RallyIQ Ratings

Volleyball is a string of rallies, and every rally has a server and a receiver. RallyIQ gives each team two numbers: a serve strength (how often it wins rallies it serves, i.e. scores break points) and a receive strength (how often it wins rallies it receives, i.e. sides out).

The chance the serving team wins a rally is modeled as the league baseline, plus the server's serve strength, minus the receiver's receive strength, plus a small home-court term, on a logistic scale. All strengths are fit together in one pass over every serving block this season, so a team that sided out 60% against Nebraska gets more credit than one that did it against a weak opponent. Nothing is a rating of ratings: it is one statistical fit.

Early in the season a few matches can swing raw percentages wildly, so every strength is shrunk (ridge regression), and not toward average: toward a preseason rating, last season's final RallyIQ regressed 10% toward the middle and nudged up or down by returning production (the share of last year's points scored by players who are back on the same roster). The pull is worth a fixed number of rallies, roughly three matches' worth of serving, so it fades on its own as games pile up: in week one it carries the forecast, and by midseason the team's own rallies outweigh it several times over.

How hard to pull, how much to regress, and a final calibration step (strengths are scaled by 0.9, the setting whose forecast confidence best matched results over the 2025 season) were all chosen by the backtest below, and each setting had to hold up on the whole 2025 season as well as 2026. Recency weighting, discounting blowout rallies and a neutral-site correction were tested too. None improved the forecasts in both seasons, so they aren't used. Halving or doubling the shrinkage barely moves the table (rank correlation 0.996 or better).

Adj Brk% / Adj SO% are the break % and sideout % a team would post against a perfectly average D1 opponent on a neutral floor. AdjHit O / D do the same for hitting percentage: the hitting % a team would post against an average D1 defense, and the hitting % an average D1 offense would post against it (lower is better). They come from the same kind of one-pass adjustment on box scores (D1 average .208). They describe a team; they don't feed the forecasts.

RallyIQ is the expected points won per 100 rallies (50 served, 50 received) above what an average team would win in the same spot. +10 means about 10 more points per 100 rallies than average.

Win vs avg runs those two rates through an exact set-and-match probability model (every score state to 25, win by two, 15 in set five, best of five). Small rally edges compound over a match, which is why the top teams sit near 100% against an average D1 opponent.

SOS is the average RallyIQ of the opponents in the games used (with its rank among D1 teams). Only games whose play-by-play reproduces the official set scores are used. Current fit: baseline break 42.7%, home edge h = 0.028 logits, λ = 60, calibration ×0.9, preseason prior from 2025.