Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Information ant! Innovation in a Networked WorIdi Da`:id Lazer John F. Kennedy School of Government The potential for the diffusion of information regarding successful goYe~mental (international. national or subnational) innovation has increased enormously in recent years. Information from geographically distant locales is often simply a click affray, where networks of intergovernmental information exchange are spontaneously ernergin ,. The intertwining of information technology and globalization - extends the pool of accessible innovations and lowers the barriers for their diffusion. (Ben~stein & Cashore, 2000; Colen~an & Grant, 1998: Coleman & PerT 1999; Evans & Davies, 1999) This diffusion process has enormous potential for increasing public welfare, by allowing Iocation B to adopt the succe.sstu] innovation in location A. There is, however, ~ potential dark side to the increased diffusion of information. First, as information diffuses more efficiently. it becomes more of a public good. As the publicness of information increases, so does the likelihood of free riding. There is an incentive for each government to allow another government to take the risks of innovation. and then to simply adopt the successful inbox ations. Second. in complex policy areas. the diffusion process may be too efficient: resulting in either pren~ature convergence on a non-optin~a] policy. or eliminating policy alternatives that while not optimal in the present, might be in the future. The governance implication is that in the networked world special attention must be given to increase governments. incentives to experiment and innovate. (Moon & Bretschneider, 1997) This paper wild be organized as follows. First, it will briefly discuss some distinctive features of the diffusion process in the public sector. Second, it will analyze the informational efficiency" of different types of networks. Third, it will examine the potential for informational free ndina in the networked world. Fourth, it will study the paradoxical possibility that the more efficient the system is at spreading, information. the less information the system mi ,ht contain. Finally, it will discuss the implication for ~o`'er~ance: how does one design a system that its efficient at "spreading the word" while encouraging experimentation'? Inter-organizational diffusion of innovation Networlced governance is in vogue (e.g.. O Toole 1997. Rhodes 19971. By '-networked governance ' ~ mean a system of interdependent sovereign units. Thus, one night think of the relationships argons nations as net~c~rl~ec] governance (although typically with the threat of ~ iolence removedKeohane and Nye 9000; Slaughter 90001. One might also think of the relationship among local ~os;ernments as .-net~orked'-, and ~ This material is based upon work supported by the National Science Foundation under Grant No. 0131923. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authorts) and cio not necessarily reflect the views of the National Science Foundation. DYNAMIC SOCIAL NETWORK MODELING AND AWALYSIS 101
even agencies within the federal government as effectively 'networked.' in that that hierarchy within the federal government intrudes little on the basic independence of federal agencies~specially where it comes to issues around coordination and cooperation ~ ith other agencies. In the US due to both shared powers within the federal government. and a system of dual sovereignty between state and fecleral goven~ment the networked nature of government has been an accepted feature of governance silence- the founding of the republic (although not with that `;ocabula~y). As we move into the 918 century there is an increased awareness that the networked nature of governance is universal- in systems that heretofore might have been considered models of hierarchy (e.g.. the British Rhocles 1997) or anarchy (i.e., the international system. Elsewhere ~ have argued that three strands of interdependence are coordinative. cooperative, and informational (Lazer 2001; Hazer and Mayer-Schoenberger 2002~. For each of these interdependencies there are large literatures which can and should be snapped into the ideas around networked government. and descriptive and normative theories of "networked governments' developed Ill this nnner r chill f]P.~rt°.iOt~ 1.h informational dimension. ~ _ . . ,~ _! . . .. ~ _ . _, ~,~ it ~ _ The presence of the informational aspect of networked government is that policy generates potential informational externalities. When a policy actor adopts ~ policy. that adoption and subsequent experience conveys information to other policy actors. Some of those choices may be a matter of public record (statutes and regulations) and in principle accessible to aIl. and other important information wild remain private. Policy actors thus sin~uitaneously suffer from information o`;erioad and information deprivation. Actors need to aclopt both network strategies selective attention to help sift the public information and access private information ant! internal filtenng strategies to eliminate the large majority of information that is publicly available. One may therefore usefully construe the universe of policy actors as a set of nodes among which there is a set of evolving connections. Over these connections flows information and attention about adoption success ant] failure and just raw ciata. It is the assertion of this paper that this architecture negaters, that some architectures are better at facilitating in:forn~ation transfer than others' and that it is necessary to understand how this structure emerges. The next section of the Daner discusses the general Ten he ~ r - r ~ ~~ ~~~ 2~ r~~~~~~~~ ~~~ _ _ ~ _ I_ _ _ ~ ~ . . . ~ . . , I. . .~ . .. .... . wnlcn networks emerge, and what the likely consequences for information diffusion in the en~er~ent struck; ret There exists. of course. a substantial body of literature on the diffusion of policy innovations - and adjacent topics such as policy networks, policy transfer, and policy convergence. berg.. Abrahamson & Rosen:kopf, 1997; Bennett, 1991; Berry. 1994; Berry & Berry 1992; Coleman' i994; Coleman & Grant. 1998; Dolow-itz, 2000; Dolowitz & Marsh, 1996 & 9000; Evans & Davies, 1999; Gray, 1973 & 1994: Hubner. 1996; Ko=ut & Zander, ~ 995; Nlintrom ~ 997a & ~ 997b; Mintrom & Vergari. ~ 998; Robertson. So an, & Newell, 1996; Salvages 1985; Schenk. Dahlia & Sonnet 1997; Seeii3er' 1996; Stone, 2000; Valente, ~ 995 8: ~ 996; Walker, ~ 969, Weeni.3, ~ 999) Similarly. there is a large literature on the diffusion of innovations through iIlter- organizational networks within and between corporations. (Atkinson & Bierling, 1998: Coleman & Grant, ~ 998; Dolowitz & Marsh. 9000; Dyer & Nobeoka~ 2000; Evans & Davies, 1999; Gupta & Go~indarajan, 199:1. Donut & Zander. 1999 & 1995; Liebeskincl 102 DYNAMIC SOCL4L NETWO=MODEL~G ED ISIS
et al. ~ 996; Nooteboom. ~ 999; Radaelli. 2000; Robertson, Swan. & Newell, 1996; Roan. Peterson, & Scheve. 1998; Seeliger, :~996; WeaTe et al.. 1996) This literature su='C'ests that a tremendous amount of information flows through inter-or~anizationa] networks (typically measured through overlap of corporate boards). This Voluminous research examines the process of diffusion, how innovation evolves as it diffuses, the characteristics of early versus late adopters. etc. (Rogers 19951. The objective of this paper is to consider what are the generic processes by which the architecture of cliffusion of emerges- the network; what the norn~atiYe implications of different architectures; and what is distinctive about diffusion among public organizations. information diffusion through intergovernmental networks is quite different on certain dimensions from diffusion in the private sector. First. in the private sectors many innovations are proprietary, thus increasing both the cost of adopting an innovation. as well as the likelihood of the innovation in the first place, since the innovator may extract most of the benefits of that innovation.~ The profit motive also means that the innovator has an incentive to spread information about the innovation. Second where innox ations are not proprietary a corporation has an incentive to keep information secret from competitors as long as possible. The public sector in contrast has relatively little incentive to suppress information about successful innovations Third. zenith survival less of an issue and relative performance more dit-f~cult to measure, bureaucratic inertia is likely a greater barrier to adopting successful innovations in the public sector than in the private. Fourth, many policy makers are likely "proselytizers" anon ed to innovate ant] to spread the word in order to increase their impact on society. There is therefore substantial potential for diffusion of successful policy innovations both intra and internationally. The question asked here is what is the impact of the shift front local to global informational networks. The next section examines the }:ole that the structure of the informational network plays in the speed with which information spreads in a system. The architecture of the network First a few definitions: A troche is a unit which may contain and pass on information. It may be an individual or an organization. I:n this paper, ~ will largely focus on public sector actors. but at the end wit! speculate what the implications are for public- p~ivate partnerships. A connection between two nodes Cleans that there is some passing of information between those nodes. At its broadest definition. it may mean that some actors are just selectively parting attention to other actors (e.g., everyone is paying attention to Caiifomia's experience in electncity deregulation). At its narrowest (and more typical) definition' it means that there is a private exchange of information among a subset far least two' actors. An i~formationc`l ''eta` ork is a set of connections among nodes. It is usetu! to distinguish among three kinds of informational networks: spatial, organizational, anal emergent. . ., ~ Although spillovers occur not just in the public sector but in the private as well. despite the protection of intellectual property law. Baumol (1999) for example. estimates that innovators retain only approxin~ate~y lOC%c of the gains from their innovations. DYNAMIC SOCIAL NETWORK A1ODEL~G ED TRYSTS 103
A .sl~atial ''eta ork is a network whose dyadic connections are determined by proximity: each actor speaks exclusively to other actors in its neighborhood. For example, in figure I~ A con~municates only with its four immediate neighbors to the north, south. east. and west. == == = ~ ~ _ _ Figure ]: Exan~ple off a Sparta/ net~-vo'~k The probability of communication between any two actors is strongly related to how close they are to each other. The relationship between distance and communication, of course, is ~ astly more complicated than characterized by the lattice in figure I. As noted above, the Great Plains excepted, geography is typically not as smooth as; characterized in figure l. These irregularities affect the costs of communication bet`~;een any two nodes. Distance is also partially a social construct. The probability that two local jurisdictions communicate is probably affected by whether they are in the same state. tor example. Finally, communication frequency is not a linear function of distance. As a general matter communication drops off precipitously with distance (McPherson 2001). An organi`.ational 'lets ork is simply the communications that result from the groupings within the organization (Mintzberg 19921. That is. the formal organizational chart is typically related to the architecture of the informational network. If faculty, for example. are grouped into departments, communication will typically be higher within those departn~ents. in part because of a functional interdependence, in part because the i-nstitc; tion then structures serendipity~epartn~ents will often be grouped together departmental ~neetin~s guarantee that paths will cross. etc. Both of these network archetypes are flawed at spreading Donation. Spatial networks are often broken by spatial 'chasms'mountains. rivers, or climate in a geographic context, buildings in an organizational context, railroad tracks and highways within communities between which little information flows. Further, even in the absence of these discontinuities a purely grid-type of network. such as in figure I, should only slowly (if inexorably) spread information. If one assumed that it takes one period to spread infolTnation to a novels immediate four neighbors. it would take eight periods for a piece of information to spread from one corner of the system to the other-. Similarly, organizational networks are often characterized by dysfunctional chasms~omn~unication within stovepipes but not between. In fact, the "networked organizations cross-func~ional teams, the matrix form. etc. etch is often seen as an antidote to the stovepipes of the organizational chart. However, as cliscussed below. organic (emergent) nets orI; structure have their own disjunctions. Serendipity is the underlying principle of el'!erge7?t networks. Emergent networks result from the myriad] of decisions by illdi`'iclual nodes to part attention or not pay attention. by pairs of nodes to form a relationship, and by larger numbers of nodes to 104 DYNAMIC SOCIAL NETWO=MODELING AND ANALYSIS
create formal. or informal groupings that then form. the basis for larger scale co~nmunication. The assumption ~ make here is that these decisions are Blade on an egoistic basis. made in a boundedly rational fashion. It is this "think locally. act locaI.ly assumption that can result in outcomes that at the systemic level. are suboptin~al.. There are a number of fairly robust patterns that have been observed in a wide range of social networks: the emergence of cliques. power-a s of connectedness. homophily. powerful core-pe~iphery tendencies. each of which is discussed in the context of governance networks. Cliques: Networks often break down into cliques. Where there is a much higher density of communication within cliques than between. Thus, for example, a tie between A and B ant! a tie between B ant] C predicts a tie between A and C (Davis 19671. There are a variety of reasons why cliques might emerge. For example, B's tie to A and C .mi~ht facilitate a tie between A and C. Cliques might also emerge out of a functional need to collectively produce something, that all benefit from, and for which a certain scale is required (e.g.. a pick tip softball. ~a.me). Inform.ationa.~.ly. it ~n.i.ght also be more efficient to share information within a group than dyaclically (reducing repetition and redundancy). Cliques might also be epipheno:rnenal: the result of ho~nophily or proximity. As discussed below, similarity and proximity predict communication. If A is sim.ila~JcI.ose to B ant! C. then it is likely that B and ~ are similar and close to each other. Scalefree net~vo'-ks: Notably, in networks where nodes hare no constraints on communication the frequency at the node [eve.' of any given level. of cor~nectedness of a node is proportional to the inverse of that lease! of connectedness. raised to some power (i.e.. Power I.a`~" di.stributeci, also known as scale free networks see other contributions to this ~ol.u~ne). In essence the snore connected. the less frequent. In a power Jaw world the well connected are vastly more connected than the amperage connected nodes, ant! thus pI.ay a hastily disproportionate role in the flow of information in the system. Such a power law ctistributi.on has been observed with.r.espect to websites. citation :.frequencies; (Price ~ 976), and, surprisingly, number of sexual partners. One suspects. in the policy world, that particular exemplar policies emerge with the bulk of attention; thus. policymakers fool; disproportionately at Calif.omia's experiences in deregulation Wisconsin's experiences with welfare reform etc. Power law frequency distributions tend to emerge from stochastic growth processes where the growth of any particular observation. is proportionate to its size (e.~. the growth rate of sn~all units is about the same as large units~.g. Simon 1956~. These hubs. in a power law world, play a disp.ropo~ionate role. One could imagine (as discussed below;) that they help sy~sten~s overcome problems in diffusing and processing information that.would likely result Tom the other types of processes enumerated here (and probably do, to a certain extentsee small world section below). One could in~a~i.ne that hubs are the nodes with the highest processing capacity anal serve as instruments to aggregate and re-disseminate information (and they probably do. to a certain extent see information aggregation section below). However. the conditions under which then tvpicaliv arise limits their potential as conduits of information since for snort of the cases enumerated above, the well connected nodes only send information, and do not receive. That is, one Night imagine a power law world where the exceptionally well connects recei`;e`1 as much as they sent; howex er, in the policy DYNAMIC SOCKS HETWO~MODEL~G ED CYSTS 105
wodd, it seems unlikely, for example, that California pays as much attention (and then retransmits) the expenences of others as much as others nav attention to Cnlifr>~,in , ~ ~ , ~ ~~ ~ ~_^ ~^ ~ ^^ ~~ _~^ ~4 ·A$~- Homopl~ily: It does turn out that birds of a feather do tend to flock together. Similanty turns out to be a strong predictor of communication across a wide vanety settings (~icPhe:£son 2001~. There are a number of likely explanations for this. First. similar actors will be more likely to have useful information tor each other. Imagine moving to a new city: would it matte any sense to talk to someone who had a n~uch larger income than you, and thus could afford a much more expensive house`? Similarly. one mi ,ht expect that polic-Ymakers would do best to pay attention to those in similar circumstance. Second. especially in political contexts. there are strategic reasons to share information more with those with similar preferences. Information assists actors in achieving their goals. If the goals of another individual are opposed to your Coarse you would be unlikely to share information with them. C07 es anc! peripheries: Emer~,ent networks often will have a "nch get lichen-" dynamic. Assume that nodes have unequal access to private information. Those nodes with noose private information will be more cies~rable as partners zenith which to exchange information. If all nocles have constraints on how much they communicate (this does not apply to attention networks, for example) then the node with the most information will be in the greatest demand to form ties with. It s~;ouId presumably choose ties with those who have the most pop ate information. Those ~ ith Iess information would be in a less of a position to be so picky. Out of a process where the most informed choose the most informed, the moderately informer] will be left With each other to choose to communicate with, and the least informed to choose each other. That is. the network Bill have a weIt- informed core, that distributes information internally, and a less-informed periphery, with occasional leakage of information frown core to periphery. Thus. while the diffusion of information has the potential to reduce informational inequalities in a system, it will potentially just replicate those inequalities. This tendency will be exacerbated by the fact that 1locles that are the most informed will often hare the capacity to form more connections. If connectedness were roughly proportional to the informedness of nodes, it is conceivable that the spread of information would (while raisins, the absolute inforn~ed~ess of ~.Y'~.lVQll~\ ~~U _ _ . ~ , ~ , . ~ ~ . . ~ J ~ J J exacerbate one ~nrormat~ona£ equates In the system. One would expect that out of suchaseenanowoulde~nergeahighI'info~'edandinterconnectedCorf: acid a hinblv uninformed and unconnected penphery. Sn~aI! worlds ~ , ~ 2 ~ Emergent networics thus tend to result in clusters of nodes which are highly similar to each other, within which there are many connections. and between which there are Rely few connections. Clearly, this is not an effective architecture for spreading information. Might networks auto~natically~ acijust the~nseIves so as to reduce the worst effects of this inefficient configuration? For example Burt ( 19951 en~-me.~-~ec the. ~ _ ~ . t , . ~ compe~l~lve am anode acloIs can acnle`;e by bridging these ';structura1 holes'. in the network. One might imagine the larger the advantage, the greater the likelihood that actors will seek to close these holes. There are a number of obstacles to this. however. especially in the public sector. 106 ~ he combination of cheap intra-cTique communication . DYNAMIC SOCIAL N~TWO~MODELING kD ANALYSIS
and lack of property rights may discourage such con~munication. By their fiery nature cliques allow for accidental (and inexpensive) collisions. The lack of property rights means that a node that seeks extra-~lique communication that brings novel information may pay a high price for novel information vY here the benefits quickly diffuse to the entire clique 3 Those nodes that do not attempt to bricige the structural holes will actually fair better, under these circumstances than those that do. An additional deterrent to inter- cTique communication is that inter-cTique information may be less reliable. because the co~n;rnunicators have fewer reputational concerns. If they pass on inaccurate information, there are Sterno consequences, since the two nodes are not embedded in the same social structure (Granovetter :~985 Uzzi ~ 9961. The above analysis suggests that it is likely that organizational, spatial. and emergent informational networks in the public sector world will tend to be inefficient at spreading inivrn~ation. However. what may be true of each of these networks play not be true of the networks together. as the "small world' findings of Watts and Strogatz (1998) illuminate. What these findings demonstrate that is that white a highly structured network (e.~., the lattice) is not effective at spreading information, and a purely random network (e.g., where the probability of a tie between A and B is uncowelated with the probability of a tie between any other dyed in the system). an Slav of strllc~llre once random networks is vers effective at spreading information. This Is indicative of a snore general phenomenon: cross-cutting types of ''eta orbs are .~,p~cc'ti~ Encore efifecti~ e cc! sp'-eatI~ng i'?formatiorz than equally, close networks c~fa single type. A simple illustration will demonstrate why. Consider the network represented by figure 2- where each actor communicates with its immediate tour neighbors. = == - - ~ l = ======== l Figure 2: I ] ~11 world Assume now, that the actor in the middle of the chess board has a successful innovations which is then adopted by its four neighbors. which is in turn adopted by each of their ne~:,noors~ and so on. ~r wall rake l U rounds or communication before the whole system has adopted the innovation. By comparison. a 'random-collision ~ network, where actors 1_ ~ ~ 1 at, 'to . ~ ~ ~ ~ r 3 Obviously. this is ~ critical assumption that will differ in different systems. DYNAMIC SOCIAL NETWORK MODE~ING~D ANALYSIS 107
randomly "collide ~ with four other actors each round Will take just ~ rounds before the `~hole (99~;j system has adopted the innovation. A spatial network is inefficient at spreading ~ntorrnation simply because the informed are spending most of their time communicating with other informed actors. Only at the periphery of the infonned set of actors is information actually spreading,. An overlay of an emergent network on a spatial network is potentially far more ettect~ve at spreaclin~ information than just a spatial network. for the simple reason that the emergent network will provide bridges between the regions (or, alternatively. the spatial network will provide budges between the cliques), thus increasing the proportion of the uninfor~nec] in communication with the inforrned.S A minor elaboration of the above example will illustrate why. Imagine. now that while all actors still communicate with all other actors, the actor in the middle of the chessboard communicates also with an actor in the distant corners What impact wit! adding just this one tie to the 949 that already exist haste on the speed with which the innovation. spreads? It will the 40C%o less time for irlHovatlon to spread{. ,~Iternat;.Yely, what is the impact of simply accentuating the existing spatial network by cloubling the neighborhood with which an actor communicates? Increasing the number of neighborhood ties by 242 is only sli.ghtI.y more effective than adding one non- neighborhooc) tie-- the tinge it would take for an innovation to spread drops by 50/c. Figure 3 presents the rate of diffusion for four diffusion models: random- colI.ision. spatial with 4 neighbors, spatial with ~ neighbors, and spatia.! zenith 4 neighbors ~ one non-spat~al tie. . ~ . ^^ . .. Note that this is ~ very different notion of a 'random netu orgy than Watts and Strogatz (1998) use. ~ Note that this general observation works with any cross-cuttin.~ networks :.g., two emergent networks, two organizational networics, etc. The key is that each network have a I.ocric that is orthogonal to the other network. 6~D this scenario ~ am. assuming that the chess board ' wraps wound' ie.. the actors on. the top con~rnunicate with the parallel actor on the bottom; actors on the left communicate with the parallel. actor on the right. 108 DYNAMIC SOCIAL NETWO~MODEL~G ED ISIS
1 it, o.9 o : <; 0.8 ° O.7 A, 0.6 Q 0.5 - o =5 O.4 -- .° 0.3-- O 0.2 - Q 0.1 O f~ 0 2 4 6 Figu' e 3: Four monies of ~iff~`s~o'~ ,~?~ · /',/~ + spatial-- 4 cliff usion random spatial+1 -¢ spatial-- 8 RolJnd 8 10 12 The above analysis highli~,hts how even the slight overlay of one network on another can dramatically increase the rate of the diffusion of information in that network. It also demonstrates how technologies that reconfigure the Togics of the networks in a system can haste effects disproportionate to their use. Producing information The second key dimension of an informational network is the production of information by the network nodes. Does the architecture of the network affect the incentives to produce information? Yes- and there is a potential downside to ~ snore info~Tnationally efficient system, however. Generally, the absence of property nights discourages investment in producing inflation in the public sector (although see caveats to this general proposition below).7 A system that is more effective at spreading information. may further aggravate this. Specif~cal.ly, governments may become mole complacent with respect to innovating. in the hope that someone else will bear the costs of ~ successful innovation. An illustration. highlights why this might happen. Imagine a potential innovation that yields $ i . I.0 worth. of benefits and costs $ i. .00 to produce if a go`~e~.ment produces the initial innovation. or is free if some other government produces the innovation. Assume further, that there are ~ 00 governments. In the absence of any information diffusion (call this the "island scena.no.'), every ~o`'ern~nent wit... spend S].00. and produce $l . IO worth of benefits, for a total of $] lO of benefits for $100 of costs. :~, the networked world where there is rapid diffusion of information. assuage that there its an initial innovator. that spends the initial $ ~ .00. and reaps S. 10 worth of net benefits. All other states then adopt the innovations for SI.10 worth of net benefits. Fro no the systemic 7 See Strum.pfforthcorni.n=; Rose-Ackerman 198O. D YNAMIC SOCIAL f JETWORK MODELING AND ANAL YSIS 109
point of YiOW, that $1.00 of cost has yielded 5109.10 of net benefits as compared to just SIO in the previous scenario. From a systemic point of view. this is an enormous success lo. ~ ~ ~ . ~ ~ groin ~~ governments point OT Yrie\~q this is an enormous success. For the forth government. this is, in absolute terms exactly the same as the island scenario. The networked world scenario is therefore Pareto superior to the island scenario. However, it is not a stable scenario if you assume that the choice to innovate is endo~envus. If you assume (:~) that each government is choosing whether to innovate: and (2) that governments are in part benchmarked by each other s performance (Besisy and Case ~ 995) and that therefore the innovation clecision over the lone run, is itself modeled on the decisions of the governments that produce the highest net benefits, the equilibrium scenario is zero innovations by any government-- O net benefits.8 The impact of free riding is particularly acute because the benefits of an innovation would be so much greater in the networked scenario-- in fact, innovations that result in net absolute losses Thor an innovator could result in welfare; ains for the system. If one assumes that the initial costs of an innovation are F. the costs of adoption for each government after the initial innovator are c. the benefits for each government from that innos ation are B. and N governments benefit from that innovation, then the innovation would result in net benefits if N*(B - c) - F > 0. For example, if N = ~ 00 c = O. B = $~.10, that innovation Louis! produce net systemic benefits even if F = $109. The innovator. however. would face net losses of $107.90. If the innovator retained rights to its information, then it could, in principle. extract many of the benefits that everyone else in the system receives. The danger of free riding, is determined. in part, by whether got ernments have different underlying preferences with respect to a potential innovation. Free riding is a ~ ~ ,~ ~ ~ ~% _ ~ ~ ~ ~ , ~ .~ ~ ~ ~ ~ ~ ~ A, . ~ 1 ~ I, _ . . ,. 1 ~ ~ . 1 ~ =~al QL£~=G! \~tICIC Urns blat: IlLb d11 -- :,vverl1menLs nave lacnucal preferences. It is no danger if each government requires a unique solution (of course. in this latter scenarios there is no benefit to the networked worIct either). The possibility of free riding may be reduced to the extent that policy makers are '~prosely~izers', valuing the possibility that their innovation will spread. If one assumes that, rather than being egoists. policy negaters are proselytizers. then the rate of innovation in the networked scenario wrill be greater than the rate of innovation in the island scenario. That is. those who seek to maximize their impact on the world rather than their jurisdiction will have greater opportunities to affect a networked world. The Rabbit and the Hare - A second danger in the hinhly networked worId is that some di`'ersitv of nolic~r . · . it, . . ~ . . ~ J ~ ^ ~ ~ J solutions will be lost, to the detriment of the systems. The decision to attempt an innovation wit] rely in Daft on a crovernment's a.~sec~nnent of the in tin ~cl~nt`~H Hal ~ ~ 1~ ~ _ ._ d. _ ~ ~ ~ _ ~ 1_ _ ~ Id _ . ~ . . ~ orner states. ally whether there Is a consensus In the system as to what best practice is. In ~ poorly networked worIcI. a government will occasionally look at what a small number of other governments are doing-- if none hare ~ clearly superior alternative, that go`'ern~nent may experiment. A successful innovation somewhere in the system will spread slowly, Galore technically. -'no innovation' is an ex;olutionarily stable strategy-- see Axel:rod 1984. 110 DYNAMICSOCIALNETWORKMOD~L~GAND~^YSIS
resulting in continued expe~in~entation in the rest of the system dunng a spoor '-take off. period. If that innovation is the optimal solution this is clearly dysfunctional; however. if it is not, the continued experimentation in the rest of the system may uncover a better solution. ~4~Iternatively, even if the successful innovation is optimal. it may not be optimal in the future. and maintaining a diversity of approaches would therefore be healthy. Heterogeneity is a systemic property that may yield benefits to all within a system. Adherence to unconventional and suboptimal policies today may provide diversity in the system for all to benefit fro no tomorrow. It also serves as a platform to experiment from. Excellent policy svIntions may only differ from policy disasters on a few din~ensions. A world where everyone rapidly cons er=,es to .'best practices' will likely have better policy outcomes in the short run than a world where everyone experiments in different ' neighborhoods ' of the policy space and then only slowly concierge to best practice. However. the latter world mill have more experimentation and may be snore likely to produce better policy outcomes in the long run.9 A classic example of premature conver~,ence is the conference on the QWERTY layout of keyboards. Early in the typewriter industry, there was substantial diversity of key layout. The QWERTY layout was originally designed to slow typing to prevent the mechanical jamming of the typewriter. and. over time through a diffusion process. the QWERTY layout became stanciard. While the mechanical jamming of typewriters is no longer a problem, the QV`rERTY standard remains. a' One might hypothesize that ~ , ~ QYV~KI Y-type at outcomes we more likely In an into~ationally efficient system. The likelihood of A adopting B's innovation should drop as the similarity of A and B's policy objectives drops, since B.s innovation would presumably be tailored to its policy objectives. Heterogeneity in underlying policy objectives should therefore help maintain a diversity of policy approaches (although limit the benefits to policy diffusion as well). . . Aggregating information The third dimension to thinking about informational networks is how well that network aggregc'~es information. Bad information as well as good spreads in informational net~orics. A more efficient network at spreading information is also a more e~:r~e~ent network at spreading rags, manias, etc. As the information cascade literature demonstrates. in a so stem where adoption is the only thing that one actor can observe about another can easily result in the spread of misinformation (Strang and Macy 9001). Essentially, if one imagines that each node in the system has private information about the value of an innovation, but that this private information can be outweighed by the observation of the adoption decisions of others then all it might take for the system to get rolling in the corona direction is for a few of the initial adopters to have incorrect signals. At that point, the priorate information of subsequent adopters is outweighed by ~ ' . . 1 . ~ ~ ~ 9 See Larch ( 1991 ) more generally on the trade off between exploration and exploitation. it' See Dasid ~ 986. Also see Liebowitz and Ma~-golis (1990) for a critique of Da`;id~s analysis, which debunks Dvorak as a superior alternative to Q~rERTY (although does not den~on.strate that QWERTY is an optimal layout of keys). DYNAMIC SOCL4L N~TWO~MODEL~G ED ^^YSIS 1 1 1
~ hat they observe others to have adopted. resulting in a potential bandwagon going in the wrong direction. These potential bandwagon effects might be ameliorated by a number of potential clynamics. First, potential adopters could pool their priorate in:fonnation regarding an innovation. Given a large enough set of nodes sharing their private evaluations. this pooled knowledge could outweigh the information conveyed by a "bandwagon ' (since, actually, bandwagons do not convey that much information). Second, adopters knight send information about their experiences. That is, not only is adoption information conveyed" but success/failure information. This would vastly increase the amount of information conveyer] in the adoption process; and every bandwagon would contain the seeds of its own destruction since the bandwagon would create a body of data about its failings. The potential of success/failure info~rnation to eliminate bandwagons depends on (:l ~ the lag between adoption and successifaiTure data; (2) whether such data are en en generated by the process (as noted above). It is in fact not in the interest of adopters to produce data that clemonstrate that they chose failing policies. There is an incentive to suppress negative feedback, and, even worse, to suppress any feedback at all in feces that it could be negative. a. Parallel processing . The vision sometimes conveyed of networks is that they distribute the , ~ informational load over the many nodes within the system tic comn~rer1 lo n'` h,~r~rt~himn,] svelter, which overloads the top node. It is not at al] clear that such ~ network would emerge organically. however, or that the ideal configuration differs greatly from an hierarchy. Imagine the following scenario: each node receives a signal about the state of the Florid [et's say ~ quantitative estimate of something of importance to the system. Each node has a certain processing capacitylet's say a capacity to "a~e~-a~e" its information and the information of 10 other nodes. What would be the most efficient organization of the system? An hierarchy. where at the bottom of the hierarchy nodes were grouped by 10, each communicating with one node above it. which averaged the 10 bits of information along with its own. This layer would be identically organized in groups of :10 passing information upwards. This structure would continue iteratively upward:;, until the sincere ton node averaged the inforn~ation trf)m its immunity In subordinates. Would such structures emerge or~anica:lly. without centralized intervention? It seems unlikely. The hierarchical structure described above is. arguably, at bests a very hard to reach equilibrium, if one posits that the nodes are seeking to maximize their own informedness. The reason for this is that there is little reason for nodes to communicate with the nodes below them. as compared to the nodes above them or at their own level. Consider the nodes one level down frown the top. A pair of nodes at this levels given the opportunity to switch one communication groin one of their subordinates each to each other. would certainly improve the quality of info~ation they there receiving. The one caveat here (and the reason why the hierarchy is a potential equilibrium) is that if one assumes that the top node .'broadcasts'! its solution once it has calculated it node of its ~ ~ _1 _ _ ~ ~ . 1 , ~ ~ r ~ =,_~ ~~_ ^^,^.~^ ~~& Ivy 4~ loll 1 \.J . . ~ . . .. . . succ3ra~nales will nave Ine Incentive to deflate trom their compunction patterns since it would be cietrin~ental to the quality of this signal. However. there is no smooth path to 112 DYNAMIC SOCKS NETWO=MODE~G ^D ISIS
that equllib~ium. because the actors that emerged as more central would have an incentive to drop their ties to less central nodes which would undermine the hierarchical structure outlined above. Institutions in the middle There are potential institutions that can play an aggregating role. In particular. one might imagine institutions that haste both high processing capacity and high levels of connectedness. For example the [etle~al government might play the role of a central node that processes the experiences of the states. develops anti disseminates best practices. However arguably, the federal government focuses more of its analytical capacities on cleveloping its own mandates lather than enabling, state and local governments develop their policies. There are also a variety of national organizations of local and state ~overnn~ents (the National Governors Association, the National Conference of State Legislatures, the National District Attorneys Association. etc). There exist also venous international organizations that serve (in a network sense) ~ similar function, such as the Organization of Economic Cooperation anc] Development, the Horace Bank, etc. In the US context, however these organizations often has e a somewhat limited desire to serve as a conduit of information in significant part because these really act as industry associations- represer~ting their collective needs to the outside worId, rather than facilitating the smooth flow of information among their members. In fact the reason for this is that these two objectives are at odds: facilitating the smooth flow of intonation, aggregating into~rnation into studies of best practice, etc. in Fact means picking the practices of a sn~all number of menders as winners and the rest as losers. Such a strategy might quickly unclerrnine the support of such an organization. ~ ~ Finally, academia plays a potential role as a central nodecollecting, comparing. and critically exan~ining the practices of many jurisdictions. The organizational nexus: networks, markets, and hierarchies Markets, networks and hierarchies offer- different mechanisms to deal with the . , . Issues amount . Nonagon diffusion, creation. and aggregation. Networks reds on reciprocity and embeddedness to regulate the behavior of their members. They are relatively poor at dealing with complex chains of exchange for exan~ple, where A has something B needs, and C has something A needs. and B has something C needs. In this seenano, the network cannot be sustained by reciprocity or reputation. but instead may at best be supported a con~plicated system of arbitrage. Networks, in the absence of markets and property rights, may do poorly at encouraging significant investment in new , . . . . . , ~ . . n~o:rmat~on, because or the lack ot control ot that Conformation once it is produced. Markets institutions standardize goods along salient din~ensions~ offer a standard for exchange (money). Supporting institutions. such as property rights, also encourage ~ ~ ~ thank Robert Behn for- this point. Notably, some of the international organizations are exceptions. Arguably, this is in part because of the political inelegance of those organizations. Further, ill many of these cases the international development . . . . organizations cnt~ca] examination is not of donor countries, but of recipients. D YNAMIC SOCIAL N~TWO~ MODELING ED ~^ YSIS 1 1 3
production of infonnation (although at the cost of efficiency in the monopoly rights conx~eyecI). People thus write books, develop new medicines, create new software, in the expectation that the properties of the goods will somehow be captured by dimensions that have emerged as salient in the market (of course these din~ensions are themselves highly contested). Markets work less well at supporting transactions where the properties of the good is not easily captured by the dimensions that the market has defined. Consider the vast array of little bits of information. gossip. etc. exchar~=ed in intormatior~al networks: '-the biggest difficult we encountered in implementation was..." .'the next division chief is going to be..." etc. These non-standard bits of information, essential to the operation of any so stem, would be impossible to put valuations on (or fair too costly). Instead, this type of information exchange is governed by reciprocity. norms. etc. Hierachies (by which ~ mean authority structures), have a vanery of mechanisms to deal with the issues ~aisecl above. At the simplest level. ir~fo~mation sharing and intorn~ation production can simply be commandecI. Thus. if A is in a position to discover something, A is ordered to do so. If A has information B would find useful. A is ordered to share information. Of course such a system would typically put demands that would quick outstrip the capacity' of the apex an organization. There are other, less bIwnt, authority tools. Hierarchies standardize behavior ant! information. facilitating information flow. Hierarchies also create the public goods for information exchange.,., infrastructure. Hierarchies can also recognize and reward information production and information sharing ex post. ~ less difficult task than doing it . . a pnorl. A particular system will generally operate by multiple institutional logics. The prescriptive question is whether there is a match between what a particular institutional logic is effective at, and what it is being asked to do. Conclusion: how to maintain innovation.go, ? The objective of this paper has to be view the intergovernmental network as an informational network;: each node producing inforrnat:ion anti potentially transferring information to other nodes. The shift from relatively geographically bound networks to global networks should greatly increase the informational efficiency of international policy networks. Yet by increasing the rate at which successful innovations spread through the systen~ globalization may discourage policy expenn~entation due to free aiding and premature policy convergence. This suggests a counter-intuitive governance prescnption: as governments receive more information regarding what other governments are doing, incentives for all governments to continue experimenting (and thus creating information) should be increased. 114 DYNAMIC SOCIAL N~TWORKA]ODE~WG ED ISIS
References: Abrahamson E.~ & Rosenkopf, :~. ~ 997. ~ Soci.a] Network Effects on the Extent of In.n.oYati.on Diffusion: A Computer Simulation.,'. Orga'~i-.c`tion Science, 8: 989- 309. Atkinson, M. M.. Bierlincr, G. 1998. -'Is there con.~;er~ence in provincial spending priorities?,. Ca'~aclian Public Policy, 24: 7.~-89. AxeIrocI. Robert (1984~. The E~iolut~o~ of Cooperation. Neat York: Basic Books. Baumol. William J. (1999~. Pareto optimal sizes of innovation spillovers. Working paper. Starr Center for Applied Economics New York University. Bennett, C. J. 199:~. "How States Utilize Foreign Evidence,' ]~`rnal of Public Policy;, Il: 31-54. Bernstein., S., & Cashore. B. 2000. "Globalization. four paths of internationalization and domestic policy change: The case of ecolorest:r-y in British." Cctnaci~a~t Journal of Political Science, 33: 67-99. Berlin, F. S. 1994. ';Seizina Up State Policy Innovation Research.' Police Studies Journal, 22: 449-~6. BerIy. F. S., & Beny, W. D. 1990. "State Lottery Adoptions as Policy Innovations: An Extent History Analysis,' Ante~icc~n Political Science Review. 84: 393-415. Burt. Ronald. 1995. Str~`ctural Holes: Tithe Social Structure of Con~petitio'~. Belknap Press. Case. Anne, and Timothy BesIcy. 1995. '-Incun~bent Behavior: Vote Seeking. Tax Settin:,, and Yardstick Competition." A'~e'-icc~n [economic Review 85: 24-45. Coleman. W. D. ] 994. 'Policy convergence in banking: a comparative study,~. Political Stt~ciies, XLll, 274-299. Coleman, W. D.- & Grant W. P. 1998. ~ Policy convergence and policy feedback: Agricultural finance policies in a globalizing era,'. Eurc~pea`~ Jo~'rnc`! of Pviitica11 Resec~rc/~, 34: 325-947. Coleman. W. D.. & PerI. A. ~ 999. '-Internationalized policy environments and policy network analysis.'. Political Studies, 47: 69 ~ -709. Davis, James A. 1970. "Clustenn~ and Hierarchy in Interpersonal Relations: Testing Two Graph Theoretical Models on 742 Sociomatrices. ' American Sociological Review 35: 843-851. Dolowitz. D. P. 900~). '~Special issue on policy transfer - Introduction." Got ernance. 13: 1 -4 Dolowitz D. P., & Marsh. D. ~ 996. '~Who learns what front Phone: a review of the policy transfer literature," Policy StucI'es, 44: 343-357. Dolowitz, D. P.. & NIarsh~ D. 2000. "Learning from abroad: The fore of policy transfer in contemporary policy-makin~,'. Grace, 13: 5-94. Dyer, J. H., ~ Nobeoka, K. 9000. Creating and managing ~ hi~h-per4-or~nance knowIedge-shann=' network: The Toyota case, Strategic ~War~c~geme'~t Jo~`rnc'/, ~ I: 345-367. Evans' M.. & Davies, J. 1999. .-Understandin;, policy transfer: A ~nulti-level. multi- disciplinary perspective e,'. Public Acl~ni''zstrc~tion~ 77: 36 ~ -385. DYNAMIC SOCKS N~TWO=MODEL~G AND ISIS 1 1 5
116 Granoveti;er, Mark. 1985. Econon~ic action and social structure: The problem of en~beddeciness. American Journal of Sociology, gI(31: 481-~IO. Gray. V ~ 973. '.Innovation in the states: A Diffusion Study,~' Americana Political Science Review. 67: ~ ~ 74- ~ ~ 85. Gray. V. 1994. "Competition. Emulation. alla policy Innovation,' L. C. Dodd & C. .lilIson. Washington, DC: CQ Press. Gupta A. K., & Go`;indarajan, V. 1991. ' Knowledge flows and the structure of control within multinational corporations." Acaclen~;y of ~~anc~geme~:t Leviers 16: 768- 792. Hubner, H. 1996. ' Decisions on innovation and diffusion and the limits of deregulation," TecI:~ova~ion. ~ 6: 327-339. Keohane, Robert, and Joseph Nye. 9000. Poser anti h~tercle.pencie'~ce' 3r~ editions Addison-WesTey. Kogut, B.. & Zande~, U. ~ ~ 99~. `'Knowled3e and the speed of the transfer and imitation of organizational capabilities - an en~pincal tests'' Organization Science, 6: 76-99. Lazer. David 2001. -`Re,~latory Interdependence and International Governance.' Jour~'c~l of European Public Policy: 474-492. Lazer. David, and Viktor Mayer-Schoenberger. 9009. ' Governing Networks,' Brooking Journal of International Law. 27. number 3: 8:19-~:~. Liebeskind. J. P., Oliver, A. L., Zucker, L. & Brewer. Hi. 1996. ' Social networks learning. and flexibility: Sourcing scientific knowledge in new biotechnology firms,'. Organizations Sciences 7: 428-443. Liebowitz. Steve and Stephen Mar`'olis. ~ 990. "The Fable of the Keys. ' Journal of Lair anti Economics 33: 1 -94. March. dances. ~ 99 ~ . "Exploration and exploitation ire organizational learning.' O)-ga':~atio'~ Science. 2~] j: 71-87 McPherson, M.. Smith-Lovin, L.^ & Cook. J. M. 2001. Birds of a feather: Homophi~y in social networks. Annual Review of Sociology, 27: 415-444. Mintrom. M. ~ 997~. "Policy entrepreneurs and the diffusion of innovation. ' Americans Journal of Political Science, 41: 738-770. Mint~om, M. I997b. "The state-Iocal nexus in policy innovation diffusion: The case of school choice,~' Is, 27: 4 ~ -59. Mintrom M.^ & Ver~ari S. ~998. Policy networks and innovation diffusion: The case of state education reforms,' Journal of Politics. 6~: I _6-l Ads. Hen~yMintzber~. 1992. Str~{ct'`rei't Fire*: DesigninOE.ffectiveOrgani~at~o'~. Prentice Hall. Moon. At. J., & Bretschneider, S. :~997. "Can state government actions affect innovation and its diffusion?: An extended communication mode! and empinca] test, Terf~nc,/~`,~ic`'l Forecasting cinch SOC[f]] Change, 54: 57-77. Nooteboom. B. i999. ';Innovation and inter-firm linkages: new; implications [vr policy. Research Panic!;, 28: 793-805. O Toole. Laurence J.. Jr. -'Treating Networks Seriously: Practical and Research-based Agendas in Public Administration. Pt'l'/ic Acin~nistrati`~z Remelt 57, ~ (January- Febmary 19974:45-59 DYNAMIC SOCIAL NETWORK MODELING AND ANALYSIS
Price. D. ]. de Solia. 1976. 'A general theory of bib~iometnc and other cun~ulative advantage processes". ]. Amer. Soc. inform. Sci. 27: 292-306. RadacIli, C. M. 2000. "Policy transfer in the European Union: Institutional isomorphism as a source of le$itin~acy.' Got ernance, i 3: 95-43. Rhodes. R.A.W. 1997. Understandi'~g Governance: Policy Metalworks, Governance, Re.fiexi~!it; acid Act ou''tabitit~, Open University Press. Robertson. M.. Swan. J. & Newell, S. ~ 996. `'The role of overworks in the diffusion of technological innovation," ~olfwlai of Ma7:age~?~ent Studies. 33: 333-359. Rogers, E. 1995. Diffusion Off Innovations, 4th edition, New York, NY: Free Press. Roan. M. C.. Peterson P. E.. & Scheme K. F. [998. "Interstate competition and welfare policy," Puts, 28: ~ 7-37. Rose-Ackerman. Susan. 1980. Risk Taking and Re-election: Does Federalism Promote Innovation?" Journal of Legal Studies 9: :93-616. Savage R. L. fed.) 1985. Special issue. ' Policy Diffusion in a Federal System, ' Plies, ~ 5: ~ -97. Schenk. NJ., Dahn~. H., & Sonje, D. :1997. The importance of social networks concerning the diffusion of new communication technology." Kol,~er Zeitschrift Fir So~ivIr~gie ``nd So ialps~cI~olog~e, 49: 35-. Seeliger, R. 1996. "Conceptualizing and researching policy convergence," Policy Stuclies Jot~r'~l, 24: 987-3(36. Simon Herbert. 1955. ' On a class of skew distribution functions,' Biomet~ika~ 49:42S- 440. Slaughter, Anne-~lane. 9000. ' Agencies on the Loose: Holding Government Networks Accountable%?' in George Be~rnann? ~Iatthias Herdeaen, anti Peter Lindseth (ecls)- Transatlantic Regulators Cooperation. Stones D. 2000. ';Non-=overnmenta] policy transfer: The strategies of independent policy institutes,- Grace, :13: 45-70. Strang David, and Michael Macy. 2001. `' In Search of Excellence': Fads, Success Stories, and Adaptive Emulation" A~f~er~cc~r' Journal of Sociology, ~ 06. Strump:f, Kenneth. Forthcoming. Does Government Decentralization Increase Policy Inno`;atic)n'?" Journal of Public Eco,~o~f~zcs. Brian Uzzi. 1996. The Sources and Consequences of En~beddedness for the Economic Performance of Organizations: The Network Effect. America Sociological Review 6 I: 674-698. Valente, T. W. ~ 996. "Social network thresholds in the diffusion of innovations." Social Nenvo~ Its, ~ 8: 69-89. Valente. T. Vyr. 1995. Network knockers of the diffusion of innovations. Cres~kill. N]: Hampton Press. Walker J. L. ~ 969. "The Diflusion of Innovations Amon it, the Ame~ical1 States ' A`?~e'~Ca'? Political Science Review, 63: 880-899. Duncan Watts and Steven Stro~atts. 1998, Collective dynamics of Small-word networks. Nature 393:440-49. Weale A.. Pri~ha~n G., Williams, A. & Porter, M. 1996. .'Environn~ental administration in six European states: Secular convergence or national distinctiveness'?," Public Acl~ni''istrc~tio~z, A: 953 974 D YNAA1IC SOCIAL NETWORK MODE:L~G AND ANALYSIS 1 1 7
Weenig, At. W. H. I 999. "Con~munication networks in the diffusion of an innovation in an organizations ' Journal of Applied So`:~c`! Ps~C/~OlOg3. 99: 1072-1099. 118 DYNAMIC SOCIAL N~TWO=MODE(~G ^D CYSTS
