Hierarchical Absorbing Markov Chain Models for Pooling Sparse Combat Sequence Data
Abstract
By viewing tactics in no-gi Brazilian jiu-jitsu (NGJJ) as discrete states, we estimate the transition probabilities of tactical progression, while simultaneously exploring the effects of weight categories on these dynamics with a hierarchical absorbing Markov chain model. We further cluster the weight categories and borrow strength using the Ewens-Pitman attraction of Dahl, Day and Tsai. Random partitions allow us to explore the underlying groupings of these weight categories in terms of their transition patterns. The hierarchical structure enables us to obtain an overall mean estimate of the transition probability matrix along with the individual estimates of the weight categories. Through simulations we find that the proposed model is more flexible than hierarchical methods with no partition or fully-pooled alternatives while sacrificing very little efficiency under model misspecification. Applying this method to NGJJ, we found a back-take is the tactic most likely to lead to achieving submission across all weight categories. In addition, we discovered the lightest (66kg) and heaviest (99kg+) weight categories vary significantly from the others in long-term tactical progression.
Related
- Polynomial Decay in Absorbing Continuous-Time Markov Chains
- Conditional Stable Laws and Rare-Event Limits for Absorbing Markov Chains
- Sparse Bayesian Neural Networks via Sticky Piecewise Deterministic Markov Processes
- A Network-Structured Bayesian Hierarchical Model for Sparse Mutation-Drug Response Associations: Application to Cancer Pharmacogenomics
- Sharp Asymptotic Minimaxity of the Gavrilov-Benjamini-Sarkar Step-Down Testing Procedure in Sparse Gaussian Sequence Models
BibTeX
@misc{halverson2026hierarchical,
title = {{Hierarchical Absorbing Markov Chain Models for Pooling Sparse Combat Sequence Data}},
author = {Brigham Halverson and Matthew Heiner and Leonardo Lamas and Gilbert Fellingham},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.31787},
eprint = {2610.31787},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.31787},
}
RIS
TY - UNPB
TI - Hierarchical Absorbing Markov Chain Models for Pooling Sparse Combat Sequence Data
AU - Brigham Halverson
AU - Matthew Heiner
AU - Leonardo Lamas
AU - Gilbert Fellingham
PY - 2026
DA - 2026/10/22
PB - arXiv
DO - 10.48550/arxiv.2610.31787
UR - https://arxiv.org/abs/2610.31787
AB - By viewing tactics in no-gi Brazilian jiu-jitsu (NGJJ) as discrete states, we estimate the transition probabilities of tactical progression, while simultaneously exploring the effects of weight categories on these dynamics with a hierarchical absorbing Markov chain model. We further cluster the weight categories and borrow strength using the Ewens-Pitman attraction of Dahl, Day and Tsai. Random partitions allow us to explore the underlying groupings of these weight categories in terms of their transition patterns. The hierarchical structure enables us to obtain an overall mean estimate of the transition probability matrix along with the individual estimates of the weight categories. Through simulations we find that the proposed model is more flexible than hierarchical methods with no partition or fully-pooled alternatives while sacrificing very little efficiency under model misspecification. Applying this method to NGJJ, we found a back-take is the tactic most likely to lead to achieving submission across all weight categories. In addition, we discovered the lightest (66kg) and heaviest (99kg+) weight categories vary significantly from the others in long-term tactical progression.
ER -