1 / 27

Social Network Measures for “Nosduocentered” Networks, their Predictive Power on Performance.

Social Network Measures for “Nosduocentered” Networks, their Predictive Power on Performance. Lluís Coromina, Jaume Guia i Germà Coenders Universitat de Girona Seminari Departament d’Economia. Universitat de Girona (UdG) 18 Gener 2004. Background. Goals:

effie
Download Presentation

Social Network Measures for “Nosduocentered” Networks, their Predictive Power on Performance.

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. Social Network Measures for “Nosduocentered” Networks, their Predictive Power on Performance. Lluís Coromina, Jaume Guia i Germà Coenders Universitat de Girona Seminari Departament d’Economia. Universitat de Girona (UdG) 18 Gener 2004

  2. Background • Goals: • “Nosduocentered Network” structure. • To assess social network measures based on complete networks measures (centrality degree, closeness…) and some tailor-made measures. • Application of these measures in different networks (advice, collaboration…). • Specification of a regression model to predict research performance of PhD students, based on social networks measures.

  3. “Nosduocentered” network

  4. “Nosduocentered” network

  5. “Nosduocentered” network cI cI cO cO aO aO bI eO dI eI dO bO aI bI aO

  6. “Nosduocentered” network • Focus in two main actors, EgoA and EgoB. • Actors who are not defined as EgoA or EgoB are called “alters”. • No relation present between “alters”. • Actors who do not have any line are considered as isolates • Lines can be distinguished between directed or undirected, valued or binary. • Examples: PhD students and supervisors (PhD students performance can not be understood without supervisor’s influence), husband and wife…

  7. “Nosduocentered” network • Advantages and inconvenient with respect to complete networks: • “Alters” are not central in the network; difficult to reach them. • Reduction of the cost and time. • Less problems for non-response and/or data quality problems because in complete networks respondents have to answer about too many people. • Less information is available. • Advantages and inconvenient with respect to egocentered networks: • In many cases a pair and not a single individuals is what is central in a study. • More information is available  a network more real without a lot of effort. • More time to reach actors in the network.

  8. Network measures • Centrality measures: • Degree Centrality. • Closeness Centrality. • Betweenness Centrality. (Cannot be used). • Measures of Centralization • Measures of Density • Tailor-made measures.

  9. Degree Centrality • It counts the degree or number of adjacencies, for a actor pk: • where: • CD(Pk) = number of contacts connected to Egok. • a(pi,pk) = contact for pi to pk. 0 or 1 in binary networks or any non-negative real number for valued networks. • n = network size

  10. Degree Centrality Undirected networks: Binary data: The count of contacts for the ego. Valued data: The sum of egos’ contacts with other actors in the network. Directednetworks: Binary data or Valued data: Outdegree Centrality CDO(Pk). Indegree Centrality CDI(Pk). Relative measure of centrality (C’D(Pk)):

  11. Degree Centrality Undirected nosduocentered network: CD(Pa) = a + c + d CD(Pb) = b + c + e Directed nosduocentered network: CDO(Pa) = aO + cO + dO CDO(Pb) = bO+ cO+ eO CDI(Pa) = aI + cI + dI CDI(Pb) = bI + cI + eI Any of these expressions can be converted into relative centralities by dividing by n-1.

  12. Closeness Centrality • Centrality is obtained using the geodesic paths to reach all actors in a network. • where: • Cc(Pk) = Closeness centrality • d(pi,pk) = number of paths that egok has to follow to reach each actor. • Undirected binary Nosduocentered network. • Cc(Pa)-1 = = 1(a+c) + d + 2b(d) + (3b(1- d) + 2(1- d))*(c>0) • Cc(Pb)-1 = = 1(b+c) + e + 2a(e) + (3a(1- e) + 2(1- e))* (c>0)

  13. Closeness Centrality Directed nosduocentered network This measure can often lead to infinite distances for directed networks.

  14. Centralization Indicator Centralization measures the extent to which the cohesion is organized around particular focal points. The general procedure is to look for differences between centrality scores of the most central point and those of all other points. We only have two egos; therefore we compare one centrality with the other. Centralization standardized. = - Relative degree for EgoA minus the relative degree for EgoB. Centralization can also be computed for Closeness Centrality.

  15. Density Density (Δ)= ratio of number of lines present, L, to the maximum possible. Undirected network: Δ = Nosduocentered undirected binary network: Each of the network members apart from both egos (n-2) can be connected to both and both egos can also be mutually connected, thus (n-2)*2+1 = 2n-3 possible lines. “nosduocentered density” ΔN: ΔN: = It counts d=e only once. A related simple measure that is not bounded between 0 and 1 could be: +

  16. Density Nosduocentered directed binary network: Density indegree and outdegree = 4n-6 = (n-2)*4+2: contact from EgoA to EgoB and vice versa is different. Nosduocentered directed binary network: Only a part of the relationships (indegree or outdegree) is observed. It becomes as (n-2)*2+2: ΔND = Nosduocentered valued network: The denominator is multiplied by the maxim intensity that a line can has. Interpretation: the mean of the strength of the contacts in the network as a whole as a proportion of the maximum possible strength.

  17. Tailor-made measures • To use measures that are as closely related as possible to a, b, c, d and e. • The centrality measures directly related to centrality of EgoAcould be: • a = number or sum of direct contacts from EgoA with alters others than EgoB and EgoB‘s contacts. • c = number or sum of shared contacts among EgoA and EgoB. • d = number or sum of direct contact from EgoA to EgoB. • (d/max)*b = the influence in EgoA from EgoB’s contacts through EgoB, where max is the maxim intensity that a contact can have.

  18. Data, sample and performance (Illustration) Structure of the nosduocentered network: EgoA are PhD students EgoB are their supervisors “Alters” actors are people who belong to the PhD student’s research group. Each PhD and supervisor pair (64 pairs) are asked, among other questions, about four networks questions. Population: PhD students who began in the academic years 1999/2000 and 2000/2001 in Slovenia. Procedure: Design a web questionnaire about PhD students’ performance in research, created within the INSOC (International Network on Social Capital and Performance) research group (De Lange et al. 2004). List of members: we defined theoretically the research group. Then, PhD students were phoned in order to know who their promoter was. Next, personally interview promoters in order to obtain a list of influential research group members.

  19. Data, sample and performance (Illustration) • There were two questionnaires, one for PhD students and other for their supervisors, with the same network questions and alter names. • Then, we create a nosduocentered network for each four different networks for each pair PhD student and supervisor. • a) Scientific Advice network b) Collaboration network • c) Emotional Support network d) Trust network • With this information we can compute the centrality, density, centralization and tailor-made measures for each network. Used as independent variables for the specification of the regression model used to predict research performance of PhD students.

  20. Data, sample and performance (Illustration) Each PhD student is asked about publications, conferences and workshops: “international articles” (int_art). “publications with review” (pub_rev). “normal publications” (pub_norm). “paper conferences” (pap_conf). Index of performance (Y) = 2*int_art + 2*pub_rev + pub_norm + pap_conf Influence of nosduocentered network measures for the networks of Scientific advice, Collaboration, Emotional support and Trust over research performance of PhD students.

  21. Data, sample and performance (Illustration) • Scientific Advice network: Consider all the work-related problems you've had in the past year (namely since 1 November 2002) and that you were unable to solve yourself. How often did you ask each of your colleagues on the following list for scientific advice? • Collaboration network: Consider all situations in the past year (namely since 1 November 2002) in which you collaborated with your colleagues concerning research, e.g. working on the same project, solving problems together, etc. The occasional piece of advice does not belong to this type of collaboration. How often have you collaborated with each of your colleagues concerning research in the past year? • Emotional Support network: Imagine being confronted with serious problems at work; e.g. lack of motivation, problematic relationship with a colleague. To what extent would you discuss these problems with each of your colleagues? • Trust network: In a working environment it can be important to be able to trust people in work-related matters (e.g.concerning the development of new ideas, your contribution to common goals, the order of co-authorship or the theft of new ideas). Consider the following opposite nouns: distrust and trust. The further to the left you tick off a box, the more you associate your relationship with a particular colleague with “distrust”. The further to the right you tick off a box, the more you associate your relationship with that colleague with “trust”.

  22. Models (Illustration) Model 1: The tailor-made measures are used. Frequency of direct contacts for EgoA (PhD student) and the importance of non contacts for EgoA which are contacts of EgoB (supervisor) depending on the frequency of the contact from EgoA to EgoB. The variable faculty is used in all models, regression models fit better. Hypothesis: closest contacts have stronger influence in the performance for PhD students but also supervisor’s contacts are influential if a rather strong relation between PhD and supervisor exists. Y = f ( a, c, (d/max)*b, d, Faculty )

  23. Models (Illustration) Model 2: Research performance of PhD students depend on variables which are relative measures and size. Model 2 for nosduocentered networks: Y = f ( , , n , Faculty ) Interpretation: using sum and difference we are testing the variation for this network. When we sum we consider all contacts between egos and the rest of the network. While when we use the difference of densities, we consider the difference between EgoA from EgoB.

  24. Models (Illustration) Model 3: It uses absolute measures instead of relative measures and size. Y= ( + , - , Faculty ) + = Absolute Density - = Absolute Centralization Each model is used in each of the four different nosduocentered networks (scientific advice, collaboration, emotional support and trust).

  25. Results (Illustration)

  26. Conclusions • In complete networks centrality measures, centralization and density can be measured. • We compute these measures adapted to nosduocentered networks. Moreover, new tailor-made measures have been created specifically for nosduocentered networks. • All three models perform about equally well, to predict performance using nosduocentered or adapted complete network measures. • Model 1: uses exclusively nosduocentered network measures. • Model 2: uses comparable (relative) measures and size. • Model 3: uses absolute measures (the most parsimonious model).

  27. Conclusions In this paper we do not present nosduocentered networks as a cure-all. The ideal situation would be to have the complete network. However, when the complete network is unavailable due to high costs, low accessibility, poor data quality or low response rate, the nosduocentered network still makes it possible to define network measures which are interpretable, which have predictive power on performance, which are easy to compute and which are richer than those would be obtained from egocentered network alone.

More Related