Two mode Network PAD 637, Lab 8 Spring 2013 Yoonie Lee Lab exercises • Two- mode ▫ Davis • QAP correlation • Multiple regression QAP Types of network data • One-mode network (M × M ) M1 M2 M1 0 1 M2 1 0 • Two-mode network (M × N ) Two mode network data • Check “davis.##h” in UCINET ▫ These data were collected by Davis et al. in the 1930s. They represent observed attendance at 14 social events by 18 Southern women. ▫ The result is a person-by-event matrix (18-by-14) • Two mode networks are also called as “affiliation" data ▫ because they describe which actors are affiliated (present, or members of) which macro structures. Two mode network data • Basic idea of two-mode in davis data: ▫ By examining patterns of which women are present (or absent) at which events, it is possible to infer an underlying pattern of social ties, factions, and groupings among the women. ▫ At the same time, by examining which women were present at the 14 events, it is possible to infer underlying patterns in the similarity of the events. ▫ Let’s do a visualization with two mode data! Go to Netdraw with davis data Ron Breiger (1974) “The duality of persons and groups” calls for attention to the dual focus of social network analysis on (1) how individuals, by their agency, create social structures (e.g., individual choice meaning of events) (2) while, at the same time, how social structures develop an institutionalized reality that constrains and shapes the behavior of the individuals embedded in them. (e.g., events choice of individuals) Two mode network data • With “davis.##h” in UCINET ▫ Convert two mode to one mode ▫ In UCINET> Data> Affiliations (2-mode to 1mode) Rows Columns Who is most active actor? Which event is the biggest one? Let’s do some centrality analysis with this converted matrix • Let’s do the Freeman’s degree centrality measure with converted women-by-women data • What does this mean? • What does 57 mean here? • Think about converted matrix! Theresa had 57 connections through attending events. Let’s see the converted one mode network for a moment… • What does this mean? • What does 57 mean here? Theresa had 57 connections with others through attending events. Let’s do betweenness analysis with converted one-mode data • Let’s do the Freeman’s betwenness centrality measure with converted event-by-event data • What does this output mean? • What does highest values of betweenness mean here? • Think about our previous visualization! Two mode network – Bipartite graph • Let’s creating a bipartite graph with “davis.##h data” ▫ Transform> Graph Theoretic> Bipartite … Finally, we have a square matrix format!!!! Two mode network – Bipartite graph • Let’s do the Freeman’s degree centrality measure with this bipartite graph What does “14” imply here? What about “8” ? Two mode network – Bipartite graph • Let’s do the Bonacich power centrality measure with this bipartite graph • Q1: Which woman has the highest power? • Q2: Which event has the highest power? And Why? Two mode network – Bipartite graph • Let’s do the Bonacich power centrality measure with this bipartite graph • Q1: Which woman has the highest power? • Q2: Which event has the highest power? And Why? • Theresa’s power centrality depends on centralities of events where Theresa attended. • E8 power centrality depends on centralities of its members. Two mode network data • Let’s do the Core-periphery analysis with “davis.##h” in UCINET ▫ Networks> 2-Mode Networks> 2-Mode categorical core/periphery model Core=A cluster of frequently co-occurring actors & events Periphery= (1) actors who are not co-occur to the same events and (2) disjointed events because they have no common actors. 0 =bad fit, 1 = excellent fit Need your judgment.. Core=A cluster of frequently co-occurring actors & events Periphery= (1) actors who are not co-occur to the same events and (2) disjointed events because they have no common actors. QAP(Quadratic Assignment Procedure) • The basic idea of QAP is: ▫ To identify systematic connections between different relations Correlation M1 M2 M3 M1 0 1 1 M1 0 1 1 M2 1 0 1 M2 1 0 1 M3 1 1 0 M3 1 1 0 Relation 1 M1 M2 M3 Causality Relation 2 QAP(Quadratic Assignment Procedure) • In UCINET ▫ Tools> Testing Hypotheses> Dyadic (QAP) … 1. QAP correlation 2. Multiple regression QAP 1.QAP Correlations • Computes the Pearson correlation all pairs of a set of equally sized square matrices (With SAME actors!!) • Then, assess the frequency of random measures as large as actually observed. • Basic question here: If two actors have a strong tie of one type, are they also likely to have a strong tie of another? • E.g., If Yoonie and Bill have a strong relation in their friendship, do they also have a strong relation in money exchange? What do you think? It might be an empirical question… 1.QAP Correlations • So, we’d like to test whether or not these two (i.e., friendship and money exchange) relations are positively related or randomly these relations are observed. • Algorithms of QAP: ▫ Calculate Pearson’s correlate coefficient between corresponding cells of two matrices ▫ Randomly and repeatedly permute rows and columns of two matrices to compute whether or not a random measure is larger than or equal to the observed relations between the matrices 1.QAP Correlations • Let’s do some exercise. • Here, we will test whether or not organizational information sharing relations is positively associated with monetary exchange relations. • Data: Knoke & Wood (1978) collected 10 organizations’ two relationships: information sharing (KNOKI) and monetary exchange (KNOKE) Result correlation between the two matrices Q: Information sharing is positively related to money exchange here? Result Descriptive results from random 5000 permutations Result proportion of randomly generated correlations that were as large (or small if they are negatively correlated) as the observed Simply, it is similar to p-value in statistics! Wait! Here, what should be our null Hypothesis? Can we randomly observe this “correlation” value (=-0.0508)? 1.QAP Correlations: Another example- Trade data • Let’s do another QAP correlation analysis! • H: Crude materials trade relations between 24 countries will be positively associated with manufactured goods trade relations between 24 countries. • Data: ▫ CRUDE_MATERIALS ▫ MANUFACTURED_GOODS • Analysis: QAP correlation in UCINET 2.Multiple Regression QAP • Now, we’d like to focus on causality between different types of relations, rather than correlations. • The independent variables should be the matrices format with the SAME actors. ▫ E.g., different relations, attributes of the pairs in the network 2.Multiple Regression QAP M1 M2 M3 M1 0 1 1 M2 0 0 1 M3 0 1 0 M1 M2 M3 Independent Relation 2 M1 M2 M3 M1 0 1 1 M2 1 0 1 M3 1 1 0 Independent Relation 1 Causality M1 0 1 1 M2 1 0 1 M3 1 1 0 Dependent Relation 2.Multiple Regression QAP • Example: • Q: Political relations would be affected by international trade. “Is diplomatic relations affected by different types of trade relations between countries?” Trade relations between countries Diplomatic relations between countries 2.Multiple Regression QAP • Data: Countries Trade Data ▫ Five types of relations 1. 2. 3. 4. 5. MANUFACTURED_GOODS FOODS CRUDE_MATERIALS MINERALS DIPLOMATIC_EXCHANGE Independent VARs Dependent VAR 2.Multiple Regression QAP • Tools> Testing Hypotheses> Dyadic (QAP)> MR-QAP Linear Regression> Double Dekker Semi Partialling MRQAP 2.Multiple Regression QAP: Another example- Friendship & Advice • Data: Krackhardt (1987), five actors’ 1 friendship and 3 types of advice networks ▫ ▫ ▫ ▫ Krac_Friend_QAP = friendship network Krac_Advice1_QAP = advice 1 network Krac_Advice2_QAP = advice 2 network Krac_Advice3_QAP = advice 3 network • H: Friendship network will be positively affected by different types of advice networks. 2.Multiple Regression QAP: Application using attribute data M1 M2 M3 M1 M2 M3 M1 0 1 1 M1 0 1 1 M2 0 0 1 M2 1 0 1 M3 0 1 0 M3 1 1 0 Independent Relation 2 M1 M2 M3 0 1 1 M2 1 0 1 M3 1 1 0 M1 Independent Relation 1 Causality Dependent Relation BUT!! Usually, it’s not easy to collect multiple relational data with the same actors. Are there any ways for us to use this approach even if we have only one relational data and some attribute data? 2.Multiple Regression QAP: Application using attribute data M1 M2 M3 M1 M2 M3 M1 0 1 1 M1 0 1 1 M2 1 0 1 M2 1 0 1 M3 1 1 0 M3 1 1 0 Attribute 2 Causality M1 M2 M3 0 2 1 M2 2 0 3 M3 1 3 0 M1 Attribute 1 Dependent Relation How about converting actors’ “attribute” into “matrix ” format? Can we create similarity or dissimilarity matrix between actors based on their attributes? 2.Multiple Regression QAP: Application using attribute data M1 M2 M3 M1 0 1 1 M2 1 0 1 M3 1 1 0 Attribute 2 Causality M1 M2 M3 0 2 1 M2 2 0 3 M3 1 3 0 M1 Attribute 1 ?? What does “1” imply here??? M1 M2 M3 M1 0 1 1 M2 1 0 1 M3 1 1 0 Dependent Relation How about converting actors’ “attribute” into “matrix ” format? Can we create similarity or dissimilarity matrix between actors based on their attributes? 2.Multiple Regression QAP: Application using attribute data M1 M2 M3 M1 0 1 1 M2 1 0 M3 1 1 M1 M2 M3 M1 0 1 1 1 M2 1 0 1 0 M3 1 1 0 Attribute 2 Causality M1 M2 M3 0 2 1 M2 2 0 3 M3 1 3 0 M1 Attribute 1 E.g., 1 = M1 and M2 have same attribute Dependent Relation How about converting actors’ “attribute” into “matrix ” format? Can we create similarity or dissimilarity matrix between actors based on their attributes? 2.Multiple Regression QAP: Application using attribute data • Question to think: ▫ To what extent can advice network between managers in a company be explained by their age, length of service or tenure, level in the corporate hierarchy and department? 2.Multiple Regression QAP: Application using attribute data • Data: ▫ Krackhardt D. (1987). collected network data from the 21 managers of a high-tech company on the west coast of the United States. This company had just over 100 employees with 21 managers. • 3 Networks = ADVICE, FRIENDSHIP, & REPORTS_TO • 4 attributes of 21 managers 1. age (in years), 2. length of service or tenure (in years), 3. level in the corporate hierarchy (coded 1,2 and 3; 1=CEO, 2 = Vice President, 3 = manager) 4. department (coded 1,2,3,4 with the CEO in department 0 ie not in a department). 2.Multiple Regression QAP: Application using attribute data • Let’s check the dataset ▫ Networks ▫ Attributes 2.Multiple Regression QAP: Application using attribute data Age difference relations Tenure difference relations Advice network Hierarchical Level difference relations Same department relations • Now, we will convert attribute data into matrix format ▫ Data> Convert Attribute to Matrix How can we interpret this result? What does it mean? Employees with relatively short tenure periods are going to those with relatively long tenure periods for advice 2.Multiple Regression QAP: Application using attribute data • Another example: Age difference relations Tenure difference relations SAME Hierarchical Level relations Same department relations Friendship network • First, create SAME hierarchical level matrix by converting attribute data into matrix format ▫ Data> Convert Attribute to Matrix For more information... • “QAP – the Quadratic Assignment Procedure” (William Simpson, Harvard Business School, 2001). • Decker, Krackhardt, Snijders, “Sensitivity of MRQAP Tests to Collinearity and Autocorrelation Conditions”
© Copyright 2026 Paperzz