FanGraphs Game Odds!
For every game this year we’ll be projecting the odds of each team winning on our live scoreboard. We’ve also added an extra experimental line on live Win Probability graphs that adjusts for the newly calculated game odds.
To calculate these game odds we’re doing the following:
1. First we take a combination of the ZiPS and Steamer projections for each team’s position players (or starting lineup once that information becomes available), and each team’s starting pitchers and bullpen.
2. We take those projections and run them through Base Runs and then the PythagenPat to get a winning percentage.
3. Then we take each team’s expected winning percentage and run it through Log5 to get the actual winning percentage of the home team.
4. Finally, we add 4% to the home team, to account for home field advantage.
To calculate the new line on the Win Probability graph, things get a little more complicated.
1. First we add 1 run to the team we have projected to win to find how much weight to give the regular win probability line and the new win probability line where the projected winning team is up by 1 run.
For instance, at the beginning of a game, the non-projected win probability is always 50%. If we project the home team to have a 57% chance of winning, and the win expectancy of a team at the start of the game being up by 1 run (with no outs and no runners on base) is 60%, we would weight the 50% win expectancy by 3 and the up by 1 run win expectancy by 7. That will be our baseline weighting for the rest of the game.
2. Then for each out, we push the weighting towards the 50% line for each out that is made. In the end, the lines converge at the actual winning percentage at the end of the game (100% for the winning team). Basically, as the game goes on, we continue to discount our projected odds.
David Appelman is the creator of FanGraphs.
Off-topic, but any idea why Jose Abreu’s player page isn’t loading?
Maybe they’re calculating game odds for him
Should be fixed now!
Sweet.
De-juiced lines (102% books anyway) from bet365 versus fangraphs:
Royals-Tigers 36.5%-63.5% (bet365) versus 37.9%-62.1% (fangraphs)
Subsequent games don’t have full lineups but…
Braves-Brewers 46.0%-54.0% versus 40.0%-60.0%
Twins-White Sox 44.0%-56.0% versus 40.6%-59.4%
Nats-Mets 56.0%-44.0% versus 55.0%-45.0%
Does the model account for managers refusing to use their best reliever in high-leverage but non-save situations?
No. It should at least account for the pitcher who is in. i.e., before Kimbrel enters, the manager might have selected someone else or left the starter in. But once he’s in, I expect a jump in the odds.
why cant i see my own guy in THE GAME leaderoards
Try setting the PA / Innings to 0 instead of qualified.
Excellent. Appelman’s posts always seem to have great news.
Any chance adjustments for opponent quality expand to other statistics? Opponent adjusted WPA (oWPA, aWPA, ???) would be a natural next step after this announcement.
Ultimately, I would love to see a stat similar to wRC+ that adjusts for pitcher quality.
Awesome – I just asked about this in the Q&A section yesterday seems promising
Does anyone know specifically why the home team has a 4% advantage? I remember I read a study saying that home field advantage in all sports is generally due to umpire/ref bias for the home team, but has anyone written about this on Fangraphs?
I’d guess it’s mostly (a) umpire bias, but there’s also (b) roster construction being optimised to the ballpark, which might not get picked up in the overall projections, and (c) a very small tactical advantage to the home team in batting 2nd (most obviously, knowing how many runs they need in the bottom of the ninth).
Dude, he asked if anyone “know[s]”, not, “hey, can I get some speculations on this?”
Oh, and (d) travel time / jet lag / sleeping in one’s own bed.
I sleep in a racing car…do you?
I sleep in a big bed with my wife.
I’m looking at the scoreboard right now. For games currently in progress, why doesn’t the graph start at the projected point? (Currently it starts at 50% for in progress games.)
Do the odds take into account handedness of starting pitchers vs batters or is it the generic Zips/Steamer aggregate numbers?
Please add a link to the live scoreboard. thanks.
Is there any way you could factor park factors into this as well? Ie fly ball pitchers pitching in smaller parks and vice versa will affect the team’s odds of winning.
Is there anyway that someone could post the formula(s) or work out an example for exactly how Fangraphs calculates the game odds?
It’s promotional so I won’t link to it, but there is a recent book dedicated to this topic with very detailed examples.
It’s promotional because I wrote it; it’s easy to find wherever books are sold.
Where within the book is this process described? Also interested in seeing an example.
techniques commonly used in Bayesian statistics would let you calculate the convolution via numerical integration and would allow a pretty rich model. Why have this static model?
Very cool. Thanks for always improving the site.
Small point, but I believe the home field advantage calculation is more accurate if you take the neutral field log5 calculation and then multiply the home team’s win percentage by 1.08, as opposed to adding 4% to their win probability. Makes no difference at the 50/50 level but as the favorite gets bigger the difference grows.
It’s based on the fact that home teams do everything 8% better (pitchers walk 8% less, hit 8% less batters, etc.) than when they are on the road. Matt Swartz, when he was at BP about five years ago did a fabulous multi-part series on this topic. If you do a multi-year study of say, .600 teams playing .400 teams, both home and away, you’ll find the 1.08 method is more accurate.
Calculating Win Probability seems like something that can be easily automated. Can you provide us fantasy players a continually updated number for the next 2 weeks of games? I know it is lineup/starting pitcher dependent, but why couldn’t you just assume standard lineups (use your depth charts) and starting rotations. You could then give us a ranking of pitchers associated with the highest probability of winning next week for streaming purposes. If a pitcher has 2 games that week you could add the probabilities together. Yeah, I know that a team Win Probability is not equal to a pitcher Win Probability and assumptions about lineups and starting rotations aren’t perfect but, as long as you gave that disclaimer I think it would be VERY useful for us fantasy players.
Jordan Zimmermann was a scratch this morning for the Nationals, and this is reflected in Roark being in the P spot in the lineup, however Zimmermann is still listed as the starting pitcher. How are those names being pulled and will they be able to handle pitching scratches, since it appears the pitcher is set on the scoreboard days in advance?
I will put in some code to make sure the starting pitcher when the lineup is announced is the same one as the projected starter. I switched it manually today and it will recalculate the odds if there are any changes up until the game starts.
Awesome, thanks for the quick turnaround too.
A step in the right direction and something that I assume will be improved upon in the future. I am glad that @Fangraphs realizes that the context neutral win expectancies don’t do that good of a job in many cases. A scale from 1 to 10, @Fangraphs has gone from a 2 to a 5 on their in game win probabilities. Thanks for beginning to tackle this and if you need any help let me know.
vr, Xei
Were can I see this game odds for daily games?