1 / 48

Ego-centered Network Analysis Meredith Rolfe, Oxford University

Ego-centered Network Analysis Meredith Rolfe, Oxford University. Using Sample Surveys to Study Social Networks Connect to wireless D ownload file http :// dl.dropbox.com / u /7685173/polnet2011.R. Whole Networks…. Source: Christakis and Fowler, 2008. ….vs. EgoNets.

woods
Télécharger la présentation

Ego-centered Network Analysis Meredith Rolfe, Oxford University

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Ego-centered Network AnalysisMeredith Rolfe, Oxford University Using Sample Surveys to Study Social Networks Connect to wireless Downloadfile http://dl.dropbox.com/u/7685173/polnet2011.R

  2. Whole Networks…. Source: Christakis and Fowler, 2008

  3. ….vs. EgoNets

  4. Pros and Cons of Egonets Whole Networks Ego Networks General Sample Included in widely available attitude and election studies (NES, GSS, South Bend, ILS, NCCS, BES, CCAP) Respondent burden Recall/Reporting errors Sample attrition • Often limited samples (college/high school students, work groups) • Political attitudes are often tack-ons, if collected at all • Subjects enjoy • Lots of well-developed analysis methods • Nice pictures • Recall/Missing Data

  5. The Traditional Approach Ego NetWork Measures: 5 Political Discussants

  6. Surveys with Political Networks data • Columbia School (Elmira) • Detroit Area Studies • South Bend 1985 (Huckfeldt and Sprague 1985) • Indianapolis-St. Louis (1996 ILS) (Huckfeldt and Sprague 2000) • ANES (years) • Spencer Foundation 2000 (Mutz) • General Social Survey (1985/1987) • CNES 1992 • Great Britain, Germany, Japan, Spain, and the United States • Snowball sample of Spouses (D2) and Other Discussants (D3) • CCAP (US, UK, Germany)

  7. Political Discussion Name Generator

  8. Structure of the Data: Original (Wide)

  9. Example: CNES & CCAP • Read in Data cnes.d1<-read.dta("http://dl.dropbox.com/u/7685173/cnes.d1.dta") cnes.d2<-read.dta("http://dl.dropbox.com/u/7685173/cnes.d2.dta") cnes.d3<-read.dta("http://dl.dropbox.com/u/7685173/cnes.d3.dta") ccap<-read.dta(http://dl.dropbox.com/u/7685173/ccap.dta) • Summary of Data summary(cnes.d1) colnames(cnes.d1) summary(cnes.d2) summary(cnes.d3) summary(ccap.nets)

  10. Structure of the Data: Transformed to Dyadic (Long)

  11. Transform the Data to Long Format • Step 1 - “line up” the variables correctly shape<-c(seq(2,10,by=2),seq(3,11,by=2)) for(i in 12:23) shape<-c(shape,seq(i,i+48, by=12)) temp<-NULL for(i in seq(1,66,5)) temp<-c(temp,list(names(cnes.sm)[shape[i:(i+4)]])) • Step 2 – reshape cnes.long<-reshape(cnes.d1, varying=temp, idvar="caseid", timevar="discnum", v.names=c("d.given", "d.name", "d.relate", "d.cowork", "d.church", "d.nghbr", "d.friend", "d.close", "d.educ", "d.discpol", "d.disagree", "d.male", "d.expert", "d.vote"), direction="long", time=1:5) • Step 3 – Do any variable recoding (see polnet20011.R)

  12. Personal Network Size • Discussant Name Given? (d1.given – d5.given) cnes.d1$netsize<-rowSums(cnes.d1[,grep(“d[1-5]_given”, colnames(cnes.d1))]==1, na.rm=TRUE) egennetsize= neqany(d1.given d2given d3given d4given d5given), values(1) • Analysis attach(cnes.d1) table(netsize) hist(netsize, breaks=6) ##alsoseelattice version summnetsize tab netsize

  13. Non-response • No one to talk to tapply(married, netsize==0,mean.na) 35% are married! • Nothing to talk about table(sp_talkpol, married) • Low political interest prop.table(table(polint, netsize==0),2) chisq.test(polint, netsize) • Forgetting • Non-compliance

  14. Online survey issues:Invalid responses • Telephone and Face to Face surveys, no invalid answers that could be verified • Online – there could be many invalid answers! attach(ccap) table(b2.pn1.cat) table(b2.pn1[b2.pn1.cat=="missing (sure)"]) • Invalid answers can increase/decrease depending on non-response format and forced/semi-forced choice options prop.table(table(b2.pnum==b2.pnum.orig))

  15. Political Discussants Named

  16. Network size:Political Discussion sub-network • Need to identify political discussants from important matters discussants table(d.discpol) table(tapply(d.discpol%in%c("often", "sometimes", "rarely"), caseid, sum.na)) • Assign back to “wide” format file cnes.d1$pnum[order(cnes.d1$caseid)]<-tapply(d.discpol%in%c("often", "sometimes”), caseid, sum.na) • Can set different thresholds for discussion table(tapply(d.discpol%in%c("often", "sometimes”), caseid, sum.na)

  17. Graphic: Network Size of Impt. Matters and Political Discussion Networks par(mfrow=c(2,2)) hist(cnes.d1$netsize, breaks=6, main="Important Matters", xlab="Network Size") hist(cnes.d1$pnum, breaks=6, main="Talk Politics Subnetwork", xlab="Network Size") hist(tapply(d.discpol%in%c("often", "sometimes"), caseid, sum.na), breaks=6, main="Talk Politics Sometimes Subnetwork",xlab="Network Size") hist(tapply(d.discpol%in%c("often"), caseid, sum.na), breaks=6, main="Talk Politics Often Subnetwork", xlab="Network Size") ###also see lattice alternative

  18. The Traditional Approach Ego NetWork Description: network Composition

  19. Personal Network Composition • Who does R discuss politics with? Family, friends, coworkers? • Is political discussion primarily a male activity? • How politically interested are R’s discussants? • Does R disagree about politics with discussants?

  20. Traditional Name Interpreters • How is [name 1] connected to you? • spouse or partner • other relative [specify] • unrelated • Is [name 1] a coworker? • Is [name 1] a member of same church? • Is [name 1] a neighbor?

  21. CCAPOnline SurveyPolitical Discussion Name Interpreter

  22. Structure of the Data: Original (Wide)

  23. Structure of the Data: Transformed to Dyadic (Long)

  24. Non-response: a potential issue • Respondents who name NO discussants • Omit Rs from all composition measures • Divide by netsize=0 or NA will omit R • Respondents who don’t provide one or more composition variables • Omit discussants with invalid or missing information from R’s netsize (must adjust manually for each characteristic) • table(d.given[is.na(d.relate)])

  25. Personal Network Composition:Who does R discuss politics with? Family, friends, etc? • Number of discussants that are family members table(tapply(d.relate%in%c(”spouse", ”family”), caseid, sum.na)) • Proportion of discussants that are family members x<-tapply(d.relate%in%c("spouse", "family"), caseid, sum.na)/tapply(d.given==1 & (d.relate%in%c("dk","rf"))==FALSE, caseid,sum.na) summary(x) mean.na(x==0) mean.na(x==1)

  26. Personal Network Composition:Who does R discuss politics with? (continued) • Proportion of discussants that are coworkers x<-tapply(d.cowork=="yes", caseid, sum.na)/tapply(d.given==1 & (d.relate%in%c("dk","rf"))==FALSE & (d.cowork%in%c("dk","rf"))==FALSE, caseid,sum.na) summary(x) mean.na(x==0) mean.na(x>.5)

  27. Personal Network Composition:Is political discussion primarily a male activity? • Proportion of discussants that are male prop.table(table(d.male)[1:2]) • Proportion of POLITICAL discussants that are male prop.table(table(d.male[d.discpol%in%c("often", "sometimes", "rarely")])[1:2]) prop.table(table(d.male[d.discpol%in%c("often", "sometimes”)])[1:2]) x<-tapply(d.male=="male" & d.discpol%in%c("often", "sometimes"), caseid, sum.na)/tapply(d.given==1 & (d.male%in%c("dk","rf"))==FALSE & d.discpol%in%c("often", "sometimes"), caseid, sum.na)

  28. Personal Network Composition:Is political discussion primarily a male activity? (cont.) summary(x) prop.table(table(x==0)) prop.table(table(x>.5)) prop.table(table(x==1)) • Frequency of political discussion with male discussants print(prop.table(table(d.discpol, d.male)[,1:2]), digits=2) chisq.test(table(d.discpol, d.male)[,1:2]) t.test((4-unclass(d.discpol))~d.male, data=cnes.long[d.male%in%levels(d.male)[1:2],])

  29. Personal Network Composition:How politically informed are R’s discussants? (cont.) • Distribution of expertise among all or political discussants print(prop.table(table(d.expert)[1:3]), digits=2) print(prop.table(table(d.discpol,d.expert)[,1:3],1), digits=2) • Do R’s prefer informed discussants? chisq.test(table(d.discpol, d.expert)[,1:3]) • How many R’s have highly informed discussants? x<-tapply(d.expert=="Great deal", caseid, sum.na)/tapply(d.given==1 & (d.expert%in%c("dk", "rf"))==FALSE,caseid, sum.na) table(x==0) table(x>.5)

  30. Political Network Composition

  31. Network composition:Do R and discussants disagree about politics? Self-reported Disagreement Self-reported Vote Choice Differences Which candidate do you think [name] supported in the presidential election this year? table(d.vote, vote) prop.table(table(d.vote== vote))[2] print(prop.table(table(d.disagree, d.vote!=vote)[1:4,],2), digits=2) When you discuss politics with [name], do youdisagree table(d.disagree) print(prop.table(table(d.disagree, d.vote!=vote)[1:4,],1), digits=2)

  32. Misperception of Political Preferences of Discussants: Snowball Sample Huckfeldt & Sprague (1987) Merge data files tp replicate detach(cnes.long) temp<- merge(cnes.d3,cnes.d2, by=c(colnames(cnes.d3)), all=TRUE) cnes.match<-merge(cnes.long, temp,by=c("caseid","discnum"), all=FALSE) cnes.match$correct<-cnes.match$d.vote== cnes.match$act.both.vote attach(cnes.match)

  33. Misperception: Analysis • Does misperception depend on discussant vote? prop.table(table(d.vote==act.vote,d.vote),2) • Does misperception depend on respondent vote? prop.table(table(d.vote==act.vote,vote),2) • Does misperception depend on agreement? print(tapply(correct, list(vote, act.vote), mean.na), digits=2)

  34. The Traditional Approach Ego NetWork Analysis: Participation and Influence

  35. Personal Network Analysis:Traditional Approaches • One time wave (with snowball) • Instrument for discussant vote choice (H&S, 1991) • Two or more survey waves (change) • Kenny (1994) • Nested/hierarchical models • van Duijn, Busschbachand Snijders 1999 • Lubbers et al 2010 • de Miguel Luken and Tranmer 2010 • Respondent driven sampling (snowball) • Gile and Handcock (2010), Goel and Salganik (2010) • Use in UCInet/whole network software if enough ties

  36. Changes in Personal Networks(from Feld, Suitor and

  37. Political Participation • McClurg 2003 • South bend, dyadic transformed • DV: index of political participation • Specification: negativebinomialregressionmodel3 • Controls for socioeconomic status, politically-relevant attitudes, generalized civic engagement, and political mobilization • See Also: McClurg 2006 (“wide” data) • See Also: La Due Lake and Huckfeldt, 1998

  38. Dyadic Particiapation:Spouses, Family, and Friends

  39. “Wide” Participation model

  40. Influence • Kenny (1994) “Microenvironment of Attitude Change” Journal of Politics • OLS on transformed dyadic data • DV is directional change in pid from pre-election to post-election (-6 to 6) • Includes national and local level context variables • Discussion network in 3rd wave

  41. Influence model

  42. The Traditional Approach Ego NetWork Measures: Network Mechanisms

  43. Network Mechanisms & Measures • Weak ties • Neighborsor Coworkers • Strong ties • Contact frequency • transitive ties • reciprocated ties (snowball) • Centrality • Self-report • Social capital • Organisationalmemberships • Network size • Network diversity • Do you know a… • % who disagree • % with diff demographics

  44. Network Styles: Activities

  45. Network Style: Time Use

  46. The Traditional Approach Drawbacks and Significant Issues

  47. Threats to analysis and inference • Mistaken perceptions of discussant political views (minimal) • Sample issues with snowball • Selection vs. influence (difficult to disentangle) • Loosely defined traditional mechanisms • Very narrow conception of how networks might impact political behavior • No network structure • Little thought to social cleavages & social groups (old school)

  48. Solutions? • Longitudinal studies with shorter measures • Formal modeling (mathematical and simulation) identifies new network mechanisms • New questions to tap into the mechanisms & cleavages • Other test implications if we think interaction matters (usually with formal models) • Network structure measures

More Related