Home

Social Science Computer Review

1. White et al KINSHIP AND MARRIAGE NETWORKS 269 into separate genealogies for each of the lineages following a chosen line of descent and the entire set of genealogical connections represented by lines of intermarriage that occur on the same page together with identification numbers for the parents in law of spouses whose families of origin can be found on a separate page White and Johansen 1998 use this method to print the entire labeled genealogy of the 250 relinked marriages shown in Figure 5 for the clan of nomadic Turkish pastoralists Another feature of Pgraph is the capability to write to different types of output files DOT graphics and UCI link lists Pgraph has two additional features that have not as of this writing been incorporated into the repertoire of Pajek algorithms One is the option noted earlier to organize genealogical data by lineages A second feature is an algorithm for reorder lineages to maximize any exist ing tendency for marriages to occur between opposing sides of a marriage network as in sys tems of dual matrimonial organization Houseman amp White 1998 in press Par Side is the accompanying auxiliary program for evaluating sidedness by calculating the probability of observed tendencies toward dual organization bipartiteness of the graph given the null hypothesis Doreian and Mrvar 1996a 1996b also provide an algorithm for finding the best approximation to bipartiteness in a graph The most important
2. 1998 UCINET 5 0 for Windows 95 NT Nantick MA Analytic Technologies Brudner L A amp White D R 1997 Class property and structural endogamy Visualizing networked histories Theory and Society 25 161 208 Doreian P amp Mrvar A 1996a A partitioning approach to structural balance Social Networks 18 149 168 Doreian P amp Mrvar A 1996b Structural balance and partitioning signed graphs In A Ferligoj amp A Kramberger Eds Developments in data analysis Metodoloski zvezki 12 pp 195 208 Ljubljana Slovenia FDV Foster B L amp Seidman S B 1979 Network structure and the kinship perspective American Ethnologist 8 329 355 Foster B L amp Seidman S B 1989 A formal unification of anthropological kinship and social network methods In L C Freeman D R White amp A K Romney Eds Research methods in social network analysis pp 41 59 Fairfax VA George Mason University Press Freeman L C 1998a Using available graph theoretic or molecular programs in social network analysis Online Available http tarski ss uci edu new html Freeman L C 1998b Using molecular modeling software in social networks A practicum Online Available http eclectic ss uci edu lin chem html Freeman L C Webster C M amp Kirke D M 1998 Exploring social structure using dynamic three dimensional color images Social Networks 20 109 118 Online Available htt
3. in the time required for computation White 1997 and Brudner and White 1997 developed both the vocabulary and our sociological understanding of the substantive relevance of identifying bicomponents as bounded subgraphs within kinship networks They defined the populations in those mar riages contained within bicomponents as bounded subgraphs or social units with structural endogamy Cohesion Problem 2 Structural endogamy has substantive implications for the second of our theoretical prob lems What are the socially cohesive subgraphs of a kinship and marriage network The social implication of structural endogamy through network links within a set of living per sons is that information or any other resource that might flow through these links is able to travel from anyone to anyone else in the set through multiple independent links This enhances the conditions for cultural sharing for the cross checking of information and enforcement of sanctions and for the maintenance of social boundaries even where cycles are of considerable length White and Harary 1998 carry these ideas further to characterize the concept of social cohesion for social networks with the following insight Because no cohesive group should 254 SOCIAL SCIENCE COMPUTER REVIEW contain a cutpoint whose removal leads to the disconnection of the cohesive group into two parts cohesive groups must at a minimum have the graph theoretic structure of bicompo nents Bi
4. will result in a new object shown in the Network window containing only the network selected Draw Draw Partition now will show this reduced network with different colors assigned according to the shortest distance to or from other connected nodes Organizing Visual Network Displays The Layers macro used to create generational levels and draw a genealogy is equivalent to two sets of options The first set Net Partition Depth Genealogy causes a descriptor to appear in the Partition object window for a partition file containing the generation level of each vertex Any such object appearing in a window may be saved to a file At this point White et al KINSHIP AND MARRIAGE NETWORKS 261 Figure 4 A Pajek Screen for the Schmidt Family Info Partition will show the number of vertices in each generation The second set of options Draw Draw Partition initiates the actual drawing of the current network according to the generational levels computed A two dimensional 2D graphic such as in Figure 4 is drawn by a slight variant from the Layers1 macro The options used are Net Partitions Depth Acyclic acyclic is the variant and Draw Draw parti
5. acommon parental root to link their multiple marriage vertices 8 Which sex is assigned solid versus dotted or other types of arcs and the coloring of arcs is a matter of con vention and may vary from study to study depending on what the researcher wishes to emphasize The default for both Pgraph and Pajek is to assign solid arcs to male links and dotted arcs to female links which is the reverse of Figure 1 9 The connectivity of a regular genealogy with a maximum of two parents for each individual can only be 0 several weak components 1 weakly connected or 2 2 connected 10 In egocentric representations there are several types of structural ambiguities such as how to represent mar riage whether the existence of children necessarily implies the presence of a marriage or pseudo marriage link and whether comparable definitions of marriage in different societies are being used definitions that can have artifac tual effects on the analysis In the P graph these problems do not occur because they do not affect the structure of the graph rather they merely affect what types of cultural labels are placed on the vertices where marriages concubi nage informal sexual liaisons and the like may occur 11 Development of the Pgraph program package was supported by a 1993 1995 National Science Foundation grant Network Analysis of Kinship Social Transmission and Exchange Cooperative Research at UCI UNI Cologne and CNRS Paris t
6. and original of Pgraph s algorithms is a random simulation option that keeps sibling sets intact but within each generation reassigns each sibling to a randomly chosen marriage partner of the same generation By saving several sets of randomized data any observed marriage regime can be compared to random baseline data a task that is done by the Par Bloc White amp Skyhorse in press auxiliary program CONCLUSION Possibilities for the study of kinship and marriage networks are radically transformed by the evolution of software for large network analysis both in contemporary sociological con texts Brudner amp White 1997 and in traditional anthropological contexts Schweizer amp White 1998 Ours is not the first team of anthropologists and graph theorists see also Foster amp Seidman 1979 1989 Hage amp Harary 1983 1991 1996 Seidman amp Foster 1978 to strug gle with large scale anthropological network analysis but our particular trio including com puter scientists has worked to create program packages both to analyze and to visualize net worked relationships On the computer science side we address the problems of fast algorithms for large networks combining different types of objects of study incorporating the fundaments of graph theoretic analysis and aesthetics of 2D and 3D graph drawing and organizing visual displays When principles of design replicate principles of thought the act of arranging info
7. available at links that begin at http eclectic ss uci edu pgraph An introduction to P graph analysis is found at http eclectic ss uci edu knhe str endo htm GIM Genealogical Information Manager by D Blaine Wasden and Brian C Madsen using GEDCOM format http www mindspring com dblaine gimhome html1 For documentation on the GEDCOM standard by the LDS Church see http www gendex com gedcom55 55gcint htm KineMage protein display and viewing language developed by David Richardson For introductions see http www elsevier com locate son and articles by Linton C Freeman and others at http tarski ss uci edu new html and http eclectic ss uci edu lin chem html as well as a dynamic sample page for kinship images at http eclectic ss uci edu drwhite pgraph mage html Kith and Kin commercial software by SpanSoft for P graph style genealogies available on a trial basis http www rocketdownload com details home kithkin htm Pajek program for large networks analysis by Vladimir Batagelj and Andrej Mrvar at the University of Ljubljana http vlado fmf uni lj si pub networks pajek See http vlado White et al KINSHIP AND MARRIAGE NETWORKS 271 fmf uni lj si pub networks pajek pajekman htm for the manual An introduction to draw ing genealogies with Pajek is found at http vlado fmf uni 1lj si pub networks doc sitges pdf For a description of the program see http vlado fmf uni 1j si pub networks pajek pajek p
8. densities for kinship and neighborhood Unpublished manuscript University of California Irvine White D R amp Harary F 1998 Cohesiveness measurements for social groups Unpublished manuscript Univer sity of California Irvine White D R amp Johansen U 1998 Social anatomy of a nomadic clan An anthropological introduction to net worked histories Unpublished manuscript Institute of Ethnology University of Cologne Germany White D R amp Jorion P 1992 Representing and analyzing kinship A network approach Current Anthropology 33 454 462 White D R amp Jorion P 1996 Kinship networks and discrete structure theory Applications and implications Social Networks 18 267 314 White D R Schnegg M amp Brudner L A in press Multiple connectivity bounded cohesion and cross cutting integration Kinship and compadrazgo in rural Tlaxcala In J Gil amp S Schmidt Eds Social networks Theory and applications Mexico D F UNAM Press White D R and Skyhorse P in press Parent Suite User s manual for analysis of kinship and marriage net works In V Burton T Finnigan amp D Herr Eds Multimedia Renaissance in social science computing CD ROM Urbana University of Illinois Online Available http eclectic ss uci edu drwhite pgraph p graphs html Douglas R White is a member of the Social Networks graduate program the Institute of Mathematical Behavioral Scien
9. example is implemented to run very quickly Pajek Menus Pajek s main menu contains pull down options at the top of a screen that is largely com posed of six small windows for showing which of the basic types of objects currently are in use in the analysis Because these screens initially are blank the place to begin is by choosing to read a network file in either NET or GED format in the uppermost window To the left White et al KINSHIP AND MARRIAGE NETWORKS 259 Finished reading input file Time spent le lines read Figure 3 Screen Snap of Data Objects in Pajek s Main Menu of each window are speed buttons for reading saving and editing of files Figure 3 illustrates a screen snap after the user has clicked open the file PPDEM NET of our illustrative input data When a speed button is clicked for opening a file a directory of files in the current direc tory with NET extensions i e network files appears To read a GED file one must click the option to change the data type and then click GED The GEDCOM files created by Ego2Cpl will appear if they are in the current directory and one may be clicked to open it Once a network file is read the options in the upper menu can be used Fil Net Nets Operations Partition Partitions Permut Cluster Hierarchy Vector Vectors Options Draw Macro Info Options Versus Macros Each of Pajek s menu choices activates submenu tre
10. family number even to unmarried children or to those individuals e g spinsters bachelors who are not recorded as having any conjoint relationships When two people marry their conjoint unit is coded under a common FAMS number This double system for identifying the fundamental units of kinship individuals 248 SOCIAL SCIENCE COMPUTER REVIEW and families leads to a double system for coding kinship relations In the ego centered sys tem the kinship relations are between identified individuals In the family centered system or P graph the kinship relations are between families Among the many commercial programs for drawing genealogies there is only one that uses the family centered system with couples at the vertices of the graph and it is not well suited for network analysis of marriages Data Transformation From Individuals to Couples Ego2Cpl The Ego2Cpl program Ego and Cpl stand for individual vs couple or family numbers is the entry level database engine of the Pgraph system White amp Skyhorse in press origi nally created by White and Jorion 1992 henceforth we use Pgraph for a software package as opposed to the P graph genealogy drawing format It converts from the numbering of indi viduals such as in Table 1 to family FAMS FAMC numbering for Pgraph and GEDCOM files Table is typical of formats used in computer files in which kinship data are collected As noted here Variables 1 to 3 are the ego identifier n
11. in the case of Individual 2 female Note that the line for Female 3 is a demiarc Harary 1971 because it does not link to a set of parents the identities of her parents are unknown Pairs of demiarcs also have been drawn for Families F4 and F2 one for the husband and one for the wife in each case An individual with plural marriages such as Individual 1 will have separate arcs for each mar riage as is the case for Individual 1 in the figure SOLVING FIVE KEY PROBLEMS OF KINSHIP NETWORKS USING P GRAPHS For the study of our five theoretical problems bounded subgraphs cohesion size social relations and groups the P graph conventions used in Figure 1 have considerable advan tages over conventional network diagrams To see these advantages we have only to break the expected Western paradigm of kinship as a tree and examine the relevance of closed social circles of marriage among different families for opening up a host of questions of interest to kinship analysis Thus we add a link to Figure to exemplify what happens when families relink through marriage 250 SOCIAL SCIENCE COMPUTER REVIEW TABLE 3 GEDCOM Coding of the Data in Table 1 Individuals 0 I1 INDI 1 NAME John Schmidt 1 SEX M 2 1 FAMS F1 1 FAMS F3 1 FAMC F2 0 I2 INDI 1 NAME Mary Finley 1 SEX F 1 FAMS F1 1 FAMC F4 0 14 INDI 1 NAME Adam Schmidt 1 SEX M 1 FAMS F2 0 I5 INDI 1 NAME Elizabeth Mark
12. where child to parent links from lower to upper vertices or if transposed to parent to child from upper to lower vertices are stipu lated to be asymmetric represent the two different types of networks that can be coded in the ordinary manner of social networks the first with individuals as vertices and the second in Pgraph format where conjoints couples or single individuals are the vertices The double horizontal lines in the first graph indicate a symmetric marriage relation This graph in Table 3 has two types of relations those with parents solid lines and those between spouses double lines and the labeling of vertices must indicate gender of individuals If we remove the marriage relation then we have the genetic graph of Ore 1963 used by geneti cists in which no graphic relation is given between couples who have no children e g Indi viduals 1 and 3 Where there are two or more children the skeleton or symmetric graph of parental relations within a nuclear family forms one or more cycles Figure 1 shows a P graph of the basic kinship data in Table 1 in terms of relations among the six nuclear family or conjoint vertices F1 to F6 The labels on the lines connecting the families indicate linking individuals Individual 2 for example is a wife and mother in Fam ily F1 but her parents are in F4 A coding of two types of lines dotted and solid shows whether the linking individual is female solid or male dotted
13. Anthropology Analyzing Large Kinship and Marriage Networks With Pgraph and Pajek DOUGLAS R WHITE University of California Irvine VLADIMIR BATAGELJ ANDREJ MRVAR University of Ljubljana Slovenia Five key problems of kinship networks are boundedness cohesion size and cohesive relinking types of relations and relinking and groups or roles Approaches to solving these problems include formats available for electronic storage of genealogical data and representations of genealogies using graphs P graphs represent couples and uncoupled children as vertices whereas parent child links are the arcs connecting nodes both within and between different nuclear families Using results from graph theory P graphs are shown to lend themselves to solutions of the problems discussed Relinking of families through marriage for example can be formally defined as sets of bounded groups that are the cohesive cores of kinship networks with nodes at various distances from such cores The structure of such cores yields an analytic decomposition of kinship networks and constituent group and role relationships The Pgraph and Pajek programs for large network analysis help both to represent kinship networks and their patterns and to solve problems of analysis Keywords kinship marriage networks large network analysis graph theory cohe sion social groups and roles genealogy graphic representation Pajek Pgraph INTRODUCTION In the st
14. C It also produces a series of compact files for Pgraph data analysis read by various Parent Suite programs the VEC or vector file extensions VE DOC or documentation file DO the NAM or names file NA and in some cases additional coordinates CO files The last letter of the extension of Pgraph files is used by conven tion to distinguish raw data D files from subset S and bicomponent C subset files as well as from reduced B files of blood marriages or kinship model M files The VEC file lists the two vectors needed for Pgraph analysis These arrays are indexed to the new series of unique family numbers as explained by White and Jorion 1992 The first 268 SOCIAL SCIENCE COMPUTER REVIEW Figure 7 P graph Combining Individual and Family Vertices for Sample Data array gives the family of orientation FAMC of a son who may be or may become a husband FAMS and the second array gives the family of orientation FAMC of a daughter who may be or may become a wife FAMS Additional arrays are just for purposes of keeping track of alternative family number and individual labels Par Calc is a Pgraph program that reads the VEC file and does statistical analysis of the different types of blood marriages compared to numbers of existing relatives of each type and ranging from siblings to fifth cousins in ego s generation as well as other types of rela tives The advantage of this procedure is that by computing
15. In the process of graphic analysis a number of partitions may be defined and saved as CLU files patrilineage and matrilineage memberships connected components and relinked bicomponents other data entered manually e g a partition of places of residence Because these are simply column variables in ASCII files they are easily read or pasted into standard statistical packages such as SAS SPSS and SYSTAT for cross tabulation and sta tistical analysis Graphic Aesthetics Aesthetic criteria are used to compute some properties of large but well organized pic tures and to draw graphics accordingly closest vertices vertices should not be too close smallest angle between lines with a common vertex should not be too small shortest longest line lines should not be too short or too long number of crossings the fewer the better and vertex closeness to line vertices should not be too close to lines Using measures such as these provided by Pajek the user can check for the quality of a picture before and after reorganization Making a partition by which vertices are colored in the graph using the Draw Draw Partition option may help to organize the graph To see a lineage as a partition of vertices for example Net K Neighbors Input vertex number input ancestral couple number 0 will color all the descendants of the selected vertex and Parti tion Binarize range of distances from ancestor input maximum distance will give
16. air is connected by two inde pendent paths Hence every pair of vertices in a bicomponent is connected by marriage cir cles formed by relinking marriages The bicomponents of a P graph of kinship and marriage are mutually exclusive sets of arcs or bounded subgraphs or social units with structural endogamy White 1997 Brudner amp White 1997 To extract cohesive sets of marriages after using Net Components Bicomponents we double click the Hierarchy object window to view a list of components and their sizes and then use Hierarchy Extract Cluster to extract by its number one e g largest clusters Operations Extract from Network Clusters the number of which already is selected restricts the network as in Figure 5 to only the bicomponent Along the way the menu Info e g Hierarchy lets us see the distributions by size pertinent to the various objects we have computed Returning to Draw Draw Partition the graph now shows only the chosen bicomponent Attributes of vertices can be constructed within Pajek or be imported as partition CLU files and similarly for permutations PER clusters CLS hierarchies HIE and vec tors VEC One way in which to select a subgraph is to use Operations Extract from Net work Partition or Cluster which asks for criteria for selection If the same or another parti tion is wanted to label or analyze the subgraph then the options to use are Partitions First partition choosing some p
17. also is available at the same location 5 Ego2Cpl allows a choice of eight formats depending on the order and type of variables 1 ego number name sex father number mother number spouse number 2 ego number sex name father number mother number spouse 3 ego number name sex spouse number father number mother number decade of birth 4 ego number sex spouse number father number mother number name other data 5 ego number sex spouse number father number mother number name other data 6 ego number sex spouse number father number mother number premarital residence decade of birth postmarital residence name other data 7 ego number sex spouse number father number mother number premarital residence decade of birth postmarital residence purchase name other data 8 ego number sex spouse number father number mother number name number decade of birth premarital residence postmarital residence Formats 4 to 7 are flexible in having the name and other data in last place in the data format so that new variables for data entry can be extended indefinitely Formats 6 to 8 exem plify special cases in which certain types of data are coded e g place of residence pre and postmatriage 6 It does not check beyond parents however for errors where egos are their own ancestors This is done by Pajek and by the Par Calc program in the Pgraph suite of programs 7 InaP graph multiply married persons need
18. and Harary 1997 1998 show the same results independently and propose that the relinking index is most relevant when it is computed on a bicomponent of a P graph that is for a structurally endogamous group Otherwise arbitrary data that happened to be collected outside of the structurally endogamous group will affect the index of relinking Network Size Radial Cohesion and Linked Vertices Problem 3 Rethinking the concept of cohesion led White Schnegg and Brudner in press to define forms of cohesion that might be suitable for the study of the types of diffuse cohesiveness that often are found in the kinship networks of complex societies where the length of marriage circles often is very large They define the radial cohesiveness R of a bicomponent as the ratio D Z where D is the density of the bicomponent and Z is the density of the sum of ties in the kth order zone of ties at distance k from each ego Thus Z is the density of the ties of those who are directly linked to acommon ego Z is the density of the ties of those who are distance 2 from a common ego and so forth For Z to be greater than zero in a P graph some mar riages must be as close as uncles marrying nieces hence the ratio D Z often will be infinite Z will be greater than zero only if some marriages are as close as between cousins For bicomponents in some populations D Z gt 1 and as a statistically significant departure from chance for up to quite large va
19. artition to be reduced for use with the subgraph and Parti tions Second partition criterion partition or the partition used to select the subgraph fol lowed by Partitions Extract Second from First The result is that the reduced graph and the reduced first partition are of the same dimension and can be used together for further analysis White et al KINSHIP AND MARRIAGE NETWORKS 267 Pajek as partly seen from the macros in Table 6 offers many other useful options such as e searching for the shortest kinship paths among persons e determining all predecessors and successors of selected persons and e extracting the neighborhood of a selected person Options we have not discussed that are of somewhat greater complexity as in our discus sion of problems for kinship analysis e g groups roles relations include e searching for interesting patterns in a genealogy e marriages among relatives where and how often they occur if a patterned type of subgraph does not occur very often in the genealogy then this option is executed quite quickly even for large genealogies e parents having many children e persons married several times and e statistics average number of children maximum number of children The ability to output UCInet Borgatti Everett amp Freeman 1998 DL files for further network analysis lets us analyze features such as e centralities of individuals and couples e block modeling of relational pat
20. as a baseline the number of times relatives of a given type exist e g the number of men who have a mother s brother s daugh ter available to marry compared to those who actually marry such a relative the rates of blood marriage are normed to control for demographic variables such as sibship size The program also gives the percentage consistency of the observed blood marriages with dual matrimonial organization statistics on generation differences between husbands and wives in blood marriages and relative numbers of male and female links in such marriages Par Link performs a similar statistical analysis on pairs of marriages that relink two fami lies such as sister exchange cousin exchange and two brothers marrying two sisters Unlike Par Calc however it currently requires considerable sophistication in marriage alliance the ory and involves a complex set of notational conventions to read the output Prior to the advent of Pajek Pgraph was the principal program for network visualization within the Parent Suite Many of the Pgraph algorithms can now be found in Pajek and need not be separately described White amp Skyhorse in press provide a manual for program options One of the continuing advantages of this program however is its options for creat ing screen images and saving coordinate files that can be converted into high quality graph ics using the HPGL language for printer control Extremely large genealogies can be split
21. aths All paths from ito j to an ancestor After bicomponents enter bicomponent number to extract HieCompL Extracts bicomponent from genealogy computes genealogical layers and draws Hierarchy genealogy Component Layers HieCINet Extracts bicomponent and selects its cluster from genealogy Hierarchy Cluster Network Predecessors and successors of a person or a couple the couple will be partitioned by distance 3Bef8Lat Three generations before and three generations after person or couple enter the same person couple twice AllBef All predecessors before selected person or couple AllLat All successors after selected person or couple AllBef and AllLat for selecting predecessors or successors all descendants all ances tors or all relatives in the case of genealogical P graph files are important macros equiva lent to Net K Neighbors Input or Output because these commands often are used to select a particular genealogical line within a potentially huge database e g selecting all of Queen Victoria s descendants from the European royalty genealogy The result of such commands is apartition in terms of distance from ego Partition 0 to the farthest connected relative Par tition k at this distance those unconnected are in the partition labeled unknown and are colored dark brown Info Partition will show the distribution of distances Opera tions Extract from Network Partition Select classes from to
22. ation Type 1 to his parents in Couple 2 The graphic instruction c Black p Solid 1 John Schmidt assigns black solid lines and a label to his arc The second entry shows the same information for John Schmidt this time from the perspective of Couple 3 where he has a different wife The first five entries are for males husband s parents and the last three are for females Relation Type 2 wife s parents to be shown as red lines com posed of dots with labels This file has been read and its data have been graphed by Pajek which has added the x y and z coordinates for each of the couples or individuals Algorithms for the Analysis of Large Networks The philosophy underlying Pajek is to analyze networks in terms of a number of different types of objects e g networks partitions hierarchies clusters permutations vectors and to implement algorithms that are subquadratic i e with run time less than O n where nis the number of vertices that is increasing less rapidly than the square of the number of verti ces The algorithm for finding the bicomponents of a network for example runs in O n m where m is the number of edges Large networks often are sparse and many sparse network algorithms run in O n O n log n or O nVn time Pajek provides very fast algorithms for selected large network problems Finding all the possible paths or shortest paths between two vertices in a genealogical database of 5 000 people for
23. bles 2 to 4 give an example Fourth it creates a file with the extension NET for use with the Pajek program and allows the user to choose different types of graphic labels Fifth it creates a series of files for specialized use with the Pgraph programs The Pgraph and Pajek packages are discussed later GEDCOM Standards The GEDCOM GED file header information produced by Ego2Cpl is shown in Table 2 The input file listed is one containing the data in Table 1 Records in GED files have a line starting with O and continue on lines starting with 1 The header data in the GEDCOM file produced as output from the data in Table 1 are fol lowed by a series of records one for each individual as shown in Table 3 Each individual is assigned two family numbers where the FAMC given to a child corresponds to the FAMS number of his or her parents A list of the families follows as in Table 4 giving the number of each family and the individual numbers for its members husband wife and children White et al KINSHIP AND MARRIAGE NETWORKS 249 TABLE 2 GEDCOM Coding of the Data in Table 1 Headers HEAD SOUR _ PAF 2 2 DEST PAF DATE 03 MAR 95 FILE p dem ged the file shown in Table 1 CHAR ANSEL S1 SUBM NAME Doug White ADDR School Social Science CONT UC Irvine 92697 CONT Internet Email address drwhite uci edu PHON 949 824 5893 Sek ND NG Sk ek Si ek ee ek SS The directed graphs drawn inside Tables 3 and 4
24. cal affective or religious ties functions that clearly can be taken on by kinship links is the underlying model of the relation of parentage in which parents precede their chil dren as point of origin for birth or nurturance in childhood Whereas the relations of parent ing are variable the potential existence of a parental couple is at least a recognized norm White et al KINSHIP AND MARRIAGE NETWORKS 247 TABLE 1 Coding Individuals and Their Relationships 5 Demo Data Family Schmidt l4 1X A21 A1 314 3x 12 1 John Schmidt M 2 4 5 1920 1 John Schmidt M 3 4 5 1935 2 Mary Finley F 1 8 9 1920 3 Sara Gonzalez F 1 0 0 1935 4 Adam Schmidt M 5 0 0 1895 5 Elizabeth Markowitz F 4 0 0 1895 6 Maude Schmidt F 0 1 2 7 John Schmidt Jr M 0 1 2 8 M 9 0 0 9 F 8 0 0 END OF DATA Variables and variable labels 1 2 3 4 5 6 7 Ego Name Sex Spouse Father Mother Marriage date modeled on sexual reproduction One can push this flexible definition in a biological direc tion to find the unique pair of genetic parents or open it up to allow a range of different types of parents and parental couples and relations to be defined in different cultural contexts Database Construction A kinship network is constructed as a database by taking individuals as a starting point and noting their gender their different conjoints sexual or marriage partners in whatever form and their children or their parents e g biological sociologica
25. ce and the Department of Anthropology at the University of California Irvine CA 92697 e mail drwhite uci edu His current work is on complexity theory and the effects of emergent struc ture in social networks on a large scale including social class large scale cohesion solidarity and exchange elites markets and global processes He does extensive collaboration on longitudinal social research projects in Europe Latin America Africa and Asia 274 SOCIAL SCIENCE COMPUTER REVIEW Vladimir Batagelj is a professor of discrete and computational mathematics at the University of Ljubljana Slovenia His main research interests are in mathematics and computer science and combinatorics with emphases on graph theory algorithms on graphs and networks combinatorial optimization algorithms and data structures cluster analysis and applications of information technology in education He may be contacted at vladimir batagelj uni lj si Andrej Mrvar is a teaching assistant of computer science and statistics with the Faculty of Social Sciences at the University of Ljubljana Slovenia where he is completing his doctoral thesis on the analysis and visu alization of large networks with the Faculty of Computer and Information Science He may be contacted at the Faculty of Social Sciences Kardeljeva pl 5 1000 Ljubljana Slovenia e mail andrej mrvar uni 1lj si Web http www uni lj si fdmrvar andrej html
26. components do not necessarily contain cliques hence the concept of overlapping cliques is not a necessary condition for bicomponents White and Harary 1997 1998 pro pose two measures of the social cohesiveness of the network graph G for a social group a the connectivity of G or minimum number of independent paths between any two vertices in the group and b the cycle density of G In a regular P graph vertices can have at most out degree 2 however the connectivity never will exceed 2 Hence the cycle density is the appropriate measure of cohesion Mrvar and Batagelj 1998 also independently propose the cycle density of the undirected version of a P graph as a relinking index for a genealogy _ kt m n k n 2M where k number of weakly connected components of the P graph n number of vertices m number of child parent links and M number of apical ancestral vertices having out degree 0 In general k m n is the number of independent basic cycles in a graph In a regular P graph where all but M ancestral vertices can have at most outdegree 2 the maxi mum number of edges is 2 n M so that the maximal number of cycles isk 2 n M n k n 2M Hence the relinking index p is the cycle density of a P graph Mrvar and Batagelj 1998 prove that 0 lt P lt 1 where P 0 if the genealogy is a forest and that there exist arbi trarily large genealogies with P e g two brothers marry two sisters White
27. dership and the formation of social economic or symbolic capital The transmission of power among Central American elites White 1996 the transmission of estates among Austrian farmers Brudner amp White 1997 the ethnogenesis of Turkish nomads White amp Johansen 1998 radial political cohesion of Tlaxcalan villagers White et al in press and the biblical kin ship charter for transmission of religious leadership White amp Jorion 1992 provide exam ples using the P graph as a framework for societal level studies A second set of hypotheses seeks to explore specific genres of social organization e g dual organization in marriage structures generalized exchange radial cohesion semicom plex kinship systems through network analysis Houseman and White 1998a 1998b argue that the bicomponent structure of a kinship network contains much of the structural informa tion for the study of marriage rules and emergent alliance structures Their study of Amazo nian societies Houseman amp White 1998a provides elements of a new vocabulary for the White et al KINSHIP AND MARRIAGE NETWORKS 257 identification of structural regularities emergent out of marriage choices and therefore open to change as a product of social interaction Their restudy Houseman amp White 1998b of Leach s monograph on Pul Eliya Sri Lanka provides a detailed case study in which the exis tence of dual organization is self organizing at the leve
28. dforhttp vlado fmf uni 1lj si pub networks pajek sunbelt 97 pajek htm UCINet networks analysis package for sale from Analytic Technologies http eclectic ss uci edu lin order html NOTES 1 The GEDCOM 5 5 standard is accessible at http www gendex com gedcom55 55gcint htm It is of course possible to enter data directly into a GEDCOM database such as Genealogical Information Manager available at http www mindspring com dblaine gimhome html that assigns both individual and family numbers automatically Pajek can read GEDCOM files directly 2 Hence one of their two prior FAMS numbers if any can be reassigned to their conjoint unit and the other can be reserved in case the corresponding individual marries again Note that the FAMC number of a child equals the FAMS of the child s parents 3 Kith and Kin currently in Version 3 02 which runs in Windows 95 and which prints descendant and ances tral trees in P graph format is authored by SpanSoft and is available on a trial basis at http www rocketdownload com details home kithkin htm Pictures maps diagrams and sounds may be embedded in or linked to a person or family A printed book can be produced with family group details and a cross referenced index 4 The Ego2Cpl program as part of the Pgraph package is available at http eclectic ss uci edu drwhite p graph ego2cpl html It is a DOS program that is compatible with Win dows but a straight DOS version
29. es that are three to four levels deep as in the manual listing menu choices available at Pajek s Internet address We refer here to menu item command sequences including sub and sub sub menu choices by command labels which are clicked separated by slashes For example Info Network shows the number of vertices in the network Net Transform Transpose is an important option to con sider at the start of analysis because it reverses the direction of arcs from child to parent the P graph convention to the more intuitive parent to child direction The options Macro Record and Macro Play allow the user to save as in a command file and to play back a series of operations as a macro command Standard macros for genealogi cal analysis included with the program are shown in Table 6 When macros are accessed by Macro Play a directory opens from which existing macro MCR files can be clicked 260 SOCIAL SCIENCE COMPUTER REVIEW TABLE 6 Pajek Macros Name Operation Compute genealogical layers and view initial picture macros of the same name in the Draw subdirectory also will draw and show the graph Layers1 2D graphic with genealogy depth partition use x and y keys to rotate the picture still can be improved using averaging x coordinate minimizing total length Layersz 3D graphic with spring embedding use x y and z keys to rotate Find connections between people enter two persons couples Path Shortest undirected path AllP
30. eters for screen editing with the mouse Move Fix y or Fix x for example lets you move vertices only in the horizontal or vertical direction whereas Move Grid allows you to change format for the grid spacing between vertices For a 3D graphic Macro Layersz clicking the LAYERSZ MCR file in a directory is equivalent to clicking in the Graph window Layers Type of layout 3D layers in z direction Energy Starting positions Given z coordinates 2D Fruchterman Reingold This begins a minimum energy configuration algorithm commonly called spring embedding The algo rithm pushes unconnected pairs of vertices with no child parent arcs toward opposite perimeters of the graph and pulls connected vertices toward one another and therefore toward the center until both the length of links is minimized and the distance between unconnected points is maximized The 3D graph can be rotated around the x y or z axis by the use of the corresponding key i e holding down the shift key along with the x y or z key reverses each of these rotations Starting configuration options may make for different out comes on each run so outcomes should be saved as files Viewing and Printing Exported Graphics Because Pajek is a Windows program pressing the PrintScreen key copies the screen image into the Windows clipboard For printing or further enhancement the clipboard image is then inserted into a word processor or graphics editor Export from the Draw wi
31. f uni 1lj si pub networks draweps htm 22 From this list of the hierarchy of bicomponents clicking on the largest bicomponent and choosing Edit Change type will toggle it to Close Then on the main menu Operations Shrink network Hierarchy select ing Yes for checking whether the network is simple and 1 for the minimum number of connections will result in the Closed group the largest bicomponent being shrunk to a single node Net Transform Remove loops will remove the loops produced by the arcs inside the bicomponent After that the shrunken network can be treated as an ordinary genealogy with the shrunken vertex having vertex number Net K Neighbors All with vertex label 1 and distance 0 no limit will sort the vertices by their distances from the largest bicomponent 23 UCINet is for sale at http eclectic ss uci edu lin order html REFERENCES Batagelj V amp Mrvar A 1997 February Pajek Program package for large networks analysis Paper presented at Sunbelt XVII conference San Diego Online Available http vlado fmf uni 1j si pub networks pajek sunbelt 97 pajek htm and http vlado fmf uni 1j si pub networks pajek pajek pdf Batagelj V amp Mrvar A 1998 Pajek A program for large network analysis Online Available http vlado fmf unk 1j si pub networks pajek Bollobas B 1985 Random graphs New York Academic Press Borgatti S P Everett M G amp Freeman L C
32. g one or more edges define a bounded social unit or subgraph a set of families of which any pair is connected by one or more marriage circles In graph theory maximal subgraphs of this sort are called the 2 connected components of a graph which we refer to here as bicom ponents A bicomponent of a Graph G is a subgraph of G with three or more vertices in which any pair of vertices is connected by two independent paths hence by aset of edges that form a cycle Two distinct bicomponents of a graph can have at most one vertex in common between them Hence any given edge in the skeleton of a P graph will belong to at most one bicomponent By definition and formal proof Harary 1969 pp 27 28 the following state ments are equivalent for a Graph G G is a bicomponent G contains no cutpoint Every two vertices of G lie on a common cycle Every two edges of G lie on a common cycle There is a path joining any two vertices of G that uses any given edge of G There is a path joining any two vertices of G that uses every other vertex of G There is a path joining any two vertices of G that avoids any other vertex of G As self bounding subgraphs of mutually exclusive sets of edges in a P graph bicompo nents are easy and quick to determine by computer because the time required for computa tion is a linear function of the size of the network unlike those many network problems that are polynomial of higher order or even exponential nonpolynomial
33. graphics and network analysis in P graph representation format The easiest way in which to enter kinship data into Pajek is to begin with a format such as that of Table 1 see Note 4 and transform the data to a NET or GED file using the Ego2Cpl program There is an advantage of having Ego2Cpl produce network or NET files indigenous to Pajek in that names or numbers of individuals and the colors and types of lines used in the graphs are more easily managed Pajek network files also can handle many other types of social relations and attribute data PAJEK A PACKAGE FOR LARGE NETWORK ANALYSIS Pajek the Slovenian word for spider is a Windows 32 bit program for network visu alization and analysis written by Batagelj and Mrvar 1997a 1997b Because it is oriented to analysis and graphic representation of large networks meaning thousands of vertices it is especially well suited to the analysis of kinship and marriage networks along with a host of other types of data that may be available on either small or large populations of individuals Genealogical Data Entry GEDCOM files are one of Pajek s standard input formats The file in Tables 2 to 4 pro duced by the Ego2Cpl program in the Pgraph package or as GED output files of numerous standard genealogical programs can be read directly into Pajek and visualized as a graph using the P graph format The P graph representation format became Pajek s default for kin s
34. he Alexander von Humboldt Foundation TransCoop program and the Maison des Science de l Homme Paris including the Maison Suger 12 Ego2Cpl allows as labels either individual names or numbers 272 SOCIAL SCIENCE COMPUTER REVIEW 13 There also are ways in which to do this in the GEDCOM format but it is difficult to find software to handle the extra data graphically 14 Pajek is available always in its most recent version as a download from the Internet address http vlado fmf uni 1j si pub networks pajek at the University of Ljubljana 15 Available at http vlado fmf uni 1lj si pub networks pajek pajekman htm 16 Draw Spin Normal and type 001 and Draw Spin Spin around and type 180 also will spin 180 in the z direc tion A 180 spin is not required if the arcs are transposed to parent to child 17 Available at ftp ftp cs wisc edu pub ghost ftp ftp cs wisc edu pub ghost rjl1 and ftp ftp cs wisc edu pub ghost aladdin 18 Available at http cosmosoftware com 19 Available at http www mdli com wel html 20 Pajek provides an option to include labels in Mage images but this might produce a crowded image The other way in which to get fixed but selected labels is within the Mage program using option Edit Draw new On the right hand side where options appear Labels must be checked so that vertices subsequently may be clicked for a label to appear on the layout 21 Embedded commands are listed at http vlado fm
35. hip data as of Version 19 in December 1997 The network or NET file however is the native Pajek file Table 5 shows the data of Table 1 plus Individuals 10 11 and 12 having been processed through Ego2Cpl to produce a NET file where nuclear family numbers identify the nodes to which individuals are attached Only one of the Pajek data formats is shown with Vert ices and Arcs as headers for two of Pajek s data types The first data 258 SOCIAL SCIENCE COMPUTER REVIEW TABLE 5 Schmidt Family Data in Native Pajek Format as a NET File Vertices 7 T2 0 3500 0 5000 0 5000 2 45 0 5341 0 0589 0 5000 3 1 3 0 7310 0 4982 0 5000 4 8 9 0 1706 0 0589 0 5000 5 10 6 0 5210 0 9214 0 5000 6 7 0 0 1864 0 9214 0 5000 7 11 12 0 3950 0 8625 0 5000 Arcs 1 2 1 c Black p Solid John Schmidt 3 2 1 c Black p Solid John Schmidt 5 3 1 c Black p Solid Ahmed Schmidt 6 1 1 c Black p Solid John Schmidt Jr 7 3 1 c Black p Solid Habib Schmidt 1 4 2 c Red p Dots Mary Finley 5 1 2 c Red p Dots Maude Schmidt 7 4 2 c Red p Dots Mathilde Finley type lists the seven vertices or families with labels in quotes giving husband s and wife s names The second Arcs lists two types of directed links between children and their par ents in P graph format Here the first entry indicates that Couple 1 is linked by John Schmidt the husband Rel
36. ight require an altogether different set of conceptual models A third and practical problem related to that of limits and concepts of diffuse cohesion is that of size Is it possible to study the structure of large social networks and up to what point is it feasible and useful to visualize network structures in terms of graphical representations A fourth problem is how to define and code the basic social relations of kinship and mar riage Genealogists typically do so in terms of the primary relations of parent and child and the derived relationships of consanguinity and affinity i e relations between two parents or connecting through a marriage or couple Geneticists are concerned only with consanguin ity Anthropologists problematize the cultural construction not only of marriage in all of its myriad forms including questions as to whether marriage is universal but also of kinship relations in general They question the universality of the assumption that biological parent age is the basic kinship relationship Besides the question of how to treat the multiple types of parentage e g biological sociological adoptive ritual cultural definitions of who are kin and who are not often depend on behavior not simply ascription so that both levels are significant A fifth set of problems is related to both cohesion and the difficulty of characterizing the social relations of kinship those of the kin groups and of the networks of social r
37. illage study they show that some children of farm families marry or are linked into a single structurally endogamous bicomponent that runs through a whole ethnic segment of the farming valley of the region The majority of children of these families how ever do not marry or have children who marry so as to become members of the bicomponent This is a social system of farmsteads however that tend to pass intact to single heirs and in which the heirs and their spouses form a distinct social class of landed families as opposed to laborers craftspersons salaried workers and professionals The relevance of the study of relinking is that they show a strong correlation between those who relink and those who inherit or marry heirs Brudner and White argue that structural endogamy is the basis for the formation of a landed family social class in this region a class division that makes sharp divi sions even within sibling groups Social Relations and Types of Relinking Problem 4 The specific ways in which relinking occurs within a population to reinforce social cohe sion poses a fourth theoretical problem how to define basic and derived social relations of kinship and marriage networks White 1997 discusses the relevance of different types of relinkings the number of distinct families involved the length of the cycles the number of generations The various types of relinking are easily characterized using P graphs The significance of the P g
38. l adoptive Because parent child is a reciprocal relation it is sufficient in a database simply to list every individ ual and his or her parents and conjoints A system of coding for individuals conjoints and parents is exemplified in Table 1 Here there are seven variables a 4 digit ego number a 20 character name a character to indicate gender three 4 digit numbers to identify a spouse and the father and mother of ego and a number for the year or decade of the marriage many other variables can of course be added to this format In this system when a person has two spouses as in the case of John Schmidt they are given multiple lines in the database each indicating a different spouse Part of the construction of genealogical or kinship databases as in the GEDCOM stan dard format for exchanging computerized genealogical data is the assignment of a unique set of numbers or identifiers not only to individuals as in Table 1 but also to nuclear families defined as conjoints and their offspring if any Every individual whose parent or parents are known can be assigned a corresponding FAMC FAMily of Child number and every indi vidual can be assigned a FAMS FAMily of Spouse number for each nuclear family in which he or she has a spouse or child Sociologists call these the family of orientation versus the family of procreation Murdock 1949 White and Jorion 1992 take family numbering a step further and assign a unique nuclear
39. l of behavior and egocentric kinship terminology in a context of bilateral kinship without a societal level coding into descent based moieties As opposed to descent based moieties which have been the sole means to recognize dual organization in anthropology to date they provide a novel network approach to the study of multiple forms of dual organization in marriage systems There are many other aspects of the theory of kinship developed through the vehicle of P graph representations White amp Jorion 1992 1996 H ran 1995 and remaining to be developed We will return to the advances in computability using P graph representations and data formats for the statistical study of bicomponents and various structural types of dual organization relinking and blood marriages Some of the major advances in this area are e new ways of controlling for confounding demographic variables when assessing the relative frequencies of various types of relinking and blood marriages e new ways of simulating comparative baseline frequencies for different constraints on random marriage regimes while controlling for confounding demographic variables and e statistics for evaluating network emergent dual organization Because here we are addressing the widest possible audience in the social sciences we next discuss the Pajek network analytic and visualization package used since 1998 displac ing earlier graphics programs in the Pgraph package as the vehicle for
40. l scientists as well as scholars of Asia Africa and the Pacific Exploration of new means of scientific visuali zation and computability have been two of the principal objectives in developing P graph paradigms for a series of empirical research projects on marriage and kinship networks in a wide variety of theoretical and regional contexts The programs and methods of visualization used in these projects between 1991 and 1998 were mostly those developed within the Pgraph program package White amp Skyhorse in press written by White In 1998 the authors of Pajek which is discussed later adapted the P graph format as the default for repre senting and displaying GEDCOM kinship network data Whereas much had been learned about graphic displays of kinship networks up to that time cf Schweizer amp White 1998 Pajek opened up a whole new set of possibilities for actively manipulating images and exporting them to PostScript VRML and chemistry visualization formats which allow color printing and three dimensional 3D rendering through Web browser plug ins for vir tual reality or molecular images One set of hypotheses currently being used in a wide number of P graph studies are those exploring the links between structural endogamy bicomponents and their linked vertices as a means of identifying social units or subgraphs and cohesive phenomena such as social class political or economic elites ethnicity succession to office or religious lea
41. lues of k Such bicomponents are cohesive in the sense of structural endogamy as defined by Brudner and White 1997 Having established that structurally endogamous clusters of kinship and marriage con nections may be cohesive another problem of boundaries and network size remains What about vertices that are outside bicomponents If they are connected to structurally endoga mous clusters then these vertices are potentially as important as those within the clusters because they may be the children parents spouses or other relatives of those within the bicomponent to which they are attached These linked vertices are ordered however by dis White et al KINSHIP AND MARRIAGE NETWORKS 255 tance from the bicomponent and they may be connected to more than one bicomponent at different distances Structurally endogamous subgraphs as indicated by the frequency of other vertices linked to them virtually never constitute populations that are endogamous to the point of clo sure or lack of marriage with those outside the structurally endogamous subgraphs It often is the case in such an empirical subgraph that the majority of the kinship links or marriages are with those outside the subgraph In no way do structurally endogamous subgraphs resemble caste like marriage structures They are merely a way of characterizing that relinking occurs within a population so as to create and reinforce social cohesion In Brudner and White s 1997 Austrian v
42. manipulate sets of objects on the graphic screen either with the mouse or by embedded commands within the NET file as exemplified in Table 5 Embedded commands to control the appearance of nodes arcs and labels are explained more fully in documentation attached to the Web site for the manual listing menu choices for Pajek The strategy for on screen editing emphasizes creating partitions that show as sets of col ored nodes on the Draw screen and then moving single nodes by clicking on the node and moving the cursor here it is useful to fix the x or y coordinate with the Move Fix x or Fix y option so as not to get nodes out of alignment with others in their generation or moving clumps of colored nodes by clicking near a node of the selected color and then moving the cursor To create partitions we already have seen a number of Pajek menu operations A direct way in which to make a new partition however is to create a cluster CLS file in ASCII format with a text editor putting the node numbers for the new partition in successive rows of a CLS file and reading the file into Pajek by selecting it within the Cluster object window A partition containing these nodes is then created by the option Cluster Make Parti tion With Draw Partition the new partition will become a set of colored nodes within the original network Clusters and Partitions can also be extracted from Hierarchies by the com mand Hierarchy Extract Cluster or Make Parti
43. marry a daughter of F4 his stepmother s sister New data would be entered in Table 1 for these two illustrations as follows 10 Ahmed Schmidt 11 Habib Schmidt 12 Mathilde Finley M 6 1 3 Ahmed marries 6 his parents are 1 and 3 M 12 1 3 Habib marries 12 his parents are 1 and 3 F 11 8 9 Mathilde marries 11 her parents are 8 and 9 252 SOCIAL SCIENCE COMPUTER REVIEW Figure 1 P graph for the Data in Table 1 NOTE Numbers on the lines are for individuals Numbers on the vertices are for families Figure 2 P graph of Figure 1 Plus a New Relinking Marriage NOTE Numbers on the lines are for individuals White et al KINSHIP AND MARRIAGE NETWORKS 253 Network Boundedness Problem 1 Relinkings are relevant to our first and second theoretical problems of the limits or boundaries of social groups in terms of the network of kinship relations Every relinking defines a bounded set of families not only connected but also doubly connected by a marriage circle In any given body of genealogical data there may be many such circles of varying lengths The one in Figure 2 is of length 4 If Ahmed s brother married a daughter of F4 then the length of that circle would be 5 Ahmed s and Habib s marriage circles would overlap in sharing three of the same families or two of the edges in the graph in their marriage circles Those marriage circles in the undirected version skeleton of a P graph that overlap in sharin
44. ndow includes options not only for printer and graphic format files but also for saving animated displays to files that can be viewed by Web browser plug ins The following are the current export formats PostScript PS and EPS Encapsulated PS files are standard formats that can be printed with PostScript or GhostScript and GhostView programs Pajek s Draw window provides Export Options for borders shapes and backgrounds the graph corresponds to what is seen on the screen VRML WRL files are produced from the Draw screen with Export Options for size of lines and vertices and for background colors and are read with Virtual Reality Internet White et al KINSHIP AND MARRIAGE NETWORKS 263 browser plug ins e g Cosmo Player that allow movement within the 3D structure and viewing labels by clicking vertices MDL MOL files are molecular structure output that make beautiful 3D images that can be animated for rotation rotated with the mouse viewed in stereo and endowed with other features Freeman Webster amp Kirke 1998 It requires the Chemscape Chime viewer Ver tex labels are lost in the current implementation however so these images are mostly useful for viewing general structures KineMage KIN files offer a spectacular molecular structure output that provides most of the features of the MOL files Freeman 1998a Freeman et al 1998 but also include options for labeling vertices A series of slides may be a
45. nimated and vertices may be clicked to view or suppress their labels The advantage for presentations is a series of different slides or views of a structure The KineMage option that we jointly conceived as a result of writing this article as with a number options discussed here and that Mrvar implemented in Pajek is an animation of birth death and marriage processes unfolding as a moving window image over generations The Draw screen option Export KineMage asks How many neighbor classes to show and Resize option the responses 4 and 2 will show moving win dows of four successive generations as a dynamic evolutionary view of the kinship structure This moving image is especially effective if the vertices are partitioned not just by generation but also by birthdate or birth intervals such as decades KineMage is the only one of the present VRML and chemistry viewers that distinguishes different types of arcs while also providing labeling for vertices Labels in Chime for example are chemical not social network labels and Chime does not distinguish different types of arcs None of these viewers at present however provides labeling for arcs Many of the chemistry graphic engines are available for modification and rapid evolution is occur ring among 3D viewers Further Strategies for Organizing Visual Information and Analyzing Pajek Variables The philosophy underlying graphic display in Pajek is to create and
46. oles that are either achieved or ascribed on the basis of kinship links Given the fluidity and frequent overlap among different kinship groups anthropologists worry a great deal about problems of validity in characterizing cohesive groups or interlocked sets of roles In this article some approaches to solving these five problems are described In the first part different available formats for storing genealogical data in electronic form are pre sented Then possible presentations of genealogies using networks or graphs are described It is shown that parentage graphs P graphs are the most suitable presentation for our pur poses Some results are explained from graph theory that can be used to solve the problems posed The final part describes the use of the Pajek program for representing kinship patterns and solving kinship problems AVAILABLE FORMATS FOR CODING KINSHIP DATA Against this background of the problematics of kinship a number of advances have been made that facilitate a network approach to kinship and marriage and that address the more general question as to why social scientists would want to study kinship and marriage net works in the first place White and Jorion 1992 address the problem of what are kinship links or how to study kinship networks with a minimal set of assumptions They take as their starting point that kinship links are culturally constructed but that what makes them kinship as opposed to eco nomic politi
47. other pairs 1926 1381 and 228 430 and by brother sister pairs 1152 1151 and 659 1928 and their descendants often relink through marriage not in one generation but rather within two or three Marriages tend to be very close and within the same generation they crosscut the patrilines although many are within the patrilineages themselves knitting them together from within Network Analysis Figure 5 tells only part of the story of the 250 biconnected Turkish nomad couples Figure 6 shows the entire genealogy for this society with 1 114 arcs connecting more than 1 600 persons in 946 couples and 23 apical ancestors The difference in the much lower relinking density of 19 is structural those who do not relink are those who do not survive to marry or who emigrate from the clan Pajek provides further information using options for Net Components Weak and Net Components Bicomponents that tell that there is a single con nected component everyone is a relative and that all are connected as children affines or ancestors to the 25 of the couples who form the single large bicomponent 266 SOCIAL SCIENCE COMPUTER REVIEW Figure 6 Optimized Drawing of the Entire Turkish Nomad Clan Society Pajek s main menu contains algorithms that provide solutions to some of the analytic problems of kinship analysis Recall that a bicomponent with three or more vertices as noted previously is a maximal set of vertices of which every p
48. owitz 1 SEX F 1 FAMS F2 0 I3 INDI 1 NAME Sara Gonzalez 1 SEX F 1 FAMS F3 0 I8 INDI 1 NAME 1 SEX M 1 FAMS F4 0 I9 INDI 1 NAME 1 SEX F 1 FAMS F4 0 I6 INDI 1 NAME Maude Schmidt 1 SEX F 1 FAMS F5 1 FAMC F1 0 I7 INDI 1 NAME John Schmidt Jr 1 SEX M 1 FAMS F6 1 FAMC F1 White et al KINSHIP AND MARRIAGE NETWORKS TABLE 4 GEDCOM Coding of the Data in Table 1 Families 251 0 F1 FAM 1 REFN F1 1 HUSB 11 1 WIFE I2 1 CHIL I6 1 CHIL I7 0 F2 FAM 1 REFN F2 1 HUSB 14 1 WIFE I5 1 CHIL 1 0 F3 FAM 1 REFN F3 1 HUSB 11 1 WIFE I3 0 F4 FAM 1 REFN F4 1 HUSB I8 1 WIFE I9 1 CHIL l2 0 F5 FAM 1 REFN F5 1 WIFE I6 0 F6 FAM 1 REFN F6 1 HUSB I7 0 TRLR Numbers on the vertices are for families Figure 2 illustrates a marriage that relinks families already linked what anthropologists call a relinking marriage To show such a marriage we added to Figure 1 a new link between Families F5 and F3 Ahmed Schmidt son of F3 marries a paternal cousin Maude Schmidt Any marriage that completes a circle of ties in the skeleton of a P graph in this example the circle is among Families F1 F2 F3 and F5 constitutes a relinking Not all relinkings need be blood marriages there also are affinal relinkings through one or more prior marriages To illustrate an affinal relinking we arrange for Ahmed s brother Habib to
49. p www elsevier com locate son Granovetter M 1973 The strength of weak ties American Journal of Sociology 78 1360 1380 Hage P amp Harary F 1983 Structural models in anthropology Cambridge UK Cambridge University Press White et al KINSHIP AND MARRIAGE NETWORKS 273 Hage P amp Harary F 1991 Exchange in Oceania A graph theoretic analysis Oxford UK Oxford University Press Hage P amp Harary F 1996 Island networks Cambridge UK Cambridge University Press Harary F 1969 Graph theory Reading MA Addison Wesley Harary F 1971 Demiarcs An atomistic appoach to relational systems and group dynamics Journal of Mathe matical Sociology 1 195 205 H ran F 1995 Figures et L gendes de la Parent Paris Institut national d etudes d mographiques Houseman M amp White D R 1998a Network mediation of exchange structures Ambilateral sidedness and property flows in Pul Eliya In T Schweizer amp D R White Eds Kinship networks and exchange pp 59 85 Cambridge UK Cambridge University Press Houseman M amp White D R 1998b Taking sides Marriage networks and Dravidian kinship in lowland South America In M Godelier T Trautmann amp F E Tjon Sie Fat Eds Transformations of kinship pp 214 243 Washington DC Smithsonian Institution Press Kauffman S 1995 Athome in the universe The search for laws of self organization and complexity Oxfo
50. raph representation for the problems of bounded subgraphs and social cohesion in a given population is that every circle of relinking marriages is reflected in a cycle of the undirected version skeleton of the P graph and every cycle of the P graph corresponds to a circle of relinking marriages This is not true for the genetic graph such as in the vertical links of the graph within Table 3 where there are cycles inside each nuclear family having two or more children It is even less true in the typical egocentric graph that contains both descent and marriage relations so that even the father mother child triad is a complete graph that contains a cycle In egocentric kinship graphs there are many cycles that do not reflect relinkings and that clutter our ability to visualize and to analyze kinship and marriage structures Groups and Roles Problem 5 As a coding of social relations the P graph is merely the most reduced possible graph of kinship and marriage networks Many of the derived relations starting with various types of siblingship full siblings half siblings can be computed from different concatenations of the primary relation of parentage e g full siblings share common parents or different paths in the P graph That different societies may emphasize different types of dyads or paths is part of the general observation that each case study may require understanding of different 256 SOCIAL SCIENCE COMPUTER REVIEW social proces
51. rd UK Oxford University Press Mrvar A amp Batagelj V 1998 May Drawing genealogies Paper presented at Sunbelt XVIII conference Sitges Spain Online Available http vlado fmf uni 1j si pub networks doc sitges pdf Murdock G P 1949 Social structure New York MacMillan Ore O 1963 Graphs and their uses New York Random House Palmer E M 1985 Graphical evolution New York John Wiley Schweizer T amp White D R Eds 1998 Kinship networks and exchange Cambridge UK Cambridge Uni versity Press Seidman S B amp Foster B L 1978 A note on the potential for genuine cross fertilization between anthropology and mathematics Social Networks 1 65 72 Tufte E 1987 The visual display of quantitative information Cheshire CT Graphics Press Tufte E 1990 Envisioning information Cheshire CT Graphics Press Tufte E 1997 Visual explanations Cheshire CT Graphics Press Watts D J amp Strogatz S H 1998 Collective dynamics of small world networks Nature 393 440 442 White D R 1996 Enfoque de redes al estudio de comunidades urbanas Estudios Demogrdaficas y Urbanas 26 303 326 White D R 1997 Structural endogamy and the graphe de parent Informatique Math matique et Sciences Humaines 137 107 125 Online Available http eclectic ss uci edu knhe str endo htm White D R amp Harary F 1997 Structural models and conditional
52. rmation becomes an act of insight Tufte 1997 p 9 see also Tufte 1987 1990 Pgraph as a program package and P graphs as a system of representation interface with Pajek the program for large network analysis to produce capabilities for solving theoretical problems in the analysis of kinship and marriage networks such as the five discussed here These involve boundedness cohesion large scale social systems network characteristics of social relations and identification of groups and roles 270 SOCIAL SCIENCE COMPUTER REVIEW Our approaches to these problems are strongly linked to graph theoretic findings e biconnected components of graphs as large scale bounded subgraphs e the relinking index as an appropriate measure of cohesive density and e problems of studying cohesion on a large scale Granovetter s 1973 work on the strength of weak ties in very large networks for the flow of information under certain social conditions pointed the way to new understandings Draw ing on earlier work in graph theory critical thresholds in large networks Bollobas 1985 Palmer 1985 Watts amp Strogatz 1998 are now understood to be central to the emergence of self organizing complexity Kauffman 1995 Kinship and marriage have long been looked at close up in terms of the solidarity of close ties and the formation of corporate like kinship groups Alliance theories of marriage opened up new theoretical paradigms from
53. ses and social emphases and that kinship relations at the behavioral or affective level may depend on patterns of behavior and sentiment that must be studied in their own right In this double sense the P graph is simply a minimal network armature for the study of kinship networks and a sparse representation to which other social networks may be added As one wag remarked Kinship is to anthropology as the nude is to drawing If compound or distant kinship relations have a redoubling effect that heightens the importance of certain kinship linkages then these effects are even more intensified if they are embedded within marriage circles or structurally endogamous bicomponents of the mini malist P graph kinship network Examining the distribution of those kinship relations that are activated in important behavioral transactions bears on our fifth theoretical problem that of identifying social roles or groups We cannot solve this problem of analysis in the abstract but we would argue that for the network study of social organization whether an analysis of social class elites or kinship or a succession to religious or symbolic economic e g material or political offices or property it is a very useful adjunct to have a complete kin ship scaffolding STUDIES USING P GRAPHS P graph studies of kinship at present have the character of a large and open cooperative research project involving French German American and Mexican socia
54. terns in the network and e scalings of clusters or relatives or families Finally Pajek is in no way limited simply to kinship and marriage networks or to P graphs or family network representations rather it accommodates all sorts of social network and attribute data There also is a P graph format for a bipartite graph including both couples and individuals so that we can study relationships between individuals between couples or fami lies or both at the same time Figure 7 shows a bipartite graph of the data in Figure 4 with two types of vertices squares to represent individuals and circles to represent their marriages the graph is drawn to show the two slanting patrilines crosscut by three opposite slanting matrilines This type of graph may be useful where many individuals have multiple spouses to separate such individuals from their siblings and to depict distinct network relations between individuals as opposed to those between couples SPECIALIZED PGRAPH PROGRAMS PARENTE SUITE The Parent Suite of programs White amp Skyhorse in press includes the Pgraph program for drawing P graphs and the Ego2Cpl program makes the files needed by Pgraph or Pajek from a raw TXT file such as the one shown in Table 1 As noted previously Ego2Cpl cre ates a GED file containing the ID numbers of individuals and assigns a new set of nuclear family FAMS and FAMC numbers to which individuals belong as either parents FAMS or children FAM
55. the same color to all the descendants up to the selected distance Then clicking not on but rather near a vertex of the color of the lineage the mouse will move all the vertices of this color in tandem to a new location on the screen Hence the simpler options such as Draw Layers Optimize layers in x direction do not limit the types of visual reorganization of the data on the screen that are possible Figure 5 represents a picture of a 2D graph made by partitioning vertices of the genealogy according to their neighborhood relative to vertices with very high degree a type of leader partition This gives a general layout in clusters of similar vertices that can be further White et al KINSHIP AND MARRIAGE NETWORKS 265 Figure 5 Optimized Drawing of the Bicomponent of Intermarrying Patrilineages in a Turkish Nomad Clan Society improved by manual editing The image contains 416 arcs and 250 vertices with 14 apical ancestors for an extremely high relinking index of 75 but the network nonetheless has a marriage porosity i e openness to the outside of 25 In spite of the density of the graph many of the numeric labels on the uppermost arcs are readable as are those in the most recent lowermost generations where marriages have only just begun The image conveys many of the structural features of the Turkish nomad social networks given by White and Johansen 1998 There are few early ancestors identified by numbers for br
56. tion 264 SOCIAL SCIENCE COMPUTER REVIEW Defining and moving matrilineages in a P graph is obtained by changing the Options ReadWrite Threshold to 1 be careful to set it back to the default 0 when done because it is otherwise reset for the next run and saving a separate file with the female net work after moving elements Reconstituting the new configuration involves the multiple net work option Nets First Network and Nets Second Network entering the file containing the female network in the first case and the entire network in the second case Defining patrilineages also involves the multiple network option Nets First Network and Nets Second Network again entering the saved file containing the female network in the sec ond case and the entire network in the first case Nets Difference will then subtract the arcs in the second case from the first case and the result will be the male lines which can be saved to a file after points are moved about In using separate files for patrilines versus matrilines the option Partition Components Weak will identify separate lineages that can be moved as blocks clicking near but not on a point of chosen color on the graphic screen and can then be saved To recombine the two images the multiple network option Nets First Network and Nets Second Network is used again entering the separate saved files in the order wanted so that the first file will define the graphic configuration of the new image
57. tion followed by Layers in y direction in the Draw screen s Graphics menu and then holding down the z key to spin the graph 180 so that descendants colored yellow are on the bottom Then Options Lines Mark labels will label the arcs and Options Mark vertices using Numbers labels the vertices The input file for Figure 4 is from Table 1 plus the added persons 10 12 using either the NET or GED output of Ego2Cpl This is a P graph where the arcs are labeled with names note that Sara Gonzalez is missing because we did not have a link to her parents for this rea son and because of multiple spouses it sometimes is useful to create dummy parents and vertices are labeled with both individual numbers Other choices e g names for couples at vertices numbers for arcs would have been possible with Ego2Cpl s NET output file as Pajek s input The Pajek Draw screen in which the drawing appears as in Figure 4 has its own menu Energy EigenValues Layers GraphOnly Redraw Options Export Spin Move Info When the Draw screen is opened it might appear at first to be empty depending on whether coordinates saved from a previous session are included with the NET file they are not with 262 SOCIAL SCIENCE COMPUTER REVIEW the GED file Clicking Layers in y direction will fix generations into layers and make a2D picture appear Once the graph is visible the mouse can be used to click and move points the lines move a
58. udy of contemporary and historical societies ethnography social demography political and economic elites social class ethnicity and numerous other fields the analysis of those social networks that are partially ascriptive such as those constructed through rela tionships of kinship might present formidable problems Computerized genealogical data are now available for tens of millions of people as well as for large genetic demographic eth nographic and historical databases Five Key Problems of Kinship Networks Even small scale kinship studies require that we consider the first and foremost problem of limits how to bound the field of study Social networks constructed through inter Social Science Computer Review Vol 17 No 3 Fall 1999 245 274 1999 Sage Publications Inc 245 246 SOCIAL SCIENCE COMPUTER REVIEW marriage unlike egocentric genealogical studies of ancestries pose problems of how to anchor and bound the network A second problem is conceptual If we look for socially cohesive networks of kinship and marriage as one of the possible ways in which to bound such a study then what do we mean by cohesion Traditionally cohesion in social networks studies has begun from the small groups approach and defined the paradigm of social cohesion as the clique in which there is a complete set of connections e g friendships between pairs of individuals Cohesion in kin ship and marriage networks is more diffuse and m
59. umber name and gender Variables 4 to 6 are the numbers of the spouse father and mother and the final variable is the date of the mar riage Many other variables can be added to such records but the first six variables in what ever order are essential The input file for Ego2Cpl conversion requires two header cards one to identify the order and type of input variables here with a Type 3 format that is dis cussed later and another to identify the format for reading the variables The format 14 1X A21 A1 314 3x 12 asshownin Table 1 specifies a 4 digit integer a space a21 followed by a 1 character field three 4 digit integers 3 spaces and a 2 digit integer that truncates the year of marriage into a decade e g 1920 into 92 The program creates FAMS numbers for couples and remaining individuals and uses these numbers to assign FAMC numbers to children It produces five sets of output First it flags errors in data entry and stores them in an output file An individual for example might be listed once as a husband and again as a mother gender error or might be listed as his or her own parent axiomatic error Second it checks to see whether any individuals who are multiply married lack references to parents and creates a file of links to virtual parents needed in Pgraph and GEDCOM formats to keep track of multiply married persons Third it creates a standard genealogical GEDCOM file with extension GED of which Ta
60. utomatically and the picture can be enhanced in a variety of ways The first option under Layers Type of Layout provides 2D or 3D choices and affects the other options under Layers In 2D the options Layers in y direction Layers Averaging x coordi nate and Layers Optimize layers in x direction will alter the picture in systematic ways In 3D the Layers options that change the parameters for the picture are Layers in z direction Layers Averaging x and y coordinates and Layers Optimize layers in xy plane Options in the Draw window menu provides for different types of labeling coloring and size of the vertices and arcs Options Mark vertices using Labels will show the direction and type of arrows as well as the names for husband and wife from the original data set If no ver tex labels are wanted then Options Mark vertices using No labels will suppress the labels but keep the direction and type of arrows arrows here not transposed are upward to parents If the NET file was made by Ego2Cpl then Options Lines Mark will label the lines with each person s name or number depending on the Ego2Cpl option chosen previously This also is true if a GED file is read using Options ReadWrite Pgraph and labels The settings chosen using Options ReadWrite in the Main window and Options Lines MarkLines in the Draw screen are saved when work with Pajek is finished and is restored when Pajek is run again Move in the Draw window provides param
61. which sprang the current representation of P graphs Studies of kinship and marriage network on a large scale as a necessary back ground and foundation for the study of other social processes may prove to have large scale properties that will change our conception of how societies market systems and major social institutions are organized The emergence of social class elite formation succession to office and political domination ownership of corporations market exchange and processes of inheritance and bequest need to be restudied and reconceptualized in a large network context in which kinship and marriage network play a critical role as part of critical threshold emergent phenomena The research tools discussed here can play a role in this development APPENDIX Additional Information on Sources Following is a list of URLs for Web sites at which software and documentation reviewed or men tioned in this article are available in their most recent versions in alphabetical order Chime Web browser plug in by MDL Information Systems featuring chemical structure displays http www mdli com wel html Cosmo Player Web browser plug in from Cosmo Software Platinum Technology featuring VRML displays http cosmosoftware com Ego2Cpl part of the Pgraph package for parentage networks analysis by Douglas R White at the University of Cali fornia Irvine http eclectic ss uci edu drwhite P graph ego2cpl html Docu mentation is

Social Science Computer Review

Contents

Download Pdf Manuals

Related Search

Related Contents