P208 (02 473P208)
Methodological Advances in the Study of Corporate Networks Fracturing and Concentration
Form of presentation:
The Promise and Perils of Using Big Data in the Study of Corporate Networks: Problems, Diagnostics and Fixes
Eelke Heemskerk | University of Amsterdam | Netherlands
Network data on connections among corporate actors and entities – for instance through co-ownership ties or elite social networks – is increasingly available to researchers interested in probing many important questions related to the study of modern capitalism. We discuss the promise and perils of using Big Corporate Network Data (BCND) given the analytical challenges associated with the nature of the subject matter, variable data quality, and other problems associated with currently available data at this scale. We propose a standard process for how researchers can deal with BCND problems. While acknowledging that different research questions require different approaches to data quality, we offer a schematic platform that researchers can follow to make informed and intelligent decisions about BCND issues and address these issues through a specific work-flow procedure. Within each step in this procedure, we provide a set of best practices for how to identify, resolve, and minimize BCND problems that arise.
Multi-layer Motif Detection Algorithms for Understanding Corporate Networks
Frank Takes | University of Amsterdam | Netherlands
Corporate networks, in which firms are connected based on for example ownership, debt, trade and shared board members, have been shown to contain a wealth of information. The analysis of these types of networks has thus far focused on analyzing one type of relationship. In this work we go one step further and study multiple layers of interaction in the global corporate network. Furthermore, we will not merely look at how micro (organization) level interaction leads to behavior at the macro (system) level. Instead, we aim to detect higher order patterns at the meso level : distinct network motifs of groups of firms interacting within and between multiple layers of the network.
A network motif is a frequently occuring subpattern consisting of more than one node. The significance of a network motif can be established by comparing such a subpattern with the frequency of that pattern in a similar randomly generated network. To attain the above mentioned goal of multi-layer motif detection, we modify an existing algorithm by means of a novel layer encoding scheme. Furthermore, the null model against which significance is tested is adjusted such that the multi-layer aspect is properly captured. This process is nontrivial because there is a large degree of interlayer assortativity and because particular layers may contain bipartite relations.
The multi-layer motif detection algorithm is applied on a large corporate network dataset consisting of millions of ownership and board-interlock relations between firms across the globe. The motifs found this network show interesting patterns, revealing for example the presence of block holdings and joint ventures. More importantly, using metrics of motif complexity, we investigate the behavior of particular industry sectors and their involvement in complex motifs. This reveals not only how the financial sector is over-represented in more complex motifs, but also how certain industry sectors more frequently re-use particular corporate structures. These results not only pave the way for a better understanding of the organization of corporations, but also for automatically detecting patterns of financial risk and complexity at a global scale.
Pruning large corporate networks for core identification
Anton Grau Larsen | Copenhagen Business School | Denmark
This paper describes challenges for constructing large, longitudinal and inclusive networks of corporate interlocks suitable for core detection analysis. We describe the three distinct steps in shaping register data to core analysis; 1) data selection, 2) data cleaning and 3) network pruning.
The goal of the core analysis a group which has a profile that is compatible with the inner circle originally identified by Michael Useem. That is the most central agents that bridge between several major corporations and which is the dominant political voice of the corporate world. The core is identified with the K-core decomposiition.
The available data lets us identify an inner circle within the Danish corporate network for each month in the period from 1987 to 2016. To not underestimate the degree of integration in the corporate network we increase the sample from the customary sample size of less than a thousand boards to more than 560.000 boards, 700.000 directors and 3.600.000 positions. This sample is substantial but far from the entire register. We discuss the problems with sampling from large registers, in particular problems with shell corporations and foundations. As a solution we propose sampling all boards that link between corporations with at least 10 employees.
We move on to discuss the problems of cleaning the time series networks for missing or unrealistic data and reducing the influence of “ultra-connectors”. As part of the solution we propose a spell graph – a graph made from spells of board memberships. In the empirical case the sample is reduced to 213.000 boards, 432.000 directors with 1.285.000 positions.
Finally we discuss the problems of redundancy in networks and the problems it poses. Redundancy in affiliation networks is when a mother corporation has several subsidiary corporations with separate but almost completely overlapping boards. This poses a problem where it becomes difficult to determine whether a director actually bridges different corporations or just holds several board positions within a corporation. Redundancy poses problems both for the the concept of the inner circle and for the core decomposition. As a solution to redundancy we propose a pruning technique dubbed betweenness decomposition. The procedure iteratively removes directors with a betweenness less than 1 in their 3rd neighborhood. The resulting network only contains directors that are essential for the connection of the network and it suitable for core identification. We apply the principle to the empirical case and identify a relatively stable core with a size of around 500 in each month in the period.
In the end we discuss the different types of affiliation network for which these problems exist and where variants of these solutions might apply and how it might be applied to weighted networks.
Delineating the Corporate Elite: Inquiring the Boundaries and Composition of Interlocking Directorate Networks
Jouke Huijzer | University of Amsterdam | Netherlands
Researchers of corporate elites typically study samples of directors and executives comprising, say 50, 100, 200 or 500 largest firms within a particular region. While these studies have revealed important patterns of corporate elite organization, the demarcating criteria of the group under study are rather arbitrary and poorly linked to the concepts that designate the group. This is problematic because decisions for demarcating the group under study likely affect empirical outcomes and thus impair a comprehensive understanding of the corporate elite, especially when they are compared over space and time (Mintz, 2002). In an effort to meaningfully determine a sample of organizations whose directors comprise the corporate elite, this study empirically compares various demarcations of the corporate elite using data from Canada. First, it is demonstrated that decisions for particular sampling criteria can significantly affect network properties and the conclusions drawn from it. Second, I explore alternative sampling strategies that align better with our theoretical understanding of the corporate elite and compare the new demarcations with conventional ones. I show that compared to conventional demarcations, our alternative strategy performs equally well at delineating a corporate elite that is connected and willing to promote its group interests. Finally replicate the landmark studies in Stokman, Ziegler & Scott (1985) and demonstrate how our understanding of networks of corporate power would have been different if alternative sampling criteria were applied. The findings enhance a more robust understanding of corporate elite organization and facilitate better comparisons of corporate elite networks over space and time.