Knowledge communities that use unique knowledge will perform better. Now we turn from the way knowledge communities use knowledge to the way they use, in aggregate, rhetoric. Knowledge communities are largely text-based—that is, the primary form of communication between members is through articles. Knowledge communities have a particularly difficult signaling problem: Members need to identify each other in order to learn from each other and benefit from the advantages of community membership, but such identification is not always easy (Braam 1991a; White 2003). Since articles do not contain explicit school identity labels (though journals can give hints toward potential identities), authors who wish to position their articles within a school may use keywords to identify themselves as a part of that school or may use language specific to that school (Abrahamson 1996; McCloskey 1998).
From a marketing perspective, we expect rhetoric to be used in a knowledge community very differently from the way a community uses previous knowledge. Members of strong knowledge communities will, we believe, for internal and external reasons, tend to converge on shared language both as a way to reduce the ambiguity of communication and because such communities use similar methodologies and similar language in presenting problems and issues (Abrahamson 1996; Bartel and Saavedra 2000; DiMaggio and Powell 1983; McCloskey 1998).
From an external perspective, literature on integrated marketing and branding suggests that firms which use a single consistent message will be more effective (Schultz et al. 1994). This should hold even more strongly for knowledge communities, where the lack of concrete labels causes authors to flag themselves as part of a school as a signaling message (Mizruchi and Fein 1999; Pfeffer 1993). Lastly, the very diversity of knowledge being presented by knowledge communities gives rise to a need for a single rhetorical lens with which to express these ideas. Successful knowledge communities, we believe, will draw on many sources for their knowledge and then present these diverse ideas under one shared rhetoric or framework of analysis.
Innovating knowledge communities that use consistent rhetoric will perform better. But the lack of formal leadership that makes knowledge communities so different from firms also limits their ability to coordinate change. The rhetoric of a strong paradigm tends to be stable over time, as it is based on a shared construction of rhetorical meaning (Price 1963). Reconceptualizations of shared mental constructions and changes in meaning of shared language, particularly in a context without any explicit leadership or coordinating mechanism, can result in inefficiency and ambiguity in communication. The advantages of shifting rhetoric over time are not as clear. Buzzwords and rhetorical fashion are not a good basis for a sustained competitive advantage, particularly in a field that relies on technical work (Abrahamson 1996). Since knowledge communities are amorphous groups without formal boundaries, they benefit from a unified use of language in order to create a recognizable sort of identity or brand (Haynes et al. 1999). Particularly in a technical context, language is not the source of innovation; rather, the ideas underlying language are. Therefore, once a knowledge community constructs a consensus in rhetoric, it remains relatively stable.
Innovating knowledge communities that use stable rhetoric will perform better. Finally, knowledge communities, if they are to be successful, must appeal to large numbers of people. This implies that schools claim and use common and recognized language as their core set of words, allowing them to grab the metaphorical ‘‘middle ground’’ (Downs 1957; Hotelling 1929). Knowledge communities whose words and meanings are difficult for others to understand tend to isolate themselves. While this has proven to be a successful strategy for professional fields that desire barriers to entry (e.g., doctors, lawyers), it hampers the cooperation and collaboration that is desirable between knowledge communities (Kripke 1982). Addressing the important issues of the field and communicating good ideas in broadly understood language will lead to the largest possible audience (Downs 1957).
Innovating knowledge communities that use mainstream rhetoric will perform better. Although knowledge communities do not have formal leaders or formal organizational legitimacy, they do have the preconditions for idiosyncratic capabilities and routines— intense socialization, repeated interactions, mechanisms for punishment and reward, an interconnected incentive structure (Barney et al. 2001; Cohen et al. 2000; Wernerfelt 1984). These differential routines and capabilities could constitute a competitive advantage for communities and lead some communities to have sustained levels of higher innovativeness. Understanding how and why knowledge communities create a culture of innovation that effectively generates and distributes good ideas is our central goal. The effect of firm involvement in knowledge communities is an interesting one. While it is not a main variable, we include it in order to explore its impact on all our models. Firms place a higher value on proprietary information, while knowledge communities often thrive on very open sharing of information, at least in the sciences and social sciences. The difference this makes is hard to fully appreciate until our analysis extends to firms or commercial patents.
The data in this paper were drawn largely from CiteSeer, a digital library of papers from conferences and journals in computer science. We cross-reference all of CiteSeer’s more than 700,000 indexed papers with the DBLP Computer Science Bibliography, a European database with over 600,000 papers that indexes a similar group of computer science papers, in order to verify existing information and gather supplemental information on journals and conferences. The match between these two databases is close; indeed, the DBLP links most of its papers to the corresponding papers in the CiteSeer database. The majority of papers in these databases are from between 1992 and 2003. They give us a rich picture of the field of computer science as it has evolved in the past decade and a half. Some limitations, such demarcation of paper type, remain.