Reimagining Institutional Diversity
A Case Study of Williams College
Welcome to Williams, a small liberal arts college located in the heart of the berkshires.
In the academic year 2017-2018, the college community was comprised of 2061 undergraduate students.
looked like*:
*Each dot represents 1% of the population
We have traditionally looked at institutional diversity through charts like these, but what does diversity look like on the ground? How does this diversity in numbers manifest in campus life?
Williams is a residential college, so a large proportion of social interactions are determined by where you live.

Let’s investigate this question by taking a look at the college dorms.
First-year students entering Williams are assigned to live in one of three dorms.

We can see the compositions of these houses compared to the college wide racial makeup to see how similar or dissimilar they are.
This comparison is quantified using a Chi-squared test and buildings are color-coded based on how likely the population breakdown is given that no factors with racial biases were involved in housing selection.
For any building, a value below 5%, indicated from here on in shades of red, can be interpreted as extremely unlikely to happen randomly — there is a significant deviation from the college-wide racial composition.
After their first year, students have the freedom to choose who they live with and where they live based on a lottery pick.

Here's a look at where the class of 2021 moved for their sophomore year. The thickness of each line represents the proportion of students that moved from a building.
The movement of sophomores predominantly towards the west side greylock quad of campus is a cultural norm at Williams. In fact, most buildings and regions on campus have associated norms and stereotypes of who lives in them.

Therefore, overlaying this colloquial knowledge with the racial composition of a building can give a deeper insight into the degree of diverse interactions on campus.

With this in mind, let's look at some specific buildings on campus.
Poker Flats is a coop senior dorm. Coop lotteries at Williams are smaller and conducted much earlier than the general housing lottery. Therefore, there is a larger degree of intentionality and selection involved in coop housing choice.

In 2017, 4 of the 5 buildings with the largest proportion of white students on campus were coops. Poker Flats was one of the most dissimilar in composition to the rest of the college with a large proportion of white students deciding to live here.

It is colored in deep orange to indicate that this composition would be very unlikely without the presence of racial factors in housing selection.
Lehman Hall is a dorm accessible through the general housing lottery. Although any student after their first year can pick into Lehman, it typically houses juniors and seniors.

Of all the dorms on campus, Lehman has the largest proportion of asian and hispanic students. At the same time, it has the smallest proprtion of white students.

Lehman is colored in deep red due to the extreme disparity in these proportions.
Horn Hall is the newest building on campus and was opened in the fall of 2017. Horn is also available on the general housing lottery to all students after their first year.

It has the largest proportion of black students on campus and the second smallest proportion of white students after Lehman. As a result, it is also colored deep red.
Scroll down to explore proportions in the rest of the campus dorms.
Data and Methodology

This project was initiated in January 2018 at Williams college with the goal of gaining a data-based insight into the lived experience of minority students on campus. The data used for this analysis was merged from several different sources and its scope extends beyond the information showcased in this presentation.

The first main data source for the project was housing rosters available from the Williams Housing portal. These rosters, one for each building, contain names and genders of students, room numbers and room types. The portal only contains information for the current school year, therefore, the rosters were collected once in January 2018 for the 2017-2018 school year and again in April 2018 for the 2018-2019 school year.

The other source of information used was archived first-year student directories. These contain student names, pictures and first-year housing assignments. The final missing piece is gathering information of student race. While self reported data is available to the college, this information was not shared with the project. As an alternative, a facial recognition AI was used to infer race from the already obtained pictures.

This collected data was then merged by name to obtain a list containing student names, race, class year, first-year room, 2017 room and 2018 room. Care was taken during data preparation for this presentation to protect student privacy.

The methodology used to invesitgate social interactions in this project can be extended to explore additional dimensions beyond race. How do students from different athletic teams, majors and economic classes interact on campus? Furthermore, this analysis can be done at different levels of granularity, from looking at roommate pairings, building regions and floors to entire neighbourhoods. Each of these layers would give access to another facet of diverse interactions in the campus community.

More information about diversity at Williams, including institutional reports, can be found here.