Football Quantitative Research
Using statistical rigor to make football problems clearer, more measurable, and easier to solve.
DM Analytics is a standalone body of football analytics work focused on turning complex football questions into structured, explainable models. The aim is not just to build technically sound work, but to make the logic, trade-offs, and findings easy to follow for a wider audience as well.
Flagship Projects
Foundational Model
Expected Goals Model
A La Liga expected goals case study built with an R-first statistical workflow, using interpretable probability models, season-based validation, calibration-led evaluation, and a small freeze-frame context upgrade to create reusable shot-quality infrastructure.
Status: ImplementedQuestion: How should shot quality be quantified from open event data so that analysts can separate chance quality from finishing and reuse the outputs in later forecasting and player models?
- Naive conversion baseline
- Logistic regression
- GAM with mgcv
- Freeze-frame context features
- Season-based validation
- Probability calibration diagnostics
View case studyForecasting Layer
Dynamic Team Strength and Match Forecasting Model
A La Liga match forecasting case study using a broader historical open-data sample, with a baseline-to-dynamic model ladder and explicit evaluation of whether time-varying team strength improves pre-match probabilities.
Status: ImplementedQuestion: How should current team strength be estimated for match forecasting and pre-match strategic priors?
- Naive outcome baseline
- Static pooled attack/defence model
- Dynamic team-strength updates
- Temporal train, validation, and test splits
- Multiclass forecast evaluation
View case studyTactical Decision Support
In-Game Win Probability and Tactical Decision Model
A minute-level live forecasting case study that updates win, draw, and loss probabilities from scoreline, time remaining, red-card state, cumulative xG, and pre-match strength priors.
Status: ImplementedQuestion: How should outcome probabilities update during a match, and how can those updates support tactical decision-making?
- Empirical state baseline
- Remaining-goals Poisson model
- GAM-based remaining-goals model
- Minute-level state table
- Pre-match strength priors from Project 2
View case studyRecruitment and Development
Player Rating and Recruitment Model
A role-aware recruitment model built from lineup-derived minutes, event-level player contributions, and team-context adjustment, designed to separate player signal from strong-team inflation.
Status: ImplementedQuestion: How should a club compare recruitment targets within role while controlling for team environment and small-sample noise?
- Role-aware component ratings
- GAM-based team-context adjustment
- Reliability shrinkage
- Season-to-season stability analysis
View case studyApplied Statistics
Model choice, testing, calibration, and uncertainty are treated as core parts of the work, not technical decoration added at the end.
Decision Support
Every project is framed around a football decision: valuing chances, forecasting matches, understanding live game state, or comparing players for recruitment.
Production Discipline
The work is built as reproducible, testable software rather than a collection of one-off charts or notebook experiments.
Research Workflow
- Start with a football question that matters in practice.
- Build a data pipeline with clear assumptions and realistic validation splits.
- Compare simple baselines against richer models instead of jumping straight to complexity.
- Explain what worked, what failed, and where uncertainty is still large.
- Package the output so it could be reused in a broader analytics workflow.