takhteyev wellman gruzd 2010

Upload: antero-garcia

Post on 02-Jun-2018

236 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    1/25

    Geography of Twitter Networks

    Yuri TakhteyevUniversity of TorontoFaculty of Information

    [email protected]

    Anatoliy GruzdDalhousie University

    chool of Information !ana"ement

    "[email protected]

    #arry $ellmanUniversity of Toronto

    De%artment of ociolo"y

    &[email protected]

    Abstract$e use a sam%le of %u'licly availa'le data on T&itter to study net&orks of mostly &eak asymmetric ties.$e sho& that a su'stantial share of ties lie &ithin the same metro%olitan re"ion. As &e e(amine ties

    'et&een re"ional clusters) &e find that distance) national 'orders and the difference in lan"ua"es all affectthe %attern of ties. *o&ever) T&itter connections sho& the more su'stantial correlation &ith the net&orkof airline fli"hts) hi"hli"htin" the im%ortance of lookin" not +ust at distance 'ut at %re,e(istin" ties

    'et&een %laces.

    1. Introduction

    ocial contacts 'enefit from %hysical %ro(imity. This fact of social life is so 'asic that for a lon"time %ro(imity &as often sim%ly taken for "ranted. ocial interaction &as "enerally understood tomean face,to,face interaction) for &hich distance acts as a %o&erful 'arrier. In other &ords) the fact that'ein" near each other facilitates the formation of social ties for the most %art &as not so much a findin"of social research) 'ut its 'asic assum%tion. $ith a fe& odd e(ce%tions) communities could 'e safely%resumed to 'e local -eller /0123. ocial net&ork analysts &ere amon" the first to ar"ue thatcommunity should not 'e assumed to 'e local -$e''er /0145 !itchell) /0105 Tilly /0675 Fischer et al.)/0665 Fischer /0485 $ellman /0605 $ellman /0605 $ellman and 9ei"hton /0603. They sho&ed thatthe social net&ork a%%roach afforded the %ossi'ility of follo&in" social ties as they cross s%ace) thusma%%in" the more distri'uted communities that &ere re%lacin" the ones 'ased on nei"h'ourhoods.

    The advent of the Internet created even "reater %ossi'ilities for maintainin" useful social tiesover lon" distances) as &ell as "reater a&areness of such %ossi'ilities. :undits %roclaimed that distance&as dead -;airncross) /0063 and that the &orld &as no& >?3) assumin" thatlo&,cost) instantaneous) content,rich Internet communication &as eliminatin" the need for %ro(imity inmaintainin" contact. !any recent studies have sho&n that &hile social ties can o%erate over distance)%ro(imity does make a difference -e.".) #utts) 8>>03. *o&ever) most such studies have looked at e,mail) &hich &as sho&n to hel% e(tend and maintain e(istin" stron" ties -e.".) $ellman *o"an) et al.)8>>13. In the recent years) ne& relational means have develo%ed on the Internet) some of &hich seemless tied to stron" ties or face,to,face,contact. Are those ne& forms of electronic interaction alsoaffected 'y %ro(imity

    $e focus on one such Internet,'ased system) T&itter) a %o%ular social net&orkin" and micro,

    'lo""in" service that allo&s users to %ost and read short messa"es) limited to /7> characters. uchmessa"es B called

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    2/25

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    3/25

    T&itter also differs from other >03 re%ort a uantitative investi"ation on the&eakness of T&itter ties) concludin" that only a small share of such ties im%ly any meanin"fulinteraction.

    This easy formation of ties makes T&itter accounts similar to 'lo"s B a term ?3 &ho found that distance has an effect. 9iveournal >? re%ort that an avera"e user in their sam%le has ei"ht

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    4/25

    "reater distance translates into hi"her cost of travel) &hich de%ends on the e(istin" trans%ortationinfrastructure. Distance is also intert&ined &ith many other factors) includin" national 'oundaries anddifferences in lan"ua"e. #oth limit interaction &hile 'ein" correlated &ith distance) and their effectscan reduce the avera"e len"th of ties. $e thus focus on four varia'les in our investi"ationE the %hysical%ro(imity 'et&een the users) the freuency of fli"hts 'et&een their cities) &hether the users are in thesame country) and the match in lan"ua"e.

    !.1 %hysical %ro&i'ity

    In the %re,Internet days) $ellmanCs second study of Hast Yorkers -a former 'orou"h of Toronto3found that only 88 %ercent of their socially,close friends and relatives &ere in Hast York and no HastYorkers had most of their active ties livin" &ithin a mileCs &alkin" distance -$ellman et al.) /0443. TheHast YorkerCs lon"er ties &ere ena'led 'y the "ro&in" use of cars and %hones) &hich allo&ed %eo%le tocoordinate meetin"s 'y %hone and the drive to them. -ee also Ito) !atsuda Daisuke) 8>>?5!cH&en) 8>>0.3 Hast YorkerCs &orlds &ere e(%andin") 'ut &ere not indifferent to face,to,faceinteractions. ather) social contacts &ere no& %rimarily 'ounded 'y the scale of metro%olitan area)reflectin" increased mo'ility and flat,rate metro%olitan %hone calls -$ellman Tindall) /0023. $hile

    the tele%hone &as im%ortant) its use &as com%limentary rather than su'stitutive &ith in,%erson contact.$hile Hast Yorkers had su'stantially more distant ties) they rarely had freuent tele%hone contact &ith%eo%le &hom they did not also often see in %erson -$ellman) /0605 $ellman and Tindall) /0023.

    A ne&er study of Hast Yorkers conducted in 8>>? discovered a"ain that the num'er of socialties dro%s shar%ly as the distance increases 'et&een / and 8> miles -!ok et al.) 8>/>3. $hile modernHast Yorkers maintain yet more distant connections) some reachin" as far as Huro%e or :akistan) their%hone use is lo&er for distant connections. Hmail B a ne& medium not availa'le to the Hast Yorkers ofthe /04>s B is less sensitive to distance) in the sense that the amount of email contact per tieis rou"hlythe same re"ardless of the contactCs location. *o&ever) since Hast YorkersC ties are local) theover&helmin" ma+ority of their email is local as &ell. ther studies have similarly found that mostemail use is local) and that there is lo&er overall contact &ith stron" ties &ho live further a&ay -e.".)

    #oase) 8>>43. tudies of >73 su""ests that they in fact do. Ties created throu"h

    cosmo%olitan interests may 'e uite &eak. This) ho&ever) can 'e their stren"th. Unlike stron" ties thatoften reuire nurturin" throu"h in,%erson contact) &ith communication technolo"ies servin" as only anim%erfect su'stitute) &eak ties can traverse lar"e distances) takin" full advanta"e of the o%%ortunitiesoffered 'y ne& technolo"ies.

    In %articular) it 'ecomes im%ortant to ask &hether distance matters at different scales. It is uite%ossi'le) for e(am%le) that distance %lays an im%ortant role at short ran"e) 'ut makes no difference %asta certain threshold.

    7

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    5/25

    !.2 Air Tra(el

    Distance) ho&ever) does not affect social interaction directly. :erha%s the most immediate effectof distance is that it limits the o%%ortunities for face,to,face interactions) &hich are im%ortant for socialties. :eo%le &ho live in different %laces are less likely to meet in the first %lace. They &ould 'e a lotless likely) for e(am%le) to take u% +o's at the same com%any or to meet at a 'ar. In cases &here they domeet -e.".) &hile travelin" or due to movin"3) lack of face,to,face interaction can lead to a decline inthe activeness of other ties) as for e(am%le found 'y !ok et al. -8>/>3. The e(tent to &hich distancetranslates into a reduced likelihood of face,to,face interaction can vary) ho&ever. Je& York is euallyfar from 9ondon and from the to&n of Hirune%L in the #razilian state of Amazonas. Gettin" from Je&York to 9ondon) ho&ever) is much much sim%ler than "ettin" to Hirune%L.

    For intercity ties) an im%ortant com%onent of ease of travel is the availa'ility of airline fli"hts.Jumerous fli"hts connect Je& York and 9ondon) &hile a fli"ht from Je& York to Hirune%L &ouldreuire many sto%s. $e must therefore ask &hether the freuency of airline fli"hts may 'e a 'etter%rediction of non,local ties than the %hysical %ro(imity. -Jote that freuency of fli"hts is stron"lycorrelated &ith %ro(imity and may to some e(tent serve as a %ro(y for %ro(imity the the shorter ran"e)&hile 'ein" a 'etter %redictor ties 'et&een %laces that are further a%art.3

    Freuency of airline connections can also 'e inter%reted as a %ro(y for the more "eneralconnectedness. In the literature on "lo'al cities) for e(am%le) airline data has 'ecome one of the%o%ular &ays of ma%%in" the "lo'al structure of the city net&ork. uch data has sho&n that cities thatare most central in the net&ork of airline connections are also im%ortant in the net&ork of relationshi%s'et&een transnational accountin" firms -#eaverstock et al.) /0003) 'ein" also the same cities that &eredeemed im%ortant in the earlier theoretical literature. !easures such as the freuency of airlineconnections can thus 'e inter%reted not +ust as a factor that reduces travel time -thus facilitatin" theformation of social ties throu"h in,%erson contact3) 'ut also as a measure of 'roader economicconnectivity 'et&een the &orldCs cities.

    !.! National )oundaries

    TodayCs &orld is or"anized as a system of nation,states -!eyer et al. /0063) that is) units that tieto"ether territory) %olitical %o&er and identity. Jation,states are states) in the sense that they control a&ell,defined territory -&hich is usually conti"uous3) and their ri"ht to control that territory isreco"nized 'y other states. Unlike other historical forms of states) ho&ever) a nation,state is associated&ith a

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    6/25

    throu"hout the United tates. esidents of Toronto) +ust a fe& hundred kilometers a&ay) ho&ever) mayhave a lot less interest in the issue) since decision of U courts have limited 'earin" on ;anada. It isim%ortant) ho&ever) to avoid >?3. ne &ould e(%ect shared lan"ua"e com%etency to 'e evenmore im%ortant for use of I;Ts) &hich are often used for %ractices that involve lan"ua"e as a sine quanon. #arnett and ;hoi -/00?3 re%ort) for e(am%le) that lan"ua"e similarity e(%lained 84M of thestructure of the "lo'al tele,communication net&ork in the /04>s. The use of T&itter) a %urely te(tualmedium) &ould similarly reuire at least some de"ree of com%etence in the lan"ua"e of the t&eets.

    9ike national 'oundaries) lin"uistic differences are intert&ined &ith distance. :eo%le livin" inthe same %lace ty%ically share a lan"ua"e -or several3. #ecause of ancient settlement %atterns) morerecent %atterns of colonization and todayCs national 'oundaries) lan"ua"e similarity is also affected 'ydistance at lon"er ran"es. An individual located in Je& York) &ho is likely to s%eak Hn"lish) issurrounded 'y a circle of around 2)>>> km &ithin &hich most %eo%le can 'e e(%ected to s%eak Hn"lishas &ell. 9ookin" %ast 2)>>> km and u% to a'out 4)>>>) such an individual &ould encounter locations&here %eo%le s%eak related lan"ua"es) such as %anish) :ortu"uese or German. The ma+ority of %eo%le&ho live yet further a&ay) s%eak lan"ua"es that are uite dissimilar to Hn"lish) for e(am%le ;hineseand a%anese. The effects of lan"ua"e and %hysical distance can thus 'e closely intert&ined.

    1

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    7/25

    n the other hand) it is im%ortant to note that lan"ua"e stands in a more com%licatedrelationshi% to distance than national 'oundaries. $hile %eo%le usually s%eak the dominant lan"ua"e ofthe city or country &here they live) they can also s%eak other lan"ua"es. In %articular) many %eo%learound the &orld today s%eak Hn"lish in addition to their local and national lan"ua"es. -ee fore(am%le) *errin" et al) 8>>6) on the role of Hn"lish in 9iveournal net&orks.3 The informal andasynchronous nature of T&itter can some&hat reduce the reuired de"ree of com%etenceE readin" -or

    even &ritin"3 /7> character messa"es can 'e less challen"in" than interactin" 'y %hone or email and ithel%s that one can sim%ly i"nore messa"es one does not understand) since no re%ly is usually e(%ected.It is therefore im%ortant to consider not +ust the similarity of lan"ua"es 'et&een %laces &here the usersare located) 'ut also the actual lan"ua"e em%loyed 'y the users.

    *. )uilding a ,a'ple of Twitter Ties

    The %rimary data source used in this article is a sam%le of %airs of "eocoded T&itter accountsconnected 'y a >0.1To ensure that &e &ould only collect t&eetsavaila'le to the %u'lic) the %u'lic timeline &as retrieved &ithout lo""in" into T&itter. $e collected atotal of 74/)874 messa"es. $e do not kno& the %recise total num'er of messa"es %osted durin" theanalyzed %eriod) 'ut &e 'elieve the num'er &as around />> million.6

    The t&eets included in the %u'lic timeline do not necessarily re%resent a randomsam%le of

    %u'lic t&eets sent in the corres%ondin" time %eriod) since &e do not kno& e(actly &hat method T&itteruses for selectin" t&eets that "o into the %u'lic timeline. Furthermore) &e have received re%orts-citation to personal communication removed for review3 that the mi( of messa"es may vary some&hatde%endin" on the account used to retrieve the %u'lic timeline. Des%ite those %ro'lems) the %u'lictimeline is commonly used to sam%le T&itter messa"es -e.".) ava et al.) 8>>65 Jaaman et al.) 8>/>5Golder Yardi) 8>/>3) due to the fact that other %u'lic methods reuire a su'stantially hi"herinvestment in hard&are 'ut similarly %rovide a sam%le of messa"es that could not 'e "uaranteed to 'erandom. -T&itter %rovides access to the full stream of messa"es B kno&n as million as the u%%er limit.

    6

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    8/25

    conseuently under,sam%le the users &ho t&eet on a Je& York schedule -&hich &ould likely includemost of those in Jorth and outh America3 and oversam%le those &ho t&eet &hen Je& York slee%s.-This distortion) ho&ever) &ould only affect the ori"inal sam%le of e"os and not the len"th of the ties)since the e"os connections &ere sam%led in a se%arate ste%.3

    *.2 Geocoding and ,ubsa'pling

    !ost -6? %ercent3 of the messa"es in this lar"e sam%le had some location value associated &iththem. In other &ords) they &ere sent 'y users &ho either s%ecified location in their T&itter %rofile or) inthe minority of cases) used a T&itter client that automatically u%dated the location field in their %rofile.ur analysis of a sam%le of those location descri%tions sho&ed that most of such descri%tions -a'out 4?%ercent3 referred to a real %lace) at the level of %recision any&here 'et&een a country name and e(actcoordinates. For the %ur%ose of our analysis &e assume that such descri%tions re%resent userCs actuallocation) either the %lace &here the user tends to 'e in "eneral or &here they &ere at the time themessa"e &as %osted) discountin" the %ossi'ility that the users mischievously mis,identify theirlocation.4$e &ere careful) ho&ever) to %ro%erly classify location descri%tions that su""est that they are&ishful -

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    9/25

    $hile most of the %rovided location descri%tions could eventually 'e ma%%ed onto a rathers%ecific location) &e found that the users em%loyed a &ide ran"e of conventions for descri'in" &herethey are. $hile many of the locations in 9os An"eles &ere identified as

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    10/25

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    11/25

    de"ree of reci%rocity and is therefore more likely to re%resent a some&hat stron"er tie.3 To sim%lify the%resentation) ho&ever) &e do not discuss our com%arison of mutual and non,mutual ties.

    *.* Aggregating Nearby +ocations

    ince s%ecific locations vary su'stantially in %recision and since users can often choose 'et&een

    a ran"e of s%ecific names for the same %lace -e.".)

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    12/25

    t&o clusters. The &ei"hts for the ed"es in the inde%endent varia'le net&orks are descri'ed 'elo&) &hen&e discuss each varia'le. $e have found that the net&ork of 241 T&itter clusters &as e(tremely s%arse)since the num'er of ties in the sam%le &as small relative to the num'er of nodes. As a result) more than00 %ercent of cluster %airs had zero T&itter connections 'et&een them) leadin" to lo& correlation-'et&een >.>? and >./3 &ith the com%arison net&orks) e(ce%t for the net&ork of airline connections. /8For this reason) &e limited our correlation and re"ression analysis to the ties 'et&een +ust the 8? lar"est

    clusters) &hich allo&ed for a much less s%arse T&itter net&ork -an avera"e of >.61 ties %er %air3.

    3.1 %hysical %ro&i'ity

    The use of T&itter is concentrated in the United tates) &hich accounts for 70 %ercent of oursam%le of e"os) ?7 %ercent of the alters) and si( of the ten lar"est clusters -see ta'le 23. At the sametime) over half of the e"os are in other countries) as are four of the ten lar"est clusters) &hich includeTokyo) So :aulo) and t&o clusters in the United in"dom. In this sense) T&itter users are distri'uteduite &idely around the "lo'e) &ith distance seemin"ly not servin" as much of a 'arrier to T&itter use.At the same time) ho&ever) the users are hi"hly concentrated in a relatively small num'er of locations.8? clusters account for ?7 and 1/ %ercent of the e"os and alters res%ectively. This level of concentration

    e(ceeds the "eneral concentration of %o%ulation in ma+or ur'an a""lomerations.

    /2

    Table ! around here. /0Top clusters.

    #ein" in the same cluster also has a stron" effect on the formation of ties. 20 %ercent of the ties'et&een e"os and alters fall &ithin the same re"ional cluster. The lar"e share of in,cluster ties can 'e%artly e(%lained 'y the su'stantial de"ree of clusterin" mentioned earlierE &hen users are concentratedin a handful of %laces) a lar"e share of ties &ould 'e local even if ties &ere formed randomly) &ith fulldisre"ard for location. The share of local -in,cluster3 ties) ho&ever) is su'stantially hi"her than &hat &e&ould e(%ect +ust due to clusterin". Given the distri'ution of e"os) only 8 %ercent of ties &ould 'e localif ties &ere formed randomly 'et&een the e"os. -An avera"e userCs cluster accounts for 8 %ercent of thetotal num'er of e"os.3

    Fi"ure / sho&s the distri'ution tie len"ths -the first "ra%h) la'eled to 8>)>>> km) &ith a %eak at />)>>> km) re%resented 'y the third "ra%h of fi"ure / -la'eled

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    13/25

    corres%ondin" to distances 'et&een ma+or clusters. In our case) for e(am%le) &e can e(%ect asu'stantial num'er of connections &ith the len"th of 2)0>> km the distance 'et&een Je& York and9os An"eles) the t&o lar"est clusters. $e can e(%ect to findfewerof the shorter ties) due to the smallernum'er of users located in the ran"e of 8)>>> to 2)>>> km from Je& York and 9os An"eles. imilarly)&e can e(%ect to find very fe& ties a%%roachin" the len"th of 8>)>>> km B not +ust 'ecause of the%rohi'itive nature of such distance) 'ut also 'ecause the anti%odal %oints of all ma+or clusters fall in the

    ocean. -The closest to an anti%odal distance 'et&een a %air of ma+or clusters &ould 'e the one 'et&eenao :aulo and a%an) at around /4)1>> km.3

    The second "ra%h of fi"ure / sho&s the distri'ution of ties in a simulation in &hich e"os)located &here they are in our sam%le) form ties amon" each other at random. The "ra%h sho&s thee(%ected %eak at 2)0>> km -see a'ove3) follo&ed 'y a valley corres%ondin" to not,uite,transatlanticdistances) follo&ed 'y a rise as &e reach Huro%e. A lar"er %eak -at 6)1>> km3) corres%ondin" in lar"e%art to the distance 'et&een Je& York and So :aulo 'ut reinforced 'y a lar"e num'er of otherconnections. The last t&o nota'le %eaks -at />)4>> and /4)?>> km3 corres%onds to the distance fromTokyo to Je& York and from Tokyo to So :aulo.

    ;om%ared &ith this 'aseline) the o'served distri'ution of tie len"ths sho&s a sur%lus of ties fordistances to /)>>> km) a some&hat mi(ed record from there to ?)>>> km and a consistent deficit of ties

    at "reater distances. $e note) thou"h) that the %eak in the num'er of ties at the Je& York 9osAn"eles distance is actually higherthan &e &ould e(%ect if ties &ere formed randomly. n the otherhand) several other e(%ected %eaks remain unrealized. In %articular) &e o'serve no %eaks at the valuescorres%ondin" to the distances 'et&een Je& York and So :aulo) Je& York and Tokyo) and Tokyo andSo :aulo.

    For net&ork com%arison &e created a )>>> km divided 'y the "reat,circle distance 'et&een the t&o clusters)calculated 'y haversine formula3. The com%arison of this net&ork to the net&ork of T&itter ties for theto% 8? clusters sho&s a correlation of >.22 for the to% 8? clusters) si"nificant at >.>>? level -see ta'le73. $e note that this correlation re%resents the effect of %ro(imity on ties 'et&een clusters) that is) after

    the shortest ties are already e(cluded.

    Table * around here. /05A% -orrelations.

    3.2 Air Tra(el

    To investi"ate the effect of the ease of travel on T&itter ties &e o'tained a dataset sho&in" anum'er of fli"hts 'et&een %airs of 2)>82 air%orts on five different days in 8>>4 and 8>>0. -The sam%le

    &as kindly %rovided 'y a researcher at a different institution &hose name is omitted for the revie&.3 $eassi"ned those fli"hts to %airs of clusters 'y matchin" each cluster to the air%orts located &ithin />> kmfrom its center. $e then constructed a net&ork 'y "ivin" each %air of clusters a &ei"ht 'ased on theo'served num'er of fli"hts 'et&een the air%orts assi"ned to each of them.

    The resultin" net&ork sho&ed a correlation of >.80 'et&een T&itter ties and fli"hts &henlookin" at the full net&ork of 241 clusters -the only case &e found a meanin"ful correlation fornet&orks 'et&een all 241 cluster3 and a lar"er correlation of >.78 for the to% 8? clusters -see ta'le 73.The net&ork of fli"hts thus a%%ears to 'e a 'etter %redictor of non,local T&itter ties than the %hysical

    /2

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    14/25

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    15/25

    incomin"E T&itter users in the United tates are often follo&ed from a'road) &ith over 2 incomin" tiesfor each out"oin" tie. For some of the other countries) on the other hand) international ties areover&helmin"ly out"oin". For users in #razil) for e(am%le) the ratio of incomin" ties to out"oin" isnearly /E?. #razilian users of T&itter actively follo& forei"n accounts) 'ut receive little attention inreturn.

    The more domestic orientation of the American users also reflects itself in ho& they descri'e

    their locations. $hen codin" the locations &e noted &hether the country &as stated e(%licitly orim%lied -e.".) ./1 %ercent of reuests comin" from theUnited tates. To "et a measure of %ro(imity 'et&een a %air of clusters &e summed the %roducts of thet&o countriesC %references for lan"ua"es. For e(am%le) Je& York So :aulo %air received a &ei"ht of

    >./7) reflectin" the match in Hn"lish ->.07V>./? W >./73) to"ether &ith ne"li"i'le terms for otherlan"ua"es ->.>>/ for the match in the %reference for :ortu"uese and a'out the same amount for thematch in %reference for %anish3. Je& York Tokyo %air received >.>2) &hile Je& York Amsterdam%air received a &ei"ht of >.20) reflectin" %rimarily the much lo&er %reference for the Hn"lish$iki%edia amon" the reuests from a%an and the much hi"her %reference amon" reuests comin"from the Jetherlands./7The resultin" lan"ua"e net&ork sho&s correlation of >.80 &ith the net&ork of

    /7 $e also constructed an alternative net&ork 'ased on the lan"ua"es s%oken in each clusters and the %ro(imity 'et&eenthe lan"ua"es in the hierarchical classification of lan"ua"es -for e(am%le) assi"nin" a hi"her de"ree of similarity toHn"lish a%anese %air than to Hn"lish Dutch3. $e have found that the t&o lan"ua"e net&orks had a correlation of

    /?

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    16/25

    t&itter ties 'et&een the to% 8? clusters -ta'le 73.

    It is im%ortant to note) ho&ever) that lan"ua"e stands in a more com%licated relation &ith"eo"ra%hy than) for e(am%le) the country effect. $hile a user located in the Je& York cluster isnecessarily located in the United tates 'y virtue of the fact that the Je& York cluster is in the Unitedtates) she may or may not t&eet in Hn"lish. It 'ecomes im%ortant) therefore) to look at the lan"ua"e ofindividual t&eets in addition to the lan"ua"e of the clusters.

    Ta'le 6 sho&s the most common lan"ua"es used in the t&eets. Hn"lish is 'y far the dominantlan"ua"e) accountin" for 62 %ercent of the messa"es. :ortu"uese is the only other lan"ua"e accountin"for more than /> %ercent. a%anese) %anish) Indonesian and German each account for //> %ercent)&ith all other lan"ua"es 'ein" under / %ercent. Ta'le 4 sho&s the most common com'inations oflan"ua"es 'et&een e"os and alters. In 44 %ercent of the cases) the e"o and the alter t&eet in the samelan"ua"e. ver three uarters of those -14 %ercent of all ties3 are cases &here 'oth are usin" Hn"lish)&ith sli"htly over one uarter 'ein" cases &here 'oth use a different lan"ua"e) most often :ortu"ueseor a%anese. ;ross,lan"ua"e ties are relatively rare.

    Table 6 around here. /07ost co''on languages.

    Table 8 around here. /0+anguage co'binations.

    The share of same lan"ua"e ties in other lan"ua"es is su'stantially hi"her for local ties -84%ercent3 and su'stantially lo&er for ties 'et&een clusters -/7 %ercent3. It falls even further if onlyinternational ties are considered -? %ercent3. The total share of same,lan"ua"e ties dro%s some&hat as&ellE from 08 %ercent for local ties) to 44 for ties 'et&een clusters) to 61 %ercent for international ties.

    This loss is made u% almost e(clusively 'y the share of ties in &hich an Hn"lish,t&eetin" alter isfollo&ed 'y a non,Hn"lish,t&eetin" e"o.

    9ookin" at the lan"ua"es used 'y e"os in each cluster or country) &e found a some&hatim%erfect match 'et&een the lan"ua"e used 'y individual users and the dominant lan"ua"e of thecluster. For e(am%le) &hile :ortu"uese is unam'i"uously the dominant lan"ua"e of #razil) only 46%ercent of the t&eets from users located in #razil are in :ortu"uese) &ith another 4 %ercent 'ein" inHn"lish. -Informal analysis of the %rofiles su""ests that many of the Hn"lish,t&itterin" users located in#razil are #razilians rather than travelin" Hn"lish s%eakers.3

    3.3 7ulti(ariate Analysis

    *avin" found that all four varia'les that &e considered have an effect on T&itter ties) &e used are"ression analysis to see &hether their effects are inde%endent. The results of the re"ressions are%resented in ta'le 1. ;om%arin" model 2 &ith models / and 8) &e see that %ro(imity and 'ein" in thesame country have an inde%endent and si"nificant effect. A com%arison of models /) 7 and ? sho&s thatthe same is true for %ro(imity and lan"ua"e. -The effect of lan"ua"e) ho&ever) is not si"nificant &hen&e control for country.3 :ro(imity and the num'er of fli"hts similarly have an inde%endent effect

    >.0? and %roduced nearly identical results. For this reason &e avoid the discussion of the alternative lan"ua"e metric)focusin" +ust on the net&ork %roduced from the $iki%edia dataset.

    /1

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    17/25

    -models 6 and 43. A model com'inin" all four varia'les sho&s a si"nificant effect of %ro(imity) countrymatch and fli"ht num'er -thou"h only at the >.>? si"nificance level3. The num'er of fli"hts a%%ears to'e overall a sin"le 'est %redictor of non,local ties and has the heaviest &ei"ht in the com'ined model.$e note) ho&ever) the residual effect of sim%le %hysical distance.

    Table 9 around here. /05A% Regressions.

    9. -onclusion

    9ookin" at the net&ork of ties in T&itter &e found that distance and related varia'les -lan"ua"e)country) and the num'er of fli"hts3 all have an effect on T&itter ties des%ite the su'stantial ease &ith&hich lon" ran"e ties can 'e formed. As a li"ht&ei"ht system that takes little effort to set u% and can 'eused from either %ersonal com%uters or mo'ile devices) T&itter offers a %romise of transcendin"distance) connectin" everyone &ith anyone. ur analysis sho&s that this is not uite so. A verysu'stantial %ortion of ties -20 %ercent3 connect users &ithin the same re"ional clusters) ty%ically the

    size of a metro%olitan area. -All such ties are also domestic and connect users in the same lin"uisticarea. !ost of them &ould fall &ithin easy drivin" distance.3 Hven for the remainin" lon"er ran"e ties)connectin" users in different clusters) distance a%%ears to matter. Ties u% at distances of u% to /)>>> kmare more freuent than &hat &e &ould e(%ect if the ties &ere formed randomly) &hile ties at lon"erthan ?)>>> are underre%resented.

    For such lon"er ties) distance) lan"ua"e differences) country 'oundaries) and ease of travel canvary inde%endently) even as they remain stron"ly correlated. This &arrants an inde%endent analysis ofsuch varia'les and a com%arison 'et&een them. $e find that %ro(imity) country and the num'er offli"ht freuencies all have an inde%endent effect. -The effect of lan"ua"e is no lon"er si"nificant &hencountry is included in the model.3 The num'er of fli"hts) ho&ever) a%%ears to 'e the 'est %redictor of

    non,local ties) hi"hli"htin" the im%ortance of the structural constraints on ties rather than sim%le%hysical distance.

    6. References

    Anderson) #. /00/. Ima"ined ;ommunitiesE eflections on the ri"in and %read of Jationalism. 9ondonE Nerso.

    Anderson) #. /007. H(odus) ;ritical Inuiry) 8>) %. 28?.

    #arnett) G. A.) ;hoi) Y.) /00?. :hysical distance and lan"ua"e as determinants of the international telecommunicationnet&ork. International :olitical cience evie& /1) 870,81?.

    #eaverstock) .N.) mith) .G.) Taylor) :..) /000. A roster of &orld cities. ;ities /1) 77?,7?4.

    #oase) . 8>>4. :ersonal net&orks and the %ersonal communication systemE Usin" multi%le media to connect) Information);ommunication and ociety //) 7) 70>,?>4.

    #or"atti) .:.) Hverett) !.G. and Freeman) 9.;. 8>>8. Ucinet for $indo&sE oft&are for ocial Jet&ork Analysis. *arvard)!AE Analytic Technolo"ies.

    'oyd) d.) Hllison) J.) 8>>6. ocial Jet&ork itesE Definition) *istory) and cholarshi%. ournal of ;om%uter,!ediated;ommunication/2) /.

    #ruck) A.:.) Dorr) D.) ;ole) D.!.) Favre) .) Gramstad) .) !onaco) .!.) ;ulek) :.X.) 8>>7. Transnational mediaconcentrations in Huro%e. e%ort %resented at the ;ouncil of Huro%e) tras'our" ) France) Jovem'er.

    #utts) T.;.) 8>>6. :ermutation !odels for elational Data. ociolo"ical !ethodolo"y 26) 8?6,84/.

    /6

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    18/25

    #utts) T.;. 8>>0. The Im%ortance of im%le Thin"s. Freeman A&ard 9ecture to the International un'elt ocial Jet&ork;onference) an Die"o) !arch.

    #utts) T.;.) Acton) .) 8>/>. %atial !odelin" of ocial Jet&orks. InE Jyer"es) T.) ;ouclelis) *.) !c!aster) . -Hds.3) a"e*and'ook of GI and ociety esearch) forthcomin". a"e) Thousand aks) ;aliforniaE forthcomin"

    ;airncross) F.) /006. The Death of DistanceE *o& the ;ommunications evolution Is ;han"in" ur 9ives. ;am'rid"e)!AE *arvard #usiness chool :ress.

    ;ard&ell) .) 8>>0. A T&itter Timeline of the Iran Hlection. Je&s&eek) htt%EKK&&&.ne&s&eek.comKidK8>20?2 -visitedE>0,/?,>03.

    ;astles) .) !iller) .) 8>>0 . The A"e of !i"ration International :o%ulation !ovements in the !odern $orld. :al"rave!acmillan.

    ;ornell) .) *artmann) D. /006. Hthnicity and raceE makin" identities in a chan"in" &orld. 9ondonE a"e.

    Derudder) #.) 8>>4. !a%%in" Glo'al Ur'an Jet&orksE A Decade of Hm%irical $orld ;ities esearch) Geo"ra%hy ;om%ass8) ??0?67.

    Derudder) #.) $itlo() F.) 8>>?. An A%%raisal of the Use of Airline Data in Assessin" the $orld ;ity Jet&orkE A esearchJote on Data. Ur'an tudies 78) 826/8244.

    Fischer) ;. /048. To D&ell Amon" Friends. University of ;alifornia :ress) #erkeley.

    Fischer) ; ) ackson) .!.) teuve) ;.A.) Gerson) .) ones) 9.!. and #aldassare) !. /066. Jet&orks and :laces. Free

    :ress) Je& York.Frank) A. G. /010. ;a%italism and Underdevelo%ment in 9atin America. Je& YorkE !onthly evie& :ress.

    Friedman) T.) 8>>? The $orld is FlatE A #rief *istory of the T&enty,first ;entury. Farrar) traus and Girou(.

    Friedmann) .) /041. The &orld city hy%othesis. Develo%ment and chan"e /6) 10,42.

    Golder) .) Yardi) .) 8>/>. tructural :redictors of 9ocal Tie Formation in a ;onversational Jet&ork. Un%u'lishedmanuscri%t.

    GuillLn) !.F.) urez) .9.) 8>>?. H(%lainin" the Glo'al Di"ital DivideE Hconomic) :olitical and ociolo"ical Drivers of;ross,Jational Internet Use. ocial Forces 47) 14/,6>4.

    *errin") ;..) :aolillo) ;..) amos,Niel'a) I.) ou%er) I.) $ri"ht) H.) toer"er) .) cheidt) A.9.) ;lark) #.) 8>>6. 9an"ua"eJet&orks on 9iveournal. InE :roceedin"s of the *a&aii International ;onference on ystem ciences) *ilton$aikoloa Nilla"e) anuary.

    *oneycutt) ;.) *errin") . ;. 8>>0. #eyond micro'lo""in"E ;onversation and colla'oration via T&itter. :roceedin"s of theForty,econd *a&aiZi International ;onference on ystem ciences. 9os Alamitos) ;AE IHHH :ress.

    *u'erman) #.) omero) D.) $u) F.) 8>>0. ocial net&orks that matterE T&itter under the microsco%e. First !onday /7) /.

    *utchinson) .$.) 8>>?. 9in"uistic Distance as a Determinant of #ilateral Trade. outhern Hconomic ournal 68) /,/?.

    Ito) !.) !atsuda) !. Daisuke) . -eds.3. 8>>?. :orta'le) :ersonal) :edestrianE !o'ile :hones in a%anese 9ife. ;am'rid"e)!AE !IT :ress.

    aco's) . /010. The Hconomy of ;ities. Je& YorkE andom *ouse.

    ava) A.) on") [.) Finin) T.) Tsen") #.) 8>>6. $hy $e T&itterE Understandin" the !icro'lo""in" Hffect in User Intentionsand ;ommunities. InE :a%er %resented at the DD $orksho% on $e' !inin" and $e' Usa"e Analysis-$e'DD3) an ose) ;A. Au"ust.

    eller) .) /014. The Ur'an Jei"h'orhood. Je& YorkE andom *ouse.rackhardt) D.) /046. RA: :artiallin" as a Test of %uriousness. ocial Jet&orks 0) /6/,/41. rackhardt) D.) /008. A

    ;aveat on the Use of the Ruadratic Assi"nment :rocedure. ournal of Rua$ive Anthro%olo"y 2) 860,801.

    rishnamurthy) #.) Gill) :.) Arlitt) !. 8>>4. A fe& chir%s a'out t&itter. :roceedin"s of the first &orksho% on nline socialnet&orks) eattle) $A) UA.

    9i'en,Jo&ell) D.) Jovak) .) umar) .) a"havan) :.) Tomkins) A. 8>>?. Geo"ra%hic routin" in social net&orks. :JA)/>8 -223) %%. //182//184 .

    9in) .) *alavais) A.) Xhan") #. 8>>6. The 'lo" net&ork in AmericaE #lo"s as indicators of relationshi%s amon" U cities.;onnections 86) 8) 88,2>.

    /4

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    19/25

    !cH&en) . 8>>0. A $orld !ore IntimateE H(%lorin" the ole of !o'ile :hones in !aintainin" and H(tendin" ocialJet&orks. :h. D. dissertation) Faculty of Information) University of Toronto.

    !c:herson) !.) mith,9ovin) 9.) ;ook) .!.) 8>>/. #irds of a FeatherE *omo%hily in ocial Jet&orks. Annual revie& ofsociolo"y 86) 7/?,777.

    !eyer) .) #oli) .) Thomas) G.) amirez) F.) /006. $orld ociety and the Jation,tate. American ournal of ociolo"y)/>2-/3) %%. /77/4/.

    !itchell) ;lyde -Hd.3. /010. ocial Jet&orks in Ur'an ituations.!anchesterE !anchester University :ress.

    !ok) D.) $ellman) #. ;arrasco) .A.) 8>/>. Does distance still matter inthe a"e of the internet Ur'an tudies 71) /8E in %ress

    Jaaman) !.) #oase) .) 9ai) ;.) 8>/>. Is it eally A'out !e !essa"e ;ontent in ocial A&areness treams. InE:roceedin"s of ;;$) avannah) Geor"ia) Fe'ruary.

    Jis'et) o'ert. /018. ;ommunity and :o&er. Je& YorkE (ford University :ress.

    n") A.) /000 . Fle(i'le ;itizenshi%. Duke.

    :arre\as) .) 8>>/ . ervants of Glo'alizationE $omen) !i"ration) and Domestic $ork.

    :arsons) T. /0?/. The ocial ystem. Glencoe) I9E Free :ress.

    :H$ Internet and American 9ife :ro+ect) 8>>0 . T&itter and status u%datin" .htt%EKK&&&.%e&internet.or"Ke%ortsK8>>0KT&itter,and,status,u%datin".as%(

    :utnam) . 8>>> . #o&lin" AloneE The colla%se and revival of American community) Je& YorkE imon and chuster.

    i'eiro) H. 8>>0. :artici%a]So do T&itter no #rasil atin"e /?M em +unho) informa I'o%e. IDG Jo&^) uly /2)htt%EKKid"no&.uol.com.'rKinternetK8>>0K>6K/2K%artici%acao,do,t&itter,no,'rasil,atin"e,/?,em,+unho,informa,i'o%eK

    a(enian) A.) 8>>1. The Je& Ar"onautsE e"ional Advanta"e in a Glo'al Hconomy. ;am'rid"e) !AE *arvard University:ress.

    chonfeld) H. -8>>03 T&itter eaches 77.? !illion :eo%le $orld&ide In une -comcore3) Tech;runch. Availa'le at htt%EKK&&&.techcrunch.comK8>>0K>4K>2Kt&itter,reaches,77?,million,%eo%le,&orld&ide,in,+une,comscoreK

    mith) D.) Tim'erlake) !.) /00?a. ;ities in "lo'al matricesE to&ards a ma%%in" of the &orld,systems city system. InE no(: 9 Taylor : -eds.3 $orld ;ities in a $orld,ystem) ;am'rid"eE ;am'rid"e University :ress.

    mith) D.) Tim'erlake) !.) /00?'. ;once%tualizin" and ma%%in" the structure of the &orld systems city system. Ur'an

    tudies 28) 846,2>8.mith) D.) Tim'erlake) !.) 8>>/. $orld ;ity Jet&orks and *ierarchies) /066,/006E An Hm%irical Analysis of Glo'al Air

    Travel 9inks. American #ehavioral cientist) 77-/>3. /1?1,64.

    te"er) !.) 8>>4. The ise of the Glo'al Ima"inaryE :olitical Ideolo"ies from the French evolution to the Glo'al $ar onTerror.(ford University :ress) (ford.

    ysomos) 8>/> . H(%lorin" the Use of T&itter Around the $orld. htt%EKK&&&.sysomos.comKinsidet&itterK"eo"ra%hy

    Tilly) ;. /067. _Introduction._ :%. /,2? in An Ur'an $orld) edited 'y ;harles Tilly. #ostonE 9ittle) #ro&n.

    Tilly) ;.) /00> . ;oercion) ;a%ital) and Huro%ean tates) AD 00>,/008.

    T&itter) 8>/>. T&itter HT A:I !ethodE statuses %u'lic`timeline.htt%EKKa%i&iki.t&itter.comKT&itterHTA:I!ethodEstatuses%u'lic`timeline.8>>0,>1,/6,8/,2?,>>

    U De%artment of ;ommerce. 8>>7. The Forei"n,#orn :o%ulation in the United tatesE 8>>2 .$allerstein) I. -/0603. The ;a%italist $orld,Hconomy. ;am'rid"eE ;am'rid"e University :ress.

    $e''er) !.) /012. rder in DiversityE ;ommunity &ithout :ro%inuity. InE $in"o) 9. -Hds.3) ;ities and %aceE The FutureUse of Ur'an 9and) #altimore) ohns *o%kins :ress.

    $ellman) #.) /060. The ;ommunity RuestionE The Intimate Jet&orks of Hast Yorkers. American ournal of ociolo"y 47-?3E

    $ellman) #.) 8>>/. :hysical :lace and ;y'er :laceE The ise of Jet&orked Individualism. International ournal of Ur'anand e"ional esearch 8?) 886,?8.

    /0

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    20/25

    $ellman) #) *o"an) #. &ith #er") .) #oase) . ;arrasco) ,A.) ;btL) .) ayahara) .) ennedy) T.9.!.) Tran) :. 8>>1.;onnected 9ivesE The :ro+ect. :%. /?6,8// in Jet&orked Jei"h'ourhoodsE The nline ;ommunity in ;onte(t)edited 'y :urcell) :. %rin"er) Guildford) U.

    $ellman #.) 9ei"hton) #.) /060. Jet&orks) Jei"h'orhoods and ;ommunities. Ur'an Affairs Ruarterly /7) 212,0>.

    $ellman) #) Tindall) D. /002. *o& Tele%hone Jet&orks ;onnect ocial Jet&orks. :ro"ress in ;ommunication cience)/8E12,07.

    $ellman) #) ;arrin"ton) :.) *all. A) /044. Jet&orks as :ersonal ;ommunities. :%. /2>,47 in ocial tructuresE A Jet&orkA%%roach) edited 'y $ellman) #. and #erko&itz) .D. ;am'rid"e University :ress) ;am'rid"e.

    $immer) A.) chiller) J.G.) 8>>8. !ethodolo"ical nationalism and 'eyondE nation,state 'uildin") mi"ration and the socialsciences . Glo'al Jet&orks 8) 7) 2>/227.

    !c9uhan) !. /017. Understandin" !ediaE The H(tension of !an. !cGra&,*ill) Je& York

    $atts) D. 8>>2. i( De"reesE The cience of a ;onnected A"e. Jorton) Je& York.

    8>

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    21/25

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    22/25

    So :aulo 2.? 2.1 64.7

    an Francisco 8.4 7./ 7/.8

    Je& erseyVVV 8.? 8./ 8>.>

    ;hica"o 8.8 /.6 28.>

    $ashin"ton 8./ 8.1 27.2

    !anchester /.0 /./ 2>.4

    Atlanta /.6 8./ 71.8

    an Die"o /.? /./ 81.2

    Toronto /.2 /.? 78.0

    eattle /.2 /.8 ?4.4

    *ouston /.8 /.> 7>.>

    io de aneiro /.8 /./ 2>.4

    Dallas /.8 /.7 1/.?

    #oston /.8 /./ 8>.>

    Amsterdam /./ >.0 ?>.>

    akarta /./ >.2 78.0Austin /.> /.2 ?>.>

    ydney >.0 >.4 24.?

    rlando >.0 >.1 /1.6

    :hoeni( >.4 >.1 //./

    York -U3 >.4 >.? 8?.>

    saka >.4 /.> 8?.>

    V identified 'y the most common %lace nameVV share of local of ties amon" all ties for e"os in this clusterVVV This cluster is centered 'et&een :hiladel%hia and Trenton) J and includes all locations identified as +ust .78>

    All % values are W >.>>?

    8.3 Table 3: Top countries

    share of e"os -M3 share of alters %ercenta"e of %ercenta"e of follo&in" forei"n country named

    88

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    23/25

    -M3 domestic ties domestic tiesamon" non,localties

    alters K 'ein"follo&ed froma'road

    e(%licitly -M ofe"os3

    UA 74.? ?7./ 0/.1 40.2 >.2 4./

    #razil />.1 />.7 42.? 68.? 7.0 ??.7

    U 6.1 6.? ?>.1 22.2 /.8 7?.2

    a%an ?.? 1.8 08./ 41.> /.7 8?.>;anada 2.6 8.0 22.2 82./ /.1 ?4.?

    Australia 8.6 8.> ?>.> 28.> 8.8 10.6

    Indonesia 8.1 /.4 1>.> 8?.> 6.> 42.2

    Germany 8./ /.2 18.0 ?4.4 2.8 ?4.1

    Jetherlands /.7 /.8 11.6 88.8 /.2 ?7.2

    !e(ico /.8 >.6 77.> 4.2 6.> ?1.6

    8.9 Table 9: 5A% Regressions

    !odels

    / 8 2 7 ? 1 6 4 0

    interce%t 1.88 >.827 >./2/ ,>.8>> ,>./0? >.>>> >.842 >.82/ ,>.>7/6

    %ro(imity >.>20>VV->.2803

    >.>868VV->.82>3

    >.>2>0VV->.81>3

    >.>/7?V->./823

    >.>/82V->./>73

    domestic /.76VVV->.2213

    /.>?VVV->.87/3

    /./7VV->.81>3

    >.24?V->.>443

    lan"ua"e /.?7VVV->.8083

    /.>4VVV->.8>23

    >.?16->./>63

    >.26?->.>6/3

    fli"hts >.>>2?VVV

    ->.78>3

    >.>>80VV

    ->.2763

    >.>>82V

    ->.8643

    >./>4 >.//2 >./?6 >.>4? >./7? >.//0 >./66 >./41 >.8>/

    Ad+. >./>4 >.//2 >./?? >.>4? >./77 >.//6 >./66 >./4? >./06

    Jum'er ofo'servations

    1>> 1>> 1>> 1>> 1>> 1>> 1>> 1>> 1>>

    i"nificance levelE V W ?M) VV W /M) VVV W >./Mtandardized coefficients are sho&n in %arentheses.

    8.6 Table 6: 7ost co''on languages

    9an"ua"e M of e"o t&eets

    Hn"lish 68.?

    :ortu"uese />./

    a%anese ?.7

    %anish 2./

    Indonesian /.4

    82

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    24/25

    German /.6

    Dutch /.>

    ;hinese >.0

    orean >.7

    &edish >.7

    ussian >.7

    8.8 Table 8: +anguage co'binations

    9an"ua"e com'inations as a %ercenta"e of...

    all ties in,cluster ties (,cluster ties intCl ties

    ame lan"ua"e -total3 44.7 0/.1 44.7 6?.?

    Hn"lish,Hn"lish 16.? 12.2 67.7 6>./

    ame lan"ua"e) non,Hn"lish 8>.0 84.2 /7.> ?.7

    ;ross,lan"ua"e

    ther,Hn"lish 6.7 8.1 4.? 8>.?

    Hn"lish,ther 2./ 7.? 8./ 2./

    Different lan"ua"es &here neither is Hn"lishV /./ /.8 /.> >.0

    V The most common com'inations &ere a%anese,;hinese) %anish,Italian) and :ortu"uese,%anish.

    87

  • 8/10/2019 Takhteyev Wellman Gruzd 2010

    25/25

    . ;igures

    .1 ;igure 1: