Let’s Talk About the Phillies’ Playoff Odds (!)
It’s May 19th, and the Phillies are in first second place in the National League East. Yes, the Phillies. The team that generated a 1,000 “tanking is ruining the sport” articles this winter has, six weeks into the season, the third-best record in the National League. As Jeff Sullivan noted this morning, their remarkably excellent bullpen has been one of the primary drivers of the early success, with David Hernandez and Hector Neris surprisingly emerging as dominant forces in the middle innings, and Jeanmar Gomez driving another nail into the coffin of the necessity of a “proven closer”. And yet, despite being in second place at this point, if you look at our current playoff odds, you wouldn’t actually know that the Phillies are off to a great start.
See that flat line across the bottom? That’s the Phillies. Their 24-17 start hasn’t moved the needle, at all, on our forecasts expectations for their chances of reaching the postseason. Okay, that’s not exactly true; they’ve gone from a 0.1% chance of winning one of the two Wild Card spots in our preseason forecast all the way up to a 0.3% chance of getting to the play-in game now. But their odds of hanging on to the NL East? Still close enough zero to round down when displaying one decimal point.
This is, to some, puzzling. A question in my chat yesterday brought up the point that our system is far more bearish on the Phillies hot start leading to postseason success than others; Baseball Prospectus gives them a 2.3% chance of winning the division and a 7.6% chance of getting a Wild Card spot, for 10% overall odds of reaching the playoffs. FiveThirtyEight is even more bullish than that, putting them at 4% to win the NL East and 13% to reach the postseason. So why is our system so stubborn relative to others attempting to look into the same crystal ball in order to see what the final standings will look like in October?
Well, it’s worth noting that there are dramatic methodological differences between the way that we (and BP) do our forecasts and the way FiveThirtyEight is doing theirs. Our forecasts are essentially an amalgamation of individual player projections, summed up at the team level; BP does this same thing too, using their PECOTA projections where we use a mix of ZIPS of Steamer. FiveThirtyEight, on the other hand, is using Elo Ratings, which adjust up or down based on whether you win or lose a game, and how much of an upset that win (or loss) was relative to their pre-game expectations. While they used individual player projections to come up with their preseason Elo ratings, their in-season adjustments are based on responding to a team’s win-loss record.
Our system doesn’t care at all about a team’s actual record at any point in the season; it only looks at the individual player performances to try and ascertain whether there have been significant changes in expected playing time or production level to adjust the team’s expected record up or down. So far, ZIPS and Steamer look at what the Phillies players are doing and think “yeah, that’s basically what we expected.” Our preseason expected win% for the Phillies was .395; 41 games into the season, our rest-of-season expected win% for the Phillies is .396. And because they were starting from such a low baseline of expected wins, even their extra banked wins through the first six weeks don’t really change how often they make the postseason in our simulations; the strong start has taken them from 64 to 72 projected wins, but nobody’s making the playoffs at 72 wins, or anything close to it.
And that’s part of the reason why FiveThirtyEight’s playoff odds are so much more optimistic than ours, even beyond the different methodologies for adjusting to in-season performance. In their preseason forecast, they had the Phillies as a 68 win team, an expected winning percentage of .420, which produced a 2% chance to win the division and a 5% chance of reaching the the postseason overall. Now, with their hot start included, FiveThirtyEight’s forecast has the Phillies finishing at 77-85, which gets them close enough to have variance carry them into the mid-80s or even low-90s over enough trials.
But to get from 24-17 to 77-85, that still means that FiveThirtyEight is only forecasting the Phillies to go 53-68 the rest of the way, which translates out to a .438 winning percentage. That’s up a little bit from the .420 preseason expectation, but not a ton; it’s the difference between seeing them as a 68-win team or a 71-win team over a full season. Their system is more responsive to in-season results than ours is, but there’s still a good amount of “don’t overreact, it’s early” baked into their forecasts.
Interestingly, it’s BP’s system that has had the most dramatic change of heart about the Phillies expected performance from this point forward. Prior to the season, they had the Phillies being as lousy as we did, putting them at 65-97 in their initial forecasts, a .401 winning percentage. If you go to their playoff odds page today, though, the Phillies are projected for a whopping .450 win%, and are now expected to finish the season at 79-83; that puts them at 55-66 over the rest of the season, two wins better than what FiveThirtyEight projects, and eight wins better than what we expect. Despite having the middle value in terms of playoff odds, BP’s system is the biggest believer in the Phillies going forward.
Well, sort of. It’s maybe more accurate to say that BP’s forecasts suggests a belief in a very small spread of talent in MLB right now; the top team (the Cubs) have a forecast .572 win%, while the worst team (the Braves) checks in at .441. We have the Cubs at .594 and the Braves at .403, so our top-to-bottom win% spread is 191 points, while BP’s is just 131 points. The Phillies are still seen by PECOTA as one of the worst teams in the league, but since the gap between teams is seen as quite small in their system, even the worst teams still have a fighting chance at the postseason.
Of course, I should mention that there’s actually a fourth playoff odds calculation out there; we have a “season-to-date” stats mode that eschews the projections entirely, and only evaluates a team based on their 2016 numbers. This isn’t a great system, since throwing out what we know about players and teams heading into the season means we’re discarding useful information, but it’s a useful thing to look at and see what we’d expect the odds to be if we really didn’t know anything about anyone beyond how they’d played this year. And in that model, the Phillies are expected to finish at 77-85, with a 5% chance of winning the division and a 10% chance of grabbing a Wild Card spot; even with their 24-17 record, their lousy offense and poor run differential makes the system think this is still not a good team.
Also fun; in season-to-date mode, the Cubs finish with 113 wins!
So, with a bunch of different systems producing wildly different results, what’s the “right” number for the Phillies chances of reaching the postseason, considering they sit in second place on May 19th? I think there’s a decent argument to be made that our system isn’t responsive enough to in-season changes; combining two projections systems together adds regression-to-the-mean by balancing out outlier forecasts, and so it’s pretty difficult for a team to move the needle in a big way in our forecasts. But, to be honest, I also have a hard time buying the high numbers put out there by BP and FiveThirtyEight; we’re talking about a roster currently running a 73 wRC+, after all.
Their pitching has been outstanding, but pitching is a fickle beast, and their current BaseRuns record even with that excellent pitching is just .410. I don’t see any real reason to think that the Phillies can keep winning one-run games at the level they have been, and with a lousy offense and young starting pitchers that are likely to have some innings limits put on them, I think expecting something closer to our 72 win forecast sounds more likely than the 77-79 win forecasts spit out by other teams.
But certainly, there are enough unknowns in forecasting that no one should be defending their projections as gospel, and perhaps the fact that BP and FiveThirtyEight are producing significantly higher playoff odds than we are should serve as evidence that our numbers are too low. After all, we certainly weren’t expecting the Phillies to be in second place in mid-May, so perhaps our forecasts could be improved by adding in some adjustment for team performance above and beyond individual player projections. With multiple systems and different methodologies out there for public review, we’ll be able to look back at the end of the season and see which one best projected what a surprising first-place team did over the final three-fourths of the season.
Dave is the Managing Editor of FanGraphs.
The Phillies are in second place
Small Nitpick (from an admitted Nationals fan) – Phillies are half a game behind the Nats as of May 19th
Yeah, I’m assuming when he looked at the standings it was after the Phillies’ day game yesterday, but before the Nationals’ night game. So technically they were tied for first place for several hours, but it doesn’t meaningfully change the point behind the post.
I feel like your model has WAY too much confidence in itself, no projection system should have any team at less than 1% to start the season. That’s insanity.
For example, what were the odds of Houston making the playoffs last year, in April?
I was coming here to say something similar but in a less nitpicky way. I think the projections are great for identifying the mean/median results but I think the variance might be a bit bigger than the model suggests. I agree that the phillies are bad but there is realistic downside for the Mets and Nats to collapse and while I think the Phillies are only better than the Braves, Brewers, and Reds in the NL, I would definitely take a bet at 100-1 odds for the playoffs.
I think the problem is that they treat the games as independent random draws, and they are not.
A breakout or injury or trade can fairly easily be a 2-3 win swing. If you threw in a very small random chance of a substantial and continuing change in team quality at each game till September, then I think you’d actually get more realistic early season playoff odds.
We had them at 14.5% to start the season, and they were at 59.9% by this date last season. A very different situation.
http://www.fangraphs.com/coolgraphs.aspx?lg=AL&div=W&stat=poff&year=2015
What would the option be anyways?
If you project two teams in the same division, one at 90 wins and one at 65 wins, how do you realistically put the latter at more than maybe a couple of percent?
You’d have to heavily regress all of the projections and call the two teams 85 and 75 win teams or the like. At that point, probably just call every team an 81-win team.
Fair enough.
I guess Texas would be the example to use from last year, who had a 0.7% chance in May. I mean, how many 1 in 200 events have to occur before we realize that those odds are NOT representative.
One look at these graphs and the changes 20+% occurring over 7 game spans in MAY. is ludicrous.
The odds are literally swinging by 20%+, during 4% of the time period. In MAY!
The problem here, is that there is no progressive information being used in the odds. If you predict a team to win 35% of their games, but they actually win 60%, the chance of winning that LAST game, is probably a lot closer to 60%, than 35%…
The team you should be asking about is the other team in Texas.
Breaking: The MLB season is 162 games long. Further updates as events warrant.
I always referred to you as “The Chef” while playing GoldenEye 007, Baron.
VIRGINS!
I’d love to know what % of the scenarios have the Phillies in the race heading into July with the Marlins clearly out of it. Within those scenarios, there’s a decent chance that (1) the Marlins would dump players, making them an easier opponent for Philly and (2) the Phils would add some marginally helpful players to keep them in the hunt.
Because their only viable chance appears to be playing either the Mets or the Nats for the wild-card game.
” I don’t see any real reason to think that the Phillies can keep winning one-run games at the level they have been”
Was there not some discussion in years past (relating to the Orioles and Royals) suggesting that having a great bullpen can improve performance in one-run games? If so, then the Phillies’ excellent bullpen (per Sullivan’s article) could be a reason for expecting continued success.
Of course, that assumes the bullpen performance is sustainable, which I suppose is not at all a given.
Basically the last part is most prudent to me. Bullpens are volatile, especially when talking about non elite relievers.
i assume it would help, sure, but not to the extent that a team as bad as philedelphia would have a record in 1-run games like they currently do. i mean, 14-3 for this phillies team? insane
Dave doesn’t credit any strategy that doesn’t fit the basic paradigm of the existing models for WAR/BaseRuns. Royals go to back-to-back World Series and its “clutch hitting”. Its a really special form of hubris.
The whole premise of a prediction model is that all actors operate under the same game strategy, in this case if one team operates under a different strategy, its outcomes may deviate from the model. Granted its rare, those outcomes are so much more successful than the model allows, but that is exactly what’s happened.
I’m not saying Phillies are going to the playoffs, I’m just saying there is another novel strategy to win games with the prediction models ignore….hence under-weighting performances like the Phillies so far.
How are the Royals doing this season? Projections have a range of outcomes. The Royals managed to outperform theirs the past 2 seasons, but it doesn’t mean they figured out how to beat the system.
“How are the Royals doing this season?”
According to BaseRuns and Baseball Prospectus, they’re beating their projections.
That being said, I tend to agree with your point–but perhaps the Royals aren’t the best example, at least right now.
Outperform is beating projections by a handful of games. Back-to-back World Series appearances is more than outperforming….when your projection is a sub-.500 record.
> Dave doesn’t credit any strategy that doesn’t fit the basic paradigm of the existing models for WAR/BaseRuns. Royals go to back-to-back World Series and its “clutch hitting”. Its a really special form of hubris.
Or, you know, it’s just the Royals beating the odds. Which is actually a more exciting story than “all the maths are wrong!”
There’s no question at this point that lockdown bullpens break baseruns in extremes (especially a large number of close games). There’s no question that this particular aspect of the model needs to be adjusted. However, if the Royals were simply beating predictions based on their bullpen, we would expect to see similar results from other teams in the recent past with superlative bullpens, and that just doesn’t happen to the same degree.
The simplest explanation is that they simply got lucky in a lot of instances, and generally played a bit above their true talent. Which is to say, exactly what happens with most championship teams.
What is more likely — the Royals got lucky a bunch of times two years in a row (three if you count beating their projections this year) OR the projections system doesn’t account for bullpen pitching, defensive range (particularly in the outfield) and baserunning? That’s the crux of the questions.
When you look at it that way your “luck” argument is the one defying statistics not my point that the model doesn’t weight factors properly. The error bars of the model are much wider than your tale scenario.
I’m not necessarily supporting one side or the other here But If something happens consistently for an extended period of time (324 games over 2 years for example) then it seems a bit flawed to call it “luck”. Unless you have a different definition of luck than I’m aware of.
The simplest explanation to me is not that the Royals “got lucky” for 2 whole years (that actually seems pretty far fetched) but that the projections simply missed something about the Royals. That doesnt mean they are inherently broken or anything. Just that they have gaps like everything else.
Pointing to the Royals performance this year is also quite flawed as they dont have the same roster as they had in the previous two years and neither do any of their opposition. So their performance against projections this year is irrelevant to their previous performance.
While I certainly wouldn’t expect the Phillies to make the playoffs, 0% does seem a little low! The Twins in 2015 come to mind, that team was not good, probably no better than the current Phillies team (better hitting, much worse starting pitching), but they stuck around all year and nearly snuck into the wildcard. If the Phillies were to hang around above .500 another 6 weeks or so and make some trade upgrades, you never know.
“…but nobody’s making the playoffs at 72 wins, or anything close to it.”
The 1994 Texas Rangers would like a word with you, Dave.
How do you figure? They didn’t make the playoffs in 1994. Nobody did. Yes, I get the joke you’re trying to make, but the Rangers were 52-62 in 1994.
Things have to break right on all counts, but perma-benching Ryan Howard in favor of someone like Tommy Joseph, calling up JP Crawford, having Maikel Franco return to form, finding out that Tyler Goeddel is actually as ready for the show as he’s looked the past couple of weaks, and a couple of savvy trades, might make the Phils a 500ish team. I can dream anyway.
Maybe Franco did return to form.
That’s the most worrisome thing about this year. When your success is hinged on swinging hard and not striking out there’s a lot of chances for that to come apart.
“The Phillies are 24-17 despite being outscored by 28 runs.”
Oh, yeah, that’s sustainable. Not.
By the way, the Phils had a 5-game stretch where they were outscored 35-8 where they went 1-4). The remaining 35 games they basically broke even. Sometimes a few blowouts over-emphasize the run differential. Maybe you get behind by 4 or 5 runs late, and you let the last reliever in the bullpen go out there and get blown up. This is not reflective of the remainder of the team. Sure, every team gets blown out from time to time (and win in blowouts too), and perhaps a team that gets behind by that many runs more often IS a reflection of the overall talent level of the team, but teams that are made up of mostly halfway decent players and a couple of real duds might make their run differential worse than other teams of mostly similar talent levels of the players that log the most innings.
So if they are basically even in the other games how would 23-13 be sustainable?
I never said it was sustainable. I was just trying to make a point that run differential in and of itself could be misleading, and I just gave one example of how that could be possible.
I’m just confused how it’s misleading, even if excusing their worst games they are still overperforming. His point was valid. Let’s remove the padres first 3 games of the year, now they are -2 in run differential and still 17-21.
Even adjusting for that, 23-13 isn’t sustainable with the adjusted run total. That dog won’t hunt: even letting you remove those games from consideration, there is no way their record is even CLOSE to sustainable.
If you take a look at http://www.fangraphs.com/depthcharts.aspx?position=BaseRuns you will see that both PythagenPat and BaseRuns have the Phillies expected record at 17-24. Argument over.
This is so wrong. Run differential has value, but its not this infallible guidepost for teams.
Just use your brain — if you were a manager, do you manage a game differently when you’re down by 1 run or down by 5 runs? Yes, unequivocally you manage differently. You leave your SP or middle reliever in the game longer, to save your bullpen, you don’t steal or hit and run as much, you play for the double play on defense vs. bring the infield in, etc.
But the true talent is what creates big deficits to begin with.
Yes, I made both these points by MLB and Jedi in my post above, but to me, it’s not as simple as either of these viewpoints, although I lean more to MLB’s view than Jedi’s. Every team has some games where they just don’t perform well (even GOOD teams fall behind be several runs in some games), and if your roster is constructed with a couple of really “bad apples” at the end of your bench that hardly ever see the field except in a blowout, that really doesn’t affect the overall team W-L outlook much at all, in my opinion, but they could affect the run differential quite a bit, when they are used in blowout games.
But like I said, I agree that a team with less talent is more likely to fall behind by larger deficits more often.
Two notes about the Phillies success:
— Part of the problem with the prediction model is that it doesn’t account for actual player changes (i.e. extemporaneous improvement), it only looks at historical performance. Its been documented elsewhere on FG, but the Phillies hired Rick Kranitz this year from the Brewers. In his tenure with the Brewers, Kranitz made them the most curveball happy team in baseball, leading the league in CB% for the last 4 years. Since Kranitz joined, PHI CB% went from 6.0% last season (3rd lowest in MLB) to 18.4% in 2016 (highest in the league by a mile). That is a real, meaningful change that appears to have positively impacted the Phillies pitching staff. The model couldn’t and doesn’t account for that.
— The Phillies are the 3rd youngest team in baseball and take that title if (more likely when) Howard gets pushed aside. For a team that has been putrid offensively this season, there is a LOT of room for improvement and considering more than half of their starting lineup have under 2 year of service time its like they will improve. I’m curious as to how accurately the model can forecast that improvement given so little MLB data to project player value? Wouldn’t it be a necessity to update the model with current player data and weight accordingly to provide meaningful projection data?
The player projections are updated daily. For example, before the season Velasquez was projected for a 3.71 ERA, and his projection now is a 3.47 ERA.
I want them to win I really do, but they had a stretch this good en route to last year’s finish. Teams have nice streaks, and some players have certainly improved, but to think any of the actual winning is sustainable is crazy. Performance might stay the same but outcomes will begin to change.
I don’t disagree with you….its just should their playoff odds not reflect improved performance? Several players are significantly outperforming their projections, maybe due to real improvement in skills, yet the projection doesn’t change at all……that seems to indicate an issue with the projection, no?
I saw in this article (http://www.fangraphs.com/blogs/the-difference-between-predictions-and-projections/) that the best projection systems have a standard deviation from actual win-loss records of 6-10 games. What is the standard deviation for Fangraphs projection systems?
The Braves are already zero across the board, not even a miracle at this point.
If it were possible to have NEGATIVE playoff odds, the Braves would be on that list. I think their .256 W-L pct is sustainable and they have a shot at eclipsing the modern-day loss record of the ’62 Mets. Especially if the Phillies play respectable baseball.
If only the Phillies had traded Cole Hamels for Brad Miller, like Dave had suggested.
Atlanta’s playoff odds look like my grandmother’s last seconds on earth.
I have noticed that whenever a team really surprises in the win loss column over projections they are a low scoring team that is very good a run prevention. This leads me to conclude that not all runs are the same. Runs prevented seem to matter more towards a team winning a game than runs scored. Logically this makes sense. A team can’t lose a game in which they give up zero runs but team loses several times a season when they have scored more than 10 runs.
Found a flaw in your reasoning. After the fact, it is easy to say, see these teams that destroyed projections, they had great run prevention so obviously they should have been projected better.
And the 10 teams, 2013*-2015 that beat theirs by 10+ wins were about 61 runs better than average on defense and only 31 runs better on offense.
These ten teams averaged a +88 run run differential yet were projected for only 82 wins on average. The problem is- the projection systems came up with 82 wins because they projected these teams to average only a +10 run differential.
So, I don’t think the problem is so much that they ignore some bonus value tied to run prevention, but that these teams beat the underlying projections. The teams scored 19 more runs on average than Fangraphs projected and allowed 64 fewer.
So if there is something to be learned from these teams, I think you have to dig down into why they were so far off on run prevention first
Here are the 10 teams, I look forward to your analysis on where Fangraphs went wrong with their underlying stats-
2013 CLE +11*
2013 LAD +11*
2013 OAK +14*
2013 PIT +11*
2014 BAL +18
2014 LAA +13
2015 CHC +13
2015 PIT +13
2015 STL +12
2015 TEX +15
* Way Back Machine’s earliest was 5/10/13, so not pre-season for 2013, but as of that date.
I am misunderstood, but I think this time it is my fault. I am not saying Fangraphs projects anything wrong. What I am saying is that teams that win more than their run differential expects tend to be teams that prevent runs extremely well based on my observation. The current Phillies team has given up about 30 or so more runs than they have scored. In other words, They should have lost 3 more games than they should have won. Yet, they are several wins up on losses. I have observed that teams that prevent runs tend to win more than you expect from run differential, and teams that are poor at run prevention rarely win more than run differential expects and often under perform. So run prevention seems better for winning than run scoring.
This article is all the words minus one too long. And that word is Nowayjose. #SanePhilliesFan
I’m the most positive and optimistic Phillies fan there is, but to talk about the playoffs…PLAYOFFS?…..before Memorial Day is ridiculous. They begin an almost month long stretch of games against very good teams next week. Let’s see where they stand after that, or on July 1st, before entertaining such possibilities. I wouldn’t be opposed to trading some players who don’t figure into the team’s future like maybe Hellickson and/or Asche, Chooch to get an established OF bat in the final year of his contract…maybe a Josh Reddick, for example (though there are other possibilities out there as well). We could even consider resigning him on a 2 or 3 year deal, but even if we didn’t I’d like it.
Realistically, this is probably the only time of year that us Phillies fans can even think about the playoffs. Kind of feels like we’re looking at their record and all holding our breath, hoping they sustain this kind of record as long as possible.
Hmm that’s odd…this article is supposedly about the Phillies and yet only one paragraph mentions actual players…and no mention at all that after this weekend their schedule is:
@ Tigers
@ Cubs
Nats
Brewers
Cubs
@Nats
and a home and home with Toronto…this Phillies fan thinks it certainly seems like the regression to what we all thought they were might start somewhere in there, no?