Features / NCAAB

---

If you've filled out a men's March Madness bracket, you know the drill: pick a few upsets, pray, watch half your picks die by Saturday. The women's tournament plays out differently. Favorites win more often. Top seeds dominate by wider margins. Cinderella runs are genuinely rare. That's not a hunch — we fitted our prediction model to 11,700 women's college basketball games and the data made it clear. The best teams in the women's game are more dominant than their counterparts on the men's side, and the model had to be rebuilt from scratch to reflect that.

Starting today, Boxscorus covers NCAA Women's Basketball with full BXS ratings, conference priors, and 10,000-tournament Monte Carlo bracket simulations. The 2024 championship drew 18.7 million viewers, more than the men's final. Even post-Caitlin Clark, ESPN's WBB viewership in 2025 was up 17% over 2023. The audience showed up. The prediction coverage should match.

Why the Men's Model Doesn't Work Here

The naive approach: take the NCAAB model, point it at women's game data, done. Same BXS framework, same conference priors, same simulation engine.

The problem: women's college basketball has fundamentally different variance. Upsets are rarer. Dominant teams win by more. The talent gap between power conferences and mid-majors is wider. A model fitted to men's game data produces probabilities that are systematically too uncertain for women's outcomes.

So we refitted everything from scratch using ~11,700 WBB games across multiple seasons.

Three Numbers That Tell the Story

MARGIN_SIGMA is the big one. 10.0 vs 16.33 means women's game outcomes are significantly more predictable. When the BXS model says a team has a 70% chance of winning in the women's game, that probability calibrates well against actual results. The men's game has wider variance around the same prediction.

Why the difference? The talent distribution in WBB is more top-heavy. Programs like UConn and South Carolina have sustained dominance that doesn't really have a men's equivalent (Duke and Kansas are great, but they lose 8+ games a season). The gap between the top 10 and the rest of the field is wider, and the data confirms it.

MARGIN_BETA is slightly higher (0.0560 vs 0.0518). A 200-point BXS gap produces an 11.2-point expected margin in the women's game vs. 10.4 in the men's. Dominant teams win by more.

MARGIN_TO_ELO dropped from 19.0 to 12.0. Cross-conference margins are a noisier signal in WBB, so the model trusts conference priors less aggressively. A conference that averages +5 in cross-conference play gets 60 BXS points of credit instead of 95.

Same Engine, Different Fuel

The structural framework is identical: conference priors from cross-conference margins, FiveThirtyEight-style margin-of-victory multipliers, opponent quality scaling, between-season regression (67% toward conference prior), and vectorized NumPy simulation at ~100k sims/sec.

Round-by-round sigma still escalates from 16.0 (Round of 64) to 18.0 (championship game). Later rounds are less predictable regardless of gender. By the Elite 8, everyone left is good.

The 2026 Field: UCLA Runs Away With It

The NCAA committee's second top-16 reveal put UConn, UCLA, South Carolina, and Texas as the four No. 1 seeds. The BXS model agrees on the names but has opinions about the order:

Source: Boxscorus 10,000-tournament Monte Carlo simulation, as of March 8, 2026.

UCLA leads at 20.8%. Compare that to the men's bracket, where Duke leads at 10.1%. That's the lower variance at work. When the best team really is clearly the best, and game-to-game randomness is lower, the model gives them a much bigger share of championships.

UCLA, South Carolina, and UConn combine for 52.1% of simulated championships. In the men's bracket, the top three teams combine for about 28%. That gap captures something real about the structural difference between the two tournaments.

The lower sigma also shows up on the 5-vs-12 line. Take the closest matchup: Maryland (1866 BXS) vs. South Dakota State (1738 BXS), a 129-point gap. Run margin_win_prob(1866, 1738, sigma=10.0) and you get 76.4%. Run the same matchup through the men's model with sigma=16.33 and it drops to 65.8%. Same teams, same gap, 10.6 percentage points of difference, all from that one variance parameter.

Here's the full 5-vs-12 picture:

So where's the March Madness chaos? The 12-over-5 upset that's basically a rite of passage in the men's bracket is much rarer here. Maryland vs. South Dakota State is the closest, and the 5-seed still wins over three-quarters of the time. The honest answer: the women's tournament just has less randomness baked in. The model isn't making that up. It's measuring it.

What This Means for Your Bracket

If you're filling out a women's bracket pool, the strategy is different from the men's. Chalk is safer. The top seeds reach the Final Four more often, and the model says picking all four 1-seeds into the Elite 8 is the right call over 70% of the time for each of them individually. In the men's bracket, you almost need to pick at least one 1-seed upset to stay competitive. Here, you need a reason not to.

That said, the model is a starting point, not an oracle. It doesn't know about injuries announced on game day, players in foul trouble, or whether a mid-major like South Dakota State shoots 50% from three on a random Tuesday. The lower sigma means favorites win more often. It doesn't mean they always win.

Build Your Bracket

The full NCAAW experience is live:

• Power Rankings: BXS ratings for every Division I women's team
• March Madness Bracket: Interactive bracket with real-time probability updates
• Games: Daily schedule with win probabilities

Click a team to lock them through a round, and all downstream probabilities update instantly. Lock UCLA into the Final Four and watch how championship equity redistributes across the other three regions.

The model's biggest bet: UCLA reaches the Final Four in 49.6% of simulations, nearly a coin flip. South Carolina is right behind at 45.2%. If either one goes down before the Final Four, it'll tell us something about whether that 10.0 sigma is still too generous, or whether even the women's game has more chaos than 11,700 historical games suggest.