Sports research digest: September 30, 2026
Six picks for the first issue: three about hockey, three from other sports or methods. Follow the links for the original work.
Hockey
2026–2027 team point projections
Evolving Hockey · September 28 · Article
The authors explain how they turn player projections into team ratings, game win probabilities, and a 50,000-season simulation. Their comparison of xSPAR, RAPM, and box-score approaches is especially useful if you care how a projection system is assembled. The roster and injury assumptions remain a source of uncertainty, and this season’s extra games make raw point totals harder to compare with past years.
2014 first-round retrospective
JFresh · September 1 · Newsletter
A pick-by-pick comparison of draft-day scouting projections with how each player’s career developed. Read it for the recurring forecasting question: which projected skills translated to the NHL, and which weaknesses mattered more than scouts expected? It is a retrospective, so the outcome is known when the projection is assessed.
Validating Expected Possession Value against coach assessments of passing intelligence
Miikka Palvalin · September 25 · LINHAC 2026 research paper
In a sample of 120 Finnish U20 players, an Expected Possession Value passing measure had a moderate association with structured coach ratings and outperformed pass-completion percentage. The association was stronger for defensemen than forwards. The sample and youth-league setting matter when considering how far the result might travel.
Other sports and methods
Comparing Statcast Fielding Run Value to DRS
Tangotiger · September 28 · Blog
Tango walks through a reproducible comparison of two baseball fielding metrics, starting with how to align their positional baselines. The hockey connection is methodological: apparent disagreement between player-value models can come from what each metric treats as its reference point.
Seiya Suzuki’s defense took a big leap forward
Davy Andrews · September 30 · FanGraphs
Three fielding metrics now rate Suzuki positively after years of disagreement. Andrews looks beyond the headline numbers at components of outfield play and notes that defensive measures take time to stabilize. It is a concrete case for checking the underlying behavior before treating a one-year rating change as a settled change in talent.
Forecasting sports outcomes through machine learning
Research review · September 2026 · Frontiers in Computer Science
This review maps 118 studies across game results, player performance, injuries, and in-game events. Its most useful takeaway for model builders is the uneven validation practice: the authors call for time-aware testing, uncertainty estimates, and probability calibration. The review is descriptive and does not score each study’s risk of bias, which limits claims about which algorithms work best.