Vous êtes sur la page 1sur 4

Eigenvector centrality: A variation and refined version of degree centrality measure.

It counts not only the number of

neighbors connected to a given node, but also weight seach neighbor by its centrality (Borgatti etal.,2013: 68). Eigenvector centrality
requires information beyond the ego-network tobe computed.Thus,it reflects more of the full network structure than degree
centrality(Borgatti, 1995).

In the formula, ei is the eigenvector centrality of actor i, is a constant, xij denotes a tie between actor i and actor j,2 and ej is the
eigenvector centrality of the node j. in other words,the centrality of actor i is dependent not only on its own connectivity but also its
neighbors centrality. For an explanation of why this is called eigenvector centrality, see Newman, 2010: 171172.)

Host-national connectivity: This measure is calculated using information obtained from the full network. It represents a persons ability to
establish and maintain social relations with members of the community from the host country. Mathe-matically, it is the number of host-
nationals one has social relations with divided by the total number of host-nationals in the community. If the egois a host-national, it will
be excluded from the denominator. In other words, this measures the percentage of actualized social ties among all possible ones with the

Co-national connectivity: This measure is calculated using information obtained from the full network. It represents a persons ability to
establish and maintain social relations with members of the community from the same country. Mathe-matically, it is the number of co-
nationals one has social relations with divided by the total number of co-nationals (excluding the ego) available in the community. Those
who have no co-nationals in the community are excluded, since the percentage is undefined when the denominator is zero. This measures
the percentage of actualized social ties among all possible ones with the co-nationals. A special note is required for Americans. By
nationality, they are host-nationals, but due to the unique geographic location of the EI community (outside of the American mainland)
and its highly mixed local cultures influenced by both the west and the east, many American sin the EI do not consider themselves as
hosts(Chi, 2014). Therefore,they are not treated differently when this measure is calculated

Seventeen items selected from the Socio-cultural Adaptation Scale(SCAS, Ward &Kennedy,1999) are used to capture the behavioral
domain of acculturation outcomes. Respondents are asked to indicate how much difficulty they find incertain situations in the hosting
culture(i.e.,Hawaii in this case) on a Likert scale of 5(1=no difficulty; 3=moderate difficulty; 5 =extreme difficulty). This aggregate scale
covers both generic acculturation situations and those pertaining to the specific context and population of this study (i.e.,university
students).Examples of the social contexts and behaviors asked about include making friends, going shopping, understanding local words
and expressions, and so forth. The reliability test shows that items in this scale are consistent in general (Cronbachs = 0.866). Changes in
Cronbachs alpha if an item is deleted were also checked and nostatistically significant difference was identified. The frequency
distributions of the responses crowded at the high erend of the scale; that is, the majority of the respondents reported having no or little
difficulty in most of the situations. The relative mean substitution (Raaijmakers, 1999) was used to replace missing values, which creates
the least bias according to a simulation study comparing four different treatments of missing values in Likert-type scales (Rodriguez de Gil
& Kromrey,2010). A considerable proportion of the respondents are not religious, so the item about religion(Item7) got the highest
frequency of N/A choice. Therefore, this item was dropped from the summative scores of SCAS. Also, to reflect these mantic meaning
better, the SCAS scale was recoded reversely so that higher scores correspond to better adaptation(i.e.,lesssocio-cultural difficulty was

Descriptive network metrics

Several network metrics commonly used to characterize networks were also computed (see Table 2). Degree, or degree centrality, has
already been discussed. In this study, the network of socialization is directed, so for each node, there are both in-degree (number of
nominations from others in the community) and out-degree (number of nominations of others in the community).In-degree indicates
social popularity and out-degree reflects the gregariousness of survey respondents. Reciprocity looks at the cohesion of a network at
dyadic level. Mathematically, reciprocity is defined as the fraction of edges that are reciprocated (Newman, 2010: 205). For directed
relations, such as help giving or advice seeking, reciprocity measures how much of the action initiated by one is reciprocated by the other.
On other occasions, reciprocity could simply mean the agreement between a dyad on a given relation that connects them (Borgatti
etal.,2013: 155). In this study,the reciprocity measure should be interpreted more appropriately in the latter way as the consensus
between two participants regarding whether they mutually consider each other as a socializing partner. Transitivity and clustering
coefficients are measures of network cohesion at the triadic level. Ego-centric clustering is a property of nodes, and is defined as the
number of pairs of neighbors that are connected divided by the number of pairs of neighbors (Newman, 2010: 201). Network clustering or
transitivity is aproperty of the network,and is defined as the fraction of paths of length 2 that are closed into a triad or loop of length
3(Newman, 2010: 199).Transitivity measures the proportion of closed triads with in a network, and reflects how density is distributed
across the network.Transitivity and density distribution are related because, given a fixed number of edges, in order to achieve high
transitivity triplets of edges must be placed near each other(adj a centto shared nodes) rather than distributed evenly across all
nodes.The mean clustering coefficient (aka WattsStrogatz clustering coefficient) is calculated by taking the average of the ego-centric
clustering across all nodes. This metric is sometimes used as a proxy for network clustering, but it is likely to be biased toward low-degree
nodes, thus providing a less accurate representation of the entire network(Newman, 2010: 203). Although density is afrequently reported
measure for networks, it is not so comparable across different types and sizes of networks. Dunbar (2008) argued that individual social
networks have a characteristics izeof around 150 and distinct structure based on his review of evidence from behavioral ecology, neuro
psychology, evolutionary anthropology,ands ocial network theory. The implication of Dunbars conclusion for interpreting network density
is that the increase in network size after a certain level is likely to result in a decrease indensity as it is getting harder and harder for
anyone to be connected to everyone. Borgatti etal.(2013) came to the same conclusion from their years of work on network analysis as
well that densities areal most always lowerin large networks than in small networks(p.151). So they recommended that network density,
mean degree and transitivity be considered together when one characterizes a network. Degree assortativity refers to the assortative
mixing of degree in the network. In other words, it indicates whether nodes with high degree are more likely to connect to other nodes
with high degree and low degree nodes with other low degree nodes. It is mathematically equivalent to the Pearson correlation coefficient
of node degree (Newman, 2010: 229), and ranges from +1 to 1, with 0 indicating lack of correlation. Components are nodes that are
connected in a network. The weakly connected components (WCC) refer to nodes that are connected somehow regardless the direction
of ties,while the strongly connected components (SCC) take tie direction into consideration. This indicator provides a rough picture of how
much the network is linked by one path(undirected or directed, respectively).

Network representation and data analysis

To construct a full network, one needs at least a node/actor list and an edge/tie list(based on a certain relation). In this case, the node list
was constructed by giving each participant on the EI directory an ID number. Choices participants made by clicking on the photos of other
people (as described in Section 2.2 procedure) provided the edge list for this network. The relational data was stored in a 280 by 280
matrix. In-degree centrality, eigenvector centrality, host-national connectivity, and co-national connectivity were calculated in the
specialized network analysis software UCINET 6.0 (Borgatti, Everett,&Freeman,2002). Although all measures are given at the nodal
(i.e.,individual) level, the regressions need to be run in a network analysis pack age such as UCINET,rather than conventional statistical
packages. There as on is that the basic statistical assumption of variable independence is violated by relational data (the connectivity of
one node affects the connectivity of a nother). UCINET uses permutations to get statistical distributions for calculating significance in a
manner that allows auto correlation.

Table 2 provides basic statistics of the communitys social network, which shows the big picture for understanding the measures and the
testing results of the hypotheses. In-degree is distributed linearly across the network with a standard deviation of 21,while out-degree
distribution centers around a few heavily sociable persons and a long tail of non-respondents of zero out going ties. The reciprocity rate is
47.7%, which suggests that among those one claimed to socialize with within this community, about half acknowledged back the
connections.The network clustering coefficient (transitivity) of the EI social network is 0.452. This means that among all pairs of people
who share a common friend, 45% of the times the two in the pairs are also connected. The degree assortativity for this communitys social
network is around 0, which suggests that in this community, the gregarious people do not preferentially socialize withot her gregarious
people. The communitys social network is all connected in one weak component, which is impressive given the size of the community.
The higher number of strongly connected components is understandable and expected, as 130 nodes of the non-respondents have no out
going ties. Taken together,the indexes reported here indicate that this community exhibits a well-connected and cohesive social network.
Field observations and interviews of community members came to the same conclusion: participants of the EI know a large number of
other fellow students and have plenty of occasions to interact with each other as most of them live in the two dedicated dorm buildings
and are encouraged to volunteer at or attend seminars and many social events. In addition, demographic information from the official
directory and the online survey confirms that this community is multicultural no matter whether it is defined by the diversity of languages
spoken, original nationality, ethnic back grounds, or religious beliefs(Chi, 2014). It is with in this context that the three hypotheses are
tested to see how much each network operationalization of ICC explains the outcome of the community members acculturation as

H1. Sociability in a multicultural community is positively correlated with socio-cultural adaptation

Model 1:SCASin-degree centrality

The average in-degree centrality of nodes is 42,which means that in general 42 others in the same community identify a participant
assocializing partner. Over 50% of the participants socialize with more than 40 community members with a minimum of 2 and a maximum
of 95. The nodal in-degree correlates with SCAS score moderately with r of .588. The regression analysis is reported in Table 3. The result
supports the conclusion that the ability to socialize with others is conducive to adaptation ingeneral. In-degree centrality is able to explain
35% of the variance in socio-cultural adaptation (F = 146.8, p < .001)(Table 4)

H2. Sociability of one self and ones friends in a multicultural community is positively correlated with socio-cultural adaptation.

Model 2:SCAS eigenvector centrality

The nodal level eigenvector centrality is similar to degree centrality,but takes into consideration the degrees of those connected to the
ego as well, so it is better at capturing the network structures impact on the ego. A node of high degree (i.e., connected with many
others) and a node connected to a few high-degree others both can score high one igen vector centrality. Translated into a sociological
understanding, eigenvector centrality is a better measure of global structure because it measures not only how many associates one has
but also the prestige of ones associates. H2 is also supported empirically.The nodal eigenvector centrality has a higher correlation with
socio-cultural adaptation with r = .701. It can explain 49% of the variance in socio-cultural adaptation of the participants (F = 268.6, p <
.001).This means that ifone is connected to many others or a few who are very sociable, it is more likely that one will adapt well socio-
culturally. This makes sense as the eigenvector centrality indicates the potential accesstore sources and support from ones connections in
a social network.

H3a. Sociability with host-nationals in a multicultural community is positively correlated with ones socio-cultural adaptation.

H3b. Sociability with co-nationals in a multicultural community is negatively correlated with ones socio-cultural adaptation.

Model 3a : SCAShost-national connectivity

Model 3b : SCASco-national connectivity

Host-national connectivity and co-national connectivity reflect sociability with two specific cultural groups given the ethnic composition of
the community in focus. Neither correlates much with socio-cultural adaptation(rhost = .22; rco = .10). The regression model test for H3a
is statisticallys ignificant (F = 14.35, p < .001), but theres ultistrivial. Only 5% of the variance can be explained by connectivity with host-
nationals. The test for H3b is not statistically significant(F = 2.54, p = .11).

Many researchers have urged that the predominantly individual-oriented assessments of ICC need tobe expanded to include more
relational perspectives (Deardorff, 2009;Fantini,2009;Lustig&Spitzberg,1993;Matsumoto&Hwang,2013; Spitzberg
&Changnon,2009;Spitzberg,1989,1994), and some effort has been made in this direction (e.g., Arasaratnam & Doerfel,2005; Deardorff,
2011; van Driel&Gabrenya,2013). This study set out to take advantage of recently developed network analysis techniques toward the
same goal. Three ways of operationalizing ICC using structural information rather than individual traits were tested with regard to their
explanatory power of socio-cultural adaptation. It was found that individual sociability as measured by degree centrality is conducive to
cross-cultural adjustment in a multicultural community. This is in line with other studies that took an ego-network approach as reviewed
earlier. However, degree centrality only captures the immediate local structures impact and does not require information from the full
network(i.e., the social context) that the individuals are embedded in (Borgatti etal.,2013: 165). The second hypothesis went further to
test how a measure that in corporates non-local network structure performs. Results show that eigenvector centrality, as a combined
measure of ones own and ones friends sociability, is a better candidate for ICC assessment if the structural information of the full
network is available. This indicates how SNA captures what is missing in a conventiona lpsychological approach (i.e.,one that is focused on
individual attributes). Scholars of acculturation have started perceiving interpersonal adjustment (Ryder etal.,2013) as a nother dimension
of adaptation. This leads to the need for relational-based measures of ICC, and the present study offers degree and eigenvector centrality
as candidates. Methodological research is quite active in SNA, so we can expect further candidates that should be explored. The third
hypothesis was split in to two parts to test whether socializing with particular cultural groups matter more to adaptation.Co-national
connectivity was not statistically supported and connectivity with host-nationals was found to have minimal effects on the outcome
eventhough it was statistically significant. Church (1982) concluded that a common finding from previous acculturation studies is that
more host-national friendship correlates with higher level of satisfaction, less homesickness, and lessloneliness. Yet, the relation between
host-nationals and home-nationals roles in facilitating ICC was not supported. This result can be understood in the context of the
culturally well-mixed and socially cohesive hosting community at the EI, where the nominal host nationals (Americans) do not play a
special role. Such a result high lights the need and importance of developing measures of ICC that take in to account the socio-cultural
environment with in which the intercultural interactions take place. The lack of contextual sensitivity of individual trait-based measures
might explain the inconsistent findings from past studies about the relationship of host/co-national connections and adaptation

Matsumoto & Hwang(2013) emphasized the significance of developing good assessments of ICC. This study and its results have
contributed new ways of conceptualizing and measuring ICC from a relational perspective. To start, when people respond to survey
questions based on social relations, they are reporting concrete interactions and their responses can be validated by their partners on the
other end ofthe relation.This is different from many previous measures that have depicted interactants as too conceptual, too rational,
too conscious, and too intentional(Spitzberg &Changnon,2009: 35) and are excessively dependent on self-judgment. Secondly, the results
demonstrate how a network approach is able to capture what is missed by previous individual-based methods.The use of information
from the full-network structure allows us to see a well-defined context in which findings should be understood and interpreted.
Previously, connections with host-nationals were perceived and used predominantly as indicators of ones ICC without a close look at the
availability and accessibility of host-nationals in the receiving community. This actually resembles the assimilation point of view,using a
term from Berrys model (Berry, 1999,2003), because relations with host-nationals are portrayed as more valuable and instrumental than
those with co-nationals. But, this full-network study shows empirically that connectivity it self rather than with whom one is connected
serves as a better predictor for adaptation in a well-mixed multicultural community. It exposes the limitation of focusing on a binary
home-host national distinction in intercultural research when the social environment is culturally more diverse. It invites further efforts to
develop theories and models that take in to consideration both individual and structural understandings of ICC. Many researchers in
psychology, anthropology, and computer science have realized that it is not possible to fully understand how people le ar nor work if the
unit of study is the unaidedin dividual with no access to other people or to artifacts for accomplishing the task at hand(Nardi, 1996: 69).
Cultural Historical Activity Theory (CHAT), initiated by Vygotsky (1978) and elaborated by others(e.g., Cole &Engestrm,1993) provides a
useful frame work for analysis beyond independent individuals. Findings from this study suggest conceptualizing ICC asasocially
contextualized construct, and a possible direction in the future is to explore the possible application of CHAT to consider further the role
of cultural artifacts and historical context in the study of ICC. Thirdly, the social network view of ICC is dynamic and opens doors to
different research questions. For example, considering that social settings change and social relations evolve overtime, how does an
individuals ICC change as a function of the individuals relational context?

How can the theoretical construct of ICC be extended from individuals to groups, and applied tounderstand collective competency in
situations in which multiple groups of distinct cultural origins are interacting with each other? What is the relation between trait-based
assessments of ICC and network-based assessments of ICC: how do these two types of ICC assessments interact with each other? Which
one provides better evaluation criteria for organizations that aim at promoting cultural diversity and intercultural communication
competence, and how can they be productively combined as theoretical and practical constructs? Finally, the relational assessment of ICC
developed from network analysis also has implications for practice. Selection of expatriates for over seas assignments and evaluation of
sojourners adaptation often depend on identifying crucial indicators of ICC and developing valid measures of them (Fantini,
2009;Jordan&Cartwright,1998;Matsumoto&Hwang,2013;Ruben, 1976; Tung,1982). Trait-oriented scales can provide researchers with a
distribution of ICC of a population. But for individuals, it is hard to translate the scores straight forwardly in to actions that individuals or
organizations can take for improvement. Network measures are particularly informative and instrumental in this aspect, because each
actor can get an intuitive idea of their social networks from visualizations of their communitys network as well as their own ego-networks.
The very act of making one conscious of the cultural composition of their own social networks and of the community they live in could
have profound impacts on their future behavior(as was observed by Chi, 2014). It allows for personal reflections and adjustments and also
gives group and community leaders a big picture about which kind of intergroup relations have been developed and what could be
fostered further. In the context of todays rapidly developing social media, such a relational-ICC assessment might be implemented online
forreal-time feed-back and benefit those involved in intercultural communication. An understanding of ICC asconstructed through social
interaction canal so inform interface design of socia lmedia(Chuang, 2014).

4.2. Limitations and suggestions for future research

This study has some limitations that confine the interpretation of its findings and generalization of the conclusions. First of all, the special
requirement of a complete name list to do full-network analysis comes with advantages and disadvantages. It provides rich information
about a greater context, but leave sout those who actually interact with community members but are not on the list (e.g.,not currently
affiliated with the EI). The effects of such an artificial boundary are hard to gauge, although the number of such people is not large.
Although various efforts were taken to boost the response rate, there are still about forty five percent members from the community who did
not contribute to the relational data and it is unknown how these missing values might change the results. A further question tobe asked is
how to reach out to those who are not sociable by nature. How should their ICC be measured in network analysis without being biased?
Robins and Kashima (2008) laments that what is too often missing from a network approach is a complex representation of the nodes in the
network; that is, of the individuals within the system. If personality traits, such as introvert or extravert nature, are factored in to the study,
there would be a chance to compare the two and see the differences or values of each. But for a practical reason these were not measured.
Questionnaires on a network of size 280 can burden respondents heavily as they need to go through everyone on the name list regarding
their social relations. The trade-off was made by keeping the questionnaire short and leaving items on personality traits out. Secondly, the
characteristics of the social networks of this community may not becommon in typical social situations and results might not be
generalizable to other socio-cultural contexts.Whether the ICC identified in the full social network translates toother contexts is a question to
explore. But it is argued that what is learned from this successful multicultural community expands the horizons of researcher sinthisarea.
Future studies can sample more types of communities to test the results ordiscover more variations of relational ICC in practice. Thirdly, the
cross-sectional design of this study does not explore the potential of a network approach for investigating the dynamics of collectively
developed ICC facilitated by a well-connected multicultural community. Future studies can continue data collection from the EI by tracking
the development of social relations to visualize the temporal development of relational ICC. For example, participants social networks
might change dramatically because of the EI experience (e.g., its long-term intercultural exposure and community building effects). It
would also be interesting to investigate the alumni networks of former members of the EI to see whether and how the network-based ICC
developed at the EI is carried on or migrated with the person after they leave the community. Although it would be nice to include other
measures of ICC to compare with the network measures, such scales were not included because of concerns for respondents fatigue. Social
network questionnaires tend to be tediousand repetitious, which has severe impacts on response rate.This trade-off is necessary in this study
but can be avoided in future studies of communities of smaller sizes. Finally, this study is based on a network of positive social relations,
but problematic or difficult relations might reveal more about ones ICC or lack of ICC. Future studies could take into consideration of
negative intercultural relations and investigate their impacts on intercultural communication competence in comparison with positive social
experiences. In conclusion, this study has shown that network metrics derived from a relational view of ICC can account for acculturation as
measured by socio-cultural adaptation, and points to the promise of further workin both theorizing and measuring ICC as
residinginactorssocial networks as well as individual characteristics.


Our special thanks to the anonymous Educational Institution foraccess to their population and roster data, and toProf. Dharm Bhawuk for his
insightful comments and careful edits during the process of writing this paper.We would also like to thank the two anonymous reviewers for
their insightful comments and suggestions.