How does SportsLine make picks and grades?

SportsLine’s simulation model was developed by Stephen Oh, a sports data science expert.

Each league has its own model that employs similar methodology. First, Oh determines the key parameters governing how much a team is going to score. In basketball, for example, those parameters include the probability of a two-point field goal attempt, the probability of making the two-point attempt, the probability of a three-point attempt, the probability of making that attempt, the probability of drawing a foul and going to the free-throw line, etc.

He determines each team's key parameters based on their expected active roster. In addition to excluding injured players, Oh makes sure to include players who may not directly accumulate stats (i.e. offensive linemen in football), but play vital roles in the team's success.

Step two involves determining the key defensive differential parameters. In all sports, strength of schedule can vary significantly team-by-team. The SportsLine simulator does not rely on the actual stats allowed, but instead calculates the plus/minus of what the defense allowed compared to what the opponent was averaging entering that matchup.

For example, say the Denver Broncos were completing 67 percent of their passes entering a game vs. the New England Patriots. In the game, the Broncos completed 60 percent of their passes. The Patriots' defensive differential parameter for that game would be minus-7 percent. We focus on differential to account for differences in schedule and the types of opponents each team has faced.

Here's another example: Two hockey goalies both have a save percentage of 93. But one goalie allowed 0.35 percent more goals to his opponents than they scored against other opponents, while the other held opponents 0.35 percent below their average goal scoring rate.

Same save percentage, but one is a Top 10 goalie, the other a Bottom 10 goalie.

Next, Oh runs a Monte Carlo simulation -- a problem-solving technique used to approximate the probability of certain outcomes by running multiple trial runs using random variables. The "computerized" offenses play a game vs. the computerized defenses. How well each offense performs in the simulation is based on the parameters being adjusted by defensive differentials. Oh repeats the game simulation thousands of times.

He then uses these simulated results to generate a projected final score, with the probability of each team winning, each team covering, and the total score going Over or Under.

When Oh's projection varies dramatically from the Vegas line or total, SportsLine calls it an "A" pick. When the projection is moderately different, it's a "B" pick. "C" picks occur when Oh's projection nearly mirrors the Vegas line.

To keep our picks updated, SportsLine runs new simulations every eight minutes.