Effects of Hierarchical Levels on Social
Network Structures within Communities of Learning
Martin Rehma,
Wim Gijselaersb, Mien
Segersb
a
University Duisburg-Essen, Germany
b Maastricht
University, the Netherlands
Article received 12
February 2014 / revised 14 May 2014 /
accepted 15 May 2014 / available online 15 July 2014
Abstract
Facilitating an interpersonal knowledge
transfer among employees constitutes a key building block in
setting up organizational training initiatives. With
practitioners and researchers looking for innovative training
methods, online Communities of Learning (CoL) have been
promoted as a promising methodology to foster this kind of
transfer. However, past research has only provided limited
data from actual organizations and largely neglected
characteristics that constitute a major obstacle to such
collaborative processes, namely participants’ hierarchical
levels. The current study addresses these shortcomings by
providing empirical evidence from 25 CoL of an online training
program, provided for 249 staff members of a global
organization. Using social network analysis, we are able to
show significant differences in participants’ network
behaviour and position based on their hierarchical rank. This
translates into higher in- and out-degree network ties, as
well as centrality scores among participants from higher up
the hierarchical ladder. Finally, based on a longitudinal
analysis of all indicated network measures, our results
indicate that the main trend develops predominately during the
first half of the training program. By incorporating these
insights into the implementation of future CoL, it is not only
possible to anticipate participants’ behaviour. Our findings
also allow to draw conclusions about how collaborative
activities within CoL should be designed and facilitated, in
order to provide participants with a valuable learning
experience.
Keywords: Social Learning
Networks; Longitudinal Analysis; Centrality; Hierarchical
Levels
1.
Introduction
Researchers have stipulated that organizations
are transactive knowledge systems, where the vast majority of
knowledge is stored in the heads of individual employees (Cross, Borgatti,
& Parker, 2001). Consequently, it has
been suggested that facilitating an interpersonal knowledge
transfer among employees constitutes a key building block in
setting up organizational training initiatives (Argote & Ingram,
2000). This notion is further supported by researchers
who suggested that knowledge is being created while
collaborating in social networks composed of diverse groups of
people (e.g. Hakkarainen, Palonen,
Paavola, & Lehtinen, 2004; Paavola, Lipponen,
& Hakkarainen, 2004).
In practice, this process of connecting people greatly builds
upon the extensive use of electronic communication tools, such
as asynchronous discussion forums. These types of
communication channels have been proposed by scholars to
effectively enable the establishment and development of new
ways in which training can build upon networked communities (e.g. Venkatraman,
1994). Yet, organizations cannot assume that once a
technology is introduced and the appropriate structure has
been designed the rest will follow. Instead, previous research
has established that for social (learning) networks to achieve
their intended goals, a clear understanding is needed of how
existing organizational structures influence not only the
adoption of electronic communication tools, but also their
implementation (Zack & Mckenney,
1995).
With practitioners and researchers starting to
increasingly look for new approaches to design and implement
organizational training programs (Yamnill & McLean,
2001), online collaborative learning has received a
growing amount of attention in recent years (Brower, 2003). In the context of this
study, we consider (online) collaborative learning as a
setting where “[participants] are
working in groups on a shared task or problem, in which they
are expected to have equal contributions and participation” (de Laat, Lally,
Simons, & Wenger, 2006, p. 103). One promising
methodology that has been developed within this framework is
the concept of online Communities
of Learning (CoL). Being defined as groups of people “engaging in collaborative
learning and reflective practice involved in transformative
learning” (Paloff & Pratt,
2003, p. 17), CoL have been proposed to foster the effective
exchange of knowledge and experience between members of an
organization’s workforce (e.g. Stacey, Smith, & Barty,
2004). Moreover, online communities, like CoL, have been
considered as an almost ready-made laboratory for analysing
collaboration in social (learning) networks over time (Haythornthwaite, 2001).
In order to conduct these types of analysis,
numerous researchers have suggested social network analysis
(SNA) as a valuable tool for describing and understanding
whether and how members of a (learning) network interact with
each other (e.g. Daradoumis,
Martínez-Monés, & Xhafa, 2004; de Laat, Lally,
Lipponen, & Simons, 2007). According to Aviv,
Erlich, Ravid and Geva (2003) a social network can be
defined as “a group of
collaborating (and/or) competing entities that are related
to each other” (p. 4). SNA has been used to analyse
various networks from several academic domains, ranging from
social sciences, communication studies, economics, to computer
networks and different other fields (Aviv et al., 2003). Moreover, Garton and
colleagues (2006) specifically suggest
using SNA methods in the context of online learning networks.
When considering their structure and development, and
following the seminal work of Erdös and Rényi (1960), social networks should
evolve according to the concept of random graph theory. In
essence, the underlying supposition of this theory is that
while some participants of a network might get in touch with
more people than others, on average everyone should have made
the same amount of contacts, similar to a random distribution
of connections. In other words, all participants of a network
should have an equal chance of making connections (Rienties, Tempelaar,
Giesbers, Segers, & Gijselaers, 2012). However, if everyone
did indeed have equal chances of getting connected with
others, why can we then observe so many biased networks in the
real world (Barabási, 2003)?
More specifically, based on numerous studies of
newly emerging online communities, researchers have found that
a small minority of participants (15%) is gravitating around
the centre of their community’s activity, while a considerable
larger group (40%) is barely engaging into communication with
their colleagues (e.g. Cross, Laseter,
Parker, & Velasquez, 2006). In order to explain
these observed patterns, some researchers have referred to the
fact that communication is an inherently social act (Pearce, 1976). New tools and
methodologies can only reach their full potential, if
organizers fully understand how existing social relationships
influence communication patterns and participants’ behaviour
therein (Wellman, 2001). Moreover, de Laat and
Lally (2003) stipulated that the
social and contextual frameworks in which the learning takes
place have a considerable influence on how participants behave
and perform within online learning networks. Furthermore, the
nature of social networks, as well as their development over
time, is significantly affected by the background
characteristics of their individual members (e.g. Barabasi &
Albert, 1999). Yet, past research has largely been concerned
with the static features of online communities (Panzarasa, Opsahl,
& Carley, 2009). While this offers
preliminary insights on the overall processes that take place
within these communities, it lacks a more refined picture of
how social relationships might develop over time (e.g. Aviv et al., 2003; Haythornthwaite, 2001).
Additionally, the vast amount of research has neglected a
particular background characteristic that can have a severe
effect on the underlying learning processes, namely
participants’ hierarchical levels (Carley, 1992; Griffith & Neale,
2001; Romme,
1996).
The present study addresses these shortcomings by
providing empirical evidence from 25 CoL of an online training
program that was provided for 249 staff members of a global
organization. Each CoL consisted of 7 – 13 participants and
was centred on asynchronous discussion forums, where
participants from different parts of the organization’s
hierarchical ladder collaboratively enhanced their knowledge
and skills. In order to analyse whether participants’ network
behaviour was influenced by their hierarchical level, social
network analysis (SNA) was employed. Based on the resulting
findings of our study, organizers of CoL will able to
anticipate (groups of) individuals holding crucial positions
and design actions targeted at participants who tend to be
situated more towards the fringe of the network (Hatala, 2006). Moreover,
incorporating our findings into the design and implementation
strategies of future CoL will allow a more refined setup that
contributes to employees’ learning experience and can foster
the knowledge creation within an entire organization.
2.
Effects of Hierarchical Levels on Social
Network Structures within CoL
One of the key elements of online
(learning) communities is that they allow for an open dialogue
between participants (Amin
& Roberts, 2006). Yet, when
considering the findings and experiences from real-life
communities within organizations, there is increasing evidence
that information flows are constrained by underlying
organizational structures, such as departments, units and
hierarchical levels (e.g.
Cross, Laseter, Parker, & Velasquez, 2004).
One possible explanation for this finding has been put forth by
authors like Drazin (1990),
who stipulated that professionals might not join communities
with the intention of learning. Instead, individuals would
primarily engage into discussion with colleagues, in order to
secure their role, and gain access to and control over
information. Holmqvist (2009)
indicated that all organizational learning processes are subject
to the influence of a dominant individual or group of
individuals. Similarly, van der Krogt (1998)
postulated that “[…]
powerful work actors will attempt to influence both the work
and the learning network” (p. 170). Furthermore, Yates and
Orlikowski (1992)
argued that top management will spent more time proactively
setting the tone, as they are concerned with losing control of
online groups, which could potentially feed through to the real
world. Considering the role of middle management, Bird (1994)
advocated that they would act as a “nexus between the real and
the ideal” (p. 333). In practice this would result in
members of this hierarchical level to “translate” information
from one level to the next, providing clarifications and
elaborating on shared information. Focusing on the lower end of
the hierarchical ladder, Edmondson (2002)
has shown that lower level management is particularly concerned
about how colleagues perceive them and their work. Consequently,
they tend to limit their interaction with colleagues from higher
hierarchical levels. Additionally, members of this group have
been suggested to be more passive in discussions within training
programs (Nembhard
& Edmondson, 2006). Fox (2000)
has described this situation as being “caught in a dilemma”
(p.856). On the one hand, individuals would like to establish a
reputation of being knowledgeable. On the other hand, they also
need to consider the existing rules of conduct. Sutton and
colleagues (2000)
follow this notion and propose that members from lower
hierarchical levels will mainly try to blend in while not
upsetting the status quo. In practice, this then translates into
activities such as flattering, where lower level management
frequently contacts their colleagues from higher hierarchical
levels (Bird,
1994).
Regarding the overall structure of a
(learning) network, it has been established that the position of
individuals within such a network is related to their access to
valued resources (e.g. Ibarra
& Andrews, 1993; Sparrowe,
Liden, Wayne, & Kraimer, 2001).
Casciaro (1998)
noted that occupying high-level positions within an organization
provides individuals with an intrinsic attraction to lower level
management. Studying three research centres of an Italian
university, the author implied that, given their position within
the organization, higher level management has privileged access
to (vital) information and knowledge sources that are relevant
for all employees. Moreover, this power can create a type of
vortex, where lower level management is trying to get connected
and, over time, stay in contact with higher level management (Krackhardt,
1990). Additionally, Borgatti and Cross
(2003)
have argued that lower level management, with only constrained
access to valued resources, will be less likely to be contacted
for information. As a result, they should hold more peripheral
network positions. Johnson-Cramer, Parise and Cross (2007)
have found empirical evidence for this argument. In their study
of a consumer electronic company, they were able to show that
higher level management held more central positions in the
organization’s information sharing network. On the contrary,
lower level management primarily occupied positions at the outer
fringe of the same network.
Based on these considerations, and
taking into the suggestions of previous studies that called for
more longitudinal research (e.g.
Haythornthwaite, 2001), we formulate
three research hypotheses:
Hypothesis 1 (H1):
Over time, participants' propensity to actively contact other
colleagues will be positively influenced by their hierarchical
level.
Hypothesis 2 (H2):
Over time, participants’ ability to attract connections from
other colleagues will be positively related to their
hierarchical level.
Hypothesis 3 (H3): Over
time, the higher a participant’s hierarchical level, the
higher her degree of centrality within CoL.
3.
Organisational setting
The data was collected from an online training
program that aimed at enhancing the capacity and skills of a
global organization’s staff, operating in the sector of
economic development. Overall, the organization has more than
7.000 employees, operates in 126 countries worldwide, and has
its headquarters located in Northern America. The training
program was delivered twice over a time-span of 14 weeks and
covered five pre-defined content modules on the general topic
of Economics.
Operating in a fast changing environment, where
new analyses and solutions are needed to address old problems,
the organization wanted to embrace these developments by
training their management staff accordingly. Participants
engaged into two types of learning activities, namely
self-study and collaborative learning. The self-study element
included (multimedia) learning materials, such as web lectures
and online quizzes. During the collaborative learning
activities, which constituted the backbone of the training
program, participants discussed real-life tasks via
asynchronous discussion forums. The forums were nested in
dedicated CoL that consisted of 10 – 15 randomly assigned
participants. Each
of the five content modules had a separate task, which were
discussed within dedicated forums in chronological order.
Participation in these forums was obligatory and assessed by
academic staff members, who facilitated the CoL. More
specifically, a team of two academic staff members was
assigned to one CoL each. These facilitators graded
participants’ contributions, facilitated the discussions, and
provided technical assistance. In practice, this could take
the form of encouraging discussions and notifying participants
when the communication departed too much from the intended
focus of the discussion. Before engaging with their assigned
CoL, all facilitators were trained on how to work with CoL and
received elaborate guidelines for all collaborative learning
activities. Additionally, regular meetings were scheduled
where facilitators could discuss their experiences and
streamline their behaviour and actions towards participants.
Next to the obligatory, content-driven discussion forums,
participants also had the opportunity to exchange private
information and socialize via a so-called “Café-Talk” forum.
Upon successful completion, participants could attain a
certificate of participation, together with academic credits
that were based on the European Credit Transfer and
Accumulation System (ECTS).
4.
Method
Overall, 337 participants were randomly assigned
to 30 CoL. However, the present study analyses a subset of 25
CoL and 249 participants (73.88%). The underlying reason for
this smaller subset is twofold. On the one hand, we had
incomplete datasets for some participants. On the other hand,
we discovered that some CoL were biased, in the sense that not
all applicable hierarchical levels were represented.
Consequently, we dropped the applicable CoL from the analyses.
The remaining 25 CoL had an average of 9.96 members (SD =
1.72, range = 7 – 13), the average age was 43.92 (SD = 7.33,
range = 27 – 58), 54.61 percent of the participants were
female, and more than 80 nationalities were represented. The
educational backgrounds of participants were categorized into
Master’s (71.37 %), PhD’s (14.51 %), Bachelor’s (7.26 %), to
other degrees (6.85 %). Particular examples of the latter
category included, Health Sciences and International Law.
Following the official job categories of the organization in
question, participants’ could be subdivided into “Low” (n =
82, 32.93 %), “Middle” (n = 93, 37.35 %) and “High”
hierarchical levels (n = 74, 29.71 %).
Following the work of Daradoumis and
colleagues (2004),
and based on the collected log-files and user statistics from
the underlying discussion forums, we subdivided the data
according two different types of network links, namely indirect and direct links. Indirect
links refer to passive connections that took the form of reading
a colleague’s contributions, but not replying to them. This type
of activity was separately recorded in the log-files captured
via Read-Networks. In
case a participant actively reacted to another CoL member’s
contribution and replied, this established a direct link,
created another applicable entry in the log-file, and was
included in Reply-Networks.
Based on this distinction it was then possible to make
inferences about the type of learning actions underlying a
certain network connection.
Participants reported their
own hierarchical level via the training’s official registration
form. The indicated options were subject to the organization’s
official job categories. Based on the target group of the
training program, three main categories were identified, namely
”Low”-, “Middle”- and “High”-level hierarchical levels.
Generally, representatives of the “Low” group were associated
with project level work, contributing to sub-parts of the
overall product. Members of the “Middle” group were leaders of
such projects. Finally, participants from the “High” group were
responsible for departments and often entire regions in which
the organization was operating.
The analyses of this study focus on
data from individual participants. However, these participants
were distributed over different CoL. Depending on the specific
composition of a particular CoL, with respect to participants’
hierarchical levels, this could have led to different dynamics
and results. As a result, the validity of comparing across
different learning networks might have been reduced. Hence, in
order to account for possible differences in group compositions
across CoL, we employed the Shannon Equitability Index (Magurran,
1988). The index ranges from 0 to fa1
and indicates the percentage share of diversity in relation to
the maximal possible diversity within a given CoL. Focusing on
participants’ hierarchical levels as a source of diversity, the
average score for the investigated 25 CoL was .44 (SD = .05,
range = .35 – .55). Based on this value and the low standard
deviation, we concluded that the CoL represented comparable
sample for our analysis.
All network statistics were computed with the
help of UCINET 6.357 (Borgatti, Everett,
& Freeman, 2002). The visualization of
an exemplary CoL network, in terms of sociograms, was
conducted with the help of the incorporated visualization
software NetDraw (Borgatti, 2002). The underlying data
was based on the log-files and user statistics from the
discussion forums within the different CoL. In order to
determine the basic nature of the networks’ structure, we
measured the CoL network
density scores. The
density measure is based on the amount of actual ties, divided
by the amount possible ties within a CoL. Consequently, it
provides an indication of how well-connected participants
within a particular CoL are (Hanneman &
Riddle, 2005). The amount and nature of an individual’s
network connections was determined via the concept of Freeman Degree
Centrality, including in- and out-degree measures. In-degree network
connections indicate how often and by how many colleagues a
particular individual was contacted from within a CoL. More
specifically, in the context of the Reply-Networks, the
measure captures how often an individual has been replied to
by their colleagues. When considering the Read-Networks, it
reveals how frequent an individual’s contributions were read
by her colleagues. Generally, a high amount of in-degree
connections has been attributed to prominent participants
within (learning) networks, with whom others would like to be
connected (Hanneman &
Riddle, 2005). Therefore,
this
constituted our main variable to check our second research
hypothesis. The out-degree measure accounts for all those
links that originate from a focal individual and summarizes
how often that individual contacted her colleagues within the
CoL. When distinguishing between Reply- and Read-Networks, the
out-degree captures how often a participant has replied to
their colleagues and read their contributions, respectively.
Scholars have often equated a high level of out-degree
connections with influential participants, who are able and
willing to shape discussions (Hanneman &
Riddle, 2005). Consequently, this measure formed the basis for
testing the validity of our first research hypothesis. For the
analysis of our third research hypothesis, we combined the
results of the previous analyses. More specifically, taking
into account that we were dealing with multiple CoL, we
determined participants overall centrality on the basis of the
normalized number of
in- and out-degree ties, which allowed to control for the
different sizes of the individual CoL (Hanneman &
Riddle, 2005). In contrast to the more general, nominal
network measures, these particular values provided more
profound insights on how an individual’s network ties affected
their overall network position within their CoL.
In order to test for the parametric assumption of
normality of the data’s distribution, Kolmogorov-Smirnov tests
(K-S) were conducted. The results revealed a violation of the
normality assumption for all measured variables, which
translated into statistically significant K-S results at the
.01 level. Consequently non-parametric tests were used to
examine the research hypotheses. More specifically,
correlations were determined with the Spearman’s rho measure
(rs). In order to assess whether mean differences
in the chosen network measures between the different
hierarchical levels could be observed, we employed
Kruskal-Wallis tests (H). Jonckheere-Terpstra tests (J-T) were
used to identify whether the potential main effect, as
assessed by H, exhibited any possible linear trends. The
results of this provided valuable information on how the
different hierarchical levels differed in their network
measures. The occurrence of possible patterns within the
underlying H-test results was determined by post-hoc
Mann-Whitney (U) tests. Being designed to only measure
differences between two independent conditions, the U-test
results were corrected by the Bonferroni method. As a result,
our adjusted critical value of significance was .016 for this
part of the analysis. In order to cater for the longitudinal
nature of the data and to test for any possible changes in
participants’ network measures over time, a range of Wilcoxon
Signed Rank test were used. The chosen points in time for the
longitudinal study were based on the work of previous studies,
who conducted similar research on networked learning within
teacher education (de Laat et al., 2007). The authors of these
studies chose for the beginning, the middle and the end phases
of online (learning) community. In the context of this study,
we decided to subdivide the overall duration of the underlying
CoL of 14 weeks into six time intervals of about two weeks
each. This allowed to capture a short “transition period”,
during which the focus of the discussions changed from one
content module to the next. During this timeframe,
participants rounded-up the discussion of the previous module
and started preparing for the next one. Following the work of
de Laat and colleagues (2007), out of the six time
intervals, we then considered Intervals 1 (beginning), 3
(middle) and 6 (end) for our analysis. Finally, we also
estimated the effect size of our findings. However, the vast
majority of effect size measures are only suitable for
parametric data (Snyder & Lawson,
1993). Consequently, we followed the suggestion of
Rosenthal (1991) and approximated the
effect size (r) on the basis of the U-results. This measure
takes on values from 0 to 1, where small, medium and large
effects are associated with .10, .30 and .50, respectively (Cohen, 1992).
Although the focus of this research is on the
impact of hierarchical levels, we acknowledge that this aspect
might only explain parts of possible observed differences
between participants. Consequently, we controlled for age,
gender, educational background, prior knowledge, culture and
motivation for attending the training, which have been
suggested to influence online collaborative learning. With
respect to age, some researchers have suggested that older
employees tend to participate less in online training
activities (e.g. Garavan,
Carbery, O'Malley, & O'Donnell, 2010). Additionally, other
empirical studies have been able to show that age similarity
had the potential to trigger emotional conflicts within
groups, resulting in lower participation rates (Pelled, Eisenhardt,
& Xin, 1999). Regarding gender, Im and Lee (2004) stipulated that if
males dominate women in a regular face-to-face environment,
this is also likely to carry over to an online environment. In
contrast, Joinson (2001) was able to show that
online training environments had an equalizing effect on
participants. When considering participants’ educational
background and prior knowledge, previous studies have
highlighted the potential impact of participants’ prior
knowledge on their behaviour within learning initiatives (Dochy & McDowell,
1997). Even more so, there has been a growing
consensus that individuals’ prior knowledge constitutes an
important variable in participants’ activity patterns (Dochy, Segers, &
Buehl, 1999). If a participant already possesses a
considerable amount of prior knowledge about a certain topic,
it can be expected that she will be more comfortable in
contributing to discussions, thereby positively influencing
her general activity and performance levels. Participants’
cultural background has also been suggested to have an impact
on participants’ behave (Jehn & Bezrukova,
2004). More specifically, researchers like Pelled and
colleagues (1999) suggested that some
cultures tend to exhibit more competitive behaviours than
others. Hence, representatives of a more competitive culture
are also more likely to proactively engage into conversations,
trying to shape discussions and thereby achieve higher
potential benefits. Finally, numerous studies have highlighted
the importance of motivation on participants’ behaviour within
the context of online learning (e.g. Rienties,
Tempelaar, Van den Bossche, Gijselaers, & Segers, 2009).
For example, Yang and colleagues (2006) conducted research in
online learning environments and discovered that motivation
was positively related with how learners perceive each other.
Consequently, when participants share a similar level of
motivation when starting a training program, they tend to “get
along” better, which in turn affects their network behaviour
(e.g. they connect more often).
In this study, participants’ age, gender,
educational background and culture, as assessed by
participant’s country of birth, were self-reported as part of
the training programs official registration form. For
educational background, participants were asked to indicate
their highest attained educational degree, including Bachelor,
Master, PhD and Other (e.g. vocational training). Prior
knowledge was measured via a diagnostic test, consisting of 25
multiple choice questions. All five pre-defined content
modules were assessed based on five dedicated questions each.
These questions were created by academic experts and related
to the working environment of the participants. The response
rate for the test was 88.76 % and the internal consistency of
participants’ answers was acceptable (Cronbach α = .81) (Cortina, 1993). Participants’
motivation for attending the training, were approximated based
on a previously developed instrument (Rienties et al., 2009; Rienties, Tempelaar,
Waterval, Rehm, & Gijselaers, 2006). The questionnaire
consisted of 24 questions, subdivided into four categories,
and was administered with a 7-point Likert scale ranging from
1 (not true for me at all) to 7 (completely true for me). The
applicable categories for this study were (the number of
questions are reported in brackets): “Reasons to join the
Training” (6), and “Expectations and Goals” (10). The response
rate was 88.51 % and the internal consistency was again
acceptable (Cronbach α = .95) (Cortina, 1993).
5.
Results
Overall, while the vast majority of posts were
placed in the forums of the five content modules (86%), only
few contributions were shared in the “Café-Talk” forums (14%).
In order to visualise the underlying data, Figure 1 represents
a graphical depiction of the final Read- and Reply-Network of an
exemplary CoL. A first glance already indicated a great amount
of divergence between these two types of networks.
Participants were highly connected and exhibited very similar
communication patterns with respect to their reading behaviour
(Fig. 1a). However, considerable differences prevailed
regarding whether and how participants replied to each other
(Fig. 1b). Furthermore, a closer look at the figure also
revealed a first preliminary sign that participants behaviour
and network position were related to their hierarchical level
within the organization.
An overall picture of the longitudinal nature of
our data is depicted in Figure 2, which captures the average
density values of the CoL across time. As can be seen from the
applicable figure, the average density per time interval of
the Read-Networks
is about 10-times higher than those of the Reply-Networks. Yet,
while the average density of the Read-Networks
declined over time, the Reply-Networks
increased in terms of density. Nonetheless, at the end of the
CoL, the average density for the Read-Networks
remained considerably higher at a value of 62.27 (range =
26.36 – 86.36), as compared to a final value of 11.54 (range =
0 – 28.21) for the Reply-Networks.
a)
Figure
1. (see pdf) Read (a) and Reply (b) Network of an
exemplary Community of Learning.
The layout of the figure has been determined
using iterative metric multidimensional scaling. The different
hierarchical levels are denoted as: “Low” – light circle;
“Middle” – grey square; “High” – dark diamond
Figure
2. (see pdf) Longitudinal
Data on Average Density Scores for the Communities of Learning.
Table 1 summarizes the results of participants’
overall in- and out-degree network ties for both types of
networks. As can be seen from the table, all measures for the
Read-Networks were
statistically insignificant, which led us to reject research
hypotheses 1 and 2 for these types of network. In contrast,
our Kruskal-Wallis tests clearly indicated significant
differences between hierarchical levels and the degree with
which participants’ either replied to their colleagues, or
attracted replies from others.
Moreover, the Jonckheere-Terpstra tests showed a clear
trend that the amount of both in- and out-degree ties were
both positively related to participants’ hierarchical level.
Additionally, an investigation of the underlying patterns
revealed that the observed differences were especially
pronounced between the “Low” and “High” groups (In-degree: U =
2,261.50, p < .01; Out-degree: U = 2,338.00, p < .05),
which is also reflected in the observed effect sizes
(rin-degree = -.23; rout-degree = -.20).
The results of our longitudinal analysis are
represented in Table 2. As participants’ behaviour within the
Read-Networks did
not show any signs of statistically significant differences,
these networks were neglected from the analysis. Our results
indicated a significant increase of in- and out-degree ties
for the “Middle” and “High” groups over the entire duration of
the CoL. The “Low” group did not exhibit a common, noticeable
trend. Moreover, the evidence indicated that the increases for
the “Middle” and “High” groups were mainly situated in the
first half of the CoL. During the second half, only members of
the “High” group showed significant signs of continued
contact-seeking with their colleagues. Taken together, these
findings indicate that, over time, higher level management was
contacted more frequently than lower level management (H1).
Moreover, our evidence also supported the supposition that
over the duration of the CoL, participants from higher
hierarchical levels were more likely to actively contact other
CoL members, than lower level management (H2).
Table 1 (see pdf)
Results of Kruskal-Wallis and
Jonckheere-Terpstra Tests for (Nominal) In- and Out-Degree
Network Measures
Table 2 (see pdf)
Results of Wilcoxon Signed Ranked Test
for (Nominal) In- and Out-Degree Measures (Reply-Networks).
Similarly to the previous findings, we again
found no significant differences between hierarchical levels
within the Read-Networks.
However, as can be seen from Table 3, our results for the Reply-Networks did
again sketch another picture. More specifically, the
Kruskal-Wallis tests revealed significant in- and out-degree
centrality measure differences between hierarchical levels.
Another set of Jonckheere-Terpstra tests was then conducted to
determine a possible underlying trend. The results showed that
whether participants hold a central position within their
network was significantly and positively influenced by their
hierarchical level. In order to determine the pattern of the
main effect, we conducted another range of Mann-Whitney tests.
Similarly to hypotheses one and two, the most pronounced
difference was again found between the “Low” and “High” groups
(In-degree: U = 2,202.50, p < .01; rcentrality-in = -.23;
Out-degree: U = 2,234.50, p < .05; rcentrality-out = -.24).
For the longitudinal analysis, based on the
described results, we again decided to focus on the Reply-Networks. Table
4 summarizes the main results of the applicable analyses. As
in the case of the more general network statistics, we did not
find any significant results for the “Low” group. In contrast,
participants from the “Middle” and “High” groups attained
higher in- and out-degree centrality measures throughout the
duration of the CoL. However, the main acceleration for this
development again appeared to be situated in the first half of
the CoL. Taking into account that the Read-Networks did
again not yield any significant results, we did not find any
support for the notion that, over time, higher level
management will hold more central positions in their CoL
network, compared to their colleagues from lower positions
(H3). However, based on the statistically significant findings
for the Reply-Networks,
we accepted our third research hypothesis for these types of
CoL networks.
Table 3 (see pdf)
Results of Kruskal-Wallis and
Jonckheere-Terpstra Tests for (Normalized) In- and
Out-Degree Network Measures
Table 4 (see pdf)
Results of Wilcoxon Signed Ranked Test
for (Normalized) Network Measures (Reply Networks).
The investigation of whether participants
differed in terms of age, gender, educational background,
prior knowledge, culture, or motivation for attending the
training, subject to their hierarchical levels, revealed no
significant results. However, we also conducted a separate
correlation analysis, where we investigated any possible,
underlying relations between all variables included in this
study. As can be seen from Table 5, in terms of our dependent
and control variables, participants’ hierarchical level was
positively correlated with age. A closer look at the control
variables revealed that age (Reply-Networks:
In-Degree), gender (Read-Networks:
In-Degree) and prior knowledge (Read-Networks:
Out-Degree) were positively correlated with some of the
network measures. Hence, in order to incorporate this finding
in our analysis, we conducted a separate partial correlation
analysis between hierarchical levels and the chosen network
measures, while holding age, gender and prior knowledge
constant. The results are presented in Table 6. While
hierarchical levels continued to be significantly correlated
with network measures, a more refined picture emerged. More
specifically, the potential influence of hierarchical levels
now seemed to be mainly applicable for the out-degree
measures. Moreover, the partial correlation analysis showed
this to be true for both the Reply- and Read-Networks.
Consequently, when interpreting the main results of this
research, these findings need to be taken into account.
Moreover, a closer look at the results also
revealed that all measured network statistics were highly and
significantly correlated with each other. In other words, if
an individual participant attained a high amount of in-degree
ties, for example, she would also be very likely to initiate a
high amount out-degree ties and achieve a comparatively high
degree of centrality within her CoL. As we have been able to
show that hierarchical levels have a strong effect on each one
of these measures, this provided additional support for our
supposition that hierarchical levels have a significant impact
on network structures within CoL.
6.
Discussion
The purpose of this study was to determine
whether and to what extend participants’ hierarchical levels
influence the network structures of CoL. We thereby were able
to address a number of shortcomings in current research and
contributed to the discussion about how existing
organizational structures can affect training initiatives. In
order to investigate the relationship between hierarchical
levels and network structures, we employed social network
analysis and conducted a longitudinal study to test for our
research.
In the context of the investigated Read-Networks, we did
not find any evidence for individuals’ hierarchical levels
influencing their network behaviour. However, when considering
the Reply-Networks,
our results clearly indicated that higher level management
attracted more attention, contacted more colleagues, and
attained more central positions within their CoL, as compared
to their colleagues from lower level positions. Additionally,
based on our longitudinal analyses of all network measures, we
were able to show that the overall impact generally increased
over time, and in particular during the first half of the
training program.
Table 5 (see pdf)
Overview of Correlation Coefficients
between Hierarchical Level, Control Variables and Network
Measures.
Table 6 (see pdf)
Correlation Coefficients
for Hierarchical Levels and Network Measures (Controlling for
Age, Gender and Prior Knowledge).
In terms of the Read-Networks, which
capture passive connections between participants (Daradoumis et al.,
2004), this can be considered as a preliminary
indication that CoL have the potential to stimulate an
interpersonal knowledge transfer among participants (Argote & Ingram,
2000). However, the observed range of density scores
across the different CoL varied considerably. Moreover, while
the average overall density score of 62.27 can be regarded as
reasonable, there still remains a considerable gap to be
filled in order to achieve a situation where “everyone reads
everything”. Regarding the Reply-Networks, we
were able to validate our second research hypothesis, which
stated that over time, participants’ ability to attract
connections from other colleagues will be positively related
to their hierarchical level (H2). This supports the work of
Krackhardt (1990), who suggested the
existence of a vortex that allows higher level management to
attract more attention and connections from their colleagues.
Additionally, our evidence suggested that higher level
management will proactively set the tone in online discussions
(H1), which confirms the work of Yates and Orlikowski (1992). We were also able to
show that higher level management held central positions,
while lower level management was located more towards the
fringe of their CoL (H3) (Borgatti & Cross,
2003). Finally, when conducting longitudinal analyses
of the underlying data, our results indicated that the
observed general patterns increased over the duration of the
CoL (e.g. Bird, 1994; Sutton et al., 2000). Additionally, this
positive trend was particularly pronounced during the first
half of the training program, which appears as a kind of
“initiation phase”. However, we also discovered that this
trend was not statistically significant for the “Low” group.
This finding can be considered as support for the work of
Nembhard and Edmondson (2006), who suggested that
members of this group generally tend to be more passive in
discussions within training programs. Additionally, it could
also be attributed to the importance of the “initiation
phase”. Once members from the “Middle” and “High” group have
established their comparatively more central role within their
CoL, it seems as if the “Low” group is content with the
situation. Alternatively, it could also be that members of the
“High” group convey such an “imposing message”, trying to lead
the group and becoming (more) central to the discussions, that
representatives of the “Low” group rather not change their
behaviour and become more active.
Furthermore, when reinvestigating the potential
influence of hierarchical levels on the chosen network
measures, while incorporating our control variables, an even
more refined picture emerged. Our results indicate that age,
gender and prior knowledge seem to have a mediating role in
determining participants’ network measures. More specifically,
participants’ hierarchical background mainly affected their
out-degree behaviour, e.g. the degree with which they reply to
colleagues in discussions. Additionally, this effect was
applicable for both the Reply-
and the Read-Networks,
which suggests two main conclusions for higher level
management. First, members of this group really try to set the
tone and actively try to shape the discussions. Second, higher
level management more carefully followed the discussions by
reading the contributions of their colleagues from lower
hierarchical levels.
Considering these findings, we can draw
conclusions about how collaborative learning activities within
CoL should be designed and facilitated, in order to provide
participants with a valuable learning experience. For example,
acknowledging the considerable influence of hierarchical
levels, organizers can device targeted interventions that
increase the potential benefits of CoL (Cross et al., 2006). More specifically,
higher level management could be stimulated to actively draw
upon the input of their colleagues, thereby allowing
participants from lower level management to gradually move
towards the centre of the CoL network. In practice, this could
be achieved via two possible approaches. On the one hand,
facilitators could try to foster a (more) active exchange of
information between members of these two opposite parts of the
organization. The potential benefit of this approach would be
that connections between participants would be initiated and
supported by an external party. This in turn could relax
underlying norms and regulation that govern how members from
different hierarchical levels communicate with each other.
Alternatively, participants could be asked to complete
assignments that build upon a type of mentoring system. With
higher level management occupying more central positions,
these participants could take their colleagues from lower
hierarchical levels “by the hand” and actively include them in
the discussions. This could create a pull-effect, whereby
participants, who generally tend to occupy positions towards
the fringe of a learning network, are drawn closer towards the
centre. This not only has the potential to make them a more
integral part of the CoL. It also would provide them with
better opportunities to share their knowledge and insights.
Using the analogy of Kozlowski and colleagues (2009), they could thereby
more easily contribute their piece to the puzzle, which can
enhance the success of the entire organization.
Finally, considering the longitudinal findings of
our research, we have highlighted the importance of the
“initiation phase” within CoL. During the beginning stages of
the learning process, participants get to know each other’s
background characteristics, including professional experience
and prior knowledge. Additionally, participants will also
exchange either directly (as part of their introduction to the
CoL), or indirectly (by making appropriate references)
information about their hierarchical levels. This in turn will
significantly influence their behaviour towards each other
throughout the CoL. Consequently, facilitators of such
communities should pay specific attention to this initiation
process, in order to be able to possibly intervene in the
discussions and assist the central participants to engage the
entire group into the discussions.
7.
Conclusions
The current study exhibits two main limitations
that should be taken into account when considering our
results. First, we have based our social network statistics
purely on observed links between participants. In contrast,
previous studies have also commonly incorporated familiarity
measures in the context of social network analysis (e.g. Krackhardt, 1990). These measures allow
to control for the degree with which participants might
already be acquainted with each other. This in turn could
influence the comfort level of participants’ and thereby
affect their behaviour within CoL. Second, connections between
participants did not take into account the content of the
shared information. Consequently, network ties between
individual participants might have reflected personal
commonalities that have no direct link with the actual content
of the training and are therefore difficult to control for by
organizers of similar initiatives.
Building upon the findings of this study, future
research should conduct (hierarchical) multilevel regression
modelling (Goldstein, 1995). Our results indicate
that age, gender, and prior knowledge also had an effect on
participants’ network behaviour. Consequently, in order to
incorporate these findings and to further contribute to our
understanding of whether and how hierarchical levels are
transferred into the network structures of CoL, future studies
should consider modelling a larger set of explanatory
variables simultaneously. Moreover, future research should
conduct a content analysis (CA) of the underlying discussions
forums within CoL. This approach is widely accepted to assess
the quality of learning processes and outcomes (de Laat & Lally,
2003) and allows to draw a more refined picture of the
actual level of content and knowledge that has been exchanged
between participants. Moreover, by mapping the CA results
against the findings of a SNA analysis, it would be possible
to provide detailed insights about who has been in contact
with whom, what they talked about, and whether this has had an
impact on their network position (de Laat et al., 2007). Additionally, future
research should incorporate the role of facilitators into the
analysis of CoL. Previous research has suggested that online
learning communities must be cherished and protected in order
to become an effective educational resource (Paloff & Pratt,
2003). In other words, facilitators’ involvement can
have a considerable influence on how learning networks develop
and evolve over time (Anderson,
Rourke,
Garrison, & Archer, 2001). Yet, although a
considerable amount of research has already investigated how
online facilitation can affect learning processes, the vast
majority of these studies has focused on the context of higher
education (Berge, 1995; de Laat et al., 2006; Garrison, Anderson,
& Archer, 2010)
and largely neglected the field of training within
organizations. By investigating the role of facilitators in
CoL, it would be possible to provide profound insights that
can serve as a springboard for facilitators to design and
implement an effective teaching strategy for CoL.
Consequently, the quality of learning process could be further
augmented.
Keypoints
References
Amin, A., & Roberts, J.
(2006). Communities of Practice: Varieties of Situated
Learning’. EU Network
of Excellence Dynamics of Institutions and Markets in Europe
(DIME). http://www.dime-eu.org/files/active/0/Amin_Roberts.pdf
Barabási, A.-L. (2003). Linked : How
Everything Is Connected to Everything Else and What It Means
for Business, Science, and Everyday Life. from http://quijote.biblio.iteso.mx/dc/ver.aspx?ns=000149871
Barabasi, A. L., & Albert, R. (1999). Emergence of scaling in random
networks. Science, 286(5439),
509-512.
Borgatti, S. P. (2002). NetDraw: Graph
visualization software. Harvard, MA: Analytic
Technologies.
Cohen, J. (1992). Statistics a power primer.
Psychology Bulletin, 112,
155–159.
de Laat, M., & Lally, V. (2003). Complexity, theory and praxis:
Researching collaborative learning and tutoring processes in a
networked learning community. Instructional Science, 31(1-2), 7-39.
de Laat, M., Lally, V., Lipponen, L., &
Simons, R.-J. (2007). Investigating patterns of
interaction in networked learning and computer-supported
collaborative learing: A role for Social Network Analysis. Computer-Supported
Collaborative Learning, 2, 87-103.
Goldstein, H. (1995). Multilevel statistical
models. Sydney: Edward Arnold.
Paloff, R., & Pratt, K. (2003). The virtual student: A
profile and guide to working with online learners. San Francisco: Jossey-Bass.
Panzarasa, P., Opsahl, T., & Carley, K.
M. (2009). Patterns
and Dynamics of Users' Behavior and Interaction: Network
Analysis of an Online Community. Journal of the American
Society for Information Science and Technology, 60(5),
911-932. doi: 10.1002/Asi.21015
Rosenthal, R. (1991). Meta-analytic procedures
for social research. Newbury Park, CA: Sage.