Wasserman, Stanley, and Katherine Faust. (2009). Scial Netwrk Analysis: Methds and Applicatins, Structural Analysis in the Scial Sciences. Chapter II: Scial Netwrk Data: Cllectin and Applicatins What are netwrk data: scial netwrk data cnsist f at least ne structural variable measured n a set f actrs. Structural variables: measure ties f a specific kind between pairs f actrs. Cmpsitin variables: measure actr attributes. Mde: a distinct set f entities n which the structural variables are measured. Bundary specificatin and sampling: a researcher must identify the ppulatin and figure ut hw t sample when necessary. Ppulatin: wh are the relevant actrs? It is assumed that we can btain relevant infrmatin n all substantively imprtant actrs and these actrs cnsist f all scial units n which we have measurements Sampling: a sample f actrs might be taken when it is nt pssible t take measurement n all the actrs Types f netwrk: netwrks are categrized by the nature f the sets f actrs and the prperties f the ties amng them. One-mde netwrk: a single set f actrs + ne r mre types f relatins between pairs f the actrs + actr attributes Tw-mde netwrk: Dyadic tw-mde netwrk: tw sets f actrs + ne r mre types f relatins between actrs in the tw sets; Affiliatin netwrk: ne set f actrs and ne set f events + attendance/membership + attributes f the actrs and the events Eg-centered and special dyadic netwrks: cuples; mthers-children; eg-centered netwrk Netwrk data, measurement and cllectin: Measurement: scial netwrk data cnsist f ne (r mre) relatins measured amng a set f actrs Unit f bservatin: actr/dyad/triad/subset f actrs/netwrk Mdeling unit: actr/dyad/triad/subset f actrs/netwrk Relatinal quantificatin: directinal vs. nn-directinal; dichtmus vs. valued Cllectin: techniques used t gather netwrk data Questinnaire: rster vs. free recall; free vs. fixed chice; ratings vs. cmplete ranking Interview: when questinnaires are nt feasible; eg-centered netwrk
Observatin: small grup f peple; when questinnaire and interview are nt feasible; affiliatin netwrk data Archival recrds: lngitudinal relatins and ties existing in the past Other: special netwrk designs Cgnitive scial structure: respndents give infrmatin nn their perceptins f ther actrs netwrk ties Experimental: selected actrs (and specified pairs) Eg-centered: egs and alters Small wrld: the length f the chain and the characteristics f the actrs Diary: persnal netwrk Lngitudinal data cllectin: hw ties in a netwrk change ver time Measurement validity, reliability, accuracy, errr: true structure vs. bserved structure Accuracy: the accuracy f the verbal reprt; lng term pattern Validity: cnstruct validity Reliability: test-retest cmparisn; cmparisn f alternative questin frmats; reciprcity f scimetric chices Measurement errr: errr in fixed chice data cllectin design Data sets fund in these pages Krackhardt s High-tech Managers: ne-mde (21 actrs); three relatins; fur attributes; questinnaire Padgett s Flrentine Families: ne-mde (16 actrs); tw relatins; three attributes; archival Freeman s EIES Netwrk: ne-mde (32 actrs); tw relatins; tw attributes; archival? Cuntries Trade Data: ne-mde (24 actrs); five relatins; fur attributes; archival Galaskiewicz s CEOs and Clubs Netwrk: tw-mde affiliatin netwrk (26 CEOs-15 events); several attributes; interview + archival Laumann, E.O., Marsden, P.V., & Prensky, D. (1989). The bundary specificatin prblem in netwrk analysis. Research methds in scial netwrk analysis (pp. 61-79) Issue under study: the prblem f specifying system bundaries The selectin f actrs r ndes fr the netwrk The chice f types f relatinships amng the actrs Errrs in the system bundary definitin will result in a fundamental misrepresentatin f the prcess under study
Appraches t bundary definitin: tw basic ways Realist apprach: setting netwrk bundaries by definitin Wrks better fr frmally cnstituted grups Nminalist apprach: the investigatr draws the bundary fr his/her wn purpses Definitinal fci fr the inclusin f actrs: Attribute based Psitinal apprach: presence/absence f sme attribute Reputatinal apprach: judgments f knwledgeable infrmants Relatinship based: whether the actrs participate in a specific scial relatinship Snwball sampling Event/activity based: whether the actrs participate in a specific event r activity Multiple fci: using tw r mre f the three fci Illustrative bundary specificatin strategies: 8 bundary specificatin appraches Realist x Attributes (I): attendance at a particular high schl small, tightly bunded grup Nminalist x Attributes (II): Directrs f xxx largest crpratins in year xxxx large netwrks Realist x Relatin (III): primary grup Nminalist x Relatin (IV): small wrld Realist x Activity (V): Street crner sciety Alternative t strategy I Nminalist x Activity (VI): invisible cllege Realist x Multiple fci (VII): Marxian cncept f class fr itself Nminalist x Multiple fci (VIII): American plitical elite Central difficulty with VII and VIII: they cnsume many theretical degrees f freedm On inclusin rules fr relatins: Partial system fallacy: a set f relatinships is analyzed withut cnsidering the entire set f actrs Issues in multiple netwrk studies Ptential generatr Absent ties Bundary specificatin fr activities: The analyst is intrinsically interested in the events The selectin is an intermediate step in describing the structure
Callway, M., Mrrissey, J. P., & Paulsn, R. I. (1993). Accuracy and reliability f selfreprted data in interrganizatinal netwrks. Scial Netwrks, 15(4), 377 398. Intr: Netwrk data in the field f interrganizatinal studies is increasingly used and useful. They help measure the envirnment in which rganizatins wrk. Reliability f data is an pen questin. One strategy t gather data is thrugh secndary data. Fr example, lking int interlcking directrates. Anther strategy is direct researcher bservatin, thugh that had nt been used fr interrganizatinal studies at the time f this paper (1993). A third strategy is thrugh surveys and questinnaires, but studies have shwn peple have difficulty remembering wh they have interacted with unless the interactin is rutine. T lk at the reliability f data, the authrs lk at mutual cnfirmed relatinships frm a large dataset. Dataset: Data were gathered in 1989 (fr a separate study) frm 31-44 mental health service prviders in 4 US metr areas. Key infrmant was identified and interviewed. The infrmant then filled in a questinnaire. Measurement: Tw questins answered n a Likert scale f 0 t 4 frm the riginal survey were used t explre accuracy f the data: T what extent des yur agency receive infrmatin fr crdinatin, cntrl, planning, r evaluatin purpses frm this agency? Hw well crdinated are the activities f yur agency with thse f this ther agency? Symmetrical relatins are assumed n these questins and then crrelatins were calculated. Reliability using rganizatinal agents: Cnfirmed presence r absence f links is used t estimate the reliability r quality f the netwrk data. 2 aspects will be cnsidered: knwledge f wh infrmants have cmmunicated with and their ability t crrectly recall the intensity f the cmmunicatin. The crrelatins were 0.44 fr the first questins and 0.43 fr the secnd questin. Perhaps this is partly because symmetry is nt a gd assumptin. Or because the data are nt reliable. (this sectin cnfused me and therefre is nt well summarized here). Overall percent agreement n an rganizatin by rganizatin basis was abut 70%. Netwrk data is relatively reliable. Systematic errr in cgnitive netwrk data: Des the presence f strng relatins between rganizatins cause infrmants t be mre aware f relatinships? If s, this wuld intrduce systematic errrs in netwrk data. The analysis cnfirms the presence f such errr, but it appears incnsequential. But the strnger the relatinship, the mre likely it is t be cnfirmed.
Marsden, P. V. (2005). Recent develpments in netwrk measurement. Mdels and methds in scial netwrk analysis, 8, 30. Article cmprehensively reviews findings abut netwrk measurement as f 2005. It can be used as a reference. Netwrk study designs: Whle-netwrk studies cnsist minimally f ne set f bjects linked by ne set f relatinships n ne ccasin. Wasserman and Faust call this a ne-mde data set. Expansins f this design include multiple ccasins, multiple sets f bjects (e.g. twmde data), multiple relatins (e.g. cllabratin, advising and friendship) and/r lngitudinal questins. This chapter wn t deal with bject attributes, but it is cmmn t cllect these. Cgnitive scial structure design (CSS) includes measurements f relatinships btained frm multiple surces. Egcentric designs cnsider a fcal bject (eg) and the actrs it is linked t (alters). Setting netwrk bundaries: 3 basic strategies: psitinal apprach (e.g. emplyment by an rganizatin) event-based apprach (e.g. peple ging smewhere 3 times r mre) relatinal apprach expanding selectin k-cre cncept, with researcher varying k. eg-centric studies use the name-generatr questins t set bundaries. Survey and questinnaire methds: Surveys ften used, but best way varies depending n study design glbal questins (like asking wh is in yur netwrk) are nt gd Name generatr instruments fr egcentric netwrks name generatrs identify respndent s alters single name multiple name name interpreters btain inf n the alters and their relatinships cmparing name generatrs varius criteria specific scial exchanges r affective criteria recall, recgnitin and frgetting recall imprves with clser ties pssible slutin is t use recgnitin (frm a predetermined list) rather than recall
Test-retest studies mre than 75% f first-ccasin alters are cited n secnd ccasin. Patterns in the free recall f persns peple tend t remember thers in clusters f scial relatins Meaning and interpretatin f name generatrs interpretatin f relatins (e.g. friendship ) can differ between researcher and respndents. Interview cntext effects cntext influences the interpretatin f a name generatr Interviewer effects interviewer affects survey quite a bit training is necessary Name interpreters mst studies fllw up with these respndents ften becme bred and s researcher must limit questins Additinal instruments fr egcentric netwrks develped t elicit weak ties, fr example instruments fr measuring extensive netwrk size summatin methd suggests that mean netwrk size is between 280 and 290 scale up methds suggest mean f arund 1700 reverse small wrld (RSW) methd suggests mean f 129 psitin generatrs measures linkages t specific lcatins directly (rather than the standard way f identifying particular alters and later identifying their scial lcatins using name interpreters) e.g. d yu have relatinships with peple having any f these 15 ccupatins? ) identifies weak ties Resurce generatr t measure scial capital resurces wned by members f an individual s scial netwrk, which may becme available t the individual measures whether smene is in persnal cntact with anyne having a specific pssessin r capability CSS data These are judgments by several perceivers abut each dyadic relatinship in a netwrk can then lk at: a single bserver s slice f judgments a lcally aggregated structure f judgments by the tw actrs invlved a cnsensus structure based n all judgments abut a given dyad Infrmant biases in netwrk perceptin
eg bias : peple think they re mre imprtant/central t the netwrk than the cnsensus perceptin but fairly reliable Infrmant Accuracy and Cmpetence Crrespndence between reprts and bservatins majr surce f inaccuracy lies in the different respnse sets r threshlds that respndents use when making citatins respndents als tend t grup relatins rather than see them dyadically. bserving scial ties is als difficult Studies f infrmant cmpetence Sme infrmants are better than thers: cmpetence varies centrally psitined infrmants tend t be mre cmpetent Prspective uses f infrmants different strategies fr eliciting CSS-type data frm infrmants Archival netwrk data Data cllected fr ther reasns are useful fr researchers and are less expensive and time cnsuming. validity f such data rests n the crrespndence between measured cnnectins and cnceptual ties f research interest Smetimes they re clse Smetimes they re nt example f research jurnal citatins, which are prblematic cmputer-mediated systems are a ptential surce f data (but they leave ut face-t-face interactin). Observatin Used less and less ften, due t difficulty Cnclusin
Kilduff, M., & Krackhardt, D. (1994). Bringing the individual back in: A structural analysis f the internal market fr reputatin in rganizatins. Academy f Management Jurnal, 87 108. Central argument: structural analyses can be imprved by bringing in individual perceptins. In ther wrds, t sme extent, scial structure in the eye f the behlder. Example f the internal market fr reputatin in rganizatins is given. Were individual perceptins mre imprtant than an bjectively measured scial structure in determining these reputatins? (the answer is yes) Theretical framewrk: balance thery says that smene wh is perceived as a friend f a persn with a gd reputatin will als have a gd reputatin. The evaluatin f reputatin in a cmpany (i.e. the pricing f the individual) is difficult, s peple lk fr signals, especially ties t peple with gd reputatins Hypthesis 1 is that an bserver s perceptin f an individual s perfrmance will be significantly influenced by the degree t which the bserver perceives that individual t have a prminent friend. We assume peple within the cmpany are jckeying fr high reputatin by publicizing their links t prminent thers. perceptual measures will be mre imprtant in this prcess than bjective measures. Each member has a cgnitive map f the scial structure based n his/her perceptins. Hypthesis 2 is that perceptual measures will be mre imprtant in this prcess than bjective measures. Methds: Site small entrepreneurial firm n 1 flr where all emplyees saw each ther regularly. 28 men and 8 wmen questinnaire cmpleted by 92% f emplyees Measures Netwrk indexes: friendship and advice netwrks asked each persn abut the perceptins f every ther persns friendship and advice netwrks Every persn in the firm was listed as chices Actual netwrk structure was fund by nly using ties that tw peple mutually self reprted (fr friendship) Fr advice, mutuality wasn t required.
Independent variable: friend s prminence matrix fcus n each persn s mst prminent friend measured each friend s prminence in fur ways perceived vs. actual netwrk measures netwrk measure f prminence vs. rganizatinal chart Dependent variable: perfrmance reputatin matrix Each persn asked t rate every ther persn n 7-pint perfrmance scale 1 st cntrl variable: jb perfrmance matrix Managers evaluated subrdinates n 7-pint scale 2 nd cntrl variable: frmal status matrix cntrlled fr member being and wner-manager, a manger r a nnmanager. Analysis and results: used Multiple Regressin Quadratic Assignment Prcedure perfrmance reputatin was significantly crrelated with all 4 independent variables when including cntrl variables, being perceived t have a prminent friend helped bst an individual s reputatin as a high perfrmer, but actually having such a friend had n significant effect. Actually being a high perfrmer (as judged by supervisr) als helped bst an individual s reputatin. Hyptheses 1 and 2 are supprted.