Insider Presents: Game Predictor

Game Predictor is a 100% objective, user-driven, quantitative prediction tool. With just the click of a button, you can get algorithmic game predictions driven by smart math and ten years of historical NCAA basketball game results. Also, you can model any real or hypothetical matchup between any two teams.

As an algorithmic tool, Game Predictor can produce some funny looking predictions if you aren't wise about choosing the stats it uses to find statistically similar past games. For example, if you only choose one stat (say, personal fouls) that in isolation is not well correlated with winning, then you can easily generate a prediction showing a clearly inferior team with big win odds to beat a clearly superior team.

So for the most powerful results, choose your stats wisely. The first column of stats, Schedule Adjusted Ratings, includes sophisticated measures that adjust for a team's schedule strength, a critical attribute when comparing team performance levels. The second column, Expert Stats, includes measures such as the celebrated Four Factors used by many hoops number crunching experts.

There are millions of stat combinations to choose from, so have fun exploring the tool and testing different strategies. But we recommend including at least two Schedule Adjusted Ratings and one or two Expert Stats in your lineup.

Choose your stats and teams using this box. Game Predictor uses the stats you choose to determine past games that featured teams statistically similar to the two teams you are matching up. The results of identified similar games determine predictions for the current game.

You must choose two teams and up to five stats in order to generate predictions. Once you generate predictions, you can click the "Start Over" button to re-initiate the choosing process from the beginning, or just click on the stats and teams buttons at the top of this table to make changes to your existing selections.

This box displays predictions for a game between the two teams you have chosen, based on the results of statistically similar historical games. For further explanation of the methodology behind Game Predictor's predictions, see the help section for the Statistically Similar Games Driving Prediction Results section.

"Odds to win game" lists the percentage chance that each team has to win the game. This win odds figure is translated into a "Pick Confidence" rating of 1-5 stars, shown in the top right of the box.

"Most likely final score" lists a projected final score based on the results of similar past games. Final score is by far the most difficult outcome to predict with accuracy.

"Odds to cover spread" lists the percentage chance that each team has to cover the current point spread (listed in the "Line" column) based on the spread cover performance of identified similar historical teams. This prediction is shown only for actual games for which a point spread is available from our odds provider.

Important notes:

(1) For very close games, predictions such as win odds and final score, or final score and cover odds, may actually contradict one another. For example, the model could show one team with 52% win odds, but a most likely final score that favors its opponent. When such a contradiction occurs, it indicates that the model's results are not conclusive. Since the scores and point spreads of similar historical games can vary, the model must take a slightly different approach to making each specific prediction.

(2) Win picks with one-star confidence ratings are essentially toss-ups, where the projected winner's edge lies within the margins of error of the model's calculations.

This box displays a sampling of past games featuring teams that are statistically similar to the two teams you have chosen, based on the stats you have chosen. The results of these past games factor into a mathematical formula that generates final predictions.

The fundamental question Game Predictor answers is, "So what has happened in the past when two teams with similar characteristics have played one another?" Its not an easy question to answer, given the number of possible games to compare to, the fact that you are comparing across multiple stats, and the need to identify a lot of similar games in order to have confidence in the final predictions.

In short, this table presents the "proof" -- the indisputable hard data that prove why the model has identified past games as being similar. What you have at your hands is your own personal NCAA basketball math wizard, so have fun. Test out your own theories and strategies and see what the resulting predictions are based on unbiased historical facts!

Important notes:

(1) This table shows a subset of identified similar games that drive predictions. In total, more than 50 games are considered in determining final algorithmic predictions.