sign in sign up
<<
Home , Diffusion of innovations

American Academy of Political and Social Science

Accelerating the of Using Opinion Leaders Author(s): Thomas W. Valente and Rebecca L. Davis Source: Annals of the American Academy of Political and Social Science, Vol. 566, The Social Diffusion of Ideas and Things (Nov., 1999), pp. 55-67 Published by: Sage Publications, Inc. in association with the American Academy of Political and Social Science Stable URL: http://www.jstor.org/stable/1048842 . Accessed: 03/04/2011 12:17

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at . http://www.jstor.org/action/showPublisher?publisherCode=sage. .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information and tools to increase and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Sage Publications, Inc. and American Academy of Political and Social Science are collaborating with JSTOR to digitize, preserve and extend access to Annals of the American Academy of Political and Social Science.

http://www.jstor.org ANNALS, AAPSS, 566, November 1999

Accelerating the Diffusion of Innovations Using Opinion Leaders

By THOMASW. VALENTEand REBECCAL. DAVIS

ABSTRACT: on the diffusion of innovations has been used to study the spread of new ideas and practices for over 50 years in a wide variety of settings. Most studies have been retrospective, and most have neglected to collect information on interpersonal communica- tion networks. In addition, few have attempted to use the lessons from diffusion to accelerate the diffusion of innovations. This article outlines a method to accelerate the diffusion of innova- tions using opinion leaders. The authors present their optimal matching procedure and report on computer simulations that show how much faster diffusion occurs when initiated by opinion leaders. Limitations and extensions of the model are discussed.

Thomas W. Valente is an associate professor at the School of at Johns Hopkins University. He has written extensively on the diffusion of innovations, network analysis, and health .

Rebecca L. Davis is an adjunct professor of and teaches research methods and statistics at the University of Maryland, College Park.

NOTE: The authors thank Estelle Young and Emily Agree for comments on earlier drafts. The research reported in this article was supported in part by National Institute on Drug Abuse grant no. DA10172.

55 56 THE ANNALS OF THE AMERICANACADEMY

ANY programs, interventions, media interventions is not uniform. M and communication cam- Part of this inconsistency stems from paigns are designed to change an or- differences in definitions of what con- ganization or community by direct- stitutes peer networks, the behaviors ing messages at mass or local being promoted, or the settings and audiences. These messages are dis- structure of the implementation of seminated to the entire audience the programs. This article addresses with little regard for the internal the definitional problem by outlining structure of that organization or the various approaches that are community. The structure of commu- referred to as peer models. nities and organizations can be We then present a methodology that thought of as a network of intercon- can be used to enhance existing peer nected individuals-a network that education models that capitalizes on can be used, rather than ignored, learning theory and the diffusion of when creating programs. This article innovations. Next, we present a com- details the theoretical and methodo- puter simulation to illustrate the logical principles underlying one net- method and report results intended work approach for promoting social to show its predicted efficacy. change within organizations and Finally, we discuss possible limita- communities. tions and extensions to the model. Many programs have been evalu- ated that use mass media and/or DIFFUSION OF INNOVATIONS interpersonal communication for VIA SOCIALNETWORKS behavior change (Valente and Saba 1998). These programs have been Diffusion of innovations theory conducted to address topics such as explains how new ideas and practices cardiovascular disease risk reduc- spread within and between commu- tion (Flora, Maccoby, and Farquar nities. This theory has its roots in 1989), (Piotrow et al. and sociology (see 1997), HIV/AIDS prevention (Janz Tarde 1903) with some principles et al. 1996), oral rehydration therapy adapted from (Bailey (Snyder 1990), and stress reduction [1957] 1975 or even Bernoulli 1760). (Hamburg and Varenhorst 1972). In The basic premise, confirmed by many settings, only interpersonal empirical research, is that new ideas communication is used to pro- and practices spread through inter- mote behavior change via diffusion personal contacts largely consisting through peer networks or outreach of interpersonal communication activities in an attempt to capitalize (Beal and Bohlen 1955; Haigerstrand on interpersonal influence to pro- 1967; Katz, Levine, and Hamilton mote and catalyze desired behavioral 1963; Ryan and Gross 1943; Rogers changes (Jemmott, Jemmott, and 1995; Valente and Rogers 1995; Fong 1998; Neaigus 1998). Valente 1995). The impact of these interventions Ryan and Gross laid the ground- is varied. The relative and absolute work for the diffusion paradigm in a efficacy of peer networks and mass 1943 publication that found that ACCELERATINGDIFFUSION OF INNOVATIONS 57 social contacts, social interaction, which masks the actual processes and interpersonal communication responsible for the spread of the were important influences on the . adoption of new behaviors (Valente Given the importance of interper- and Rogers 1995). Their ground- sonal contacts in diffusion, scholars breaking study was followed by sev- have sometimes relied on formal eral hundred diffusion studies con- methods of measuring who talks to ducted in the 1950s and early 1960s whom within a community. Such to examine the diffusion process in methods are known as network more detail across a wide variety of analysis (Scott 1991; Wasserman topics (Rogers 1995). Most studies and Faust 1994; Rogers and Kincaid supported the idea that interper- 1981). Network analysis is a set of sonal contacts were important methods that enables researchers to influences on adoption behavior. locate individuals who are more cen- Researchers sought to understand tral to a community and thus per- how information created in gov- haps more influential. The basic dif- ernment or other organization- fusion network model uses these sponsored programs had been dis- individuals, or opinion leaders, to ini- seminated within an interpersonal tiate the diffusion of a new idea or communication environment. practice. They can function as cham- Although many factors influence pions for the new practice and accel- innovation diffusion, scholars have erate the diffusion process (Valente consistently found that interper- 1996; Katz 1957; Katz and Lazars- sonal contacts within and between feld 1955). The opinion leader often communities are very important functions as the theoretical under- influences on adoption behavior pinning to peer education programs. (Valente 1995). Although many scholars agree on OPINIONLEADER the importance of interpersonal com- MODELSOF DIFFUSION munication to the diffusion process, few studies have successfully traced Interventions designed to use an innovation through a network of interpersonal communication for social contacts. The lack of data on promoting behavior change are often diffusion within an entire network referred to as peer influence, peer stems largely from the difficulty of education, interpersonal counseling, collecting data over a time period outreach, or peer networks. Implicit long enough for diffusion to occur. As in the peer promotion model is the a consequence, most studies have assumption that some individuals relied on retrospective data, which will act as role models for others. might introduce some bias (Coughe- These role models act as opinion nour 1965; Nischan et al. 1993).1 A leaders within their communities more serious limitation of retrospec- and can be important determinants tive data is that they may capture a of rapid and sustained behavior post hoc explanation for diffusion, change. Research findings support 58 THE ANNALS OF THE AMERICANACADEMY this principle. In one study, opinion of having agendas different from leaders were shown to be effective at those of the members of the commu- decreasing the rate of unsafe sexual nity or even agendas harmful to com- practices (Kelly et al. 1991). In munity members. Equally damaging another study, opinion leaders were to the potential for influence is effective at decreasing the rate of the perception that the leader is cesarean births (Lomas et al. 1991). unaware of the community's needs or These findings imply that maximiz- that the leader may not be suffi- ing the effectiveness of these opinion ciently knowledgeable about the leaders can further accelerate the innovation. Finally, persons not rate of diffusion. selected by community members An important corollary to the may use persuasion tactics that are recognition of the function of opinion not effective in that community. leaders is to determine how they The third technique, the snowball are selected. Several potential approach, avoids the selection bias recruitment procedures exist: problem by allowing all community members to participate in the inter- 1. Individuals select themselves to vention (regardless of their leader- be peer leaders. ship status) by being both a recruiter 2. Program staff or project teams and a "recruitee." This snowball select the leaders. approach may be used to recruit indi- 3. Community members recruit viduals to receive a service (such as a participants, not leaders, who in turn clinical screening) or to disseminate each recruit new participants (Broad- information. One problem with the head et al. 1995). snowball approach is that complex 4. Some selected individuals ideas and behavior change recom- within the community nominate oth- mendations may not be effectively ers to be opinion leaders (Kelly et al. communicated by everyone, and 1991). hence the strategy may be limited to 5. All community members are easily communicated messages (a invited to nominate opinion leaders chain is only as strong as its weakest (Lomas et al. 1991; Wiist and Snider link). An additional limitation is that 1991). there is no opportunity to capitalize on the networks since the dynamic There are several limitations to nature of the snowball is temporary. the effectiveness of each of these Allowing community members to approaches. The degree of influence nominate leaders, as in the fourth wielded by an opinion leader is predi- technique, overcomes these disad- cated in part on the potential adopt- vantages by providing a pool of recog- ers' assessment of his or her credi- nized community leaders. Using only bility and trustworthiness. Self- a select few individuals to nominate selected leaders and those selected leaders may, however, decrease the from outside the community (meth- validity or reliability of the process. ods 1 and 2) could each be suspected Moreover, the desired outcome (that ACCELERATINGDIFFUSION OF INNOVATIONS 59 the leaders be effective) may be intervention. The leaders can then be highly dependent on the persons cho- instructed to disseminate informa- sen to do the selecting. tion to the general community or Allowing all community members used in a one-to-many matching so to nominate leaders (the fifth tech- that leaders train or teach those com- nique) overcomes most of the short- munity members that specifically comings of the other techniques. The nominated that particular opinion list of opinion leader nominations leader. The one-to-many strategy that it produces provides an accurate provides an optimal match of com- map of who goes to whom for advice munity members to recognized com- within the entire community. The munity leaders in a form suitable for strategy exploits the existing accelerating diffusion ofinformation, structure of information dissemina- innovation, and community change. tion within the community and The approachtaken in this article relieves the need to impose an artifi- is to develop a model designed to cial information flow network from identify opinion leaders as desig- above. This technique has been suc- nated by sociometric techniques cessfully employed in several arenas. (Rogers and Cartano 1962). Leaders For example, rotating credit- can be chosen as those who received associations in developing countries the most nominations, or, alterna- take a census of all community mem- tively, more complicated centrality bers and permit them to nominate algorithms can be used (Borgatti, program leaders. The nomination Everett, and Freeman 1998; technique follows the principle Freeman 1979; Valente and Fore- underlying democratic forms of man 1998). These leaders are then government. matched with those who nominated A community or organization them to create an optimal interper- attempting to initiate behavioral sonal pairing. The leaders can then change ensures the credibility and be given educational materials to trustworthiness of opinion leaders by educate or train those with whom allowing the entire community to they have been paired. This diffusion select opinion leaders. A second network perspective thus capitalizes advantage of this approach is that on the principles of learning theory the number of leaders selected for (Bandura 1986) and diffusion training can be varied depending on (Rogers 1995; Valente 1993; Valente the needs of the intervention. Third, and Rogers 1995), which dictate that the boundaries used to define leaders learning occurs most efficiently can be varied to account for group when individuals are trained by their membership properties (for example, "nearpeers" whom they have chosen opinion leaders can be recruited as their models (Rice 1993). based on gender, ethnicity, geogra- Once leaders are identified, they phy, or the like). can be matched to the persons in the The nomination method identifies community who nominated them. If the leaders to be trained in the someone did not directly nominate a 60 THE ANNALS OF THE AMERICANACADEMY

FIGURE1 NETWORKOF PHYSICIANSIN AN ILLINOISCOMMUNITY

SocioQgrambased on ties Ontimalleader/mentor matching

x"33"' / 32 4 2?, 10 Jo '. i / 28 ...... t... 3 / 14 30 .. 1 ,\

32 ...... •"

...... I i l \ --...... ""•' ./ ... ,,, 15...... ,30 29 ...... 7...... "...... i

2 1 31 . 1(b) l(a)

SOURCE:Coleman, Katz,and Menzel 1966. leader, that person is matched to a tetracycline prescriptions among leader who is closest to him or her. If physicians in one Illinois community an individual is equally close to two in the mid-1950s. This graph shows or more leaders, then indirect paths the network of connections between connecting this person to these lead- 21 physicians in the community and ers can be used to determine the opti- is a picture of who goes to whom for mal pairing. For example, if person X advice about medical matters.2 In nominates both leaders A and B, but Figure 1(b), the sociogram has been has more or shorter indirect paths to redrawn so that central nodes (physi- A rather than B, then X can be cians) are matched to the nodes that matched to A. Computer algorithms nominated them or are closest socio- that optimally match community metrically. This provides an optimal members to leaders can then be con- matching of opinion leaders to the structed using direct and indirect community members who look to paths between members and leaders. each of them for advice and thus can For example, Figure 1(a) displays be used to accelerate the diffusion the sociogram of network nomina- process. tions for the study by Coleman, Katz, As shown in Figure 1(a), members and Menzel (1966) of the diffusion of 8, 13, and 15 are central to the ACCELERATINGDIFFUSION OF INNOVATIONS 61 network, receiving the most nomina- community exist. In decentralized tions. After the network has been situations, the researcher is faced reconfigured, the exact leader-to- with the challenge of developing learner matchings have been carried opinion leaders who can assume out. The matching provides a strat- leadership roles for innovation egy that can be used to implement diffusion. behavioral promotion programs. The proposed process of opinion Everyone in the community is leader identification consists of the matched to a leader, some from their following three steps: direct links and others via their indi- rect links. 1. Identify the 10 percent of indi- Partitioning a network into these viduals within a community who leader-follower pairings is relatively received the most nominations by straightforward. In many settings, other members of the community networks will partition into leader- and designate these individuals as follower pairs unambiguously, while the opinion leaders. (This step may in others the partitioning may be be varied to create greater or fewer more ambiguous for a number of rea- leaders and can take advantage of sons. One impediment to optimal other methods of centrality, not dis- matching is if the network has one or cussed in this article, to identify the a small group of people who receive a opinion leaders.) preponderance of the nominations. 2. Match opinion leaders to the In network terms, such a network is community members who are closest centralized. In a centralized highly to them in the chain of information network, most members (or all) flow. That is, each individual would be to one a assign assigned (or few) to the leader whom he or she nomi- in cen- leader(s). Matching highly nated or to whom he or she is con- tralized networks will have too many nected through the smallest number individuals to the same assigned of intermediaries. leader. The solution to this problem 3. isolates who will be to have some of the Assign (individuals persons nominated no one and whom no one assigned to leaders nominated indi- nominated) to leaders or The of these randomly rectly. percentage based on a rule that would a proportionally nonoptimal pairings provide allocates isolates to more measure of the of the popular fidelity opinion leaders. leader model implementation. Fortunately, however, centralized networks are usually efficient con- The major limitation (or barrier) duits for information and thus, to the optimal opinion leader match- rather than being impediments to ing procedure is the occurrence of implementation, will generally extremely centralized networks. In facilitate diffusion. It may be the case such cases, one or few opinion lead- that a network is decentralized and ers will be identified and would opti- no clear opinion leaders within the mally be paired with all other 62 THE ANNALS OF THE AMERICANACADEMY

FIGURE 2 DIFFUSION NETWORK SIMULATIONWITH DIFFERENT INITIALADOPTERS (Threshold set at 15 percent, N= 100)

100 90 E 80 ? 70 o 60 < 50 40 2E 30 1 S 20...... 10 -4--- OpinionLeaders 0 -U-- Random 1 2 3 4 5 6 7 8 9 10 --- Marginals Time

members of the community. Unfortu- who received the fewest nomina- nately, this result can be impractical tions. Each condition was simulated for some programs or innovations. 1000 times and the results averaged For example, if there is only one over the 1000 runs. expert on a topic for an entire organi- Figure 2 displays the average zation, the network map will show cumulative adoption curves for the everyone linked to this one person. three conditions. If the first adopters are opinion leaders, then diffusion SIMULATIONSOF accelerates rapidly and everyone has OPINIONLEADER MODEL adopted by time period 3. In contrast, if the first adopters are selected ran- To illustrate the model, we gener- domly, the middle curve in Figure 2, ated hypothetical networks in which the rate of diffusion is slower, with the ties between members were only 30 percent of the network hav- randomly allocated. Each network ing adopted the innovation by time represented 100 people, with each period 3. Similarly, if the first adopt- person making seven random nomi- ers are those individuals who are on nations. We then simulated diffusion the margins (those with fewest nomi- by assuming that each person would nations), the rate of diffusion is slow- adopt an innovation when his or her est, with only 15 percent of the net- personal network exceeded a set work having adopted the innovation threshold (15 percent). We then com- by time period 3. pared three diffusion conditions As many studies (Becker 1970; based on whether the first 10 adopt- Mendez 1968; Rogers 1995) have ers were (1) opinion leaders, or those shown, it is not opinion leaders who who received the most nominations; are early adopters, but instead mar- (2) randoms, or persons chosen at ginals or individuals who are bridges random; or (3) marginals, or those to other networks who first adopt an ACCELERATINGDIFFUSION OF INNOVATIONS 63

innovation. When diffusion starts on the innovation. This informal with these individuals, the innova- approach has broad appeal as it pro- tion must percolate through the motes a collegial feeling concerning network before it reaches opinion the training. Alternatively, a more leaders who are in the position to set formal structure can be created, with the agenda for change. Consequently, the opinion leader meeting with a critical mass typically occurs late, at group of learners for a training or time periods 3 through 5 in Figure 2. informational session. By intervening directly with the The final decision is whether the opinion leaders, the lag time between learning process should be static or introduction and critical mass is dynamic. Will the opinion leader eliminated. meet once or many times with his or her peer learners? This may be dic- STRATEGYFOR USING tated by the innovation being dif- THE OPINION LEADERMODEL fused, its complexity, and the risk associated with adoption. For exam- When considering opinion leader ple, opinion leader diffusion of com- model implementation, at least three plex medical practice guidelines factors should be considered: (1) should consist of one-on-one training opinion leader recruitment, (2) loca- and should be accompanied by peri- tion of training, and (3) timing of odic follow-up. In the diffusion of a training. Getting buy-in from the simpler innovation, such as organ- opinion leaders is very important. izational reporting procedures, the These leaders must believe in the training may consist of a one-time innovation that is being diffused and training session. A third option is to be willing to be active participants in have the opinion leader meet with the diffusion process. Leaders may his or her group of peer learners for a appreciate being recognized as training session and then casually opinion leaders, which validates inquire about the learners' compli- their position in an organization or ance in follow-up chance meetings. community, but this designation Supplemental aids can also be may carry with it added responsibili- used in the process to trigger conver- ties. As such, compensation for sations. Such aids include informa- their time may facilitate increased tional posters in common areas, or participation. buttons with a related logo to be worn A second consideration is how and by the opinion leaders (Kelly et al. where this learning process should 1991). These aids can help prompt take place. Two options that are most interpersonal communication, which viable are a one-on-one training or a in turn reinforces the adoption one-to-many training. In the former, process and accelerates diffusion. the implementing agency would notify each opinion leader of whom DISCUSSION he or she is to train or inform and makes an appointment (either for- Although logic, computer simula- mally or casually) to train the learner tions, and prior experience indicate 64 THE ANNALS OF THE AMERICAN ACADEMY that this approach can be successful A second advantage of this model at accelerating innovation diffusion, is that in most organizations and a few cautionary notes should be communities, different individuals sounded. First, our model and much will be seen as opinion leaders in dif- of the literature assume that indi- ferent domains. The opportunity or viduals are influenced by their direct burden of can be ties (referred to as a cohesion model shared by a diverse set of individuals of influence [Valente 1995]), while within the community. Over time the other scholars (Burt, 1987, 1992; community can formulate itself into Granovetter 1973) have noted that a dynamic learning community that one's position in the network may relies on itself and distributed sys- also influence adoption.3 tems of monitoring to continually Second, our results may be sensi- enhance its performance. tive to missing data or the inability to The influence of interpersonal interview all, or even most, members persuasion in behavior change has of a community. The ability to effec- been repeatedly noted by scholars for tively implement this technique over 50 years. This interpersonal when a full census of the community influence is sometimes aided by mass is not possible is unknown. media and other communication Finally, how much opinion leaders strategies and is sometimes inde- enjoy being opinion leaders remains pendent of them. Rarely, however, to be seen. While most opinion lead- has the power of interpersonal influ- ers appreciate the acknowledgment ence been systematically incorpo- that comes from being recognized as rated in scientific studies of behav- such, some may find it an intrusion ioral promotion and diffusion. or may be resistant to the innovation Empirical studies designed to pro- being proposed. spectively measure the diffusion of In spite of these limitations, we innovations and to explicitly observe believe that this model provides a interpersonal communication pat- means to create and chart the opti- terns would complement the simula- mal path of diffusion within a com- tions discussed in this article to more munity. If diffusion cascades from fully explain and understand effec- the most central to the more periph- tive diffusion strategies. The eral members of the network, it can challenge for us now is to harness do so optimally by moving from the these tools in strategic ways that are persons with the most nominations meaningful to diffusion efforts to those with the fewest. Such a path- designed to promote desired social way is the optimal diffusion path. change. of this can Computation trajectory Notes provide a standard against which to compare actual diffusion processes 1. The retrospective data are often used to and a useful for estimate the rate of growth over time and pro- provide diagnostic vide the opportunity to compare diffusion measuring the relative speed of rates within and between communities (Va- diffusion. lente 1993). ACCELERATINGDIFFUSION OF INNOVATIONS 65

2. Generally, sociograms are drawn with nary Results of a Peer-Driven Inter- of and distances between positions nodes vention. Journal of Drug Issues based on a multidimensional scaling solution 25:531-64. (Krackhardt,Blythe, and McGrath 1994). Burt, Ronald S. 1987. Social Contagion 3. For example, an assistant professormay and Innovation: Cohesion Versus be influenced by other faculty with whom he or she talks regardless of rank (cohesion),or the Structural Equivalence. American assistant professormay be influencedby other Journal of Sociology 92:1287-1335. faculty who are of his or her rank (structural 1992. Structural Holes. Cam- equivalence), irrespective of whether or not he bridge, MA: Harvard University or she talks to them. Press. Coleman, James S., Elihu Katz, and Her- References bert Menzel. 1966. Medical Innova- tion: A Diffusion Study. New York: Bailey, Norman T. J. [1957] 1975. The Bobbs-Merrill. Mathematical Theory of Infectious Coughenour, C. Milton. 1965. The Prob- Diseases and Its Applications. Lon- lem of Reliability of Adoption Data in don: Charles Griffen. Survey Research. Bandura, A. 1986. Social Foundations of 30:184-203. Thought and Action: A Social Cogni- Flora, J. A., N. Maccoby, and J. W. Far- tive Theory. Englewood Cliffs, NJ: quar. 1989. Communication Cam- Prentice Hall. paigns to Prevent Cardiovascular Dis- Beal, George M. and Joseph M. Bohlen. ease: The Stanford Community 1955. How Farm People Accept New Studies. In Public Communication Ideas. Cooperative Extension Service Campaigns, ed. R. E. Rice and C. K. Report 15. Ames, IA: U.S. Department Atkins. 2d ed. Newbury Park, CA: of Agriculture. Sage. Becker, Marshall H. 1970. Sociometric Freeman, Linton C. 1979. Centrality in Location and Innovativeness: Refor- Social Networks: Conceptual Clarifi- mulation and Extension of the Diffu- cation. Social Networks 1:215-39. sion Model. American Sociological Re- Granovetter, Mark. 1973. The Strength view 35:267-82. of Weak Ties. American Journal of So- Bernoulli, Daniel. 1760. Essai d'une nou- ciology 78:1360-80. velle analyse de la mortalit6 caus6e Haigerstrand, Thorsten. 1967. Innova- par la petite v6role et des avantages de tion Diffusion as a Spatial Process. l'inoculation pour la pr6venir. Mem- Trans. A. Pred. Chicago: University of oirs of the Royal Mathematical and Chicago Press. Physical Academy of Sciences, pp. Hamburg, B. A. and B. B. Varenhorst. 1-45. 1972. Peer Counseling in the Secon- Borgatti, Steve, Martin Everett, and Lin- dary Schools: A Community Mental ton Freeman. 1998. UCINET 5 for Health Project for Youth. American Windows: Software for Journal of Orthopsychiatry 42:566-8 1. Analysis. Natick, MA: Analytic Tech- Janz, N. K., M. A. Zimmerman, P. A. nologies. Wren, B. A. Israel, N. Freudenberg, Broadhead, Robert S., Douglas D. and R. J. Carter. 1996. Evaluation of Heckathorn, Jean-Paul C. Grund, L. 37 AIDS Prevention Projects: Success- Synn Stern, and Denise L. Anthony. ful Approaches and Barriers to Pro- 1995. Drug Users Versus Outreach gram Effectiveness. Workers in Combating AIDS: Prelimi- and Behavior 23:80-97. 66 THE ANNALS OF THE AMERICANACADEMY

Jemmott, John B., Loretta S. Jemmott, Nischan, Peter, Klaus Ebeling, David B. and Geoffrey T. Fong. 1998. Absti- Thomas, and Ulrike Hirsch. 1993. nence and Safer Sex HIV Risk- Comparison of Recalled and Validated Reduction Interventions for African Oral Contraceptive . Ameri- American Adolescents. Journal of the can Journal of Epidemiology American Medical Association 138:697-703. 279:1529-36. Piotrow, P. T., D. L. Kincaid, J. Rimon, Katz, Elihu. 1957. The Two-Step Flow of and W. Rinehart. 1997. Health Com- Communication: An Up-to-Date Re- munication: Lessons from Family port on a Hypothesis. Public Opinion Planning and . Quarterly 21:61-78. New York: Praeger. Katz, Elihu and Paul Lazarsfeld. 1955. Rice, Ronald E. 1993. Using Network Personal Influence: The Part Played Concepts to Clarify Sources and by People in the Flow of Mass Commu- Mechanisms of Social Influence. In nications. New York: Free Press. Progress in Communication Sciences, Katz, Elihu, Martin L. Levine, and Her- ed. W. D. Richards and G. A. Barnett. bert Hamilton. 1963. Traditions of Re- Vol. 12. Norwood, NJ: Ablex. search on the Diffusion of Innovation. Rogers, Everett M. 1995. Diffusion of In- American Sociological Review novations. 4th ed. New York: Free 28:237-53. Press. Janet S. St. Lawrence, Kelly, Jeffrey A., Rogers, Everett M. and D. G. Cartano. Yolande E. Diaz, L. Yvonne Steven- 1962. Methods of Measuring Opinion Allan C. Hauth, Ted L. Brasfield, son, Leadership. Public Opinion Quarterly Seth C. Kalichman, Joseph E. Smith, 26:435-41. and Michael E. Andrew. 1991. HIV Everett M. and D. Lawrence Kin- Risk Behavior Reduction Rogers, Following caid. 1981. Communication Networks: Intervention with Key Opinion Lead- A New Paradigm Research. New ers of An for Population: Experimental York: Free Press. Analysis. American Journal of Public and Neal C. Gross. 1943. Health 81:168-71. Ryan, Bryce The Diffusion of Hybrid Seed Corn in Krackhardt, D., J. Blythe, and C. Two Iowa Communities. Rural Sociol- McGrath. 1994. Krackplot 3.0: An Im- ogy 8(1):15-24. proved Network Drawing Program. John. 1991. Network A Connections 17(2):53-55. Scott, Analysis: Handbook. CA: Lomas, Jonathan, Murray Enkin, Geof- Newbury Park, Sage. frey M. Anderson, Walter J. Hanna, Snyder, Leslie B. 1990. Channel Effec- Eugene Vayda, and Joel Singer. 1991. tiveness over Time and Knowledge Opinion Leaders vs. Audit Feedback and Behavior Gaps. Journalism to Implement Practice Guidelines: De- Quarterly 67:875-87. livery After Previous Cesarean Sec- Tarde, Gabriel. 1903. The Laws of Imita- tion. Journal of the American Medical tion. Trans. E. C. Parson. New York: Association 265:2202-7. Holt. Mendez, Alfredo D. 1968. Social Struc- Valente, Thomas W. 1993. Diffusion of ture and the Diffusion of Innovation. Innovations and Policy Decision- Human Organization 27(3):241-49. Making. Journal of Communication Neaigus, Alan. 1998. The Network Ap- 43:30-45. proach and Interventions to Prevent . 1995. Network Models of the Dif- HIV Among Injection Drug Users. fusion of Innovations. Cresskill, NJ: Public Health Reports 113:140-50. Hampton Press. ACCELERATINGDIFFUSION OF INNOVATIONS 67

. 1996. Social Network Thresholds Interdisciplinary Social Science Jour- in the Diffusion of Innovations. Social nal 16(3):238-69. Networks 18:69-79. Valente, Thomas W. and Walter Saba. Valente, Thomas W. and Robert K. Fore- 1998. Mass Media and Interpersonal man. 1998. Integration and Radiality: Influence in a Reproductive Health Measuring the Extent of an Individu- Communication Campaign in Bolivia. al's Connectedness and Reachability Communication Research 25:96-124. in a Network. Social Networks 20:85- Wasserman, Stanley and Katherine 109. Faust. 1994. Social Networks Analy- Valente, Thomas W. and Everett M. sis: Methods and Applications. New Rogers. 1995. The Origins and Devel- York: Cambridge University Press. opment of the Diffusion of Innovations Wiist, W. H. and G. Snider. 1991. Peer Paradigm as an Example of Scientific Education in Friendship Cliques: Pre- Growth. : An vention of Adolescent Smoking. Heath Education Research: Theory and Prac- tice 6:101-8.