Approaching Intersectionality In Analysis Means...

Welcome to data uncollected, a newsletter designed to enable nonprofits to listen, think, reflect, and talk about data we missed and are yet to collect. In this newsletter, we will talk about everything the raw data is capable of – from simple strategies of building equity into research+analytics processes to how we can make a better community through purpose-driven analysis.

?

An analysis in my work this week made me deeply uncomfortable, so you and I are talking about it today.?

To share?some?context: this project involves analyzing a huge data set representing small to mid-size nonprofits on their leadership, needs, challenges, and successes. These nonprofits in the data are led by and for various communities (including several historically equity-denied communities). There are detailed data points around identity. So obviously, the data is complex.

Finding insights about the communities, for the communities made me realize the power a researcher or analyst holds. To aggregate or not, to pull themes or not, or even to count – that’s a lot of power for one role. Not to mention, by the time data arrives on an analyst’s table, the list of requirements from funders around data representation is not far behind either.

So, today, you and I will think about what it takes to analyze when your data offers intersectionality (and missing values at that). Because unless we speak of it – enough times - we risk bringing biases into the data and our algorithms.

Intersectionality refers to the complicated mix of identities each of us carries. For example, I am a woman, BIPOC, South Asian Indian, and first-generation immigrant with anxieties. That’s intersectionality.?

Let me give an example of the risk I am talking about. Take analyzing data for women. Without discussing how intersectionality could/should be included in an analysis, we risk highlighting the exact needs of women. The already questionable state of data about women's needs, tastes, life experiences, and concerns affects everything from their discomfort to bodily injury. Algorithms then use this insufficient data to encourage repeated actions. Simply speaking, that only leads to more undervaluing and neglecting of women’s needs. This is why - to allow data not to create harm - you and I need to have repeated dialogues around the analysis of intersectionality – from different viewpoints and contexts.?

?

Now, let’s take a scenario to create a list of items for exploring intersectionality.

?

Say, you have a dataset of 35,000 rows and 50 columns. Each row represents a unique member of your nonprofit, and each column represents a feature about the member (e.g., their interests, perspectives, identity-related information, etc.). The identity-related information includes features like gender-identity, age, ethnicity, etc. Your task is to familiarize yourself with this data and analyze it for insights about the needs of various communities that live in your membership. The dataset's quality is decent – meaning there are missing values under each feature, but you have something to work with. One of those data points is me – a BIPOC woman, South Asian Indian, and first-generation immigrant with a young business.

?

So, on the specific question of this edition –

how do we look at the intersectionality in this data?

Here are 5 ways you and I can do when doing the first-round data x-ray:

1. Identify interests of identity-related data interactions in advance.

????Every identity-related data can have a unique effect on the derived insights. For example, how I am perceived can differ depending on whether you focus on BIPOC, Women, immigrants, or structures around my entrepreneurial activities. Deciding which interactions/intersections of identity-related data support the overall objective of research can help in the analysis.

2. Approach analysis with a sample-level understanding.

??A common scenario in the data is an under-representation of the population of interest. There can be likely situations where your identified intersectionality-related data points of interest have way less representation in the data. Employ initial sample-level checks to determine how you want to treat the under-representation. Some of those initial sample-level checks can include meta-analysis like:

• Do the count of observations (that is, the number of unique members in this case) are under, equal, or over the community size at a specific location and point of time (i.e., an external benchmarking number)? This can indicate under-representation.
• What is the percentage of missing values for each feature? This can indicate the analysis approach or the need for further data collection.
• What percentage of records have zero/missing/invalid values for the key features needed for the analysis?
• Do the missing values show any pattern, or are the data points missing at random?
• Are there duplicates in the records? How do you intend to handle those duplicates?

?

3. Engage with communities for interpretation.

Bringing a few representatives of the community existing in the data as you analyze and derive insights can help make more sense of the data. Those representatives can share context about their community, like historically denied opportunities and social power/privilege pre-existing in the community structures. For example, for the dataset above, say you are coming close to the first phase of your analysis. At this point, you are familiar with the data and understand what it says – at least at a high-level. Now, as you begin your second phase of analysis (that is, creating reports for the next steps), include a few community members – you can call them on your website, newsletter, or social media. The purpose here is to bring people who can share their knowledge and experience as you put your expertise into the analysis.

?

4. Check for correlation in the data.

Often the data we collect comes with correlation. Correlation expresses the relationship between two features. For example, in our members' dataset, job, zip code, and donation amounts “may” correlate. Until proven, that’s a hypothesis, and you can use correlation techniques (statistical or machine learning-based methods) to test whether the job, zip code, and donation amounts are indeed correlated.

?

5. Share all assumptions made for your analysis.

????At every step of the research - objective design, data collection, analysis, dissemination – assumptions will be made. These assumptions will come from different actors – funders, managers, data collectors, analysts, etc. Including all assumptions supports building transparency in the insights from the analysis. For example, if you are analyzing the dataset above, include assumptions like:

• How are you approaching disaggregated data?
• Are you creating groups from text analysis (such as open-ended self-identification)? If so, what is your approach?
• How are you dealing with multiple-choice data points (like race and ethnicity)?

6. Commit to continuous learning.

????This is an unavoidable commitment – continuously learning the strategies, techniques, and methodology of looking at intersectionality in the data with care and purpose.?

?

*********************************

When it comes to recognizing and acknowledging our complex identities in the data, someone, somewhere, raises a hand, saying, “but we don’t have enough data there. We need to collect more…” The most common remedy when it comes to data pains is more data collection. And, while that may be in good intentions, that data collection in itself is not enough. But, how we approach analysis is equally important.

You and I need to enable each other, so we can articulate the why and the what of the data that exists and goes missing.?

Every step of working with data presents an opportunity to recognize where power should lie.

?

?

***?So, what do I want from you today (my readers)?

• Share if and how you have approached looking at identity-data in your analysis. What is the biggest challenge you have experienced in looking at such data?
• Reminder – ‘data uncollected’ is now a bi-weekly newsletter. If you want to stay in touch weekly, consider joining the bi-weekly email newsletter, ‘dear human’ on www.namastedata.org .