The‘‘two-stepflow’’modelofcommunication(KatzandLazarsfeld,1955;Katz,1957)statesthatmediainfluenceisexertedthroughtheintermediaryactionsof‘‘opinionleaders’’,individualswhoarehighlyengagedwithmediacontentaroundanissueandwhoacttointerpretanddisseminatenewinformationtoothers

wefindthatmostindividualsengagedinonlinediscussionsareembeddedwithincommunitiesoflike-mindedusers;suchself-reinforcing‘‘echochambers’’canpreventengagementwithalternativeviewpointsandpromoteextremeviews

However,therewerealsoaminorityofuserswhofrequentlyinteractedwithotherswithdifferingviewsandweidentifiedcoherentmixed-attitudecommunitieswheresuchinteractionsoccurred

wefindthat98%,94%and68%ofusersaremembersofanechochamber,and2%,3%and28%aremembersofanopenforum,forfollower,retweetandmentionnetworks,respectively

suggestingthatthescepticminorityhasdisproportionatelyhighvisibilityindebateusingthesehashtags

Networkswerefilteredforvisualisation;followernetworkswerefilteredbyremovinguserswithtweetvolumebelowaspecifiedthreshold,whileretweet/mentionnetworkswerefilteredbyremovingedgeswithweightbelowaspecifiedthreshold.NetworkswerevisualisedasdirectedgraphsusingtheForceAtlas2force-directedlayoutalgorithm(Jacomyetal.,2012)providedbyGephi,suchthatcloselyconnectedusersareplacedneareachotheronatwo-dimensionalsurface

‘Follower’’networksconsistofdirectedlinksbetweenuserswho‘‘follow’’eachother,i.e.alinkA!BindicatesthatuserAisfollowedbyuserB.OnTwitter,usersreceiveallmessagestransmittedbyusersthattheyfollow.‘‘Retweet’’networksconsistofdirectedlinksindicatingthatoneuserhasre-transmittedamessagereceivedfromanother,i.e.alinkA!BindicatesthatamessageoriginallytransmittedbyuserAwasretweetedbyuserB.‘‘Mention’’networksconsistofdirectedlinksindicatingthatoneuserhasreferredtoanotheruserinoneoftheirmessages,i.e.alinkA!BindicatesthatuserAwasmentionedbyuserBinanoriginaltweet.Mentionsaresometimesusedtodrawattentionorengageinconversationwithaparticularuser.Followernetworkswereunweighted.Edgesinretweetandmentionnetworkswereweightedbynumberofoccurrencesofinteraction

Afterapreliminaryinvestigationusingakeywordsearchontheterms‘‘climatechange’’and‘‘globalwarming’’,wechosethethreemostwidelyusedhashtags(#climate,#climate-change,#globalwarming)forTwittercommunicationaboutclimatechange.Wealsochosetwohashtags(#agw,#climaterealists)thatshowedhighusagebyusersexpressingscepticorcontrarianviewsaboutclimatechange;thesewerechosentoensurerepresentationofadiversityofviewsinourdataset

Ourstudyaimedtocharacterisesocialmediadiscussionsofclimatechangebymappingthestructureofusersocialnetworks,measuringthedistributionofuserattitudesacrossthosenetworks,andexploringuserinteractionsandbehaviours

Forexample,Postmesetal.(2000)examinedsocialnormsinstudentemailcommu-nicationsaroundaweb-delivereduniversitycourse,findingthatnormswereestablishedbyaniterativeprocessofobservationandactive(re)negotiation,andthattheyplayedanimportantroleindefiningnewlyemergentsocialgroups.

Thedecentralisedandparticipatorynatureofonlinesocialmediaoffersanovelopportunitytostudypreviouslyinaccessibleaspectsofsocialinteractionaboutclimatechange(Aueretal.,2014),includingthesocialnetworkstructuresthatlinkindividualsengagedinonlinedebateandthatarelikelytoaffecthowattitudesevolveovertime

What do the structural properties suggest about theorganization of the hate groups? As mentioned in point(a), the structure of the network suggests that the hategroups in blogosphere have not formed into centralizedorganizations

The largest connected component, often called agiant component in graph theory (Bolloba ́s, 1985),contained 273 nodes connected by 1115 links. This giantcomponent was a rather dense graph with an average nodedegree of 8.2

In arandom network the probability that two randomlyselected nodes are connected is a constantp. As a result,each node has roughly the same number of links and nodesare rather homogenous. In addition, communities are notlikely to exist in random networks. Small-world networks,in contrast, have a significantly high tendency to formgroups and communities. Most empirical networks rangingfrom social networks (Newman, 2004a), biological net-works (Jeong et al., 2001), to the Web (Albert et al., 1999)have been found to be nonrandom networks. In addition,many of these networks are also scale-free networks(Baraba ́si et al., 1999), in which a large percentage ofnodes have just a few links, while a small percentage of thenodes have a large number of links. Thus, nodes in scale-free networks are not homogenous in terms of their links.Some nodes become hubs or leaders that play importantroles in the operation of the network. The Web has beenfound to have both small-world and scale-free properties(Albert and Baraba ́si, 2002).

For example, the central nodes often play a key role byissuing commands or bridging different communities. Theremoval of central nodes can effectively disrupt a networkthan peripheral nodes

However, it does suggest the possibility that some donations to research-oriented institutions support the research enterprise more than undergraduate education. By contrast, donations to teaching-oriented institutions may more directly influence under-graduate education.

Whereas 10–20% of undergraduate students receive Pell grants at the institutions with the highest eigenvector centrality, the same is true of 50–80% of students at the institutions with the lowest eigenvector centrality.

hus, Rogers (2015) concluded that foundations are now involved in a multitude of ways in education policy reform: identifying problems, funding research, creating solu-tions, selecting and training the individuals implementing policy, influencing government appointments, funding start-up programs, and financing selective media coverage.

This critique, now over thirty years old, has not substantially altered the nature of the foundation-institution relation-ship in higher education

Another significant finding is that efforts of the membersof religious networks—in spite of their relatively closedcharacteristics—in terms of being at the center of a net-work and taking the brokerage role are, contrary to theliterature, highly developed

congregational network members take abrokerage role much more often than the secular; accord-ingly, they are more active in providing relations and thetransfer of information between networks

sity rate ofnetwork (%)0.773

Density rate ofnetwork (%)0.88

While previousstudies presented that social networks have an influence onreligiousness, they did not respond to the question of howreligious ties structure attitudes and relations in the orga-nizational field

We deleted non-directed tweets, but itshould be acknowledged that non-directed tweets may also bear implications for knowledgesharing and can be examined in future studies. The data cleaning procedure also excludedretweets and tweets that serve as quoting.

The second goal of SNA is to identify theflow of information across various types of actors. Knowledge sharing involves participantsfrom a diverse set of professional backgrounds, with varying degrees of healthcare experience,expertise and expectation (Stewart and Abidi 2012)

Nevertheless, they allcarry out a common practice—that is contributing health information that potentially impactsthe knowledge flow

Twitter also facilitates collective actions such as organizingweekly Q&A sessions. For example,#AlzChatis for a live tweet chat that takes place everyMonday on the topic of Alzheimer’s disease. The conversations and joint activities form thecommunity component in CoPs.

Still, it may not be others' actual behavior that drives ourown ad-dictive behavior, but our perceptions of their behavior, where the twoconflict.

Outsiders in-volved in resource management or disaster mitigation must becognizant of the these kinds of underlying dynamics of knowl-edge generation in indigenous communities and how knowl-edge may put be put into practice by individuals who are at theintersection of local and global knowledge.

It is important to note that the bivariate screening forthe reduced sample (N=32) found that none of the socialnetwork variables had a statistically significant relation-ship with the dependent variable. We acknowledge, how-ever, that time constraints prevented the collection of so-cial network data from every individual and, therefore, amore thorough analysis of the global pattern of connec-tions between nodes. Of the two measures of centralitythat were calculated, we have much more confidence inthe degree measurement than the betweenness measure-ment because degree is not affected by an incompletenetwork sample as it measures direct interaction betweentwo individuals. Our betweeness measurement, on theother hand, has limited explanatory power when calculat-ed for an incomplete network and should be viewed asonly a rough guideline.

wo degree and two betweenness cen-trality measures

We also asked people to indicate whom theywould turn to find out something they did not understandabout the marine environment. We determined this to be theirexpert network.

To generateour subsample for the social network study, we employed asnowball sampling method where we initially asked 17 re-spondents about their social and expert networks. This in-volved asking informants to name the seven most importantpersons in their lives, starting with the most important, outsideoftheir household

As with most team cohe-sion research, socially desirable participant responses may be a limitation.

This means that the focus of academic capitalism (as atheory) is not only restricted to commercialisation of research, but also takesinto consideration other aspects of universities (e.g. instruction and adminis-tration) and changing relations within universities (e.g. decreasing status ofmany social sciences and humanities), as well as between universities and theirsocial environment.

interstitial organisations that facilitate collaboration between universitiesand private companies;intermediating organisations and networks between public and privatesector;