I calculated the definitive USC frat rankings

Sidechat, the anonymous social media platform that dominates university discourse, is routinely flooded with contentious rankings of USC’s frats and sororities.

And I’m sick of it.

So to put the issue to rest, I scraped the more than 140,000 posts on USC’s Sidechat server, fed them into a large language model that pulled out rankings, then used code and AI-aided data analysis to calculate the overall scores. Here’s what I found.

This data is inherently flawed. Sidechat’s anonymous users are free to post as much as they want, and snarky internet rankings — which are often a reaction to the “faceness” of pledge classes or a weekend of parties — are a poor way to judge whether you'll enjoy the brother or sisterhood of a house.

But the colloquial rankings on Sidechat, a platform that is dominated by discussion of Greek life, offer insight into the pecking order of USC’s frats and sororities that cannot be found elsewhere.

Sidechat users rank houses in wildly different formats. Many use simple lists — others use tiers or more complex mechanisms. My scoring system attempted to fairly account for these variations.

Take sororities specifically. There are 10 in USC’s Panhellenic Council. For simple ranked lists, the first-place sorority received 10 points, the second received 9, the third received 8, etc.

The math gets more complicated when it comes to tiered lists.

Here’s one way of looking at the above example:

T1 - DG (10), Pi Phi (9), ADPi (8)

T2 - Theta (7), APhi (6), Tri Delt (5)

T3 - Kappa (4)

This would hardly be fair, because most posters likely don’t intend to give meaning to the left-right order of houses within tiers.

So instead, I averaged out the points. If the first tier of a ranking included three houses (points 10, 9, and 8), each house in that tier received the average, 9. (This is why some rankings have half points.)

T1 - DG (9), Pi Phi (9), ADPi (9)

T2 - Theta (6), APhi (6), Tri Delt (6)

T3 - Kappa (4)

More interesting stories begins to emerge when you look at the data by semester. So-called top houses (“touses”) and bottom houses (“bouses”) tend to see relatively stable rankings.

But there’s a fair amount of semester-to-semester fluctuation within the middle cohorts, and some houses like the ZBT fraternity have seen steady increases while others, like the AXO sorority, have seen steady drops.

The overall rankings largely comport with popular understandings of the hierarchy of houses — one that is often cynically attributed to wealth and attractiveness for sororities, and party-throwing prowess for frats.

Still, the data collection was not perfect. I first culled any post from the dataset that did not include a reference to at least one frat or sorority and did not include multiple lines of text, then fed the remaining posts individually into ChatGPT’s API — an interface that lets your code interact directly with ChatGPT so you don’t have to manually input text — with a detailed prompt explaining what should count as a ranking.

That is: Only posts where users clearly intended to rank frats or sororities from best to worst, using criteria generally understood as essential to Greek life. ("Houses ranked on wokeness” did not count but “faciest pledge classes” and “best parties” did.)

I spot-checked the posts returned by ChatGPT and found that the selection was generally accurate, then fed them back into the API with a separate detailed prompt explaining the scoring mechanism.

Notably, of the 790 Sidechat posts identified as ranking Greek life, three-quarters ranked only frats.

Have tips about Greek life? Reach me at [email protected].