Seeing Double: Rethinking Housing Selection

Like last spring, and the spring before that and all springs since time immemorial, it’s time for Amherst students to “choose” where we’ll have the pleasure of living for the next school year. As standard parts of the school year go, housing selection might be my least favorite. It’s stressful and full of disappointment for almost everyone involved. The Office of Residential Life (ResLife) is the butt of ire and jokes for months, and we’re stuck with rooms we inevitably have problems with.

It’s time to leave the lottery system behind and move to a new system. Though it’s standard for colleges, the lottery doesn’t assign rooms in a fair and effective way.

For the moment, let’s imagine housing selection as an abstract problem to be solved. We have a fixed number of rooms, each with a certain number of beds. And we have a certain number of students to place into those rooms. Our goal is to maximize happiness across the entire group. This problem — assigning a set of things (people) to another set of things (beds) — is known in computer science as an “assignment problem.” Luckily for us, computer scientists have dreamt up a number of algorithms to find a good solution.

Right now, ResLife uses what’s known as the “greedy algorithm.” First, this algorithm randomly arranges all students into a list by class year, placing students with accommodations slightly ahead of their classmates. Then, the first few room groups (of one, two or three students) choose the best rooms according to their personal preferences. Once they’ve chosen, the next set of students choose the best remaining rooms — and so on through the whole list.

The nature of the choice here is what makes the algorithm greedy. Every person chooses the best room for themselves without any thought about the people who have yet to choose, lowering the maximum happiness that our algorithm can achieve across the entire student body.

Let’s say there are two rooms, Frost 102 and Chapin 506, and two students, Carl and Maria in the housing process. Carl is relatively indifferent about which room he receives. On a scale from zero to ten, he would rate Chapin a seven and Frost an eight. Maria, however, hates that Chapin does not have a common room. She gives Chapin a two and Frost a nine. During housing selection, Carl goes first and takes his first choice, Frost. Then, Maria takes the best remaining room — Chapin, even though she hates it. Our group happiness, measured by the rating each room’s inhabitant gave it, is ten. If instead Carl had taken a slightly worse room, allowing Maria to live in a room she likes a lot more, then our group happiness would be a stellar 16. That’s 60 percent happier!

Carl and Maria represent a real problem with the current housing assignment algorithm. When everyone gives the algorithm one piece of information — the best possible room out of those remaining, according to their preferences — the algorithm often returns a poor solution.

Instead, we could use an algorithm that takes more preferences into account. Each student has a unique set of preferences that determines how much they like a given room on campus. We should record those preferences and use them to optimize our room distribution.

Unlike the incoming first-years, who will be given housing in the first-year quad, the center of campus, upperclassmen are subject to the lottery system for rooms both near and far.

Here’s the new plan: in our room groups, we’d rate specific characteristics we desire. We could choose a specific building that we would like to be near and mark how much that matters to us. We could record a preference for gender neutral bathrooms on our floor and weigh that preference heavily. And we could say that while we would prefer a kitchen in our dorm, it doesn’t matter that much. The entire set of preferences could be extensive, ranging from space to amenities to dorm size to location and beyond.

Once each group fills out its preferences, ResLife would rate each room on campus for a particular group using their preference and how they weigh their preferences. Then, the algorithm would assign people to rooms to maximize the number of preferences satisfied. The current exceptions to the system (off-campus housing, theme houses and suite selection) would act the same as they do now.

The new technique can adapt to a wide variety of goals. For example, accommodations could be represented in the system as preferences that can’t be denied. And if it’s desirable for seniors to get better rooms than juniors, who then get better rooms than sophomores — the current algorithm does this — ResLife can weigh the preferences of a room group depending on its class year composition.

Ultimately, the algorithm would incorporate far more information into housing selection than we do currently. At the same time, it would eliminate both the stress of finding a room during your housing time slot and the arbitrary nature of the lottery system. Finally, it would prevent over 1,800 students from taking out their annoyance, frustration and anger with housing selection on ResLife.

Instead of scrambling to assign housing over four days packed with back-to-back ten-minute time slots, ResLife could set the algorithm on its way and recline peacefully while the computer spits out assignments.

It’s impossible to fix all of our problems with a new approach — after all, no room will ever be perfect for its inhabitants in every possible way. But we can approximate perfection better than we do now. The lottery system, while it claims to allow students to choose their rooms, actually significantly constrains our choices: we can only pick rooms that haven’t yet been picked. And the greediness of the lottery leads to unnecessary poor outcomes.

If we’re serious about assuring that every student is able to live in the best possible room and if we’re serious about building a fair and inclusive campus, then we should move to a new, more comprehensive and altogether more effective algorithm for housing selection.