OddsRX brings together a broad range of sports and market information, statistical modeling, and simulation-based research to explore how games may unfold. The information and methods used depend on the sport, the model, and the question being studied.
The work is ongoing. Dr. Cover refines and retools the models as needed, revisiting assumptions and approaches as the research develops. This guide explains the ideas behind that work without presenting the models as a fixed list of inputs or a single permanent formula.
Our areas of research
OddsRX has done extensive work on the NFL, MLB, and NBA. College football and college basketball are newer research initiatives.
Start with the question
A model can address team strength, an expected margin, the probability of winning, or the probability of covering a particular line. Each question requires its own interpretation. A projected winner and a projected spread are related, but they are not the same forecast.
For readers, the first step is to identify what the number describes. Check the sport, event, market, selected model, and any accompanying assumptions before comparing results from different views.
A broad range of data
OddsRX works with many inputs across its research. The information relevant to an estimate varies with the sport and the problem being examined. Inputs and methods can change as the work evolves.
The value of an analysis depends on how information is interpreted and used. A large amount of data can still contain gaps, noise, or patterns that do not hold up in another setting. Every estimate has limits arising from the information available and the assumptions used to connect it to an outcome.
Multiple approaches and possible outcomes
Different modeling approaches provide additional perspectives on the same question. Comparing those perspectives can help identify agreement, disagreement, and assumptions that deserve a closer look. Methods may share information and make similar errors, so agreement alone does not establish accuracy.
Simulation and distribution-based analysis are ways to explore a range of possible outcomes around an estimate. A projected margin summarizes one aspect of a game; the range around it helps explain why many different results remain possible. The use and interpretation of these methods depend on the particular model. A distribution chart should not automatically be read as a separate simulation.
Compare a forecast with a price
A forecast about an event and the price offered on that event answer different questions. A team can be the more likely winner while still offering an unfavorable payout relative to its estimated chance of winning.
A useful comparison matches the event, side, market, line, and observation time. A full-game forecast cannot be substituted for a first-half forecast. A price captured earlier may no longer be available. If a quote or other necessary information is missing, that limits the comparison that can be made.
Probability edge and expected value also use different units. A probability difference compares estimates; expected value combines an estimated probability with a payout. Both depend on the quality of the underlying estimate. A disagreement with a market price is a reason to examine the evidence and assumptions.
Refinement is part of the work
OddsRX retools its models as needed. Assumptions, calculations, and approaches can change as Dr. Cover continues the research. An explanation or evaluation needs to be read in the context of the model and information it concerns.
A revision is a change to be evaluated. It does not, by itself, establish greater accuracy or better future results. When comparing analyses from different periods, consider whether the same method, market definition, and assumptions were used.
Read every estimate with uncertainty in mind
A 60% probability in a two-outcome example still allows a 40% chance of the other outcome. A forecast can be reasonable and still lose. One successful prediction also cannot establish that a method will work consistently.
Small changes in an estimated probability can change a price comparison. In a hypothetical example at decimal odds 2.00, a 52% estimate gives a calculated expected return of +4% per unit staked; a 48% estimate gives −4%. The arithmetic is precise, but the probability supplied to it remains an estimate.
The purpose of OddsRX is to help readers make more informed decisions and understand the mathematics behind them. Treat the models as research tools, consider their assumptions, and distinguish a forecast from a guarantee.