Distance Is Not Dead: Social Interaction and Geographical Distance in the Internet Era

Dista nce Is Not D ead: Social Inter act ion an d Geo grap hical Dista nce in the Inter net Era Jacob Gold enb erg 1 an d Mos he Levy 2 1 The Hebre w Univ ersi ty, Jerus alem 91 90 5, Isra el 2 The H ebrew Un ivers ity, J erusa le m 919 05 , Isr ael ABS TR AC T The Internet revolution has made long-distanc e communication dramatically faster, easier, and c heaper than eve r before. T his, it has been argued, has decreased the importance of geographic pr oximity in social inter actions, transforming our world into a “global village” with a " borderless soc iety". We argue for the opposite: while technology has undoubtedly increase d the ove rall level of communication, this increase ha s been mo st pronounced for local social ties. We show t hat the volume of electronic communications i s inversely proporti onal to geographic di stance, following a Power Law. We directly study the importance of physical proximity in social interactions by a nalyzing the spatial dissemination of new baby names. Counter- intuitively, a nd in line with the above argument, the im portance of geographic proximity has d ramatically increased with the internet revolution. Acknowledgments We are ver y g ra teful to Sh irly Bi tansky , Ass af H ef ler, Y an K eren, T al L evy, A biga il Rako ver , G reg Ravik ovich, Rom Schr ift, Rach el V ov no bo y, and Jacob Sh apiro, f or their ass istanc e and man y h elpfu l comm ents . We are g ra teful to Jonah Berger, Hai m Levy , Barak Libai , Oded Netz er, Eran Rub in, and Oliv ier Toub ia for the ir helpfu l com men ts and su ggest io ns. This res earch h as b een fin anc ially supp or ted by th e Isr ae l Sci enc e Foun da tion, and by the Dav i dso n, K-mar t and Zag agi Fu nd s. Introductio n Information Technology (I T) allows us to do things that we re undreamt of onl y a few decades ago. Today, as we s it in our favorit e coffee shop in Manha ttan, we can chat, view, and exchange files, or play che ss with our friends anywhere acros s the globe. Common w isdom is t h at the IT revolution has reduced the importance of geographical proximity, creating a “borderless society,” and a “global village” (Green, and Ruhleder, 1995;Farazmand, 1999) . I t also makes shortcutting ties between clusters leading to a dr amatic increase of the short cuts in the "s mall worl d” typical social network structure ( see Watts , and Str ogatz, 1998). We argue that t he opposite is the case: in our contemporary IT-intensive world, geographical proxi mity has become an even stronger f orce than ever befor e . Far before I T attracted attention, McLuhan and Fiore (1967) predicted that the development and integration of electronic technol ogies would lead to the e mergence of a global village. I n 1992, befor e the explosion of t he IT r evolution, when the Internet was still m ainly used by universities and research institutes, Cleveland already argued that traditional notions “o wnership” and “control” will have to change as local resources , markets and physical pr oximity will lose their i mportance . Many others have argued that these technologies will make it possible to think of the world as a single global village. Information technology is a key element in achieving these purposes. In f act, these technologies may serve to define a whole new paradigm of organizations: “The borderless economy discussed here is really a whole inform ation society arising out of the spread of new computer/communications technologies” (Dobell, and Steenkamp, 1993). But since the vision of the I T revolution was formed before the tr ansformation actually occurred, could it be that i ts affect on the border less s ociety vie w is a myth, or is it a r eal f act? The r ationale of a borderless society appears to be sound. R esearch on social networks cons istently shows that human societies tend to organize thems elves in a “small world” structure ( Watts, and Strogatz, 1998) a scale fr ee structure (i.e. , consis ting of hubs, see (Barabasi, and Albert, 1999;Albe rt, et al., 2000) that lead to high clustering, and very short, efficient paths of comm unication. I n 1967, Milgram’s well-known experiment tested the hypothesis that members of any large social network (in his case, the population of the United States) are connected to each other through s urprisingly short chains of intermediate ac quaintances. Hi s famous r esult i s that the average l ength of the resulting acquaintance chains was roughl y six, where the final member of the chain wa s the ta rget itself. This result led to the phrase “ six degrees of separation.” Without going into the heated debate on whether Milgram's study supported his hypothesis (it is argued that only several dozen chains w ere eve r completed), the f inding that social structures facilitate f ast communication through short paths is l argely acc epted. In their famous study, Dodds and W atts (2003) actually performed the f irs t large scale, global verification of the small world hypothesis, using the modern Email equivalent of an earlier version of Milgram’s experiment. The detailed analyses which includ ed not only average acqua intanc e chain lengths, but als o the distribution of lengths, show ed that the world is indeed small, a nd as far as I T is concerned, ac cess to inf or mation is e ffectively independent of geographical proximity. However, even Dodds and Wa tt s’ findings were inconclusive: Only 384 out of 24, 163 ch ains were completed: the va st major ity reached a “dead end,” which m ay imply that even if the world has become smaller , borders may still exist . On the o ther hand , the r e se ems t o b e evid enc e th at shows di stan ce ef fec ts on net wor k tie fo rm atio n (Powel l, et al., 2005 ). At the same time in m an y case s the f act that a netwo rk ex ist s doe sn’t pr e clud e, d e spi te th e ea sine ss of co mmu nica tio n, th e im po rt an ce o f distan ce. Fo r exa mpl e, it ha s be en sho wn tha t d istan ce mak e s tech no log ic al col lab ora tio n mor e d ifficu lt (O lson, an d Ol son , 2 00 0;K ie sler, and Cumm ings, 200 2) . In this paper, we study the effect of t he IT revol ution on social interactions. Our social interactions have a l arge par t in defining who we are. Other than satisfying a basic human need, they influence what we thi nk, what we buy, and how we invest (Hong, et al ., 2 005). Has phys ical proximity become a less impor tant f actor influencing our social interactions? The re is no doubt that the IT revolution has intensified communications. Long-distance com munications that were techni cally challenging and prohibiti vely expensive 20 years ago now entail simple, user fr iendly procedures that, in many c ases , are almost cos tless. T his seems to intuitively s uggest that physical pr oximity now plays a more limited r ole in social interaction. However, while the effect of the new communi cation technologies on the long-distance communications is muc h more dramatic than its e ffect on local communications, we should keep in m ind that most of our acquaintances are local. IT does not typically help us for ge new acquaintances; it mostly h elps us communicate with existin g acquaintances . M ost of us c ontinue to establish new social relations in the traditional manner, through social ac tivities and face to face meetings. Perhaps trust, friendship and bonding do not disse minate in the internet as easily as information. T hus, while the IT revolution increased th e volume of all com munications, it is po ssible that it ha s intensified our loca l communications to a greater extent than it intensified ou r global communications, simply because we maintain a greater number of local contacts . If this is correct, physical proxi mity may have become an even more important factor in social dynamics compared to the pre-IT era. This is the main issue we address in this paper. We present two different studies to support the pr oposition that distance is plays an import ant rol e in s ocial inter action even dur ing the IT e ra. In the fi rst study we describe the inten sity of electronic communications as a function of phy sical proximity by examining the volume of email tr affic as a f unction of geographical distance. We also investigate the density of social network contacts as a function of distance, us ing an extensive d ataset of 100,000 Facebook network us ers. We show that both types of electronic social communication decrease inver sely with th e distance. Thus, we use electronic communications much m ore intensively locally, and this suggests that the IT revolution may have ac tually increased the importance of distance. In the second s tudy, we directly investigate the i mportance of physical pr oximity for social dynamics as a function of time by comparing the pre- and post- IT revolution eras . Using the Social Security datas et of baby names in the US, we tr ace the propagation of new baby names over the period extending f rom1970 t o 2005. For the entir e pe riod we show that new names tend to diffuse geographically, s uggesting that physical proxi mity is an important factor in the new name adoption dynamics. However, we show that the influence of physical p roximity on baby name diffusion i s much greater af te r the IT revolution occurred in the 1990s than before. It does indeed seem, perhaps surprisingly, that physical proxim ity has becom e even more important for social dynamics in the IT era than befo re. The final part o f the paper discusses the r esults and their impli catio ns. Study 1: The scal e-f ree dis tan ce distr ibu tion of el ect ronic com munica tions By definition, t here is no data on the dependency of IT us age in social a ctivity with geographical distance before the I T r evolutio n took place. It is possible however to examine whether the geographical proximity indeed plays a prime r ole in social ties and communication. This is the purpose of study 1. In this study we exa mine the r elation between social interaction a nd phys ical distance. We analyze two set s of data: links of members on t he Facebook social network, and email communications. In traditional forms of communications, physical distance typically af fects the time, cost, and ease of communicating. Thus, data about traditional communication does not neces sarily r eflect the underlying distribution of social l inks: even if we have the s ame number o f short-range links and long-range links, because of the higher cost and longer tim e required f or l ong-distance communications we may communicat e less with our l ong-range contacts. In contrast, for electronic communications phy sical di stance is technologically ir relevant: it i s ju st as easy to em ail someone who is 2000 mile s away as it is easy to email our next-door neighbor. T hus, electronic communications provide an "uncontaminated" way t o investigate the underlying distribu tion o f social links. We study two types o f s ocial electronic communication: Facebook li nks and email. Facebook is one of the largest a nd fas test-growing electronic social networks in existence today. I n Facebook, users r egister, cre ate personal prof iles, and manage real-life and Internet friendships with other users. 1 T hird parties are of ten invited t o develop different applications for the Facebook platform. “My Personality” is such an 1 See: http: //ww w.f ac eboo k.com. application, which allows users to share their personal da ta, including their zip code. By examining pair s of friends (i. e. pairs of linked users) in the database, we constructed the di stribution of link distances. The distanc e between any two linked users was calculated by converting each u ser's zip code to geographical longitude and latitude and ca lculating t he distance. We collec ted data of 100,000 Facebook users, and found 1,297 linked pairs with repor ted zipcodes. The dis tribution of Face book links as a function of physica l distance i s shown in Figure 1. Figure 1: The Distribution o f P hysical Distance s of Facebook Contacts . Panel A des cribes the empiri cal density function as a his togram. The empi rical distribution is in very good agreement with a scale-fr ee power-law distribution. The solid line shows the be st power-l aw fit to the e mpirical data, with an exponent of -1.03 and a standard err or of 0. 03. Thus, the e mpirical di stribution is in agreement with Zipf ’s (1949) Law , a ccording to which dens ity is proport ional to 1/r , where r is the distance ( of cour se, for any finite-sized system, a truncated Z ipf distribution must be considered). Panel B shows th e empirical cum ulative di stribution (solid) and th e best fit t o t he cumulative Z ipf distribution (dashed), i.e. t he best-fit l ogarithmic distribution ( recall that the cumulative distribution for a tr uncated Zipf density function is the log function). While the den sity estimate provides a good pi cture for most distances , for large distances the number of observations is insuffi cient to obtain clear results using t his method. T herefore, for large distances we also employ the rank-distance m ethod, with findings presented in Panel C of Figure 1. In this method, we ranked the links from the one with the greates t distance (rank n =1) to t he one with the lowest dis tance. The figur e presents the distance r(n) as a function of rank, on a semi-logarithmic s cale. The linear fit is a gain c on sistent with Zipf’s Law: The linear relationship Bn A )) n ( r log( - = implies that the number of l inks exceeding di stance r is = ) r ( n B ) r log( B A - . If we denote the total number of links by N , t he proportion of the distances greater than r is n(r)/N , and therefore the cumulative probability function F(r) is 1- n(r)/N : BN ) r log( BN A ) r ( F + - = 1 . The probability density function is the derivative of F( r) with respect to r , i. e. BNr ) r ( f 1 = . Thus, a linear r elationship between )) n ( r log( and n implies Zipf’s Law. The correla tion we obtain between )) n ( r log( and n is R = -0. 997. To exa mine whethe r the obtained Zipf Law is a specific property of the Facebook network, below we employ a similar methodology to investigate the distributi on of emails by distance . Email is the most widespread for m of electronic communications, with mor e than a billion users across the globe 2 . As it is extremely difficult to obtain detailed d ata on email communi cations, due to obvious privacy issues , littl e is known about the volume of email communications a s a function of the geographical distance between correspondents. In t his study we collected the se d ata by asking subjects t o report the locations of the recipients of their last 50 e mai l messages, and their own city of residence (the complete questionnaire is provided in the supplementary material section). Overall, we collected data for 4,455 email message s. The distributi on of e-mail dis tances is reported in Figure 2. 2 An Octob er 200 7 repo rt by the tech no logy mark e t research f irm Th e Radica t i Group ( http: //www .rad icat i.com / ) estim ates th at there w ere 1.2 bi llion em ail users in 20 07 , and exp ects th is num ber to r is e to 1.6 b illion by 20 11 . Figure 2: The Distribution o f Email Distance s Panel A r eports the density function. As the geographic l ocation is given only at the city level of detail, the r esolution of the email data is less detailed than the Facebook data. Out of the 4,455 messa ges, 1818 ( 41%) where sent within the same city, yielding a distance measure of ze ro. Obvio usly, this low res olution limits our ability to characterize the density function, especially for short distances . A s a r esult, Panel A of Figure 2 provi des only a r ough d escription of the den sity function, which appears consistent with Z ipf’s Law, but is not co nclusive. Panel B provides a mor e detailed picture by presenting the cumulative distribution. The solid li ne describes the empirical cumulative distribution, while the dashed line shows the best logarithmic fit. The cumulative distribution is i n good agr eement with Z ipf’s Law. T o complete the picture, P anel C describes the rank-distance r elationship, with a correlation of R = -. 989. 3 It is striking that both email and Facebook communications depend on distance i n a very similar way, and that thi s dependence is given by the simple Zi pf Law, encountered in many other branches of s cience (Z ipf, 1949;Mandelbrot, 1963;E gghe, 1991;Gabaix, 1999; Axtell, 2001;Li, and Yang, 2002;Yook, et al., 2002). A similar dependence has been found for internet routers (Lambiotte et. al. 2008) 4 . What does this regulari ty imply about the probability of two individuals, with a distance r between them, being socially linked? Let us begin by making the simplifying ass umption of i dentical "represe ntative" individuals homogeneously spread over t wo- dimensional space. T his implies that t he number of individuals at di stance r from a given individual is pr oportional to 2 π r (the circu mference of a circle wit h radius r around this indi vidual). Denote the probability of a so cial link between two specific 3 N ewm an (2 00 5) su gg ests an estim a te o f the p ow er ex po nent b ased o n log- likel ihoo d maxim iza tion ( se e h is eq .(5 ) on p . 3 27 ). W hi le th is es tima te is g enera lly su perior to the slop e e st imat es typical ly u sed in the lit eratu re, it i s sensi tiv e when the exp o n ent v alue is clo se to 1 ( see eq. B2 in his appen dix). Em ploy ing New man 's metho d to o ur d ata s ets yields expo nen t valu es sl igh tly larg er than 1 : 1.12 and 1.2 0 fo r the facebo ok and em ail da t a, respe ct ively . 4 Liben -No we ll et. a l. (2 00 5) study th e effe ct of d istan ce in th e Liv eJou rn al ne twor k. Th ey too find a power- law d ependen ce, but they find tha t the l ink p rob abili ty de crea ses a s r 1.2 . individuals with a distanc e r between them by p(r ) , and the number of links of distance r that th e individual will hav e by f(r). f (r) is the number of "neighbors" a t distance r multiplied by t he probability of a link given this distance, i.e. f(r)=2 π r ·p( r). As we empirically fi nd f(r) =c/r , this implies that: 2 1 2 r c ) r ( p p = . (1) Eq.( 1), der ived under t he assumption of identical individuals, may be con sidered analogous to the gravitational law: the probability of a social link (the for ce) between two i ndividuals ( bodies of mas s) is propor tional t o 1/r 2 . Relaxing the assumption of identical individuals, by allowing individuals to ha ve different susce ptibilities (or unconditional probabilities) of being linked, m i and m j , this analogy c an be further extended to: 2 r m Gm ) r ( p j i = , (2) where G is a constant given by 2 2 m c p . Several scholars have predicted that electronic communications for which physical distance is completely irrelevant will fundamentally tr ansform our social structure creating a "borderless society" ( McLuhan, and Fi ore, 1967;Ruhleder, and Green, 1993;Farazmand, 1999). Our empirical f indings show that even though physical distance is technologically practi cally i rrelevant f or electronic communications, we use these communication paths pr imarily for short distances, becaus e most of our social contacts are local . This result im plies that the IT revolution may have made distance even more important. Study 2 is designed to directly examine whether this is indeed the case . Study 2: Has Geogra phic Prox imity Be come Mor e or Le ss Important? – E mpirical Evide nce from t he Dis semination of Ne w Baby Names Our goa l is to examine whether physical proximity has become m ore or less important for social interactions after the I T revolution in the 1990s. Such an investigation requir es both spatial and t emporal d ata about a social ph enomenon that reflects social i nteractions, for a period spanning several decades , to cover the periods before and after the IT revolution. One such rare dataset is the d etailed records maintained by the U.S. Social Sec urity Service on new baby names. When parents choose names for their new-born babies they are typically affected, at least to some degree, by their social interactions. This positive-feedback interaction can explain both the stellar climb up the popularity chart of several new “invented” baby names, 5 and als o the fact that new baby names tend to s pread spatially. Baby names offer a stringent test of our hypo thesis because n ames are as equ ally communicable by phone or e-mail as by personal contact. The Data The S ocial Security Service publishes the list o f the 1,000 most popular baby names in the US every y ear, and the number of b abies given each of these names. In addition, the Social Security Servi ce publishes the list of the 100 most popular baby 5 A r ecent fro nt-p ag e sto ry in th e New Y ork Tim es d escr ibed the incred ibly f ast climb in th e po pu larity of the bab y g ir l nam e “N eva eh ” (H eaven sp el led b ackw ards). I n 199 9 on ly 8 new bo rn girls in th e U S were g iv en th is n am e. By 20 05 , N evaeh b ec ame the 70 th m ost p op ular n am e for n ewbo rn gir ls (ah ead of Sara, V an essa , and Aman da). In 2006 , N eva eh contin ued its stellar clim b and rank ed 43 on the pop ularity char t ( be coming mo re po pu lar than A llison and Maria) . S ee “An d if I t’s a Boy , W ill it b e Lleh?” by Jenn ifer Le e, Ne w Yor k T im es , May 18 , 2 006 . In 2 0 08 , Neva eh rank ed 34 . names in each state in each year, including the number of babies given each name in that particular state. T he data are available at: http://ww w.ssa.gov/OACT/babynames. To analyze this data set, we ran a crawler softwar e application t hat allowed u s to track the development of e ach name through time and across sta tes. T his software, which appears in the supplementary materials section, is available for public use. Results Our results are compr ised of t wo parts. In the first part we show graphically that physical pr oximity plays an impor tant r ole in the dissemination of new bab y names. In the second part, we develop a formal measure of the importance of physical proximit y t o the dynamic s, and track the behavior of this mea sure over th e last 35 years. A. Physical Proximity is Important The impor tance of physical pr oximity i n the dis semination of new baby name s can be il lustrated graphically by the dynamics of the name “Ashley” described in Figure 3. Figure 3: The Dynamics o f t he Name “Ashley”. The first panel in the fi gure is a snapshot of the y ear 1970. In t his year, the nam e “Ashley” appeared in t he T op 100 name list in onl y four states: Alab ama, Arkansas, Louisiana, and North Carolina. The sec ond panel of Figur e 1 show s a snapshot of ”Ashley” for 1974 (for simplicity of the exposition, the percentage of Ashleys in each state out of the total number of new babies in the state is expressed by shade coding). The fact that "Ashley" spread to neighbori ng states between 1970 and 1974 suggest s that physical proximity played an im portant r ole in the process of name adoption. Th e next two panels, representing 1978 and 1982, furt he r support this notion. Has phy sical proximity continued to pl ay a n impor tant role aft er the I T communications revolution of the 1990s? While we address this que stion methodically in t he next section, one example for this case is provided in Figure 4. Figure 4: The Dynamics o f t he Name “Kaden” This figure describes t he dynamics of the (boy) name “Kaden” betw een 1995 an d 2003. The dynamics again s uggest that phys ical proximity plays an important role. There is a c lear similarity to the dissemination of an epidemic, with strong spatial correlations. These ar e just two examples. The pr ogram in the supplementary materials section is available to examine th e spatial development of any other name. The analysis in th e next section employs the data of all names. B. How Has the Importa nce of Physical Proximity Changed Over Time? In order to address thi s question, we develop an i ndex to mea sure the size of the physica l proximity eff ect, and we t hen track the value of this index over t ime. Several dif ferent m easures have bee n employed in previous resea rch to capture th e importance of proximity or ef fects i n the diffusion of social phenomena (Mandelbrot, 1963). The m easure w e employ is i n the same spi rit, and i s adopted to the context of baby-names. I f physical proximit y pl ays no role i n t he dynamics, all else equal, we would expect the pr oportion of new babies with name i to be distributed uniformly across the country. I n contras t, if physical proximity does play a role, we expect the proportion of new babies with name i to be higher in the vicinity of the states where this name ha s already appeared. We separate st ates int o two groups: Group A i (t): states i n which the name i appeared on the Top 100 list before time t, plus th e immediately bordering states; Group B i (t) is th e group of all the r emaining states. W e employ the f ollowing notation: ) t ( N i A : T he number of babies with name i born in year t in Group A states. ) t ( N i B : T he number of babies with name i born in year t in Group B states . ) t ( N total A : T he total number of babies (all names) born in yea r t in the group A states. ) t ( N total B : T he total number of babies born in year t in the group B states. If physica l pr oximity plays no role in the diffusion of name i in the U S, we expect the number of babies with name i born in Group A states to be ) t ( N ) t ( N ) t ( N ) t ( N i total B total A total A ˜ ˜ ¯ ˆ Á Á Ë Ê + , (3) The term in equation (3) r eflects the expected pro portion of babies named i in Group A states, in the absence of proximity effects , i.e., the proportion of i babies in Group A and Group B is equal. In contra st, if physical proximity is i mportant, we expect ) t ( N i A , the number of babies nam ed i born in Group A states, t o be gr eater than t he express ion in (3) . We define the Proximit y-Effect Index (PE I) of name i at ti me t as follows: 1 ) t ( N ) t ( N ) t ( N ) t ( N ) t ( N ) t ( PEI i total B total A total A i A i - ˙ ˙ ˙ ˙ ˙ ˚ ˘ Í Í Í Í Í Î È ˜ ˜ ¯ ˆ Á Á Ë Ê + ≡ . (4) PEI measures the percentage by which the actual number of babies named i i n Group A states exceeds the expected number of babies named i a ssuming n o proximity eff ects. If, f or example, 2 . 0 ) t ( PEI i = , this means that the number of babies named i born in Group A s tates exceed ed the expe cted number of babie s named i in the absence of pr oximity effec ts by 20%. As we ar e interested in examining the overall tr en d in the proximity effect, we calculate the PEI for all t he names. T he development of the m edian PEI over t ime is depicted in Figure 5. 6 T he fact that the median PEI is positive implies that physical proximity has played a r ole in the baby name dynamics over the entire 35- year period (as figures 3 and 4 suggest). In fact, over 99% of the PEI meas ures for individual names are positive. T he typical value of approxim ately 0.2 f or the median PEI in the period up to 1995 i s rather large: t his value implies that the proportion of babies with a particular name in states wh ere this name a ppeared in the past and in t heir neighboring states is 20% higher t han would be expected in the absence of proximity effects. Figure 5: The Importance of Physical Proximity ove r T ime- Baby Names. The most striking result revealed in Figure 5 is the fact that the median PEI increased dramatically over the decade of t he IT revolution. While the importance of physical proximity was rather stable between 19 70 and 1990, the communications 6 We u se the m edian r ather than th e aver ag e PEI , b ecause th ere ar e so me large po s itive o ut liers in the PEI, dr iven b y sma ll s ta tes. Fo r examp le, assum e tha t a new name firs t app ears in a sma ll stat e w ith a small nu mb er of neigh bo rs, o r st ates o ut side the con t iguo us U nited St ates such as Haw a ii, o r A lask a. In this case, Gr ou p A co mpris es on ly tha t p art icular st ate, and ) t ( N ) t ( N i A i = . Be caus e the s tate is small , ˜ ˜ ¯ ˆ Á Á Ë Ê + ) t ( N ) t ( N ) t ( N total B total A total A is v ery sm a ll, and ther efor e P EI i can be ex tre me ly lar ge . Wh ile t he average PEI can b e dram a tica lly affe ct ed by such sp ecia l cases, the med ian P EI is not s imi larly affected , and therefo re is a mu ch mo re robu st measur e. Simi lar r esu l ts ar e obt ained wh en differ en t percen tiles of the P EI are emp loyed. revolution in the 1990s was ass ociated with an increase in the import ance of physical proximity. Between 1995 and 2002, t he median PEI jumped to about 0.3, a 50% increas e relative to the pr e-revolution period. This surprising increase is in accord with the counter-intuitive argument made in this paper: While the communica tion s revolution has no doubt made it easier to tr ansmit inform ation over great distances, its effect on local contacts may have been ev en m ore dr amatic, enhancing proximity effects mor e than befor e. Although it declined somewhat after 2002, the median PEI level remained higher than typical pre-1990 levels. The exact point in time where the I T r evolution took place i s hard to define, and may be s omewhat subje ctive. Howe ver, it is gen erally acce pted that this revolution took place somew here in the 1990’s. If w e take the middle of this dec ade as a reference point, we fi nd that before 1995 the average value of the P EI shown is 0.203. In the post-1995 period the average value i s 0.267. T his difference is highly significant, with a t- value of 6.76. T hus, the baby- name data sugg est that the IT revolution was accompanied by a significant increas e in t he importance of geographical proximity for social interaction. Discuss ion The Inform ation Technology revolution has dr amatically changed our abilit y to communicate with each another. T he new electronic communications opti ons hav e rendered geographic proximit y completely irrele vant by allowing us to transmit information across the globe instantaneously, with great ease. I n the past, people believed that this te chnological revolution wo uld transform society and social relations, leading to a borderless society in a "global village." T hese beliefs, however, were never empiri cally put to a test. While the IT revolution has changed many impo rtant aspec ts of our lives, it seems that it has not fundamentally changed the structure of society and the organization of social relations. E mpirical evidence suggests that contrary to expectations, geographical proximity has bec ome more important for social interactions and dynamics than ever. We argued that one poss ible reason i s t hat the major par t of our electronic communications ar e perf ormed with local counterparts. We demonstrated that while t he IT revolution has cle arly increased the overall volume of communications, it has increased local comm unications to a greater degree than long-distance communi cations: t he volume of electronic communications as a function of the geographical distance follows Z ip’s inver se proportionalit y law for both E mail and Facebook inte ractions. The IT r evolution may have transformed the world into a global village, but it has not changed us into a bor derless soc iety in which geographical proximity plays a more limited role in social relations. On the contrary, the importance of geographical proximity appears to have i ncreased. Refer ences Albert, R., et al. 2000 "Err or and attack tolerance of complex networks. (Cover story)." Nature, 406: 378-382 % U http://search.ebscohos t.com/login. aspx?direct=true&db=aph&AN= 3382459&site=eho st-live . Axtell, R. L. 2001 "Zipf Distribution of U .S. Firm Si zes." Scien ce, 293: 1818-1820 %U http://www.sciencemag.org/cgi/content/abstract/1293/5536/1818 . Barabas i, A. L., and R. Albert 1999 "Emergence of Scaling in Random Networks ." Science, 286: 509-512. Dobell, R., and P. Steenkamp 1993 "Preface to the Sym posium on Public Management in a Borderles s Economy: National Governments in a Wo rld of Tr ans-nation al Networks." International Review of Administrative Sciences, 59: 569-577 %U http://ras.sage pub.com . Dodds, P. S., et al. 2003 "An Experi mental St udy of Search in Global Social Networks." Science, 301: 827-829 % U http: //www.sciencemag.org/cgi/conte nt/abstract/301/5634/5827 . Egghe, L. 1991 "The exact place of Z ipf's and Pareto's law amongst the class ical infor metric laws." Scientometrics, 20: 93-106. Farazmand, A. 1999 "Globalization and Public Administration." Public Administration Review, 59: 509-522. Gabaix, X. 1999 "Zipf's L aw f or Cities: An Explanation." The Quarterly Journal of Economics, 114: 739-767 %U http://www.jstor .org/stable/2586883 . Green, C., and K. Ruhleder 1995 "Globalization, bor de rless worlds, and t he T owe r of Babel." Journal of Organizational Change Manage ment, 8: 55- 68. Hong, H., et al. 2005 "Thy Neighbor 's Portfol io: Wor d-of-Mouth Effects in the Holdings and Trades of Money Managers." The Journal of Finance, 60: 2801-2824 %U http://www.jstor. org/s table/3694804 . Kiesler, S., and J. N. Cummings 2002 "What do we know about proximity and distance in work groups? A legacy of research." Distributed work, 1: 57-80. R. L ambiotte, Blondel V., de Kerchove C., Huens E. , Prieur C., Smoreda Z., and P. Van Dooren 2008. "Geographical Dispersal of Mobile Communication Networks". Physica A . 387: 5317-5325. Li, W., and Y . Yang 2002 "Zipf's L aw in Impor tance of Genes for Cancer Classification Using Microarray Data." Journal of Theoretical Biology, 219: 539-551 %U http://www.sciencedirect.com/scienc e/article/B536WMD-474G567S- 538/532/538a581ec55874a496536b496590b486228b496538b . Liben-Nowell, D., Novak J., Kumar R., Raghavan P., and A. Tomkins 2005 "Geographic Routing in Social Networks". PN AS . 102: 11623-11628. Mandelbrot, B. B . 1963 "New Methods in Statistical Economics." The Journal of Political Economy, 71: 421-440 % U http: //www.jstor.org/stable/1829014 . McLuhan, M., and Q. Fiore 1967 The Medium Is the Mas sage: John Wiley, New York. Newman, M.E.J., 2005 "Power L aws, Pareto Distributions and Zipf's Law." Contemporary Physics, 46: 323-351. Olson, G. M ., and J. S. Olson 2000 "Distance Ma tters." Human-Computer Interaction, 15: 139-178. Powell, W. W., et al. 2005 "Network Dynamics and Field Evolution: The Growth of Interorganizational Collaboration in the L ife Sciences ." American Jou rnal of Sociology, 110: 1132-1205. Ruhleder, K., and C . Green 1993 "Globalization and the Tower of Babel." 460-468. Watts, D. J., and S. H. St rogatz 1998 "Collective dynamics of `small-world' networks." Nature, 393: 440-442 %U http://search.ebscohos t.com/login. aspx?direct=true&db=aph&AN= 698663&site=ehos t-live . Yook, S.-H., et al. 2002 "Modeling the Internet's large-scale topology." Proceedings of the National Academy of Sciences of the United States of America, 99: 13382-13386 %U http://www.pnas.org/content/13399/13321/13382.abs tract . Zipf, G. K. 1949 Human behavior and the p rinciple of least effort; an introduction to human ecology. Cambridge, Mass .: Addison-Wesle y Pr ess.