Network Theory and Network Data Volker Schneider • • • Philosophical background • Data between theory and facts • Competing Concepts of Networks • Network analysis – Theory or Method? The specificty of network data • Network analysis ~ structural analysis • Relational data vs. attribute data • Representation of relational data • Areas and levels in relational analysis Four ways of getting data • Surveys • Statistics • Documentary analysis • Observations From Facts and Data to Theory Naive epistemology Theory Generalization of Empirical rules and patterns More complex and realistic theory of scientific knowledge production General, formal, specific background knowledge, background theories Special Theories and Models, Hypotheses Problematique Operationalisation Facts = Reality = Experience = Data Data = representational forms of facts Mario Bunge (1996) Finding Philosophy in Social Science Facts = Reality No observation and inquiry starts from complete ignorance • • We must know something before we can observe something We must know something before we can formulate a problem and investigate it Prior Knowledge Background knowledge, theories, concepts Research Problem Theories Concepts Models Hypotheses Indictors Es gibt … keine 'voraussetzungslose' Wissenschaft" (Nietzsche) Reality Data Unobservable Facts observable Quelle: Bunge (1967: 456); leicht modifiziert Network data as representational form of network facts • Network data • Numbers: matrices, tables • Graphs: lines, arrows, configuration of lines and arrows • Animation: Movie Observation, Measurement • Network facts (examples) • Biological: resource dependencies in a cell, predator-prey relations in a an ecology • Physical: chemical interaction and reaction • Technical: Power or telecommunication network • Societal: • Social: trust between persons, network of support between community organizations; migration flows between countries, etc • Economical: control of capital among corporations, trade between national economies. • Political: Subordination among persons, military interventions among countries Network ≠ network! At least two competing network concepts • • Network1: Most abstract concept; configuration unspecific → A certain number of relations connecting social entities. • Entities may be individuals, groups, organizations; words, ideas, concepts; production units, resources, etc. • Relations (or links) may be communication, resources exchange and control, participation, membership, complementarity, etc. Network2: Only special classes or forms of configurations → A network is a class of configurations of communications and transaction links between entities of social transactions which are neither pure markets nor pure hierarchies. • Markets are systems of short-term economic transactions where completely independent actors are coordinated through “thin” communication links • Hierarchies are systems with a minimum of transactions and thick communication links Examples for network1 - based on graph theory Multiplex network B C A Dense all channel network B C A F D F B E Star network C D E D Relation a Relation b Relation c E Sparse Network - Chain - A F B C A F D E Examples for network2 based on governance theory Market – communication transactions B communication & A potential transaction C F Hierarchy E Network B B A C A F F D D E D C E What is the “social” in social networks? • • • • Narrow vs broad view • Narrow : Non-hierarchical (horizontal) configuration of intense personal relations that build on trust • Broad : Any configuration involving social entities and relations. In a systemic view a society may be subdivided into • The political system • The economic system • The cultural system In such a perspective, a political network consists of political entities that are interlinked by political relations: • Political entities: Actors, resources, institutions, ideologies, etc. • Relations: Communication, participation, exchange, membership, etc. As the “political” may be differentiated into polity, politics and policy, there may be • Polity networks • Networks of politics • Policy networks Network analysis and social theory There is no unified network theory: Many theoretical orientations that are compatible with “network analysis”: • • • • • • • • • • • • Structural-functionalism Interest intermediation Institutionalism Governance theory Resource dependence Exchange theory Bargaining theory Communication theory Power theory Rational choice Structuralist-individualist approach etc etc Network metaphers: Net, web, circuit The Phillips Machine as Iconic, mathematical and practical model for an economic circuit Approaches, Metaphors and Images of Policy-Networks amorph configurations ordered configurations contacts hierarchy circuit field of power forces zwei Netzwerke Participation, access governance structurs structures of interest intermediation ecosystem netzwork Ecosystem Network analysis, a theory, a method, or both? Network analysis → specific theory with specialized method? Barry Wellman: „ ... broad intellectual approach, not only narrow set of methods. ... Network Analysis implies asymmetric world view“ → Network analysis as “structural analysis”? or Network analysis → Toolbox for specific type of structural analysis - Sociometry - Graph theory - Blockmodelling - Clusteranalyse - Multidimensionale Skalierung (MDS) Network analysis ≈ structural analysis? There are many levels and types of structures in the social sciences: • Super structures: cultural perceptions, beliefs, ideologies • Institutional structures: social differenciation, norms, rule systems • Distributional structures: access, resources, • Relational structures: exchange, communication, friendship, etc. • Infrastructures: relations of production, technical infrastructures Network Analysis is a methodological and conceptual toolbox for the measurement, systematic description and analysis of relational structures Relational data vs. Attribute data and their representation • Attribute data: Properties of reserach objects → Variable analysis Cases Members Staff Budget FoundYear Org A 1000 10 1.000.000 1950 Org B 250 4 500.000 1972 Org C 300 2 100.000 1995 Relational data: Links between research objects → Network analysis Representation by graphs and matrices • • Influence matrix Organisations Graph B Attribution of influence A Communication C Communication matrix A B C A 0 1 1 B 0 0 C 1 0 A B C A 0 1 1 0 B 1 0 1 2 C 1 1 0 (valued matrix) (binary matrix) One-mode network and two-mode networks One-mode network Two-mode network Different classes (sets) of linked entities Organisations Events A 1 B 2 Organisations B A Communication C participation C 2 Egocentric network 3 ego-centric networks: Relations focused and “around” Egos Examples: Support in families Friendship Sexual contacts and HIV Innovation networks around firms a1 a2 a3 A a4 b1 d2 d1 B b2 b3 d4 C d3 d5 Representation of relational data: Influence relations Graph links edges B points Matrix nodes C A Edge list (e.g. relational databases) A,B,1; A,C,1; C,A,1; C,C,2; 121 131 311 332 Nodes are numbered A B C A 0 1 1 B 0 0 0 C 1 0 2 Edge list (for binary matrices) 1 23 3 13 C → C is binarized (2=1). Types and Sources of Relational Data Types • subjective vs. objective • episotic vs. recurrent Sources • Surveys research → Questionnaires • Ethnographic research → Observations • Documentary research → Texts and documents • Statistical research → Statistical data bases • Experimental research → Experiments, simulations Surveys and questionnaires • Row-based: each actor is asked about relations going to other actors (e.g. to give advice). • Row and column based: each actor is asked not only whom they give advice, but from whom they receive it. • • If relations are not confirmed, need for coding criteria: Minimum, maximum, average method. Consensus method: each actor is asked to indicate the relationships among every pair of actors • Since there are always many estimation (N = number of experts) about existing relations, there is need for coding criteria. Based on Borgatti (http://www.analytictech.com/networks/data.htm Boundary specification: Delineation of networks & identification of actors • Complete/total vs. partial networks (Barnes) • Sometimes natural boundaries exist • But in the last instance, networks are global • Approaches for the specification of boundaries (Laumann/Knoke) • Nominalist • Positional • Decisional • Reputational • Expert panels • Event linkages • Realist => Network Reputation • Name generators vs. lists of actors/organizations Perceived relations measured by questionnaire Binary relations. Example Chemicals Control Data Influence reputation in a policy network: The case of German chemicals control Question: “Which organizations in this list had particular strong influence in the policy-making process of the Chemicals Control Law?” The entries had been stored and representation in a binary square matrix Perceived relations measured by questionnaire. 2-mode network R1 A R2 B R3 Control of resources in the German chemicals policy-making network Question: “In order to influence political decisions, organizations need influence resources. On this card I have listed some potential resources” These informations produce a rectangle matrix in which rows are organizations and columns are resources Org Res1 Res2 Res3 A 0 1 0 B 1 1 1 Institutionalized access through committee memberships Two-mode network • • Q: In which of the following committees members of your organizations had representative positions? BUT: Documentary analysis is more reliable • Result: 30 x 20-Matrix of committee memberships (C) Committee memberships BMA BMLEF BMI .. IPU 1 1 1 .. BLAU 0 1 1 .. AGU 0 0 1 .. DFG.E 0 1 0 .. DFG.M 1 0 0 .. .. .. .. .. .. Committees IPU BMG.BIO AGU BLAU DFG.ENV DFG.MAK AgA BGA.AC BMI.ST IMA.CHEM UBA.AG BMT.AC OECD.CP EC.WG ECCOR GDC.OS BGC.AC AC.NS CC.OS EC.APP Communication measured by a questionnaire. Example: The „organizational state“ of Laumann and Knoke (1987:473) Question Would you please make a check in front of the name of all organizations of this list with whom your organization regularly and routinely discusses national energy policy matters! List OIL AND GAS INDUSTRY Production Companies ___001. American Natural Resources Company ___002. Ashland Oil Inc ___003. Atlantic Richfield Co. We do ___ ___ ___ Both ___ ___ ___ They Do ___ ___ ___ One-mode binary network: Sending & receiving scientific information Question: a. “… From which organizations that are named on the list your Organization (ORG) is receiving scientific information?” b) “To which organization you are sending scientific information?” Valued relations by survey: Research on BIAs Please mark those organisations, which from the perspective of Your association are • relevant (1) or • especially significant (2) to activities within Your sector. 1 1 2 3 … ANGA - Verband privater Netzbetreiber - Satelliten- und Kabelkommunikation – e.V. BDZV - Bundesverband Deutscher Zeitungsverleger e.V. 2 BITKOM - Bundesverband Informationswirtschaft, Telekommunikation und neue Medien e.V. …. Continous observation in the Bank wiring observation room • • Principle • observer in room or area records all interactions and relations that exist or take place over a continous periode in front of the observer Example • Study done at the Western Electric Company, Hawthorne Works, Chicago (Roethlisberger/ Dickson) • Records: • Categorial: Quality, physical examinations, mental tests • Relational: Friendship, conflicts, support, etc. Bank wiring observation room Helping and gaming relations Observation of randomly chosen time intervals in ethnology (time allocation method) Principles Time allocation method. Observer shows up at random times and records who is doing what to whom over a short interval. Goal is to have large number of these snap-shots Example Hames, R. (1994) Ye'kwana Time Allocation Data Base Community in the Amazon: 18,000 records of 88 residents of the village of Toki (1974-1976). Reference: Borgerhoff Mulder, M. Most individuals were recorded more and T. Caro (1985). The use of than 300 times at random hours of the quantitative observation techniques day over a period of many months. in anthropology. Current Anthropology 26: 232-262. Documents -Texts - Statistics Principle • Relations are coded on the basis of written or stored documents • Statistical data bases Examples • Interlocking directorates (→ handbooks, cd-roms, ect. ) • Joint ventures, alliances among companies • Diplomatic and military interaction among countries • Migration flows • Historical marriage records among 15th century italian families • Intracampus email Experiments Principle • Relations between research objects are triggered by measures of experimental control • Computer simulations Examples • Planting of rumors in schools or colleges and observed the spread over time. • Small World Experiments • Stanley Milgram: Examination of how many links are required to connect any two randomly chosen persons in the US. S. Milgram. and J. Travers. ``An experimental study of the small world problem.'' Sociometry, 32(4), pp. 425-443, 1969 The problem of random sampling Can you use a sampling method to study complete networks? • Possible for multiple egocentric networks • Very problematic for complete network, because of very unequal distribution of network relations Policy Network in Chemical Control BEUC ICE BGC CEF BAM EC DIN ILO FHG GDCH DFG OEKI BUND AGV BBU BMA IGC WHO BEE VDB BAU DAG VAA FDP VCI DNR UBA CDU BGA GSF DH SPD DKFZ BBA BMFT BMI BMG AGU BT BMW Legend 0ECD BML UNEP KFA BR SRU IHT Policy actors Information exchange Centrality & influence within policy actor system 90 Centrality based on closeness information exchange VCI 80 IGCPK DFG 70 BMFT UBA BMI BMA BAU BGA BR Zero influence repuation: BMWI EG GDCH AGV GSF 60 BBA 0ECD CDU FDP BUND BAM BT FHG SRU DNR DIN BBU BGCHE AGU CEFIC SPD DKFZ UNEP VAA DIHT KFA WHO ILO VDBWA 50 DH ICEF DAG BMJFG AGU DH DNR DAG DKFZ KFA FHG DIN BMELF BEE BEUC OEKOI UNEP ILO WHO BEUC BEE ICEF 40 0,0 1,0 2,0 3,0 Normalized influence reputation 4,0 5,0