How Our MLB Predictions Work
FiveThirtyEight has been publishing interactive graphics since 2016. They are the MLB Predictions dashboard and the Complete History of MLB charts. They show the chances that each team will make the postseason or win the World Series.
The Retrosheet has collected game results and box scores going back to 1871. We used that data to create an Elo-based rating system and predictive model for baseball. Elos are a simple way to rate competitors. For our purposes, each MLB team carries a rating that estimates its current skill level.
Before every game, we adjust each team’s rating based on whether it has home-field advantage, how far it's traveled to the game and how many days of rest it's had.
How MLB Predictions Work: Home-field advantage is worth 24 points, for games played without fans in attendance, 9.6 points and for travel 4 points. Starting pitchers have a much larger effect on team ratings and win probabilities than the other adjustments. Pedro Martínez was worth 109 points each time he started in 2000.
Home-field advantage is worth 24 rating points. For games played without fans in attendance, home-side advantage worth 9.6 ratingpoints. The penalty for travel is 4 points and is calculated with miles_traveled**(1.0/3.5) * -0.31.
How MLB Predictions Work. To generate pitcher adjustments, we're using a version of Bill James’s game scores. They're adjusted to take the opposing team's offensive strength into account. A pitcher's adjustment to his team's rating is all about his rGS relative to the team. The pitcher who is better than the other pitcher gives the teams a bonus when they start.
The addition of starting pitcher adjustments gives the model about a 1 percentage point improvement in the percentage of games correctly called.
Starting with the 2022 season, MLB's model will incorporate openers. Openers are pitchers who start a game but are expected to face only a few batters. The team will use its average pitcher adjustment rather than the starting pitcher's adjustment.
Before a season begins, we have to come up with a set of starting ratings for each team. 67% of the team's preseason rating comes from the win projection according to three computer projection systems: Baseball Prospectus’s PECOTA, FanGraphs‘ depth charts and Clay Davenport's predictions. 33% comes form the final rating at the end of last season.
67 percent of the team's win projection comes from the preseason win projections. 33 percent comes form the final rating at the end of last season.
Starting with the 2019 season, team ratings change at three-quarters of the speed they previously changed. Monte Carlo simulations play out the season thousands of times. The simulation's uncertainty is decreased a touch. For the postseason, we assume teams use a four-man rotation.
The historical Elo ratings for MLB teams are based on a simplified system that doesn't take pitchers, travel or rest into account. The Elos don't match the team ratings in MLB Predictions. The historical ratings will update with the results of each game this season.
How Our MLB Predictions Work. Pitcher adjustment added for starters designated as openers. Home-field advantage reduced for games played without fans in attendance. Team ratings change at three-quarters their previous speed.
