Ethnography ant;'Network Analysis 211 nized critical relationships differently. Cross- nized or structured, and how that structure has cultural studiesidentified unexpecteddifferences an impact on individual lives. They describe the that led to the first studies of kinship groups, structural environmentof individuals, organiza- marriage patterns, and social and political orga- tions, and soci~ties. nizations in small societies. These were critical There are three primary anthropological findings, and are still a part of the overall infor- approachesto network studies.These are ethno- mation that is important for understanding cul- graphic descriptions of social networks (Bernard tural differences in people's daily lives. and Killworth 1973;Morrill 1991),personal net- As ethnographic research progressed, two work studies or ego-centerednetworks (Hammer different approaches to understanding cultural 1983; Shelley 1992),and the analysis of whole networks were explored. Pasternat (1976) networks (e.g., Rage and Rarary 1991).2 summarized the initial systematic approaches The social network approach has changed to the exploration of kinship groups, and rapidly during the last ten years, due to the ElizabethBott (1971)produced an ethnographic development of sophisticated data collection exploration of social networks in England. and analysis techniques,especially the develop- Theseworks representtwo theoretical and meth- ment of new statistical approaches to under- odological anchor points fo~ anthropologically standing complex relationships.3 Wasserman driven social network analysis. Pasternak and Faust (1994)provide an excellentintroduc- describesmethods for collecting and then com- tion to both the descriptiveand the probabilistic paring the ways that different cultures identify, statistical methods used to analyze spcial rela- label, and understand the genealogicalrelation- tionships. All of thesetechniques are based on ;'ships that are part of their culture. The Bott an attempt to find order in the relationships that ?~tudyprovides an in-depth exploration of the people create. The variety of important ques- Lintimate support networks that most people tions asked in different social science research use to survive in their culture, and provides a paradigms (anthropology, sociology, geography, (model for exploring these relationships across political science,and psychology)have resulted cultures..Following these studies, anthropol- in the social network approaches described in i;ogists conducted and systematically refined this chapter. The authors provide examples of +their examinations of informal and formal a wide range of fruitful research questions ~humangroups and associations,in conjunction answeredby a social network approach, includ- ~"Withwork go~g on in sociology, social psychol- ing studies of occupational mobility, the impact Jogy, and political science (Galaskiewicz and of urbanization on individual well-being, world

.':ilWasserman1993; Johnson 1994; Wasserman political and economic systems,community elite ,~and Faust 1994). The combined approaches power and decision making, social support ~.~xpand.VJ our know.ledg.e of the ~ffects and research, group problem solving, the diffusion ::4ynalnlcs of both kinship and nonkin networks and adoption of innovations, interlocking cor- : all parts of human culture. This research porate directorates, cognition and social repre- ~i!g~~fr!?~ purely qualitative descriptions of sentation, markets, exchange relationships, (groups and associations,to quantitative social social influence,and the formation of coalitions, ~etwork schemasthat create network descrip- among others (Wassermanand Faust 1994:5-6). .on the algorithms of both graph There are three levels,of analysis that can be Each approach pro- simultaneouslyapplied to social networks. These insights into human cultures. In are analysesof the individual, the subgroup, and they provide powerful explana- the whole systemcharacteristics. At the indivi" the ways that hUmans think, act, and dual level, the analysis consists of describing their daily lives within their personal the relationships, position, and roles of the indi- . vidual in relation to other people in the network. network research describes relatIon- Each individual can be describedin terms of how relationships include physical con- his relationships connecthim with other people,

--violence, supportive how information and influence can flow to or social contact (friendship, work from him (or through him to others), and how

(com- his place in the network affectshis life by making

impact, e-mail), or even him similar to, or different from, others in similar Different types of rela- or different kinds of roles and positions in his different cultural contexts in own or other networks. Individuals can also be Social network analysis defines described in terms of their membershipin sub- as kinship or friend- groups in the network, and their closenessor influence, communication, physi- distanceto other individuals. Analysis of the sub- or social support), and then group structure of the network consists of dis- those relationships are orga- covering, describing,and analyzing the effect of 212 Handbook of Social Studiesin Health and Medicine

subgroups in the network and the connectionof those subgroupsto other groups and indiViduals. APPROACHESTO STUDYINGRELAnONSHIPS Iri larger networks, people tend to cluster into smaller groups. An example would be an extendedfamily. Kinship ties connect the entire Social network theory has commonly developed family to eachother. Each nuclear family would from the analysisof relationships, rather than an tend to have the closestties in our culture, but a priori theory of relationships. Anthropologists would still maintain contact with other nuclear noted differences in family structure in different families. The adults would tend to have the most cultures, and developed theory to account for direct connectionsand the most frequent direct those differences,rather than haVing the theory contacts,while their childrenwould be connected first, and finding the difference afterwards. The to the rest of the group through indirect ties same condition applies to many network (their mother lets them know what is happening approaches,where researchersfirst focused on to their cousins, aunts, and uncles) with much describing relationships, and then createdmeth- less frequent direct contact (family reunions). ods and theories to make those descriptionsand Network analysis allows researchersto identify analysesstronger over time. these subgroups within larger connected net- This observation-based approach to social works, and to analyze the impact that these relationships has produced different but comple- groups have on people'slives. Finally, a network mentary methodological approaches that are can be characterizedas a whole and compared 0used in network analysis.These are the explora- with other networks. Network density (the num- tion of personal networks, egocentric networks, ber of connections between people compared chained or snowball network studies, and the with the number of potential connections),net- analysis of whole networks. These approaches work centrality measures,and transitiVity mea- are summarized in Table 1. sures (a measure of whether the connections of Each of these types of study has its own the- one indiVidual are also connected with each ory, methods, and appropriate researchinstru- other) are some of the technical measures of ments attached to it. The basic approachesare total network conditions. describedin the following sections.

Table Approaches,foci, and methods of network analysis " Approach Focus Methods, Instruments ,',cO'i- I 'cj Questions about personal networks Standard questions about relationships and relationships from the perspective (McCallister and Fischer 1983) of the informant. (Burt and Minor ~t [c 1983) 1 Egocentric Description of individuals in personal Name generatorsand questions about networks and the relationships of interactions of those named both egq and the individuals named (Burt, 1984;Marsden 1990, 1993) by them to each other (Sarasonet al. 1983) Chainedor snowball Descriptions of linked and Survey instruments and name overlapping personal networks and generatorstied to chained sampling the relationships betweenindividuals designs(palmore 1967) and the whole population drawn from snowball samples,random walk designs (KlovdahlI989) Full network Identification of relationships in a Relationship matrix, membershiplists, bounded community (Knoke and questions about relationships between Kuklinski, 1982) all member~of the co=unity (Wassermanand Faust, 1994)

~ Ethnographyand Network Analysis 213 PersonalNetworks naire, collects additional information from the informant's perspective about the relationships The personal network approach focuses on indi- betweenthe other people mentionedby the vidual informants and their personal relation- informant, as part of the informant's personal ships. The focus of this type of study is to network, These two approaches,combined or identify similarities and differencesin individual singly, answer many important questions about relationship environments. This is often called cultural conditions beyond the individual level.. ego-centerednetwork analysis. Each individual is assumedto exist in a structured social context. That context may have very similar effects for PersonalNetwork Questions individuals who have the same type of contex- tual environment, and be very different for indi- One of the easiestand most productive ways of viduals who have significant differences in their finding out about the social context of an in- personalnetWorks. Anthropologists (Bott 1971; formant is to ask a series of questions about Kapferer 1973; and other British social anthro- his relationships with the people around him pologists)who were studying urban systemspi- (McCallister and Fischer 1983). Personal net- oneeredthis approach. The approachesused in work analysis provides information from the personalnetWork analysis fit very nicely into the perspectiveof the personproviding the informa- " small community, ethnographic interview, key tion. This createsan important strength to this I( :;i j i; inforDiant format of ethnographic research approach and a weakness.Two" people who are ! designs(Johnson 1994: 135), and it works well in a relationship may have widely different views with participant observation conditions. This of the relationship. One may be in love with the approachallows the ethnographerto collect per- other, while the other may merely like the first :~j sonal netWorkdata both through interviews and person. Based on their perceptions,the assump- ' ;'\; by directly observingthe behavior of individuals tions that eachone makesand the decisionsthey in key social settings. make about ways to follow through with the :' Personal network analysis concentrates on relationship can be very different. The strength :' ! asking questions or recording observationsSoaboutindividualin the personal network approach is that the ., behaviors, attitudes, and be- questionsasked allow the researcherto discover ~Ji liefs. It is an attempt to introduce information the individual perspectiveof each person inter- fabout the context of individual lives into survey viewed. However, the weakness is that this approa'Chesto understanding culture and method cannot resolve differences of opinion society. Early on, the theoretical underpinning about a relationship. It can only identify those .came from the rational choice differences because it is focused on individual or structural functionalism, or a combi- perceptions and how those perceptions are llilll"ill,..l~"i of both these approaches (Galaskiewicz related to behavior, life events, or the predict- Jj , Wasserman 1993). These approaches ability of some type of social interaction, for that humans are actors in a larger social the individual. Within this limitation, the and that the actors can be assumedto approach can produce very important insights and independence.Individuals into individual lives. treated as solitary or relatively solitary Personal network questions identify an in- situations. This means that individual formant's social context from a structural per- spectiveand a role or meaning perspective.The on behavior or culture. However, structural questions focus on the size,shape, and personal relationships and interac- organizing principles of the person's relation- ships. These include information about the size describing and predicting patterns in of the person'spersonal network, the strengthof lives. The personal network approach her relationship to other irrdividuals, the close- to encompassinfluence patterns ness or intimacy of the relationships, and the overall shape of her connections to different I" !\ ~ to solitary conditions, and kinds of people as well as different individuals. ,\ , actions, in addition to independence The role or meaning questions ask about the cultural labels and meanings of the relation- network research approaches h~ve ships, such as kinship roles, gender roles, status two types of questions about questions,and the like. The combined questions are analyzed to produce a description of each n,etworks,just as informant's personal network. This information can then be aggregatedto detertnine if there are , age, preferences,or knowledge. patterns to personal networks that are closely approach, the egocentric question- related to the critical questions that the

iI~ 214 Handbook of Social Studiesin Health and Medicine

researcheris trying to uncover. For example, approacharises from the ways people remember doesthe sizeof an individual's personal network things. Memory is sometimesaffected by emo- have an important impact on his influence or tion, bias, and biological processes (injury, power in a communi~? Do people in different aging). Reporting is affected by all of these cultures have different averagesize networks or issuesplus socialpressures to hide embarrassing different networks on the basis of their composi- or harmful information or to lie about relation- tion (roles, demographics)?Are personal net- ships by bragging or exaggerating.Asking a per- works made up mostly of family members son who has had a traumatic experience to more supportive than ones that are made up remember that experience may produce only exclusively of friends and acquaintances? If partially accurate memory, and asking a person someonewanted to get a job, what is the best who is married to name all of his sex partners kind of personal network to have?Do the struc- may produce an incorrect list, due to lying. Yet, ture and the role relationships in a personalnet- there are times when this information is the only work determine the successof an individual in information that can be collected about impor- business,love, or longevity of life? Each of these tant relationships. questions can be answered by asking people This issue produced a very powerful discus- about their personalnetworks and then compar- sion and analysis of elements of informant ing those network answersto answersabout the accuracy (Johnson 1994: 122-7). The early critical life experiencesof the informant. experimental research(Bernard and Killworth 1973,1977; Killworth and Bernard 1~76, 1979- 80) called into question the general accuracy of Egocentric Network Questions self-report data when people's reports about their contactswere compared with actual obser- The egocentric appr9ach collects judgements vations (also Bernard et al. 1980, 1982a, 1982b, about relationships among the people in ego's 1984).As other scholarsconducted further ana- personal network, from the informant's per- lysis on the original data sets, or conducted spective. This changes the data from a view further experiments, it became clear that this of how the informant sees her relationship to complex question of accuracy had a number each person she identifies in her network, to her of answers,ranging from conditions that pro- perceptions of other people's relationships to duce very little accurate information, through one another and to herself. Some of the ques- conditions that produce accurate information if tions that could be answered with this certain biases are taken into account, to approach include the following. Do networks situations in which the aggregated information where everyone knows everyone else provide from inform-ants produces very accurate data better social support for ego than networks (cf. Johnson and Miller 1983; Romney and where the only connection among members is Faust 1983; Romney and Weller 1984). These directly to ego, not to anyone else? What effect studiesfound that the accuracy of an informant does dividing up your social life into several depends on the level or intensity of the inform- different tight subgroups (cliques) have on a ant's participation in a group, It also showed person's success in business, compared with tbat informants who have similar backgrounds having one homogeneous network? If the and experiencestend to exhibit regular or pre- people in ego's network are in conflict with dictable biases in accuracy depending on the each other, how does that affect the strength characteristicsof the informant (Freeman et al. or the longevity of their relationship with ego? 1987a,1987b). It also found a range of accuracy Are members of ego's networtc more likely to in any group of informants which correlated be connected with each other if they are the with the informant's ability to describea consen- same gender, ethnicity, social status, or religion, sual model of the relationships being studied and if so, what impac~does that have on ego's (Romney and Weller 1984; Romney et aI, success, happiness, or future health? How do 1986). Finally, researchersdiscovered that the personal networks form and how stable are ability of individuals to describe some social they over time? interactions, compared with behavioral observa- Asking informants about their relationships tions, produced the most accurate picture when without the cross check of asking the people it was analyzed by aggregating individual named about the same relationship raised a cri. reports, rather than looking at the reports singly tical question for anthropologists engaged in (Bernard et al. 1982a, 1982b; Freeman et aI, persoI;1al network analysis. The primary ques- 1988). This indicates that the most accurate tion, if this approach is used for any type of pictures of social interactions often come from critical decisions, is: 'are informants accurate an aggregate group view, rather than from an when they provide information about their rela- individual perspective.The related question of tioBships?' One potential problem with this how people remember,not just what they report, Ethnographyand Network Analysis 215 was also explored (Boster 1986a 1986b; Personal and Egocentric Data Collection Killworth and Bernard 1976), followed by explorations of the impact of an individual's Personal or ego-centerednetwork data is col-lected social positions on his recall and reporting of in the form of questionnaires or inter-views. social relationships (Boster et al. 1987; Individuals are asked a series of Michaelsonand Contractor 1992).These studies questions about the structure, composition, on the relation between social position in a net- and relationships they have with a defined per- work and reporting indicate that there are reg- sonal networLThe most common way that data ularities in perception of networks and are collected is to ask a set of questions about interactions based on similarities in the roles the size and composition of a personal network,to that people have in acoxnmunity. People'sposi- ask the person to list the namesof the people tion in socialnetworks influencesthe accuracyof in that network, and to answerquestions aboutthe their reports. Personal network data also show person's relationship with each of thosepeople that the networks reported by individuals, and the relationship between thosepeople whether they are accurately measured against and each other (cf. Bernard et al. 1990)(Figure another form of data or not, are a key condition 1). In many cases,the actual names are for understanding the social world of any given not asked becauseof ethical considerationsand informant, from that informant's perspective. legal issues,but each personal network memberis Theseportrayals are useful in both understand- identified by it label or nickname, so the per- ing and predicting the behavior of individuals sonSanswering the questions do not get con-fused beyond the information that an informant about whom they are describing,and can givesus about herself alone. easily talk about the relationships.

~ 216 Handbook of Social Studies in Health and Medicine

Analytical Strategies for Personaland curately to determine the impact of events that Egocentric Data affect everyoneby colle!;ting a random sample of data from informants iIn a society. The second There are four broad anal~ca1 strategiesused assumptionis that then~is regularity in personalows to describepersonal networks. The first provides networks; everyone kn a relatively narrowding composite views of the personal network char- range of people, deperu on the type of rela-tudied acteristics of different groups of people. The tionship that is being s (knows someone, most common descriptionsare th~ size,age com- has contact with someone, has someonein their position, gender composition, socioeconomic home). The result is that if you know the size of characteristics,and other demogr~phicsof the the population, know the average connections networks, compared by key subgroups,such as between people, and k now the average size of differences between the personal networks of personal networks for Ii particular relationship, different ethnic groups, gender differences,age, you can accurately estimate the impact ofrd or socioeconomic differences. This allows the a specific event (Berna: et al. 1989; Freeman researcherto deterinine if men or women have and Thompson 1989; Freeman et al. 1989; the sameaverage size of personal networks, or if Killworth et al. 1990).The reverse small world;onducted young people have networks that are predomi- experimentshave beenc to estimatethe:ted nantly only people their own age, or if the rich number of people infec with the HIV virus,'ho have much contact with the poor. The average the number of people w died in natural disas- profile of personal networks can be compared ters (suchas earthquak~~s,where the death toll isI with the profiles found in contrasting social so high that there is no way to get an accurate groups. account),and for events(such as rape) for which If there are significant differences in the per- the actual occurrence is under-reported for.)ns. sonal networks found in a researchpopulation, social and cultural reast then the researchercan create a typology of personal networks and can describe the simi- larities and differences in the cultural context Examples of Persol Network Studies of people's lives by comparing and contrasting the individuals who are involved in eachtype of The most common studies of personal networks personalnetwork. have been social support studies (Cohen and The data that describeseach person's relation- Syme 1985; Hays et al. 1990; Ostrow et al. ship with each other person can be analyzed to 1991; Williams 1993)and diffusion of innova- determine the structure of each relationship set. tion studies (Coleman et al. 1966; Palmore Some personal networks are dense and tightly 1967; Valente 1993, 1995, 1996). In recent connected,with everyoneknowing and interact- times, egocentric network studies have been ing with everyone else, while others are very very valuable in helping prevent AIDS and loosely connected or only connected by the other infectious diseases(Latkin 1995; Latkin informant. Thesestructural differencesare likely et al. 1996; Trotter et al. 1995a, 1995c, 1996). to changethe way that the informant lives, loves, The personal network studies for th¥se projects and takes care of himself and others. The struc- have focused on 'risk networks.' Risk networks tural analysis of networks, described in more are the personalnetworks of individuals who are detail below, includes connections among at risk for some kind of problem. The HIV risk people, the subgroups,and the overall structure networks of drug addicts include the sex part- of personal networks. ners and the needle-sharingpartners of the drug Becausepersonal network data are normally addicts. Both of thesegroups put the drug addict collected along with lots of other data about at risk of being infected by HIV. The personal individual informants, it is possible to correlate network studies in this area have been valuable the network data with other individual variables. for identifying the most common problems as- This allows questionsto be asked about the rela- sociated with HIV transmission in networks, tionship between personal networks and other and have provided evidence that it is possible life outcomes. One very interesting example of to changeboth individual and network behavior this approach combines personal network ana- and slow the AIDS epidemic (Needle et al. lysis and the concept of the small world experi- 1995). ment, describedin the next section. It is called a Ego-centered personal network information reverse small world experiment (Bernard et al. allows anthropologIsts to quickly establish 1989; Killworth and Bernard 1978). This som~of the important traits of the social context approach uses assumptions from several other found in a given culture. Thesetraits, suchas the forms of network analysis. It assumesthat every- size of networks, the closenessand duration of one is ultimately connected (small world reiationships,and the impact of peer norms, can phenomenon) and therefore it is possible ac- be collected from each person or a selected

~al

218 Handbook of Social Studi~Sin Health and Medicine able to show that the program recruitment order ment of each chain, and the ability to demon- data (the rank order in wlrich individuals were strate actual connections among the people recruited into the project for their network) not being studied. only correlated with network structure measures In general, the problems with this approach (how they were connectedor to which subgroup include running into a social box canyon and they belonged), but they were also related to problems generalizingto the rest of the culture. increased risk taking (Trotter et al. 1994, The box canyon effect occurs when the chains or 1995c, 1996) and correlated with higher risk snowballs start with too few points and become behavior. Early arrivals in each network were trapped in social loops, while the real relation- more likely to have tried a drug treatment pro- ships actually extend much further out into the gram than the subsequentrecruits in the same community. The problem of generalizability i~ network, while later arrivals were less likely to produced by the sampling methods that need have injection drug user sex partners. Those to be used for this type of research. If the recruited earlier in networks were very likely to respondents are discovered through a biased have sex partners who were also injection drug selectionprocess, or a process that only identi- users (that is, they participated in double-risk fies part of the full range of people who should relationships). be interviewed, then the researchercannot gen- We also hypothesized that participation in eralizeto the community as a whole. New meth- two or more networks involved more potential ods associatedwith probabilistic sampling risk and risk taking than membershipin a single (Wassermanand Faust 1994), new approaches network. Most of the people in the project (321) for generalizing from snowball samples (Frank only participated in one dnignetwork (66.5 per 1979;Snijders 1992),and sequenceddata collec- ceilt), while 162individuals (33.5 per cent)were tion (Klovdahl1989; Klovdahl et al. 1994)help members of two or more drug networks. Our avoid some of the common snowball pitfalls. data indicated that simply asking individuals to They also help to make the information more self identify either single or multiple network easilygeneralizable to the population as a whole. membership provided a direct indication of both their types of risks aI:ld the levels of risks that the individual was most likely engagedin, Chainedor Snowball Network Examples within their personal drug-using networks. As a result, this type of personal, or ego-centerednet- The small world experiment(poole and Kochen work data collection was very useful for reveal- 1978) is the classic example of a chained or ing important facts about the most common snowball type of social network research personal groups (network of individuals) in our approach.In this type of experiment,individuals projects. The same type of information could are asked to find the best way to get a message easily be collected for educational programs, (or an object) to a randomly chosen person eConomicdevelopment conditions, or participa- somewherein the world. The only things typi- tion in any other form of cultural programs that cally known about the target person are his would be enhanced by understanding the cul- name, the town he lives in, and his occupation. tural context of people's lives. The rules are that the person must pass on the messageto someonehe or she has personalcon- tact with, with the request to keep passing it along until it reaches the target person. Then, CHAINED OR SNOWBALLED NETWORKS the number of links betweenthe original person and the target person are counted and described. The average number of links betweenpeople is The second approach to network analysis was usedto define the difficulty of anyone personin created to move beyond the individual and to the world contacting any other person through attempt to study larger social structures in a direct socialconnections. The types of links used culture. In this approach,relationships are stud- (a friend in the town, an acquaintancewith the ied by following a trail of connections from each same occupation) allow the researcher to selected key informant outward irito the larger describe the type of strategies that work, and society. The approach is similar to a chain letter the ones that do not work, for carrying out the or to sno\\(ballsampling. Each person leads the linkage task. The experimentshave shown that a researcherto another person or a set of persons successfulattempt at this type or contact takes who are all connected by a particular kind of an averageof about five links within a specific relationship. There are several forms of this cultural or racial group, about six links if cul- sequential collection of data on relationships, tural or racial boundaries are being crossedin a eachwith its own advantagesand disadvantages. single society,and about sevenlinks if si~caPJ: The advantagesare the relative easeof recruit': international boundaries are being crossed. Ethnographyand Network Analysis 219 There is some variation in the number of links described by them (alters or personal network found, depending on the importance givento the members of those interviewed, some of whom message,but these approximate averagesappear will also be named in other networks). The to be fairly stable. The broken chains, ones that first persons can be called the seeds for the did not complete the linkage, also provide very chain. The people they identify are their per- important information about the size of the sonal networks. The person chosen to move world and the fact that while everyoneis theor- the chain or snowball out one level can be con- etically connected, there are lots of ways that sidered a link. Since each person interviewed people can be isolated or kept out of the picture. identifies a personal network, it is both possible K.inshipanalysis is another interesting form of and likely that the same person or persons will chained network analysis. The description of a be identified by more than one of the people who kinship network begins with a central character are interviewed. This means that even though (ego) and that person's kinship relationships. the data collection is moving from one person From that point, individuals connected to ego to the next, the overall data sets will have net- are interviewed (often sampled) and more con- work linkages between both the interviewed ~ectionsare added to the original, forming first a individuals and the alters that they name. The chain of relationships,and ultimately a model of analysis of these data can then show a large both individual and complete kinship systemsin number of connections between individuals in a community. Early attempts to create an over- the community, eventhough only a small num- lap between sociometric network analysis and ber of people were directly interviewed. Good kinship analysis were unsuccessful,but in more sampling strategies and statistical analysis recentstudies, the two types of approach to net- allow the researcherto describe a much larger work data have been both complementary and segmentof the community than just the people haveadded new dimensions to our understand- who were interviewed about their networks, ing of kinship networks and the impact of kin- ship on larger social networks (Barnes 1980; Rage and Harary 1991; Plattner 1978; Sampling and ChainedNetworks Schweizer 1988; Seidman and Foster 1978; White and Jorain 1992). Choosing the starting point or points for net- work studies (i.e., sampling)and then interview- ing all of the people named, or sampling from ChainedNetwork Data Collection the persons named by the network seed, has becomea critical consideration for both collect-

'The basic model for chained data collection is ing and analyzing all forms of network data,

.the same in each type of snowball or chained but especially chained or snowball data.

cnetwork.. One person, or a small number of per- Anthropologists have begun to pay increasing

('sons, is chosen as the starting point. This person attention to the selectionof ethnographic infor- , , . " can be chosen randomly from a commUnIty, or mants (cultural consultants) in their research

,'if"J",chosen because of some special characteristic (Johnson 1990; Werner and Schoepfle 1987).

",,;,:';;,];,'iliat makes him or her a good starting point Johnson (1994) points out that for many net-

';':'""",10r the study. These individuals are asked to

"c" c, ... work studies, the sampling frame must be ,,' 'name and describe their relationships WIth ;;lc"mdi- extended beyond individuals to include places,

victuals in thejr personal social network. In some events, and subgroups as points for beginning

",","""cases, they are asked to name everyone they can both observational network studies and inter-

i,f,"think of, and in other cases they are asked to views.

[)i!, ','name individuals with whom they have a specific

'~~;!r:",relationship or a specific set of relationships. The

~~"i"next person or persons to be interviewed can be Chained Network Analysis 'j',t*',either,- randomly selected from ego's original per- ;ti:n' "sonal network (random walk approach) or canjtl;':\J' The analysis of chained network data can be be nominated by the first person interviewed,'.j~{t(,,'based accomplished in a number of ways. It can

on their relationship to ego (snowball or '""'CC' r " _1 involve an attempt to discover how far the e.err", approac h)\,\iT!.!!;'\, . chains extend before they truncate. This analysis c;~,ijt4c,," Each person to be interviewed is chosen from provides information about how people are con- "\' "t~e personal network of the last person inter- nected in a community, and how far those con- "\\ ,,!Iewed. This process is followed for as long as nections go before they disappear, based on 8:

"",',need~d, allowing more and more people to be some critical kind of relationship. Through this ";,,' ,~ommated and then interviewed. The data set type of research,you could ask how far a com- ci""":,,,_~I~cludes a set of interviewed people (known munity health educator's influence extends, or "!,~.idlrect c, ," connection~) and a set of persons you could determine how health prottiotion

~ 220 Handbook of Social Stud,esin Health and Medicine

ideas or values diffuse through a social system. affected and chaining out to other people who Another analytical strategy attempts to discover have the same condition or problem. how the characteristics of a network chain change as the chain proceeds outward from its source. For example,if you start a chain with an active group of drug users,how far along a chain of connections do you have to go before the WHOLE NETWORK STUDIES impact of that behavior is no longer visible or felt by the people who are ultimately connected The third approach to network studies attempts with the first segmentof the chain? The small- to describe and analyze the reciprocal relation. world experimentindicates that everyoneis con- nected in the world, but it is also obvious that shipsamong all of the me~bers of a social group. Many of these studies have focused on small not every single line of relationships can or will extend from anyone individual to any other in communities or organizations. It takes time and effort to ask eachperson about her relationships the world. Sometimesthe connections run out, with every other person in the group. There is a or are deliberately chopped off by personsusing limit to patience that must be accommodatedby their role as a 'gatekeeper'to limit contact with these studies but, within that methodological the people they are protecting. This happens when parents prevent their children from seeing, boundary, there have been a large number of important discoveriesabout the nature of com. hearing, or reading things that they think are munities, businesses, self-help organizations, harmful, such as sex education material or vio- social clubs, and interlocking power groups lent television shows. Therefore, one line of that make this approach very valuable. research is to determine all of the different lengths of chains betweenpeople, based on spe- cific types of relationships, and to determine Flill Reciprocal Networks what causesthose chains to be maintained and to be broken. The researchquestions that are important for One of the interesting issuesthat can be stud- whole network studies require that everyone in ied through chain types of network researchis the network (or virtually everyone, ,given the how networks form, how stable they remain problems of real-life data collection) ,can be over time, and how they disintegrate or change. asked and can answer questions about every Severalresearchers have investigatedthe reasons other member of the network. Sometimes the why people get to know certain people and not answersare about relationships and sometimes others,and why they form network relationships about the absence of relationships, but both (Bernard et al. 1982a, 1982b; Killworth and kinds of information are needed.Once the rela- Bernard 1979). These and other studies include tionship questionshave beenasked of everyone, an analysis of cross-cultural variation in the size it is possible to identify a large number of con- and characteristics of networks, how people ditions that have an important impact on the respond to different kinds of network or rela- network. tionship questions (giving different answers for the sizeof network for the same individual), and attempts to estimate the size of personal net- Whole Network Questions works. This type of information would be very valuable in trying to understand why people join Whole network questionscan cover a wide vari- or leave self-help organizations, or participate in ety of contextualrelationships. Some of the ques- charitable groups that are trying to find cures for tions focus on person-to-person relationships. problems such as leukemia, cystic fibrosis, or They include asking people if they have physiCal geneticdisorders. ties, suchas sexualrelationships, or physicalcon- Another line of analysis is to use the chained tact. Other questions investigate emotional or or snowball data as an approximation of the social relationships by asking people how much distribution of critical problems or relationships they are influenced by or influence other peopl~'s in a total population. This is common in decisions, or how often they meet socially or researchon hidden populations and on rare con- communicate with one another. Whole network ditions such as HIV infection or specific genetic studies have significantly contributed to our disorders. Becauseit is hard to find all instances knowledge about the AIDS epidemic and how of a rare or hidden condition, snowball and to combat it (Needle et al. 1995).Whole network chained network approaches can provide data questions can also focus on geographical net- that are analyzed through statistical estimations works by asking everyoneif, or how often, they for the culture or community as a whole, finding visit locations (businesses,museums, organiZ- one or a small number of people who are ations) or are at the same social events (Darties,~ . Ethnographyand Network Analysis 221 visits), where the location is the focus for identi- Whole Network Data Collection fying the social context. Whole network analysis Techniques can also be extended to organizations or even societies.Instead of focusing on the individual, Whole network data can easily be collected in these whole network studies ask how organiz- the form of a matrix. Each person is given a ations are.related to one another. Some studies questionnaire or interviewed about his or her have looked at the overlap in the membershipof relationships with everyoneelse in the network. corporate boards of directors, and have shown In one of our studies about HIV transmission how these supposedly separate companies are risks, we asked people to fill out a matrix ques- actually interlocked systemswith advantages tionnaire about the people in their drug-using that individual corporations do not have. Other network. Figure 2 is an example of the questions studieshave $hownhow companiesthat produce we used in that study, and an example of the parts or services for other companies can be types of answers that were given. Respondents made more efficient and profitable depending answered the questions using a scale that was on the types and frequency of contacts between labeled appropriately for each question, with the organizations that all lead toward the com- zero being the 1ow end of the scale and five pletion of a final product. being the high end. The blank lines in the ques-

Jim Jenny Tom Marge

How honest is 0 5 1 3 3 with you?

How close a 2 5 4 3 friend is ?

If you had AIDS, 0 4 4 how willing would you be to tell?

How often do you 0 3 5 3 use drugs with _?

How comfortable 3 1 4 3 wou\d you feel discussing AIDS with _?

~~ ; figure 2 Ma_trix-stylequestionnaire about Drug and Trust Relationships

~ICilie 222 Handbook of Social Studiesin Health and Medicine

tion were used to indicate that the respondent was acting as a go-between,which is very com- should answer the question about each person mon in teenage relationships. This is typical representedby the names in the columns. whole network data, and when it is recorded as This form of data is then tr~sferredto a setof a matrix, can be analyzedto show a large num- matrices (one fot each question) for analysis. ber of important characteristicsabout the group. Each question (or relationship) matrix consists Matrix data has two important attributes of rows and columns that identify all of the people that are part of the data recording process. in the network. Each person is identified with The data can be either binary or valued, and both a row and a column. The square (cell) that it can be either symmetrical or asymmetrical. is the intersection of a row (one person) with a Binary data indicates only the presence or column (another person) is used to record the absenceof ~ relationship. It is recorded as a answerto the question given by the first person, one or a zero. Sometimes the questions that about the second person. The row and column are asked in network -analysis only focus on intersection betweenthe second person and the whether a relationship exists, not how often first is used to record the responsein the other or how forcefully something happens. The direction. Each final matrix representsa single alternative is to record a value for the relation- type of relationship, so if data is being collected ship. In the previous example, the value rep- on multiple relationshipslike the example above, resentsthe number of phone calls in a week. multiple final matrices are needed. Instead of numbers, the informants could have The data recording is easier to understand been asked to rate how important the call was with a simple example. If we wanted to under- (from unimportant to very important) or they stand the relationships between five teenage could have been asked to rank the five people friends, we might want to ask how often they in terms of the frequency they call the other call each other on the phone during the week. person. Each of these is recorded as a value, This would provide us with information about rather than a one or zero. The statistics th~t calling patterns that may also indicate other help analyze network structures often require characteristics of their relationships. The five one or the other type of data. The same is friends consist of two males and three females. true of symmetrical and asymmetrical data. They all know each other, but their social rela- Some relationships automatically involve both tionships are different, so they have different individuals, so the data are identical for the calling patterns. If we asked each person how pair (the top half and the bottom half of the often he or she had called each of the other matrix are identical). In other cases,like the four people in the past week, we might get a example above, the relationship exists between grid that looked like Table 3. In this example, the two people, but more things are happening Jim was asked 'how many times in the last week in one direction than in the other direction. Jim did you call Jenny?,' and then asked the same calls Jenny more than Jenny calls Jim, even question for the other three people. He called though they both participate in all of the tele- Jenny, who is his girlfriend, five times, Tom phone calls. This produces a matrix where the once, and did not call either Cille or Marge, number in any cell may be different from any who are very good friends of Jenny's. When other cell. This helps identify asymmetricalrela- Jenny was asked the same question for the tionships and conditions. For example, there other four, she had called Jim, Cille and Marge may be a million calls each day that go to (probably to talk about Jim), but not Tom, who New York City, and only a thousand that go is Cille's boyfriend. The three girls all called each to a small town in Arizona. This means that the other during the week, several times. In this structure of the phone network will need far example,Marge also called both boys, in addi- more connections into one place and fewer tion to her girlfriends. In both cases,she was into another. The same structural differences trying to help her friends get information, and happenIn social networks.

Table 3 Phone-callpatterns amongfive friends

Jim Jenny Tom Cille Marge Jim Jenny 0 Tom 0 Cille 0 0 Marge 0 Ethnographyand Network Analysis 223 Whole Network Data-Analysis Techniques' Each of thesetypes of analysis may be impor- tant in answering different kinds of questions The simple-looking data described in the last that researchersmay be asking about the rela- sectioncan produce very complex and interest- tionships within a community. For example,if a ing results. Network relationships can be ana- person is very central in a network, is that per- lyzed from several major perspectives. These son more likely to be infected by a communic- include an analysis of every person's interaction able diseasethat is moving through the network pattern, place,and role in the network, as well as than a person who is peripheral to the group? an analysis of the overall structure of the net- How long does it take tobacco prevention infor- work. mation to saturate a network from any given The first part of any network analysis is to starting point? If you can only talk to one person look at the relationships betweeneach individual in a network and you want that person to in the network and all other individuals. These change everyone else's eating behavior, who is relationships are often called actor interactions. the best person to talk to? If there are two fac- Formal descriptions of the way these network tions in a network, and you want to separate elementsare analyzed are available in methodo- them to keep them from fighting, which bridges logical books, such as Social Network Analysis in the network allow you to split the groups by (Wassermanand Faust 1994). These analytical eliminating the smallestnumber of connections? processescan describe how each actor in a net- If you have a community that is made up of work is connected with each other actor, how many different networks (such as teenage eachperson can or cannot be reached through gangs), are there certain positions or roles in other people, how much influence each person each of the networks that allow you to predict has in the network, and how central, powerful, that people in those positions will have very or well conne9teda person is. A second part of similar life conditions and be more like each the actor analysis allows the researcherto deter- other than they are like other members of their mine if. there are subgroups in the network, and gang?You might be able to predict, and prevent, fatalities by knowing the conditions attached to important impact on people's lives. each position in the gang. elements of network analysis are often One key strategy for analyzing a whole net- social cohesion measures, structural work data is to create both a visual and a statis- and regular equivalence tical display of the presenceand absenceof key The final analysis looks at the overall relationships. One of the two most common visual displays is a sociometric diagram where types of whole individuals (actors) are representedby symbols, suchas circles for femalesand squaresfor males, relationships,are shown in Table 4. and lines or arrows representthe relationship or

Examples of network measuresfrom matrix data sets5

Analyticalstrategy Examples of measures Graph connections Geodesicdistance, reach, flow, volume, walks, influence Centrality,power and prestige Degreecentrality, closeness measures centrality, betweenness,flow betweenness,information, Bonacich power Social cohesion Cliques, components, K-cores, n-cliques,n-clans, K-plexes, factions, connectivity sets Structural equivalence Profile similarity, tabu search, CONCOR, CATIJ, automorphic equivalence Regular equivalence Categorical, continuous, tabu search Structural measures transitivityDensity, network - centrality,

I~ 224 Handbook of Social Studi~Sin Health and Medicine

connection betweenthem. The thickness of the points out the need to understand multiple rela- line may representthe strength of the relation- tionships in a group, not just a single relation- ship (for valued data) and arrowheads may ship, if you want a full understanding of a represent the direction of the relationship (for group. directional or potentially as~etrical) data. The overall structure of the group is also The second most common display is to provide important when compared with the structure a data matrix that indicates the relationships by of other drug-using networks. In this case,the numbers in a person-by-personmatrix. Finally, comparative data showed that this is a tighter each type of relationship can be described by than normal network, due to the kinship ties accompanying stati~tics, including such ap- among members,and that it has a better struc- proaches as cluster analysis and multidimen- ture for solving social and drug-use problems sional scaling,as well as summary statistics. than networks that are.based on friendship and Figure 3 is an example of a sociometric repre- acquaintanceshipalone. It is more stable an,d sentation of the relationships described in a net- has resisted change longer than other types of work where the lines representsharing drugs. An drug networks. This condition makes this tyPe arrow indicates one way sharing, while a line of network harder to change, for HIV risk without a point indicates sharing in both direc- reduction, than networks where people can tions. The thickness of the lines representsthe more easily be isolated from the influence of frequenCyof sharing. their peers. In this diagram, Aida (#7) is a nonuser, clearly representedin her lack of connections with the group. Anita (#13) shares the primary influence in the drug network with her son, Marcos (#5), THE FUTURE OF NETWORK STUDIES and with Jaime (#9), who is central becausehe is a bilingual communication bridge between the Spanish (only) and the English (only) portions The future of network studies in anthropology, of the network. Marcos (#5) scoresdrugs for this and in the other social sciences,combines the network, keeps track of drug-related conditions, continual exploration of the questions asked and influencesthe network through his mother's above with some new areas for the expansion close connections with everyone else. The rela- of the paradigm. Anthropologists have not tionship chosen for this display was sharing come close to exhausting the limits of the three drugs. Other data we collected indicated that network paradigms (personal,chained, or whole #4 and #5, who are married to each other, are network approaches)in their studies,either from strongly tied by social relationships but are not a geographical or a subdisciplinary focus. connectedby drug use. The kinship ties between Medical anthropology, cognitive anthropology, #9 and #2 (living as married) are not visible in and economic anthropology have made the most the drug relationship diagram either, nor is the substantial uses of this approach (Johnson aunt-niece connection between #6 and #7. This 1994).However. political anthropology and edu- cational anthropology have negiectedthis area of investigation, as have anthropologists who are interested in expressionand symbol, in the (4) impact of different schools of anthropological theory on one another, and in the impact of 0 graduate programs on the structure and devel- opment of the disciple, to name just a few. There is a need to re-explore and provide further information on informant accuracy. 13 Policy decisions are being made on a daily basis, with global impact, based on what our informants tell us. While we know more than we did before, there are still areas of both mem- ory and reporting that make this information more vulnerable and less predictable than is desirable. The diffusion of information and innovation has held the anthropological imagination for ~ decades,and promises to have very important implications for future studies, at the personal Figure 3 Sociometricdiagram of relationships and at the whole network level. More needsto in a drug-usingnetwork be done to define the impact of networks o~

~ c ,c Ethnographyand Network Analysis 225 "'," contagious diseases(KlovdaW 1985), on social regeneration or their dissolution. What expecta- support for victims of all types of problems tions should someonehave in terms of knowing (Sarason et al. 1986), and for directing power, people as adults that they knew as children? influence,and information. This is an areawhere How many friends from grade school do you ethnography, mass media, and social network still talk to? When and how should organiza- theory can be combined for very powerful tions or corporations, cities, states, or nations findings. create alliances or change alliances, and when There is still the need to understand the rela- should they avoid them? tionship betweenanthropological kinship analy- Finally, there is II, very important unexplored sis and network analysis. The importance of area in social network research.This is the area understanding kinship systems in a global geo- of research ethics and the ethics of research. political economy has been highly underrated From one direction, we know very little about and needs to be revisited. The news is full o( the impact of socialnetworks on ethical conduct stories about developingcountries where kinship or ethical reasoning. This is an area that could plays the key role in distributing power and be explored combining network methods,ethical influence over important sectors of the national inquiry, and anthropological theory. An impor- economy, the national armed forces, and the tant parallel issueis the exploration of the ethi- national political system. The news is also full cal conditions that need to be applied to (sometimesonly on the society page) of indica- conducting network research.Most of our ethi- tions that the rich, famous, and powerful marry cal principles for research,such" as the rules for their children to people from their own social the formation of institutional review boards or networks, further controlling power and fame the Belmont Report (which sets out the basic in the next generation. The impact of kinship principles of ethical research in the United on social forces has a great deal of potential States),are based on assumptions that may be for further analysis. either violated by, or in conflict with, the kinds On the other hand, there are some gaps in our of relationships and interconnections that exist knowledge that should be closed or addressed, in social networks. One of the three principles of even though they do not have the historical researchethics is autonomy. Single individuals importance of those above. There is a great are assumedto be capable of deciding whether dealthat we do not know about the boundaries it is appropriate for them to participate in a of networks, or the impact of the borders of researchproject, based on their being informed nations. There has been a lengthy debate on about the nature of the researchand its risks the definition of the term 'community,' and all (informed consent). However, this autonomy is of its associated terms and processes. Social potentially violated at the network level, when network approaches can make a significant people are identified as part of a social network contribution not only to identifying rational and information is given about them by the boundaries, but also to showing how those other members of the network, even if they boundarie$change depending on the nature of themselvesrefuse to participate. Their participa- relationshipsand the level of analysis that is tion may not achieve full autonomy, due to the being undertaken. We are still struggling with influence of the network. The other primary conceptsof power, information sharing, shared " principles, justice and beneficence,can also be governance,and other forms of social interac- challenging to apply when researchis directed tion,.. Social network approaches should be at a group, rather than at individuals. These ~pplled to these critical, political, and philo- conditions raise questions such as, 'Who should sophicalconditions. make the decision for a community to partici- ;';A~o~herneglected area is understanding and pate in a vaCCinetrial, individuals who want the predlc~mgthe changes that occur in networks over time, from the micro-level per~nal net- vaCCine,or community leaders who have an appreciation of the needsand the vulnerability wo!k.to macro-networks of society and the of the corilmunity?' Who should benefit most wo!ld,. Evolutionary theory in anthropology from research on a group? Should it be indivi- P~tentlallyhas a great deal to contribute in this area, and the issues of studying change duals, the community, society at large, or the researchers?How dQ we make ethi<;:aichoices ~h.roughtime are central to the discipline. Most to conduct or not to conduct researchthat will ,?:tworksare analyzedon the basis of a one-time benefit societyas a whole, but willbe destructive Pl.Cture,even though we know that our personal to individuals or to small groups? How do we m'I' f~~nds have changed over time, just as global:,~)ance~ fJj, "'" make judgements about researchthat will poten- i :"I.!! "'" , have changed. We need to know the~~echanlsmstially harm a community (destroy its positive that cauSethe formation .of net- reputation, create stereotypes)but will be ex- ",works.co", all ow them t0 be sta b"'-llUe d an d mam-. ~I:;" c.~med , and cause t helr. dlsruption., and.W;,;tremely beneficial to all of the individuals in ~.co" the community who need certain services or

~ 226 Handbook of Social Studies in Health and Medicine

ne~d to know the dangers that confront them? Triadic Methods,and in Part VI, Statistical Dyadic Bott, E How do we handle the ethical dilemmas posed Interaction Models,by Wassermanand Faust (1994). Nor" by research that is viewed as positive, needed, Urba and wanted by one subgroup in a culture, and Burt, ] viewed as negative, unwanted, and unneeded ;"'l' AmuJ by another subgroup? There have been a ~;lf;i Burt, R number of studies that have benefited males ,l_~-REFERENCES SUfV' but not females, one cultural group over Burt, F another, and so on. Research, reporting, eval- ; uating, and defining ethics beyond the level JBarnes,J. (1980) 'Kinship studies: Some impressions Anal of the current state of play', Man, 15: 293-303. Cohen, of the individual has been seriously under- and, representedin the literature and deserves Barnes,J.A. and Harai-y, F. (1983) 'Graph theory in network analysis', Social Networks,5: 235-44. Colem. attention in the future along with the other Bernard, H.R. and Killworth,'-"P.D. (1973) 'On the l\1ed questions raised above. social structure of an ocean-going researchvessel Bob! and other important things', Social Science Faust, Research,2: 145-84. anal Bernard, H.R. and Killworth, P.D. (1977) 'Informant Net~ accuracy in social network data II', Human Frank, CommunicationsResearch, 4: 3-18. " use Lein Bernard, H;R. Killworth, P.D., and Sailer, L. (1980) Rest 1 Funding for some of the researchreported in this 'Infonnant accuracy in social network data. IV. A Freem chapterwas provided by two grants from the National comparison of clique-level structure in behavioral Con Institute on Drug Abuse (NIDA). These two grants and cognitive network data', Social Networks, 2: 39. were Grant #UOI-DA07295, the Flagstaff Multi- 191-218. Freem cultural AIDS Prevention Project, and Grant #ROI Bernard, H.R., Killworth, P.D., and McCarty, C. 'Est DA09965, Small Town Drug Networks a~d HIV: (1982a) 'Index: An infonnant defined experiment Koc TransmissionDynamics. in social structure', Social Forces,61: 9~133. cAb/. 2 There are severalreviews of social netwqrk theory Bernard, H.R., Killworth, P.D., and Sailer,L. (1982b) Freem that are important historical accounts of the theories 'Infonnant accuracy in social network data. V. An experimental attempt to predict actual co=unica- (19! and methods used in these studies and which form the HilI backdrop for this chapter, but are not reviewedhere. tion from recall data', Social ScienceResearch, II: Freen These include: Boissevain 1974; Burt 1980; Burt and 30-66. (19~ Minor 1983; Harary 1965; Johnson 1994;Knoke and Bernard, H.R., Killworth, P.D., Sailer, L.. and rac~ Kuklinski 1982; Marsden 1990; Marsden and Lin Kronenfeld,.D. (1984)'On the validity of retrospec- Freen 1982; Rogers and Kincaid 1981; Sl;ott 1991; tive data', Annual Review ofAnthropology, 13: 495- (19: Wassermanand Faust 1994; Wellman and Berkowitz 517; 1988. ' Soc Bernard, H.R., Johnsen, E.C., Killworth, P.D., and Freen Robinson, S. (1989)'Estimating the size of an aver- 3 Some of the basics are reviewed or identified by (19 Freeman (1979), and details are provided in age personal network and of an event subpopula- the Wassermanand Faust (1994). While not close to an tion', in M. Kochen (ed.), The Small World. An. exhaustive list, some of the seminal approaches that Norwood, NJ: Ablex. pp. 15~75. Gala! are of interest in anthropology, sqciology,psychology, Bernard, H.R., Johnsen, E.C., Killworth, P.D., wo and political science are discussedin: Barnes 1980; Shelley, G.A., McCarty, C., and Robinson, S. tio Barnes and Harary, 1983; Bernard et al. 1990; Faust (1990) 'Comparing four different methods of meas- Re. 1988; Freeman et al. 1989; Hage and Harary 1983, uringpersonal social networks', Social Networks,12: Hage 1991; Hammer 1980; Johnson 1986,1994; Johnson et 17~215. ~ An al. 1989; Wellman 1988. Boissevain, J. (1974) Frien~ of Friends: Networks. Hage 4 In order to participate, individuals had to be 18 Manipulators and Coalitions... Oxford: Basil A, years of age or older, could not have beenin treatment Blackwell. Ham in the past 12 months, and had to havea positive urine Boster, J.S. (1986a) 'Selection for perceptual distinc- tio screenfor cocaine or heroin, or needle marks (tracks) tiveness: Evidence from Aguaruna Jivaro varieties Ham and a positive urine screenfor some other irijectable of Manihotesculenta',Economic Botany, 39:310-25. ne illicit drug. Therefore, this category includes only 18- Boster, J.s. (1986b)'Exchange of varieties and infor- Sc and 19-year-olds. mation between Aguaruna manioc cultivators'; Har. 5 This table, and the accompanyingchapter, ignores American Anthropologist,88: 42~36. dyadic and triadic network methods..These methods Boster, l.S., Johnson, J.C., and Weller, S. (198'7) Hay are rarely used in anthropology, although they do 'Social position and shared knowledge: Actors' pc;r: so form a well-establishedarea of social network analysis. ception of status role and social structure', Socia[ These methods are discussedin detail in Part V, Dyadic Networks,9: 375-87.

NOTESand Ethnographyana r Network Analysis 227 Bott, E. (1971) Family and Social Network: Roles, Johnson,J.C. (1986)'Social netWorksand innovation Norms, and External Relationships in Ordinary adoption: A look at Burt's use of structural equiva- Urban Families.London: Tavistock. lence',Social Networks,8: 343-64. Burt, R.S. (1980) 'Models of network structure', Johnson, J.C. (1990) Selecting Ethnographic In- Annual Review of Sociology,6: 79-141. formants. Newbury Park, CA: Sage. Burt, R. (1984)'Network items and the general social Johnson, J.C. (1994) 'Anthropological contributions survey', Social Networks,6: 293-339. to the study of social netWorks: A review', in S. Burt, R. and Minor, J. (1983) (eds) Applied Network WassemIanand J. Galackiewicz (eds),Advances in Analysis. Beverly Hills, CA: Sage. Social Network Analysis. Beverly Hills: Sage. pp. Cohen,S. and Syme, S.L. (eds) (1985)Social Support 113-47. and Health. San Francisco, CA: Academic Press. Johnson,J.C. and Miller, M.L. (1983) 'Deviant posi- Coleman, J.S., Katz, E., and Menzel, H. (1966) tions in small groups: The relation betWeenrole and Medical Innovation: A DiffUsion Study. New York: individual', Social Networks,5: 51-69. Bobbs Merrill. Johnson, J.C., Boster, J.S., and Holbert, D. (1989) Faust,K. (1988)'Comparison of methods of positional 'Estimating relational attributes from snowbalI analysis:Structural and generalequivalences', Social samples thrO~gh simulation', Social Networks, 11: Networks,10:313-41. 135-58. Frank, O. (1979) 'Estimation of population totals by Kapferer, B. (1973) 'Social network and conjugal role use of snowball samples', in P. Holland and S. in urban Zambia: Toward a refomIulation of the Leinhards (eds), Perspectives on Social Network Bott hypothesis', in J. Boi~sevainand J.C. Mitchell Research.New York: Academic Press.pp. 31H7. (eds), Network Analysis: Studies in Human Freeman, L. (1979) 'Centrality in social networks: Interaction. The Hague: Mouton. pp. 83--110. Conceptual clarification', Social Networks, 1: 215- Killworth, P.D. and Bernard, H.R. (1976) 'Informant 39. accuracy in social netWork data', Human Freeman, L.C. and Thompson, C.R. (1989) Organization,35; 269-86. 'Estimating acquaintanceship volume,' in M. Killworth, P.D. and Bernard, H.R. (1978) 'The rever- Kochen (ed.), 11Ie Small World. Norwood, NJ: sal sn;lal1-worldexperiment', Social Networks, I: Ablex.pp. 147-58. 159-92. Freeman, L.C., Romney, A.K., and Freeman, S.C. Killworth, P.D. and Bernard, H.R. (1979)'A pseudo- (1987a) 'Words, deeds and social structure', model c;>fthe small world problem', Social Forces, Human Organization,201-19. 58: 477-505. Freeman, L.C., Romney, A.K., and Fre~man, S.C. Killwc;>rth, P.D. and Bernard, H.R. (1979-1980) (1987b);Cognitive structure and informant accu- 'Informant accuracy in social network data. III. A racy', American Anthropologist,89:310-25. comparison of triadic structures in behavior and Freeman,L.C., Freeman, S.C., and Michaelson, A.G. cognitive data',Social Networks,2: 19-46. '{1988) 'pn human social intelligence', Journal of Killworth, P.D. and Bernard, H.R. (1982) 'A techui- Socialarid Biological Structure, 11: 415-25. que for comparing mental maps', Social Networks, Freeman,L.C., Freeman, S.C., and Michaelson, A. 3: 307-12. (1989) 'How humans see social groups: A test of the Sailer-Gaulin models', Journal of Quantitative Killworth, P.D., Johnsen,E.C., Bernard, H.R., Shelly, G.A., and McCarty, C. (1990) 'Estimating the size Anthropology,I: 229-38. of personal netWorks',Social Networks,12: 289-312. Galaskiewicz,J.and Wasserman,S. (1993) 'Social net- Klovdahl, A.S. (1985) 'Social netWorksaqd the spread work analysis: Concepts, methodology, and direc- of infectious diseases:The AIDS example', Social tions for the 1990s', Sociological Methods & ScienceMedicine, 21: 1203--16. Research.221: 3-22. Klovdahl, A.S. (1989) 'Urban social netWorks: Some Hage,P, and Harary, F. (1983) Structural Models in methodological problems and possibilities', in M. ,Anthropology.London: Cambriage University Press. Kochen (ed.), The Small World. Norwood, NJ: Hage,P. and Harary, F. (1991) Exchangein Oceania: Ablex. A Grap"h '7'1-neoreticAnalysis. New York: Clarendon. Hammer,M. (1980) 'Predictability of social connec- Klovdahl, A.S., Potterat, J.J., Woodhouse, D.E., ,;ionsover time', Social Networks,2: 165-80. Muth, J.B., Muth, S.Q., and Darrow, W.W. (1994) }{ammer,M. (1983) , "Core" and "extended" social 'Social networks and infectious disease: The "networks in relation to health and illness' Social". Colorado Springs study', Social Science and " Medicine,38: 79-88. 'H~"SclenceandMedicine ,. 17. 405-22. Knoke, D. and Kuklinski, J.H. (1982) Network ,a!ary, F., Norman, R.Z., and Cartwright, D. (1965) !~c$lructuralModels. New York: Wiley. Analysis. Newbury Park, CA: Sage. Latkin, C.A. (1995)'A PersonalNetwork Approach to , .--:, S., and Tobey, L.A. (199(» 'The AIDS Prevention', in R.H. Needle, S..L. Coil, S. I jOclal support networks of gay men with AIDS', G~~r, and R.T. Trotter, II (eds), Social Network ,~~.'nal of CommunityPsychology, 18: 374--85. , Analysis, HIV Prevention and Drug Abuse. NIDA

~ 228 Handbook of Social Studiesin Health and Medicine

Monograph 151, Bethesda,MD: National Institute The social support questionnaire', Journal of on Drug Abuse. pp. 181-95. Personalityand Social Psychology,44: 127-39. Latkin, C.A., Mandell, W.;V1ahov, D., Oziemkowska, Sarason, I., Sarason, B., and Shearin, E.N. (1986) M., and Celentano,D.D. (1996)'The long-tenn out- 'Social support as an individual difference variable' come of a personal network-oriented HIV preven- Journal of Personalityand Social Psychology, 50; tion intervention!or injection drug users:The SAFE 84s-'55. study', American Journal of CommunityPsychology, Schweiz;er,T. (1988) 'Detecting positions in networks: 24: 341-64. A formal analysis of loose social structure in rural Marsden, P.V. (1990) 'Network data and measure- Java', American Anthropologist,90: 944-53. ment', Annual Review of Sociology,16: 435-63. Scott, J. (1991) Network Analysis: A Handbook. Marsden,P. V. (1993)'Reliability measurement',Social Newbury Park, CA: Sage. Networks, 16: 435,.-63. Seidman,S.B. and Foster, B.L. (1978a)'A note on the Marsden, P.V. and Lin, N. (eds) (1982) Social potential for genuine cross-fertilization between Structure and Network Analysis. Newbury Park, anthropology1: 6s-'72. and mathematics', Social Networks' CA: Sage. McCallister, L. and Fischer, C. (1983) 'A procedure Seidn;1an,S.and Foster, B. (1978b) 'A graph theoretic for surveying personal networks', in R. Burt and J. generalization of the clique concept', Journal of Minor (eds), Applied Netw~rk Analysis. Beverly Mathematical Sociology,6: 139--54. Hills, CA: Sage. Shelley,G.A. (1992) The social networks of people Michaelson, A. and Contractor, N.S. (1992) with end-state renal disease:Comparing hemodialy- 'Structural position and perceived similarity', sis and peritoneal dialysis patients. Unpublished Social PsychologyQuarterly, 55: 300--10. Doctoral Dissertation, University of Florida, Morrill, C. (1991) 'The customs of conflict manage- Department of Anthropology. ment among corporate executives', American Snijders, T.A.B. (1992) 'Estimation on the basis of Anthropologist,93: 871-93. snowball samples: How to weight', Bulletin dce Needle, R.H., Coyle, S.L., Genser,S.G., Trotter,R.T., Methodologie Sociologique,36: 59--70. II (eds) (1995) Social Networks, Drug Abuse, and Trotter, R T. II, Potter, J. M., Bowen, A. M., and HIV Transmission. NIDA Research Monograph Jiron, D. (Tradduction Arturo Ortiz) (1994) Series,No. 151. Enfoques Etnograficos y analisis de !as redes Ostrow, D.G., Whitaker, R.E.D., Frasier, K., Cohen, sociales,para la creacionde programs de prevencion C., Wan, J., Frank, C., and Fisher, E. (1991)'Racial del uso de drogas y de VIH, en ursarios activos. differences in social support and mental health in In Las .Adicciones: Hacia un Enfoque men with HIV infection: A pilot study', AIDS Multidisciplinario. Consejo Nacional contra las Care,3: 55,.-62. Adicciones. Secrataria de salud, Mexico. Mexico, Palmore, J.A. (1967) 'The Chicago snowball: A study DF: pp. 4s-'54. of the flow and diffusion of family planning in- Trotter, R.T. II, Bowen, A.M., and Potter, J.M. fonnation', in D.J. Bogue (ed.), Sociological (1995a)'Network models for HIV outreach and pre- Contributions to Family Planning Research. vention programs for drug users', in R.H. Needle, Chicago: University of Chicago Press.pp. 272--63. S.L. Colle, S. Genser, and R.T. Trotter, II (eds), Pasteruak,B. (1976)Introduction to Kinship and Social Social Network Analysis, HIV Preventionand Drug Organization.Englewood Cliffs, NJ: PrenticeHall. Abuse. NIDA Monograph 151, Bethesda, MD. Plattner, S.M. (1978) 'Occupation and kinship in a National Institute on Drug Abuse. pp. 144-80. developing society', Human Organization,37: 77-83. Trotter, R.T., II, Rothenberg, R.B., and Coyle, S. Poole, I. De S. and Kochen, M. (1978) 'Contracts and (1995b)'Drug abuseand HIV prevention research: influence', Social Networks, I: 5-51. Expanding paradigms and network contributions to Rogers, E.M. and Kincaid, D.L. (1981) Com- risk reduction', Connections,18: 29--46: munication Networks: A New Paradigm for Trotter, R.T. II, Baldwin, J.A., and Bowen, A.M. Research. New York: Free Press" - (1995c) 'Network structure and proxy network Romney, A.K. and Faust, K. (1983) 'Predicting the measuresof HIV, drug and incarcerationrisks for structure of a co=unications network from active drug users'..Connections, 18: 89--104. recalled data', Social Networks,4: 285-304. Trotter, R.T., Bowen, A.M., Baldwin, J.A.,and Price, Romney, A.K. and Weller, S.C, (1984) 'Predicting L.J. (1996) 'The efficacy of network-based HIVf infonnant accuracy from patterns of recall among AIDS risk reduction programs in midsized towns individuals', Social Networks,4: 59-77. in the United States', Journal of Drug Issues,26: Romney, A.K., Weller, S.C., and Batchelder, W.H. 591-606. (1986)'Culture as a consensus:A theory of culture Valente, T.W. (1993) 'Diffusion of innovations and and infonnant accuracy', American Anthropologist, policy decision.making', Journal.ofCommunication, 88: 313-38. 43: 30-45. Sarason, I.G., Leveine, H.M., Basham, R.B., and Valente,T.W. (199?)Network Models of the Diffilsion Sarason, B.R. (1983) 'Assessing social support: of Innovatians.Cresskill, NJ: Hampton.

~ Ethnographyand Network Analysis 229

c T.W. (1996) 'Social network thresholds in Structures: A Network Approach. Cambridge: the diffusion of innovations', Social Networks,18: Cambridge University Press. 69-89. Werner, O. and Schoepfie,G.M. (1987) Systematic --S. and Faust, K. (1994)Social Netlvorks Fielmvork (2 ¥ols). Newbury Park, CA: Sage. Analysis: Methods and Applications. Cambridge: White, D.R. and Jorian, P. (1992)'Representing and Cambridge University Press. computing kinship: A new approach', Current Anthropology,33: 454--63. and metaphor to theory and substance', in B. Williams, H.A. (1993).A comparison of social support Wellman and S.D. Berkowitz (eds), Social and social networks of black parents and white par- : A Network Approach. Cambridge: ents with chronically ill children', Social Scienceand Medicine,12: 150~20. B. and Berkowitz,S.D. (eds)(1988) Social

~