Increased Connectedness As a Function of Organized Inequality in Directed Networks: Illustrated by the Network of Placements Among U.S. Sociology Departments Scott L. Feld, Purdue University and Michael G. Bisciglia, LSU • Connectedness is Important • Inequality Makes Connections • This paper is about 4 particular structures of inequality, and 3 particular types of indirect connections in directed networks. • In directed networks, each node has an indegree, the number of ties directed towards the node, and an outdegree, the number of ties directed away from the node. • In a closed network, the mean outdegree and mean indegree are necessarily equal to one another and equal to the total number of ties divided by the number of nodes. • However, outdegrees and indegrees can have very different amounts of variation from one another. The four properties of structure of inequality: 1)Inequality of Outdegree. The variance of outdegree is an indicator and the standard deviation divided by the mean is a standardized indicator of inequality of outdegree in the network. 2)Inequality of Indegree. The variance of indegree is an indicator and the standard deviation divided by the mean is a standardized indicator of inequality of indegree in the network. 3)Correlation of indegree and outdegree. The correlation between indegree and outdegree over nodes is a normalized indicator of the extent to which nodes that receive more ties also send more ties. 4)Correlation of outdegrees. The correlation between outdegrees of nodes that are connected by ties is a normalized indicator of the extent to which nodes that send more ties tend to send ties to nodes that send more ties. Implications of These Properties for The Overall Pattern of Relationships:_ Undifferentiated Network That Has None of the Designated Properties Figure 1: Contrasting Networks With and Without the Designated Properties An Undifferentiated Network That Has None of the Designated Properties A Differentiated Network That Has All of the Designated Properties This second network is a core-periphery situation. According to Borgatti and Everett (1999), the defining features of a core-periphery situation are a dense core, a less dense connection between core and periphery, and a still less dense periphery. The network displayed here meets these conditions. We suggest that a network that has all four properties that we describe will always have the form of core-periphery with the high outdegree nodes in the core. However, a)there are networks that have some of these properties that are not coreperiphery networks, and b)there are core-periphery networks that do not have all these properties. The four properties together seem to be sufficient but not necessary for a core-periphery network. These logical connections require further investigation. We use the language of family and kinship to facilitate discussion of the various types of short paths. The primitive direct relation: Parents A–>B Three indirect relations: Siblings B<--A-->C Grandparents A–>B–>C Uncle/Aunt C<--A-->B–>D These are the two and three step relations that we consider most likely to be consequential. We do not consider any in-law relations or great-grandparent relations. The salience of indirect relations is ultimately an empirical issue in each context. Numbers of sibling relationships (1) s = x2 + o2 s = mean number of sibling relations per node x = the mean number of ties per node o2 = variance in outdegree among nodes Numbers of grandparent to grandchild relations (2) g = x2 + rio * i * o g = mean number of grandparent relations per node x = the mean number of ties per node rio = correlation between outdegree and indegree across nodes i = standard deviation of indegree among nodes o = standard deviation of outdegree among nodes Numbers of uncle/aunt to niece/nephew relations (3) u = s * g / x + rout * s * g * x u = the mean number of uncle/aunt relations per node s = mean number of sibling relations per node g = mean number of grandparent relations per node rout = the correlation of outdegrees over the set of ties s = the standard deviation in the number of siblings of all ties g = the standard deviation in the number of grandparent to grandchild relations mediated by all ties We use the network among PhD granting departments of sociology in the United States as an illustration of how the properties of inequality in networks is related to the connectedness of networks. We specifically examine the network among the training institutions that is created by the graduates of training organizations who currently staff other training organizations. Data from the 2001 Guide to Graduate Departments of Sociology, published by the American Sociological Association. We examine the network among 111 PhD schools formed by the 1,881 faculty who received their own PhDs from another of these same 111 schools. A node is a sociology department at a university. Its indegree is the size of the department. Its outdegree is the number of placements of its alumni as faculty in other departments. The mean department faculty size is equal to the mean number of department placements; in the system of sociology departments, these means are both 16.95. Variation in Numbers of Placements The placements of these schools in other departments range widely from 0 to 138 with a mean of 16.95 and a standard deviation of 25.89. The top 5 schools accounted for 30% of the placements. These were Wisconsin, Chicago, University of California, Berkeley, Michigan, and Harvard. Variation in Department Size The sizes range from 4 to 50 with a mean of 16.95 and a standard deviation of 7.52. Correlation of indegree and outdegree Larger departments have more placements; this correlation is +0.39. Correlation of outdegrees There is a strong tendency for all departments to place their alumni in departments that make fewer placements than they make (see Feld, Bisciglia, and Ynalvez, 2004). Nevertheless, departments that make more placements tend to place their alumni in other departments that make more placements; r = +0.213. Variation in Placements and the Number of Sibling Relations There are 233% more sibling relations among departments than if there was no variation in outdegree; that is more than 3 times the number of sibling relations expected with no variation in placements. Variation in Placements and Size, Correlation Between Placements and Size, and the Number of Grandparent Relations There aret 26% more grandparent to grandchild relations would be expected with no variation in indegree or outdegree or no correlation between indegree or outdegree. All Four Properties and the Number of Uncle/Aunt Relations The large numbers of sibling and grandparent relations imply large numbers of uncle/aunt relations, over 4 times what would otherwise be expected. In addition, the correlation of outdegrees increases that another 24%. The number of uncle/aunt to niece/nephew relations is more than 5 times what would be expected in the absence of this structure of inequality. Inequality is Magnified with Connectivity Our central conclusion has been that the prominent structure of inequality increases the connectedness of the system as a whole by increasing the frequencies of consequential short indirect relations among the nodes. However, the increased connectedness that results from inequality is not equally shared by all the nodes. Rather, the nodes that have more direct parent to child connections tend to have more of every other type of connection that is built on those parent child connections. We have not presented any analytic results that show a formal connection between the four properties of inequality in direct relations and the inequality of expressing these various indirect relations. However, the empirical results show much greater inequality in participation in many indirect relations than in the direct relations. Summary and Conclusions We have presented several properties that we expect to be typical of networks of training organizations. These are inequality of outdegree, inequality of indegree, correlation of outdegrees and indegree, and correlation of outdegrees. We show that when these properties are present, the result is a much larger set of connections in the network than would otherwise be present. We use our particular example of the network of sociology PhD granting departments to illustrate the extent of the properties and of the consequent connections in the network in one situation. We suggest that networks among all kinds of training organizations are generally characterized by the structures of inequality that we have described that lead to high levels of connectedness. Empirical research is required to explore the extent to which these indirect connections have important consequences in terms of both overall increased communication and influence, and especially disproportionate influence of certain actors. Finally, we suggest that similar types of implications should be explored for other types of directed networks; e.g. networks of affect among persons, citations among journals, and officer directors among corporations. Table 1: Frequencies of Indirect Relations Among Sociology Departments Parents A–>B Siblings B<--A-->C Grandparents A–>B–>C Uncle/Aunt C<--A-->B–>D PARENTS A–Top 5 Bottom 45 B–Top 5 Bottom 45 SIBLINGS GRANDPARENTS UNCLE/AUNT 563(29.9%) 65,423(62.0%) 17,164(42.6%) 1,962,130(70.2%) 78( 4.1%) 262( 0.2%) 541( 1.3%) 1,792( 0.1%) 125( 6.6%) 9,575( 9.1%) 14,833(36.8%) 1,121,783(40.1%) 626(33.3%) 27,465(26.0%) 1,173( 2.9%) 60,121( 2.2%) SAME AS B 2,498( 6.2%) 300,954(10.8%) SAME AS B 13,283(32.9%) 638,666(22.9%) C–Top 5 Bottom 45 D–Top 5 182,964( 6.5%) Bottom 45 TOTAL 111 876,820(31.4%) 1881 105605 40313 2794824