Research Interest:
Network Data & Graphical Models
Multivariate Analysis & Machine Learning
Time Series
Bayesian Theory
Computational Social Science
Statistical ERGM analysis for consulting company network data
Park YJ, Um JM, Hong SB, Han YJ, and Kim JH*
Published in The Korean Journal of Applied Statistics, Oct 2022
Abstract
This study aims to investigate the factors influencing the formation of advice-giving relationships within a consulting company's workplace network. Using the Exponential Random Graph Model (ERGM), we analyze network data from 44 employees across offices in the United States and Europe.
The main objectives of this research are:
To identify which employees play a significant role in the advice network.
To determine which employee characteristics, such as gender, rank, and geographical location, influence the likelihood of forming advice-giving relationships.
To explore whether homophily (similarity between individuals) or heterophily (differences between individuals) exists in the advice network.
Through this research, we aim to provide insights into improving workplace network efficiency and communication by identifying key factors that influence the creation of advice relationships.
Keywords: company network, ERGM, network data, reciprocity, transitivity
Figure 1: Company network graph including all node and link characteristics.
Male employees are shown as green nodes, and female employees as white nodes. There are more green nodes than white, indicating that there are more male employees than female employees. It is also evident that male employees are more densely clustered together compared to female employees. Notably, male employees numbered 6, 26, and 39 received a higher number of strong connections and were also the ones providing the most advice.
Figure 2: Company network graph focusing on gender characteristics.
The employees are represented in the following order, from the highest rank to the lowest: Partner (white), Managing Consultant (orange), Senior Consultant (light orange), Junior Consultant (yellow), and Research Assistant (green). The second-highest rank, Managing Consultants, frequently sought advice, and the male employees numbered 6 and 26, who provided the most advice, are also Managing Consultants.
Figure 3: Company network graph focused on rank characteristics.
Figure 4: Company network graph focusing on city characteristics.
Figure 5: Company network graph focused on region characteristics.
In Figure 4, employees' city attributes are represented as follows: Boston (green), London (light green), Paris (yellow), Rome (orange), Madrid (pink), Oslo (light pink), and Copenhagen (white). In Figure 5, the region attribute is represented as the U.S. (green) and Europe (white). Through Figures 4 and 5, it is evident that many employees work in Boston, U.S., and they interact most frequently with each other. Notably, employees numbered 6 and 26, who provided the most advice based on gender and rank, are also from the Boston office in the U.S.
Employees numbered 20, 24, and 26 ranked highly in three out of four centrality-related statistics, indicating their significant influence within the network. Notably, these employees share similar attributes: they are male, hold high positions as management consultants or partners, and are based in the Boston office.
A final model was constructed by adding significant variables through forward elimination. To explore the variables influencing advice-seeking among employees at the consulting company, the basic ERGM model (Equation 2.1) was fitted as follows. First, Model 1 (Equation 4.1) was fitted, which includes only the network connections. Next, Model 2 (Equation 4.2) was fitted, considering not only the connections but also the structural characteristics of the network, such as reciprocity, transitivity, and advice provision. Finally, Model 3 (Equation 4.3) was fitted, which includes significant node attributes and their effects.
The network graph derived from the final model(Model 3) is based on a simulation. Similar to Figure 1, the node shapes differ by gender (female: circle, male: square), the node sizes vary by rank, and the node border colors represent geographical regions (U.S.: red, Europe: blue). However, weight and city attributes, which were not significant during the final model variable selection process, are not displayed. The distribution of employees between the U.S. and Europe is clearly distinct, and the simulation network reflects the geographical homogeneity, which had the most significant influence in the final model. Since male employees tend to be more centrally located in the network than female employees, the gender attribute is well represented. Additionally, employees with larger nodes are located at the center, indicating that the rank attribute is also effectively reflected.
First, it is likely that more frequent communication within a single office, rather than across multiple offices, would lead to more creative results through collective intelligence. If, for specific reasons, a project group must be formed across multiple offices, measures should be taken to reduce the physical distance and minimize any associated losses.
Second, since employees in higher ranks are likely to be key contributors to communication, forming small groups that include Managing Consultants or Partners, rather than those with lower ranks, can promote more efficient skill development among members.