has [26]. Given that we constructed the di-
rected social graphs, we considered separately the in-degree and 3.3 Message coding
out-degree centrality measures. They represent the total number In order to assess students’ social presence, all messages were man-
of incoming and outgoing relations for a given individual, re- ually coded by two coders in accordance with the coding scheme
spectively. Degree is the simplest centrality measure, very easy defined by Rourke et al. [45]. As the individual messages can
measures. To evaluate different regression models for a particular
Table 4: Social Presence Indicators centrality measure, we used the popular Akaike Information Crite-
Category Code Indicator Count Percent rion (AIC) [35]. In order to control for the inflation of the Type-I
Agreement
error rate due to multiple statistical significance testing, we used
Affective A1 Expression of emotions 288 (16.5%) 84.4 the Holm-Bonferroni correction [37], also known as the sequential
A2 Use of humor 44 (2.52%) 93.1 rejective Bonferroni correction. It provides a control for Type-I er-
A3 Self-disclosure 322 (18.4%) 84.1
Interactive I1 Continuing a thread 1664 (95.2%) 98.9 rors at a prescribed significance level – in our case α = 0.05 –
I2 Quoting from others 65 (3.72%) 95.4 while providing a substantial increase in the statistical power over
messages the commonly used Bonferroni correction [22]. In the case of test-
I3 Referring explicitly to 91 (5.21%) 92.7 ing the family of N null-hypothesis and significance level α, the
other’s messages Holm-Bonferroni method proceeds as follows:
I4 Asking questions 800 (45.8%) 89.4
I5 Complementing, expressing 1391 (79.6%) 90.7
appreciation 1) Hypothesis with the smallest observed p-value, is tested using
I6 Expressing agreement 243 (13.9%) 96.6 the adjusted significance level α0 = α/N , in the same manner
Cohesive C1 Vocatives 1433 (82%) 91.8 as in the traditional Bonferroni procedure.
C2 Addresses or refers to the 144 (8.24%) 88.8 2) However, the next smallest observed p-value is tested using dif-
group using inclusive ferently adjusted significance level α0 = α/(N − 1).
pronouns 3) The same process repeats up to the hypothesis with the highest
C3 Phatics, salutations 1281 (73.3%) 96.1
observed p-value which is tested using the unadjusted signifi-
cance level α.
Table 5: Social Presence Categories. 4) The important additional rule is that if any of the hypothesis in
Category Count Percent Agreement
the family gets rejected, then all the subsequent hypotheses are
rejected as well regardless of their observed p-values.
Affective 530 (30.3%) 80.8
Interactive 1030 (59%) 86.2
(Excluded I1 and I5)
By using differently adjusted statistical significance levels, Holm-
Cohesive 1326 (75.9%) 93.4 Bonferroni method guarantees that the family-wise error rate is
(Excluded C1) kept at the prescribed level, while providing a significant increase
in the statistical power over the more commonly used simple Bon-
ferroni correction [22]. We used the Holm-Bonferroni correction
be simultaneously classified into more than one category of so- for testing the overall significance of the regression models, and for
cial presence, each message was coded with three binary codes testing the significance of the individual predictor variables. In our
indicating whether the message belongs to a particular social pres- case, with five hypothesis tests, the values of the adjusted statistical
ence category. However, early in the coding process, we observed significance levels were α = [0.01, 0.0125, 0.0167, 0.0250, 0.05].
an extremely high frequency of some of the indicators in the co-
hesive and interactive categories. Because of this, almost all of We also inspected the QQ-Plots for the signs of the severe deviation
the messages could be classified as both interactive and cohesive, from the normality of residuals, and we assessed the multicollinear-
which would limit the discriminatory power of those two cate- ity of the three predictor variables using the variance-inflation fac-
gories. Thus, to resolve this issue, instead of coding on the levels tors (VIFs). The QQ-Plots did not reveal deviations from the nor-
of categories, the coding was done on the levels of the individual mality of the residuals and VIF values were substantially lower than
indicators, so that each message was coded with the twelve binary the typically used thresholds such as 4 or 10 [10]. Thus, we con-
codes (i.e., three indicators of the affective category, six indicators sidered the use of the multiple linear regression appropriate for our
of the interactive category and three indicators of the cohesive cate- study.
gory) each indicating an occurrence of a particular social presence
indicator within a given message. This enabled us to look at the
distribution of the individual indicators and to be more selective in 4. RESULTS
the type of the indicators that we wanted to investigate. Overall, the The results of the regression analyses are shown in Table 6. The
coding agreement was high, with all of the indicators reaching per- models for betweenness, in-degree, out-degree and in-closeness
cent agreement of at least 84%, and all the coding disagreements centralities were significant, while the model for out-closeness was
were resolved through discussion between the coders in a follow- marginally significant.
up meeting, after they first coded the messages independently. The
coding results are shown in Table 4. The results show that some of In the case of betweenness centrality, the multiple regression model
the indicators were recorded in a disproportionately large number explained 32% of the variability in the students scores of between-
of messages. Thus, in order to evaluate different aspects of social ness centrality. The backwards-stepwise regression analysis selec-
presence captured by those three categories, we omitted some of the tion using the (AIC) criterion resulted in a regression model con-
indicators from our analysis: i) Continuing a thread, ii) Comple- sisting of the affective and interactive categories of social presence,
menting, expressing appreciation, and iii) Vocatives. We intention- and both variables were found to be statistically significant predic-
ally kept the “Phatics, salutations indicator” as its removal would tors of betweenness centrality. In terms of their relative importance,
render the cohesive category in only 8.24% of the messages. By us- the interactive category had a slightly larger standardized β coef-
ing the remaining nine indicators, we categories all of the messages ficient than the affective category of social presence, indicating a
in the corpus, and the final results are shown in Table 5. slightly larger effect on the students’ betweenness centrality scores.
3.4 Statistical analysis With respect to degree centrality, the regression models explained
In order to investigate the relationships between the three cate- 86% and 83% of the variability in the measures of in-degree and
gories of social presence, as defined by the CoI model, and so- out-degree centralities, respectively. All three predictors were pos-
cial capital, as operationalized through the five network centrality itively associated with the degree centrality measures, and all three
measures, we conducted backward-stepwise multiple linear regres- reached the statistical significance. In terms of their relative impor-
sion analyses [35] for each of the five extracted network centrality tance, in both models, the interactive category of social presence
Table 6: Regression results for selected centrality measures after stepwise model selection using AIC criterion.
Betweenness In-degree Out-degree In-closeness Out-closeness
β SE p β SE p β SE p β SE p β SE p
Affective 0.27 0.12 0.024 0.18 0.054 0.001 0.23 0.059 <0.001
Interactive 0.38 0.12 0.002 0.65 0.064 <0.001 0.65 0.07 <0.001 0.27 0.11 0.015 0.37 0.15 0.017
Cohesive 0.2 0.061 0.001 0.14 0.066 0.041 -0.23 0.15 0.137
F (3, 77) 19.6 <0.001 159 <0.001 130 <0.001 6.24 0.015 3.03 0.054
Adjusted R2 0.32 0.86 0.83 0.061 0.048
had the largest standardized β coefficient, while the affective and ideas and opinions, that would in turn lead to more affective expres-
cohesive categories had roughly the same standardized coefficients. sion, and eventually to the development of the sense of community
belonging. Still, this hypothesis warrants further investigation, and
Regarding the two closeness centrality measures, the regression in the future we plan to analyze the evolution of the students’ so-
model for in-closeness was statistically significant, explaining 6.1% cial presence and the corresponding social network structures over
of the variability in the students’ in-closeness centrality scores, time, which would shed new light on this important question.
while the model for out-closeness failed to reach the significance by
a very small margin. The model for in-closeness consisted of only The results of individual network centrality measures revealed that
the interactive category, which was found to be a statistically signif- both in-degree and out-degree centrality measures were significantly
icant predictor of in-closeness centrality. Similarly, the regression predicted by all the three categories of students’ social presence.
model for out-closeness consisted of the interactive and cohesive By looking at the description (Section 2.2.2) and the indicators (Ta-
social presence categories, and explained 4.8% of the variation in ble 1) of the interactive category of social presence, we can see
the students’ out-closeness centrality scores. In the model for out- that interactive social presence is mainly about stimulating open
closeness centrality, the only statistically significant predictor was and direct communication between the students. Thus, the students
the interactive category of social presence, while interestingly, the who exhibit a high level of interactive social presence have higher
cohesive category of social presence was negatively associated with chances of “provoking” a response from the other students. Activ-
the change in the out-closeness centrality values, although statisti- ities such as asking questions, explicitly referring to other students
cally insignificantly. by name, quoting their messages, complementing them or agreeing
with their messages, are all activities associated with an interactive
and open communication, and can be used to elicit a response from
5. DISCUSSION the other students. It would be interesting to further investigate the
One finding immediately stands out of the regression analyses re- relationship between different indicators of social presence and so-
sults: Interactive social presence is the most strongly associated cial capital, as certain indicators – such as I4 “Asking questions” –
with all of the network centrality measures, indicating a signifi- seem to have more impact than the other indicators. Besides the in-
cant relation with the development of the students’ social capital. teractive category, the regression model revealed that the affective
A possible explanation of this lies to some degree in the nature of and cohesive categories of social presence were also significant pre-
students’ social networks. Given that the primary goal of social dictors of in-degree and out-degree centralities. These findings are
networks in online courses is to serve as a communication medium even more interesting, as affective and cohesive exchanges are not
for fostering of collaborative learning [27], it is reasonable to ex- directly stimulating discussions in the same manner as the interac-
pect that interactivity in communication can explain a significant tive category. Further investigation is needed to examine particular
proportion of the differences in network positions, and ultimately time periods over the duration of a course in which those different
the differences in the development of students’ social capital. The dimensions of social presence contribute to the degree centrality
reason why the interactive category is had the strongest associa- measures of students.
tion might be that only after the students have gott familiar with
each other through focused, on-task interactions, and after they With respect to betweenness centrality that is most closely related
have started developing trust within a learning community, the ex- to the notion of social capital [13, 12], the regression model was
pression of emotions and the sense of group belonging begins to statistically significant and explained 32% of the variability in the
emerge. This is aligned with the findings of Garrison [27] who sug- betweenness centrality scores. This corresponds to Cohen’s f 2 =
gested that interactive social presence is dominant at the beginning 0.47 effect size, which is considered to be a large effect size [16].
of a course, but decreases over time, while affective and cohesive Both the interactive and affective categories of social presence were
social presence increase over time [27]. However, as Garrison [27] statistically significant predictors of the betweenness centrality, with
points out, too much of the interpersonal and affective interactions the interactive category having a bit greater standardized β coeffi-
undermine the productivity of the collaborative learning activities. cient. This might be due to the nature of student communication
There is a certain amount of social interactions that is beneficial networks and their focus on collaborative learning, which resulted
for learning [27], and the focus of the instructional interventions in the emphasis on information exchange. Still, these are very in-
should be on: i) stimulating the right amount of the different social triguing findings, given that betweenness centrality is not directly
interactions that support productive and purposeful collaborative related to the number of interactions the student has, but more to the
learning activities, and ii) the development of trust and the sense of overall diversity of the interactions within a group of learners. In a
community among the group of learners [17]. follow-up study, it would be very interesting to investigate whether
there are any particular ways in which the students with the high
One practical implication of these results is that they suggest the betweenness centrality differ from the other students (e.g., asking
effective way for fostering the productive social climate – and that many questions or exhibiting higher self-disclosure).
is focusing on the student interaction and open communication. In
order to guide the development of the social relationships in a learn- Regarding the closeness centrality measures, the regression model
ing community, it seems that the instructional emphasis should be for in-closeness was also statistically significant. The model ex-
on the interventions that require engaging in an open exchange of plained 6.1% of the variability, and the stepwise model selection
using the AIC criteria resulted in a simple regression model with [2] T. Anderson and J. Dron. Three generations of distance ed-
only the interactive category of social presence. In contrast to de- ucation pedagogy. The International Review of Research in
gree centrality, which considers only direct relationships, closeness Open and Distance Learning, 12(3):80–97, 2010.
centrality also considers the indirect relationships. Such indirect re-
[3] T. Anderson, L. Rourke, D. R. Garrison, and W. Archer. As-
lationships could be the reason why only interactive category was
sessing teaching presence in a computer conferencing con-
rendered as important. The affective and cohesive exchanges be-
text. Journal of Asynchronous Learning Networks, 5:1–17,
tween students A and B, although very important, provide very lit-
2001.
tle, or no influence on the indirect relations of student B and the rest
of the students. The similar findings we could see in the model for [4] J. Arbaugh, M. Cleveland-Innes, S. R. Diaz, D. R. Garrison,
out-closeness, which was marginally significant with the p-value of P. Ice, J. C. Richardson, and K. P. Swan. Developing a com-
0.054. However, it could be expected that the significance of this munity of inquiry instrument: Testing a measure of the com-
model would be conformed in a larger replication study. munity of inquiry framework using a multi-institutional sam-
ple. The Internet and Higher Education, 11(3–4):133–136,
The major limitations of this study is the sample size and the use 2008.
of the single course from a single institution. Even though there
were six offerings of the course taught by the two instructors, there [5] A. W. Astin. What Matters in College: Four Critical Years
might still be significant effects of the adopted pedagogical ap- Revisited. Jossey-Bass, 1 edition edition, 1997.
proach, which could have shaped a specific social dynamics, and [6] R. S. Baker and K. Yacef. The state of educational data min-
thus, potentially distort the findings of our study. Likewise, we con- ing in 2009: A review and future visions. Journal of Educa-
sidered all interactions among the students as contributing to their tional Data Mining, 1(1):3–17, 2009.
social capital, it is very likely that the certain interactions (e.g.,
adversarial interactions) might have a negative effect on the stu- [7] T. T. Baldwin, M. D. Bedell, and J. L. Johnson. The social
dent social capital. In the future work, we plan on replicating our fabric of a team-based M.B.A. program: Network effects on
findings on a bigger sample and with more diverse courses from student satisfaction and performance. The Academy of Man-
different subject matter domains. Finally, we plan to investigate agement Journal, 40(6):1369–1397, 1997.
the temporal aspects of the relationship between social capital and [8] J. P. Bean. Conceptual models of student attrition: How the-
the social presence, which might give us a deeper insight into the ory can help the institutional researcher. New Directions for
complexity of the social interactions in learning communities. Institutional Research, 1982(36):17–33, 1982.
[9] R. A. Bjork, J. Dunlosky, and N. Kornell. Self-regulated
6. CONCLUSIONS learning: beliefs, techniques, and illusions. Annual review
The study presented in this paper investigated some of the social of psychology, 64:417–444, 2013.
processes that can contribute to the development of students’ social
capital. We have looked at the relationship between students’ so- [10] B. L. Bowerman and R. T. O’Connell. Linear Statistical Mod-
cial presence, operationalized through the Community of Inquiry els: An Applied Approach. Duxbury Press, 1990.
model, and students’ social capital, operationalized through the
[11] S. Buckingham Shum and R. Ferguson. Social learning ana-
three network centrality measures. The implications of our findings
lytics. Journal of Educational Technology & Society, 15(3):3–
are twofold: First, our results indicate that a significant part of the
26, 2012.
variability in network centrality scores can be explained using the
three dimensions of the social presence, and this in turn indicates [12] R. S. Burt. Structural holes versus network closure as social
the existence of the relationship between the development of social capital. In N. Lin, K. Cook, and R. S. Burt, editors, Social
presence and social capital. All three categories of social presence Capital: Theory and Research. Aldine Transaction, 2001.
were significant predictors of in-degree and out-degree centrality
measures while interactive and affective categories were significant [13] R. S. Burt. The social capital of structural holes. In M. F.
predictors of the betweenness centrality. Also, interactive category Guillen, R. Collins, P. England, and M. Meyer, editors, The
of social presence was significantly predictive of the in-closeness New Economic Sociology: Developments In An Emerging
and out-closeness centrality measures, although the overall regres- Field. Russell Sage Foundation, 2005.
sion model for out-closeness was marginally significant. A possible [14] B. V. Carolan. Social Network Analysis and Education: The-
explanation is that given the task-oriented nature of discussions in ory, Methods and Applications. SAGE Publications, Inc.,
online courses, students’ social presence develops mostly through 2014.
interactions focused on learning, and then over time, with the de-
velopment of trust among a group of learners, the other dimensions [15] H. Cho, G. Gay, B. Davidson, and A. Ingraffea. Social net-
of social presence start to emerge. Second, the study shows the works, communication styles, and learning performance in a
significant relationship between the interactive category of social CSCL community. Computers & Education, 49(2):309–329,
presence and betweenness, in-degree, out-degree, and in-closeness 2007.
network centrality measures. This provides an empirical basis for [16] J. Cohen. The analysis of variance. In Statistical power anal-
fostering the productive social climate in discussions through inter- ysis for the behavioral sciences, pages 273—406. L. Erlbaum
ventions that increase interactivity and open communication among Associates, Hillsdale, N.J., 1988.
the students. By engaging students to participate in discussions
with the clearly defined expectations, students develop social rela- [17] S. Dawson. A study of the relationship between student social
tionships which can in turn have positive impact on the attainment networks and sense of community. Journal of Educational
of the learning objectives and their overall academic experience. Technology & Society, 11(3):224–238, 2008.
[18] M. F. De Laat, V. Lally, L. Lipponen, and R.-J. Simons. In-
References vestigating patterns of interaction in networked learning and
[1] Special issue on the community of inquiry framework: Ten computer-supported collaborative learning: A role for so-
years later. The Internet and Higher Education, 13(1–2), cial network analysis. International Journal of Computer-
2010. Supported Collaborative Learning, 2(1):87–103, 2007.
[19] B. De Wever, T. Schellens, M. Valcke, and H. Van Keer. Con- [36] C. Haythornthwaite. Social network analysis: An approach
tent analysis schemes to analyze transcripts of online asyn- and technique for the study of information exchange. Library
chronous discussion groups: A review. Computers & Educa- & Information Science Research, 18(4):323–342, 1996.
tion, 46(1):6–28, 2006.
[37] S. Holm. A simple sequentially rejective multiple test proce-
[20] J. Dewey. My pedagogical creed. School Journal, 54(3):77– dure. Scandinavian Journal of Statistics, 6(2):65–70, 1979.
80, 1897. [38] K. H. Krippendorff. Content Analysis: An Introduction to Its
Methodology. Sage Publications, 2003.
[21] R. Donnelly and J. Gardner. Content analysis of computer
conferencing transcripts. Interactive Learning Environments, [39] N. Lin. Social resources and instrumental action. In P. V.
19(4):303–315, 2011. Marsden and N. Lin, editors, Social structure and network
analysis, pages 131—145. Sage Publications, 1982.
[22] O. J. Dunn. Multiple comparisons among means. Journal of
the American Statistical Association, 56(293):52–64, 1961. [40] P. Long and G. Siemens. Penetrating the fog: Analytics in
learning and education. EDUCAUSE Review, 46(5):31–40,
[23] J. E. Eckles and E. G. Stradley. A social network analysis 2011.
of student retention using archival data. Social Psychology of
Education, 15(2):165–180, 2011. [41] R. Luppicini. Review of computer mediated communication
research for education. Instructional Science, 35(2):141–185,
[24] S. L. Feld. The focused organization of social ties. American 2007.
Journal of Sociology, 86(5):1015–1035, 1981. [42] A. Mehrabian. Some referents and measures of nonverbal
[25] R. Ferguson and S. B. Shum. Social learning analytics: five behavior. Behavior Research Methods & Instrumentation,
approaches. In Proceedings of the 2nd International Confer- 1(6):203–207, 1968.
ence on Learning Analytics and Knowledge, LAK ’12, page [43] T. Rizzuto, J. LeDoux, and J. Hatala. It’s not just what you
23–33, New York, NY, USA, 2012. ACM. know, it’s who you know: Testing a model of the relative im-
portance of social networks to academic performance. Social
[26] L. C. Freeman. Centrality in social networks conceptual clar- Psychology of Education, 12(2):175–189, 2009.
ification. Social Networks, 1(3):215–239, 1978.
[44] C. Romero and S. Ventura. Educational data mining: A re-
[27] D. R. Garrison. E-Learning in the 21st Century: A Frame- view of the state of the art. Trans. Sys. Man Cyber Part C,
work for Research and Practice. Routledge, New York, 2 40(6):601–618, 2010.
edition edition, 2011.
[45] L. Rourke, T. Anderson, D. R. Garrison, and W. Archer. As-
[28] D. R. Garrison, T. Anderson, and W. Archer. Critical in- sessing social presence in asynchronous text-based computer
quiry in a text-based environment: Computer conferencing conferencing. The Journal of Distance Education, 14(2):50–
in higher education. The Internet and Higher Education, 71, 1999.
2(2–3):87–105, 1999.
[46] J. Scott and P. J. Carrington. The SAGE Handbook of Social
[29] D. R. Garrison, T. Anderson, and W. Archer. The first decade Network Analysis. SAGE Publications, 2011.
of the community of inquiry framework: A retrospective. The [47] P. Shea, S. Hayes, S. U. Smith, J. Vickers, T. Bidjerano,
Internet and Higher Education, 13(1–2):5–9, 2010. M. Gozza-Cohen, S.-B. Jian, A. Pickett, J. Wilde, and C.-
H. Tseng. Online learner self-regulation: Learning presence
[30] D. R. Garrison and J. Arbaugh. Researching the community viewed through quantitative content- and social network anal-
of inquiry framework: Review, issues, and future directions. ysis. The International Review of Research in Open and Dis-
The Internet and Higher Education, 10(3):157–172, 2007. tance Learning, 14(3):427–461, 2013.
[31] R. Garrison, M. Cleveland-Innes, and T. S. Fung. Explor- [48] P. Shea, S. Hayes, J. Vickers, M. Gozza-Cohen, S. Uzuner,
ing causal relationships among teaching, cognitive and social R. Mehta, A. Valchova, and P. Rangan. A re-examination
presence: Student perceptions of the community of inquiry of the community of inquiry framework: Social network
framework. The Internet and Higher Education, 13(1–2):31– and content analysis. The Internet and Higher Education,
36, 2010. 13(1–2):10–21, 2010.
[32] D. Gasevic, A. Olusola, S. Joksimovic, and V. Kovanovic. [49] R. A. Smith and B. L. Peterson. “Psst . . . what do you
Externally-facilitated regulation scaffolding and role assign- think?” the relationship between advice prestige, type of ad-
ment to develop cognitive presence in asynchronous online vice, and academic performance. Communication Education,
discussions. The Internet and Higher Education, (submitted), 56(3):278–291, 2007.
2014.
[50] S. L. Thomas. Ties that bind: A social network approach to
[33] D. Gasevic, A. Zouaq, and R. Janzen. “Choose your class- understanding student integration and persistence. The Jour-
mates, your GPA is at stake!”: The association of cross-class nal of Higher Education, 71(5):591–615, 2000.
social ties and academic performance. American Behavioral [51] V. Tinto. Leaving College: Rethinking the Causes and Cures
Scientist, 2013. of Student Attrition. University of Chicago Press, 1993.
[34] M. Granovetter. The strength of weak ties. American Journal [52] S. Wasserman. Social Network Analysis: Methods and Appli-
of Sociology, 78(6):1360–1380, 1973. cations. Cambridge University Press, 1994.
[35] T. J. Hastie, R. J. Tibshirani, and J. H. Friedman. The elements [53] Y. Woo and T. C. Reeves. Meaningful interaction in web-
of statistical learning: data mining, inference, and prediction. based learning: A social constructivist interpretation. The In-
Springer, New York, NY, 2013. ternet and Higher Education, 10(1):15–25, 2007.