College hoops tiers

The brief: a college basketball account wants an end-of-season Big Ten tier list, the kind fans argue about, but backed by a power rating instead of vibes, as a 1080 x 1080 post plus a 1200 x 675 cut. The rating is SportsDataverse's adjusted efficiency margin for men's college basketball through sportsdataverse.mbb, and team_tiers (sdvplot's port of sdvplotR's Tiermaker) draws the list.

import tempfile
from pathlib import Path

import matplotlib.pyplot as plt
import polars as pl
import sportsdataverse.mbb as mbb
from IPython.display import Image
from PIL import Image as PILImage

import sdvplot
from sdvplot.matplotlib import team_tiers

SEASON = 2026  # the 2025-26 season, named by the year it ends
CONFERENCE = "Big Ten Conference"
OUT = Path(tempfile.mkdtemp(prefix="sdvplot-recipe-"))  # where the exports go; use your own folder

1. Get the data

One row per team: adjusted offense, defense and their margin (points per 100 possessions better than an average Division I team, adjusted for opponents). The ratings carry ESPN team ids; sdvplot's team table supplies each team's conference and abbreviation for the same ids.

ratings = mbb.load_mbb_ratings(SEASON)
teams = sdvplot.teams("mbb").select("team_id", "abbr", "short_name", "conference")
assert ratings.schema["team_id"] == teams.schema["team_id"]  # ESPN ids as strings on both sides
league = (
    ratings.join(teams, on="team_id")
    .filter(pl.col("conference") == CONFERENCE)
    .sort("adj_em", descending=True)
    .select("team_id", "abbr", "short_name", "rank", "adj_em")
)
league

team_id	abbr	short_name	rank	adj_em
130	MICH	Michigan	1	47.352765
356	ILL	Illinois	4	40.779914
2509	PUR	Purdue	8	37.843323
127	MSU	Michigan St	11	35.216251
158	NEB	Nebraska	17	32.265985
…	…	…	…	…
135	MINN	Minnesota	78	15.001971
2483	ORE	Oregon	92	12.254012
120	MD	Maryland	121	7.168668
164	RUTG	Rutgers	124	6.900419
213	PSU	Penn State	130	6.3738

2. The first draft

Five tiers of equal size, straight into team_tiers.

equal = pl.int_range(pl.len()) * 5 // pl.len() + 1  # five tiers of (nearly) equal size
draft = league.select(team="team_id", tier_no=equal)
fig = team_tiers(draft, "mbb")
plt.show()

It looks like a tier list, but the tiers are wrong. Equal-sized groups put Michigan, the best team in the country, in the same tier as teams more than ten points worse per 100 possessions, and split neighbors who are barely a point apart. The default labels ("Elite" ... "What are they doing?") are jokes, not information.

3. Let the ratings draw the lines

Tiers should break where the ratings do. Sorting by rating and measuring the gap to the team above makes the widest gaps easy to find; cutting at the four widest gives five tiers whose members are close to each other.

league = league.with_columns(gap=pl.col("adj_em").shift(1) - pl.col("adj_em"))
cut = league["gap"].drop_nulls().sort(descending=True)[3]  # the fourth-widest gap
league = league.with_columns(tier_no=(pl.col("gap").fill_null(0) >= cut).cum_sum() + 1)
league.filter(pl.col("gap") >= cut).select("short_name", "adj_em", "gap", "tier_no")

short_name	adj_em	gap	tier_no
Illinois	40.779914	6.572851	2
Washington	20.961117	3.22975	3
USC	15.505063	3.014913	4
Maryland	7.168668	5.085345	5

Each row above starts a new tier: Michigan stands alone, and the gaps at the top (6.6 points) and near the bottom (5.1) are the conference's real dividing lines.

4. Labels that say something

Each tier's label becomes the rating range of its members, so the list carries its own evidence, and the title states the finding. tier_rank keeps teams in rating order inside a tier. The subtitle explains the rating and the rule for the cuts; the caption names the data.

tiers = league.group_by("tier_no").agg(lo=pl.col("adj_em").min(), hi=pl.col("adj_em").max()).sort("tier_no")
tier_desc = {
    row["tier_no"]: f"{row['hi']:+.1f}" if row["lo"] == row["hi"] else f"{row['hi']:+.1f} to {row['lo']:+.1f}"
    for row in tiers.iter_rows(named=True)
}
best = league.row(0, named=True)
data = league.select(team="team_id", tier_no="tier_no", tier_rank=pl.int_range(1, pl.len() + 1).over("tier_no"))


def tier_list(height=0.12, alpha=0.8):
    return team_tiers(
        data,
        "mbb",
        title=f"{best['short_name']} stood alone in the Big Ten",
        subtitle=f"Tiers by adjusted efficiency margin, {SEASON - 1}-{SEASON % 100:02d}.\n"
        "A new tier starts at each of the four widest gaps.",
        caption="Margin: points per 100 possessions better than an average D-I team.\n"
        "Data: SportsDataverse adjusted ratings via sportsdataverse-py  |  sdvplot Tiermaker",
        tier_desc=tier_desc,
        height=height,
        alpha=alpha,
    )


fig = tier_list()
plt.show()

5. Fix the contrast

On the dark Tiermaker background, three logos almost vanish: Iowa's black hawk, Penn State's navy lion and Michigan State's dark green Spartan. Most college logos are drawn for a white page, so the fix is a light background. team_tiers returns an ordinary matplotlib figure, so restyling it is a few lines: the background, the tier lines, and the white text turned dark. The logos go to full opacity too (alpha=1); the theme's default 0.8 softens them against the dark background but washes them out on white.

INK, MUTED = "#1d1d1d", "#6b6b6b"


def light(fig):
    """Recolor a team_tiers figure for a white background."""
    ax = fig.axes[0]
    fig.set_facecolor("white")
    ax.set_facecolor("white")
    for line in ax.lines:  # the tier separators
        line.set_color("#d4d4d4")
    for label in ax.get_yticklabels():  # the tier labels
        label.set_color(INK)
    for text in [ax.title, *ax.texts]:  # subtitle, title and caption
        text.set_color(INK if text.get_color() == "white" else MUTED)
    return fig


fig = light(tier_list(alpha=1))
plt.show()

6. Export at social sizes

team_tiers returns an ordinary matplotlib figure with constrained layout, so set_size_inches re-lays it at each export size. The logos are a fraction of the panel's height, and the square's panel is taller but narrower than the wide cut's, so nine logos in one tier would collide there: the square gets a smaller height. Nine teams in a row is also why the wide 1200 x 675 cut is the better post here.

exports = {"big_ten_tiers_1200x675.png": ((8, 4.5), 0.12), "big_ten_tiers_1080x1080.png": ((7.2, 7.2), 0.075)}
for name, (size, height) in exports.items():
    fig = light(tier_list(height, alpha=1))
    fig.set_size_inches(*size)
    fig.savefig(OUT / name, dpi=150)
    plt.close(fig)
    print(name, PILImage.open(OUT / name).size)
Image(OUT / "big_ten_tiers_1200x675.png", width=700)

big_ten_tiers_1200x675.png (1200, 675)

big_ten_tiers_1080x1080.png (1080, 1080)

The 1080 x 1080 cut:

Image(OUT / "big_ten_tiers_1080x1080.png", width=540)

Run it yourself

Download the notebook (outputs cleared) or open it on GitHub.