We use a variety of methods to make game predictions. However, all our methods share the same two foundations, objective data and sophisticated analysis. We're data geeks, and we believe that quantitative information and smart math lead to better picks.

Over the years, we've realized that some Team Rankings users love geeking out over numbers as much as we do, but others don't want to get their hands dirty with too much mathematical mumbo jumbo. So we've devised an approach for presenting prediction information that we think well serves both groups.

This approach involves publishing three types of information:

1. Team Rankings Picks and their associated Confidence ratings
2. The raw outputs of our various predictive models
3. Matchup related information, statistics, and analysis

## 1. Team Rankings Picks and Confidence Ratings

The Team Rankings Pick ("TR Pick") represents our "all things considered" most recent opinion on a game or certain type of bet. If all you want to know is our best guess for what's going to happen, the TR Pick is for you. It is driven by objective analysis, but also has some subjective judgment thrown in.

In making TR Picks, we consider a lot of information, from the results of our various predictive models to game related trends and stats to breaking news. Only rarely does all of this comprehensive data point to the same conclusion -- especially when we need to make picks against betting lines. In most cases, as any seasoned prognosticator will tell you, we have to process conflicting information and make our best call given what we know at the current time.

Associated with every TR Pick is a Confidence rating of 1 to 5 stars. These ratings are pretty self-explanatory; the more stars shown, the higher confidence we have in that TR pick. You will notice that game winner picks typically run the gamut from 1 to 5 stars, while betting related picks rarely go over 3 stars. These results are by design. It's much easier to pick game winners than it is to make accurate predictions against betting lines.

For game winner picks, our Confidence ratings roughly correspond to win odds increments of 10 percent. In other words, we believe 1-star game winner picks have at least a 50% chance of winning the game...pretty much a toss-up. We think that 2-star game winner picks have roughly a 60% or better chance of winning, and so on, up to 5-star picks who we think have about 90% or better win odds.

Confidence ratings for betting picks are a different story, as the bar for success is much higher. So we've designed our betting pick stars around a different model. 1-star betting picks indicate which side we think will come out on top if you held a gun to our head, but we have low confidence in these picks. 2-star picks aim to be correct at least 52.4% of the time for standard -110 odds bets, or alternatively, at a rate that would result in profitability over the long term. 3-star betting picks target around 55% performance, although year-to-year variability in results is a higher risk since there are typically fewer of these highly rated picks during a season.

It is important to note that TR Picks and/or associated Confidence ratings are subject to change at any time and for a variety of reasons. These include betting line movements, breaking news, and/or new analysis we conduct. As mentioned above, TR Picks are our most recent opinion, until the game starts of course.

For the purposes of tracking the accuracy of TR Picks, we consider our last posted pick before game time to be our "final" pick. We report TR Pick records based on those final picks.

## 2. Raw Outputs of Our Predictive Models

We have developed a number of sophisticated prediction models (also known in the industry as "math models") that leverage our data sets to make game predictions. These models include our Similar Games Model, Decision Tree Model, Power Ratings Model, and Simulation Model.

Our math models are 100% objective and quantitative; they incorporate no subjective opinions or assumptions about teams, players, or matchups. Furthermore, they were all designed and coded by some seriously nerdy Stanford mathematicians and engineers who tend to have nothing better to do on Saturday nights than tinker with predictive algorithms.

In addition to our TR Picks, we publish the raw mathematical outputs of our math models on our Predictions and Betting Picks pages. It is not uncommon for two different models to reach different conclusions about a game or a betting line, since each model employs a distinctly different methodology and mathematical design.

It's also entirely possible for one of our algorithmic models to do better at picking winners over the short (or even long) term than our TR Picks do. If we knew in advance which models would be on fire at any given time, we'd of course just default our TR Picks to use that model's picks unadjusted. Unfortunately for us, it's not that easy!

## 3. Matchup related information, statistics, and analysis

Finally, in addition to our TR Picks and math model predictions, we publish a growing amount of matchup related stats, trends, and other information on our Matchup pages. We try to incorporate as much of this information as we feel is meaningful into our TR Picks. Still, deciding how much weight a compelling trend or large statistical mismatch should have on one's final opinion of a game is often a subjective call.

Consequently, users who are willing to get their hands dirty with some stats and data should give our matchup pages a thorough review before making any calls on a game.