Reliability of measures of centrality and prominence

This paper evaluates the reliability of measures of centrality and prominence of social networks among high school students. The authors present and discuss results from eight experiments. Four types of social support: (1) instrumental support, (2) informational support, (3) social companionship, and (4) emotional support-were measured three times within each class. Four measurement scales: (1) binary, (2) categorical, (3) categorical with labels and (4) line production-were applied. Reliability of in- and out-degree, in- and out-closeness, betweenness and flow betweenness was estimated by the Pearson correlation coefficient. Meta analysis of factors affecting the test-retest reliability of measures of centrality and prominence was done by multiple classification analysis. Results show that, - Global measures (considering direct and indirect choices) are more sensitive to measurement errors than local measures (considering only direct choices). - In-measures are more stable than out-measures. - Among types of social support, emotional support gives the least stable measures of centrality and prominence,whereas social companionship gives the most stable results. - The reliability of centrality and prominence measures is higher when the network is denser. (C) 2004 Published by Elsevier B.V.