TwoSeventyOne's models are built on peer-reviewed statistical methods and validated continuously against real-world outcomes. We hold ourselves accountable to our predictions — every probability we publish has a track record we stand behind.
Individual polls are aggregated using time-decay and sample-size weighting. Systematic per-pollster biases are estimated and corrected for, so that the aggregate reflects the underlying voting intention rather than any one firm's house effects.
Trend Modelling
Smoothed Two-Party Preferred
The model tracks a smoothed two-party preferred trend, cross-checked against publicly available poll aggregation services. Where the sources diverge, the model favours the series with the broader underlying dataset and flags the discrepancy.
Structural Factors
Economic & Political Fundamentals
The model incorporates economic conditions — including cost of living, employment, and consumer sentiment — as structural inputs alongside the polling trend. Leadership approval ratings and state-level historical patterns are included as additional signals.
Seat Simulation
Monte Carlo Seat Outcomes
The national two-party preferred forecast is translated to seat-level outcomes via simulation, accounting for local margin variability. Each run produces probability distributions over seat counts and government formation scenarios — majority, minority, or hung parliament.
Uncertainty
How We Quantify What We Don't Know
Forecast uncertainty grows significantly with distance to polling day. Early in a parliamentary term, wide probability bands are expected and appropriate — a 65% probability is a strong signal, not a certainty. All probabilities represent long-run frequencies. We do not publish point predictions without uncertainty ranges, and we do not round probabilities to 0% or 100% until an outcome is known.
AFL Season Forecast
Ratings-based prediction & Monte Carlo season simulation
Ratings System
Opposition-Adjusted Team Ratings
Teams are rated on a continuous scale that updates after every game, accounting for the quality of opposition faced. Venue and travel conditions are factored into each game's prediction. Ratings reflect genuine team strength, not raw win-loss record.
Season Simulation
Full Season Monte Carlo
The remaining season is simulated thousands of times from the current state, using live ratings to generate win probabilities for every remaining fixture. Finals seeding, bracket outcomes, and premiership probabilities are aggregated across all simulations.
Validation
Historical Walk-Forward Testing
The model is validated on historical seasons using walk-forward cross-validation — it is never tested on data it was trained on. Game margin error and binary win/loss accuracy are tracked live and reported on the AFL page each round.
Contextual Adjustments
Roster & Conditions
Significant player absences are incorporated as rating adjustments based on historical impact data. Weather conditions at outdoor venues are factored into margin predictions. These adjustments are applied before each round's predictions are finalised.
NBA Playoff Model
Machine learning game prediction & bracket simulation
Game Prediction
Statistical Learning on Game Data
Game outcomes are predicted using a machine learning regression model trained on multiple seasons of historical data. Win probability is derived from the predicted point margin, with uncertainty calibrated against held-out seasons. Recent games are weighted more heavily to reflect current team form.
Features
Broad Feature Set
The model draws on a broad set of team performance indicators including offensive and defensive efficiency, schedule strength, recent form, and clutch-time performance. Features are selected and weighted through rigorous cross-validation rather than manual tuning.
Market Integration
Blended Probability
Where available, live betting market data is incorporated as a supplementary signal. Markets can reflect sharp, timely information — such as late injury news — that the model's statistical features may not yet capture. The blend is calibrated to minimise out-of-sample error.
Playoffs
Bracket Simulation
Playoff series are simulated game-by-game using adjusted win probabilities that account for home court advantage. Championship odds reflect the probability of winning every series from the current bracket state, aggregated across tens of thousands of simulations.
Our Principles
The commitments that govern every model we publish
1
Probabilistic outputs, always. We never publish a single-point prediction without an associated uncertainty range. Forecasting is inherently uncertain — our job is to quantify that uncertainty honestly, not to hide it behind false precision.
2
Validation before publication. No model goes live without historical walk-forward validation on data it has never seen. We report accuracy metrics publicly and update them in real time so readers can assess model performance themselves.
3
Peer-reviewed foundations. Our statistical methods are grounded in published academic literature on political forecasting, sports analytics, and probabilistic modelling. We build on established methods rather than inventing proprietary frameworks from scratch.
4
Continuous improvement. Every election result, every completed season is a calibration dataset. We conduct post-mortems after major forecasting events and publish our findings — including where the model was wrong and why.
5
For entertainment only. Our models are analytical tools for understanding probability — not betting advice. We are a data journalism and research publication, and our outputs should be read in that context.
NRL Model Coming Soon
Rugby League season simulation & finals odds
The NRL model is currently under development. Follow our analysis blog for updates on new model launches.
