Diversity Does Not Equal Inclusion.
Using Network Analysis To Identify Organisational Biases
By Purvi Rastogi, supported by Riccardo Volpato, Laura Weis and Ted Lappas
This summer the London School of Economics and Political Science (LSE) and Satalia joined forces to work on a problem of pressing social importance: how can we find the biases that truly guide interaction patterns in a workplace?
In most cases, when companies investigate their diversity and inclusion agendas, they track metrics relating to composition of the workforce, diversity quotas and attrition rates. These statistics, though important, do not reveal the complete picture about the general workplace attitude. It is also necessary to observe if minority-group employees have the same level of acceptance, opportunity, and utilisation in the workplace as their less diverse counterparts.
Analysing the social networks in place can reveal significant insights into the general organisational environment, showing how communication, decisions, and information flow among the employees. Moreover, a person’s level of acceptance and exposure to opportunities in the workplace can be linked to the informal relations they develop. These informal relations correspond to the networks that employees foster for acquiring knowledge, sharing information, innovating and creating value at work.
For instance, the figure below shows two scenarios with identical diversity ratios but having higher levels of integration in scenario 2 than scenario 1. The bias in scenario 1 is due to a property called homophily, which plays an important role in the social relationships we develop. It refers to the propensity of individuals to associate with others who are like themselves. Homophily can make it easier for some to get access to certain colleagues. However, it may also lead to neglecting relationships with others, reducing the diversity of information people have access to through their networks.
This can be worrisome for some employers. Hence, it is essential to recognise if the homophily characteristics govern the patterns of interaction and communication in the organisational structure. Consequently, given the need to analyse the presence of network patterns such as homophily, as HBR described, using social network analysis is quickly becoming the primary approach to understand whether a company is inclusive.
Social Network Analysis (SNA)
SNA techniques aim to assess the outcomes that are direct consequences of social relations. It is widely used by social and data scientists to map and study networks in fields such as health, technology, telecommunication and fraud detection. For our study, we use the employee social relations network by treating the employees of the organisation as nodes and the relationships between them as links connecting these nodes. In this manner, a decision-making network is generated by asking employees to nominate individuals whom they refer to the most for important decisions and generating a graph from their answers. Additionally, for each employee, we also have to collect details corresponding to their personal attributes, such as race, gender and seniority level.
After constructing the network data, the next step is to understand the processes that are responsible for the structure of the given network. Does homophily predict the existence of relationships? Do individual attributes have a strong influence on the network structure? Upon quantifying these values, we can answer specific questions such as “are men more likely to associate with men for sharing important information thereby placing them at an advantage in the organisation compared to women?” or “are all employees from different ethnic minorities effectively engaged in the organization’s social network?”. Building upon the existing approaches, during our collaboration we developed techniques and experiments to quantify the biases that might exist in a network, thereby measuring the integration of distinct people in the workplace networks.
Our Model and Experiments
Let’s think about some of the reasons that can intuitively lead to formation of connections in a social network:
- If your friend was friends with someone, you are more likely to become friends with them;
- If someone approaches you for advice, you might go to them to share information and advice;
- If someone is popular for giving advice among colleagues in the workplace, you may approach them too; and
- If someone is part of the same team as you or has the same job function, you are more likely to connect with them.
In these cases, network links are the result of endogenous processes influenced by existing network ties rather than as a result of individual attributes or other exogenous factors. Several of these effects interact and operate simultaneously to form our organisational networks. It is essential to note the need for including these effects, since failure to account for them may result in model misspecification, which risks confounding individual-specific characteristics with structural processes influencing network generation.
In our experiments, we used an approach called Exponential Random Graph Model (ERGM). This allowed us to factor in both the endogenous effects as well as effects relating to individual attributes (i.e the exogenous effects). This statistical approach counts the number of ties in the network that have specific patterns and compares them to a random graph. Thus, it gives an estimate of the likelihood that a specific pattern significantly affected the network structure. Therefore, ERGM provides a powerful, flexible tool to separate various social processes operating concurrently and evaluate the relative contribution of each on the formation of the observed network structure.
We tested our model on several artificially generated datasets of 200 employees with biased network connections. The model was quickly able to find the dependencies of the exogenous attributes that were present in the dataset. For instance, for the network shown in Figure 2, we found that the male employees are 1.5 times more likely to be involved in the interactions as compared to their female counterparts, with a similar result emerging for white ethnicity. Comparing our results with the ground truth values of the dataset biases, we saw that the model picked up the general patterns of introduced biases in the analysed networks. This result can be interpreted as employees subconsciously prioritizing the views and opinions of white-male employees in the workplace. Hence, our study illustrates the power of social network analysis to extract essential information by mapping and analysing the social structures.
Drawing Conclusions
To conclude, we can make two significant observations. Firstly, to understand workplace attitudes towards diversity we need to have adequate representation of people from distinct demographic backgrounds in the organisation. Firms can address this issue by amending their recruitment and hiring policies to ensure that unconscious bias is not an attribute influencing hiring decisions. Secondly, it is not enough to have a diverse workforce, we simultaneously need to evaluate perceptions of belonging and utilisation of minority actors in the workplace. Here is where our work comes into place: it helps identifying any form of bias existing in the workplace. If unchecked, in the long run, bias can slow down employee development, drive up attrition, and lead to an overall dip in productivity and performance.
Overall, conducting a diversity and inclusion analysis, not only brings to notice the unconscious biases guiding the workplace interactions but also helps understanding the belonging and immersion of different demographic groups. This can influence employers to be more conscious of their actions and take necessary goal-oriented measures and implement specific initiatives to help improve immersion in the organisation. Additionally, an in-depth analysis of the social networks at the individual level can help identify the subgroup of employees who are role-modelling unbiased and diverse networking behaviour.
