Intersectionality and reflexivity-decolonizing methodologies for the data science process.

Using intersectionality as a methodology illuminated the shortcomings of the data science process when analyzing the viral #metoo movement and simultaneously allowed me to reflect on my role in that process. The key is to implement intersectionality to its fullest potential, to expose nuances and inequities, alter our approaches from the standard perfunctory tasks, reflect how we aid and abide by systems and structures of power, and begin to break the habit of recolonizing ourselves as data scientists.

Main text

How can data science be more inclusive when using methodologies that promote linear thinking on complex fluid populations? People and societies are complex, dynamic, and fluid. Yet, we use a linear formulaic process to generate data reducing them into generic, flattened results, distorting views of populations in the process. Moreover, these methodologies encourage using linear tools that possess systemic oppressions that are recycled and focused on whiteness and the status quo. As a result, intersectional communities are harmed in the process and lost and erased in translation. How do we begin to decolonize the data science process and, by extension, ourselves as data scientists using methodologies that appreciate and thrive in complexity?

This article calls to use alternative methodologies for data science processes such as intersectionality and reflexivity, a practice of intersectionality.

My inspiration to use intersectionality as an alternative methodology stemmed from my investigations of the #metoo movement. Several erasures were happening that were imperative to this research. The first erasure was the conflation of solidarity #metoo narratives occluding the experiences of sexual assault and violence to existing multiple social intersections (e.g., race, gender, sexual orientation, religion, class, ability, location, etc.). The impact of this subsumed narrative is that some survivors of sexual violence were excluded from the movement and did not pertain to them because the story was focused on affluent white women. To call attention and highlight their existence, people altered the #metoo hashtag to reflect their experiences with sexual assault and violence using hashtag derivatives such as #metooblkchurch, #metootrans, #metooqueer, #metoochina, #metook12, etc.

The second erasure is the “me too” origins and widespread misconceptions about who it was designed to help. For instance, activist Tarana Burke created the phrase “me too” to identify and signify the need to create an empathic, safe healing space for Black girls and women to share their sexual assault and violence stories. Unfortunately, many did not learn about Burke’s work until after the #metoo hashtag went viral. Personally, I did not accurately understand her work until I attended a presentation at a local university. Afterward two things became clear to me: first, the origin of the “me too” movement is often ignored to discuss the viral hashtag; second, the totality of the movement was not being captured.

Knowing these two erasures, I sought after a methodological framework that would do two things: (1) capture the multidimensional intersectional social identities layers, and (2) complement Burke’s “me too” movement. I was looking for a framework that started with Black women, encompassed other social identities located on the margins, appreciated complexity, decentered whiteness, and critiqued structures and systems of power. Thus, intersectionality became the framework that guided my research. Researchers have used this metaphoric framework to solve analytical problems where there are multiple social identities in relation to systems and structures of power. Even though the term “intersectionality” was coined in the 1980s, it draws from and carries a long genealogy in Black feminist spaces. “Intersectionality is a way of understanding and analyzing the complexity in the world, in people, and in human experiences [...] intersectionality as an analytic tool gives people better access to the complexity of the world and of themselves” outlines the two tenets of intersectionality: first, it seeks to understand how multiple social identities are not independent and unidimensional but rather multiple, interdependent, and mutually constitutive. Second, it emphasizes how these multiple interlocking identities confront structures and systems of power.

While intersectionality is not new to the data science landscape, intersectionality was not utilized to its fullest potential nor as a sole quantitative methodological framework for investigation. Thus, the challenge was to incorporate intersectionality in all aspects of the data science process (e.g., design, data collection, cleaning, exploration, modeling, and interpretation), not to only confine intersectionality as a way to understand intersectional social identities such as race, gender, sexuality, etc. Coincidentally, as I was applying this intersectional lens to the steps of the data science process, I found myself mindful and thoughtful about which techniques and tools were and were not applicable for intersectionality. In other words, when using intersectionality, it calls attention to those interlocking social identities located on the margins who are not defaults yet confronted by systems and structures of power. Instead, these systems and structures of power are represented in the bias entrenched in these algorithms and machine learning and how we determine to collect the data (i.e., sampling methods), how we choose to clean the data (i.e., normalization), etc. describe this self-reflexive process when implementing intersectionality as a critical inquiry and praxis; they call this concept Reflexivity. Reflexivity allows data scientists to call attention to their own practices in the context of intersectionality while going through the steps of the data science process.

Approaching quantitative work such as data science with an intersectional lens can provide researchers with a formal structure that enables them to be more conscientious when critiquing, questioning, developing, and designing their data science processes while also pushing them to evaluate their own position in relation to the data. In this way, intersectionality is inclusive in terms of data equity and adds an additional layer of accountability to the researcher. The intersectional framework guided and widened my scope to those sidelined to the margins, and I call on other researchers to evaluate their work from this perspective. There is a need to be held accountable for our role in the process. If we are to be allies and intentional in our work and practice, we need to ask ourselves how we incorporate bias, reinforce bias, and stay silent in our bias. While fairness, accountability, and ethics are admirable, feel-good efforts to mitigate bias, these concepts are superficial. They do not challenge the whiteness entrenched in these methodologies used to analyze the data or the researchers participating in the analysis.

Using intersectionality as a methodology benefits in a few ways (1) critique and interrogate the data science process’s design, collection, algorithms, tools, and techniques, and (2) holds the appreciation for fluidity and complexity in people. Incorporating the intersectional lens into the data science process is incongruent, since both concepts are different. Data science can be described as rigid, linear, and formulaic given its abstraction of general trends away from concrete context, whereas intersectionality is a critical framework well equipped to explore how overlapping identities structure social reality and concentrate power on the hands of few privileged actors. Reflexivity is used to link these concepts synergically as an iterative process of producing knowledge in the data science process, impacting those not traditionally centered. Hence Quantitative Intersectional Data (QUINTA) is formed as a methodological framework where we can reconceptualize what it means to step away from a prescribed and perfunctory approach to a problem and consider how processes and algorithms impact vulnerable communities.

QUINTA has been crucial for navigating the #metoo data analysis and for understanding how the narratives of cis-gendered white women overtook the hashtag at the expense of sidelining those outside the white norm. Therefore, QUINTA is not a passive activity; instead, it’s a continual process requiring intentional conscientious efforts to stay engaged. It was a personal decision to be intentionally inclusive and deliberate in my work and not replicate the same harm.

Using intersectionality as an alternative methodology is one step toward appreciating complexity in data and allowing it to be consistently present throughout the process. The goal is to step away from linear thinking, methods, and application because the world, even though some resist, is indeed diverse and ever evolving. So why hold onto things that prevent growth and challenge the way we think and see the world? As a data scientist, it is crucial to tell the story that captures the picture’s fullness, not solely from a dominant perspective or privilege a few narratives as the generalized standard.