Holliday and Home/Road Splits
Coors Field is a good place to hit. That’s been true since it opened, and even the most casual baseball fan is generally aware to take numbers put up at altitude with some grains of salt. Thus, it wasn’t a big surprise that Matt Holliday’s abilities to hit away from elevation were greeted with skepticism, especially after he got off to a slow start in Oakland this year. But now, with his playing time in 2008 and 2009 nearly identical, let’s take a look at how he’s performed without playing half his games in a hitters paradise.
2008: 623 PA, .321/.409/.538, 38 2B, 2 3B, 25 HR, 12.1% BB%, 19.3% K%, .217 ISO, .361 BABIP
2009: 612 PA, .311/.389/.522, 37 2B, 3 3B, 23 HR, 10.6% BB%, 17.1% K%, .210 ISO, .341 BABIP
It would be challenging to find a player who has had more similar statistical seasons over the last two years. His numbers are practically identical, with the rates being marginally lower due to a slight reduction in batting average on balls in play. There’s no evidence there whatsoever that he was traded by the Rockies over the winter.
This doesn’t mean that Coors Field has no impact, of course. We know it’s a good place to hit, and one data point doesn’t disprove that. It should, however, serve as something of a reminder that not every player who puts up good numbers in a hitters park is going to immediately start performing at the rate at which they played on the road in previous years.
For whatever reason, it has become normal for people to adjust for park effects by looking at a player’s historical road numbers. Lots of people did this with Holliday, who had massive splits while a member of the Rockies. Those projections, based on his personal home/road numbers, significantly undershot how well he has played this year.
Personal splits can be quite enlightening, but by definition, a “split” is a fraction of a dataset. By making the sample smaller, you’re inherently making it less reliable. A home/road split gives us the effect of a player’s home park on his performance, but jumbles it up with a lot of other stuff that gets in the way.
Splits can be interesting, but be careful with them. When projecting future performance, you’re better off using one of the well tested systems, such as ZIPS or CHONE, which include park adjustments, rather than relying on that player’s previous home/road history.
Print This Post
I think the splits were right but we didn’t look deep enough. Last year, on the road he got progressively better as the trip got longer suggesting that it took him time to acclimate.
No, no, no. Now you’re taking a small sample, cutting it even smaller, and coming up with a post-hoc explanation from that?
No. That’s bad.
While Dave is completely right, I do remember BPro doing a study on this issue several years ago and found this was the case with hitters on the Rockies in general. The longer the trip, the better the team hit.
Out of curiosity, what are his home and road splits this year?
http://sports.espn.go.com/mlb/players/splits?playerId=5940
Home: 309 PA, .323/.397/.573, 15 HR, 21 2B, 2 3B
Road: 287 PA, .299/.380/.465, 8 HR, 16 2B, 1 3B
Players in general perform worse on the road. Using road splits as an evaluation of “true” or “context-neutral” hitting ability is, at best, nonsense.
The difference tends to be pretty uniform between players at least, so you could make that adjustment in a world without reliable park factors.
Holliday’s home/road splits were enormous early in his career, and the gap closed pretty steadily with age. I just always figured it took time for a developing player to figure out the adjustments he needed to make for a mile’s difference in elevation.
Nonsense, maybe, but you have to take it into account at some point. I can understand a 20-30 point wOBA swing from home to road, but when a guy is an all star at home and a bench player on the road, there’s probably more to it than simply a familiarity issue.
That said, there is a decrease of .035 in terms of OPS from home to road in the NL, .038.
He seems to excel at parks named after beer companies…except Miller Park where he is just about average.
Dave,
If your biggest argument against the home/road splits within a season is sample size, why not combine years? That would double the sample size. What other arguments you have against the home/road comparison?
For the record, Holliday had an even BIGGER h/r split (+200 points OPS) in 2007. He had a 100 OPS split last year. And in 2006, he had a 300 point split. Maybe people should’ve looked at the trend, perhaps?
Travis – the point that Dave was trying to make is that intentionally splitting up a players numbers will always lead to a shakier conclusion. It’s entirely possible that Holliday has simply been hotter at home in his career than on the road, correlation doesn’t imply causation. Of course, there obviously is some true meaning in the home/road stats, but it’s hard to separate the two.
You are never going to get as big a sample size as you’d like to evaluate any one player. No matter how many games or PAs you track over a player’s career, there will always be sampling issues; they just diminish gradually as the sample increases. One full season is obviously better than half a season. Two seasons are better than one season. Ten seasons are better than five. If you could have twice as large a sample over the same amount of time, you’d always gladly take it. So you never want to take half of your sample for a single player and just throw it out without a really good reason, because no matter how big the remaining half is, it’s not big enough that you don’t have to worry about sampling issues. Looking at a trend over multiple years is not going to overcome that issue. It’s basically just what Dave talked about in his post above; taking the same bad data and fitting it to your conclusion after the fact. Just because you know what happened and then go looking for trends that appear to fit that does not mean that whatever you find had any predictive value.
There are additional issues with home-road splits besides sample sizes, however. Looking at a player’s road stats, even after adjusting for HFA, which is absolutely necessary, is not the same as looking at what he should do in a neutral park. You don’t know for certain that the player’s personal HFA is the same as average, and it could be that the player will naturally have a higher home-road split than normal for any park he plays in. A player’s road environment is not always neutral if he plays in an extreme home park (this is why park factors use an other parks corrector). Some parks exaggerate splits by not only inflating home stats, but by deflating road stats as well. For example, pitches move differently at Coors than at other parks because of the atmospheric differences, so that makes it more difficult to adjust to hitting on the road for Colorado players than for players on other teams.
Whenever you have home road splits that differ from what park adjustments say for a player, there’s not really any reason to just take the road splits and throw out the park adjustments. You would need more data than you’re going to get from one player even over multiple years to trust the road splits more, and you’d still have to make adjustments to those splits before they were usable.
Is there a way to look at the entire sport’s home and away splits? Could we determine what an average home/away split is, and use that to figure out how different people’s splits are compared to the norm?
I am assuming the average player is better at home. Maybe this is wrong.
Go to Baseball-Reference and look at the splits for all MLB (you can go to either the AL or NL splits page for a given year, and there should be a link to see all of MLB near the top of the page). That will tell you the league-wide home/away split for that year. You can look over several years to get a better idea of what the true split should be, but it’s probably similar most of the time. Some have argued it’s gone up in the past couple years because of a crackdown on amphetamines, so there might be a difference from earlier years if they are right.
That would be the minimum adjustment to make when looking at home/road splits, but you still run into lots of problems, particularly that you are still throwing out half your sample.
NL (home in 2009): .264 / .338 / .421
NL (away in 2009): .255 / .324 / .399
Colorado (home in 2009): .285 / .365 / .481
Colorado (away in 2009): .236 / .319 / .399
NL (home in 2008): .265 / .338 / .421
NL (away in 2008): .256 / .325 / .405
Colorado (home in 2008): .278 / .350 / .454
Colorado (away in 2008): .249 / .322 / .377
Roughly taking Colorado out of the splits, NL teams have an extra 17 points of slugging and 12 points of OBP at home in 2009. Not an insignificant difference over the course of every plate appearance this year by a non-Rockie. The AL has similar differences to the NL, so this isn’t a pitcher-batting strategy related split.
Previous road numbers aren’t a great predictor if you don’t adjust them for the new home park and the 15 point OPS jump an average player gets over the course of the year due to playing better at home. (30 points extra at home = 15 points on the year).
And that little ZIPS projection glitch in Holliday’s ISO seems to have resolved itself.
Do we have any stats for BABIP at Coors vs. other stadiums? I’d be curious if it’s affected (one would assume raised) by the size of the park, and we could potentially attribute some of the drop in BABIP this year to that.
That is going to measure defense more than anything. If you take defense into account, using UZR or something, you could get a BABIP park factor, but I’m guessing it will be too noisy to be useful.
ZIPS ISO projection for the rest of the year is .167. So ZIPS actually does predict a huge drop off right? What does that mean?
Well, Holliday has a career .229 ISO, and it’s never been below .198. Furthermore, it’s been over .200 each year since 2005. He is also 29, which is when power starts to peak.
I love ZIPS; however, I have a lot of trouble believing in that projection.
Oakland elevation: 42 ft
St. Louis: 466 ft
There you go. Holliday needs 200 ft minimum to perform optimally ; )
What about comparing Raul Ibanez’s numbers this year to last year’s?
While people just looking at the home/away splits were obviously using poor methods and underselling Holliday’s abilities, I think many people were doing more than that.
They were factoring in the fact he would be playing in Oakland and saying his numbers would likely be a little below his “true talent” level. In those peoples defense his numbers while on oakland were similar to most of the projections. Now he moves to St. Louis and goes on a hot streak (I cannot find his babip split pre and post trade but it was .301 at home in oakland vs .400 in busch). Now would he have had a hot streak regardless of team he was on and brought his numbers up who knows but Holliday is still a difficult player to get a read on his true offensive talent level.
The method you’re talking about, (using park factors for Colorado and Oakland) is what Dave is advocating. Using a well established park factor gets rid of the small sample size issues that we have.
I agree with Dave on the main issue like I said but was disputing this paragraph mostly:
“For whatever reason, it has become normal for people to adjust for park effects by looking at a player’s historical road numbers. Lots of people did this with Holliday, who had massive splits while a member of the Rockies. Those projections, based on his personal home/road numbers, significantly undershot how well he has played this year.”
First, I am not sure how normal that behavior is, yes it is overused and its meaning is oversold but it does not seem like this was the only reason people were down on Holliday, even the less informed were down on him for other valid reasons. Also, like I said were those “projections” and others that far off when he was in oakland before he hit his recent, at the very least luck aided, streak?
This isn’t just Holliday, this holds true for most hitters that leave Coors. They get used to the ball moving one way at home and it moves differently on the road so they always have a larger home/road split than reality.
It’s one thing to take a system-wide response to explain a player’s numbers. It’s entirely something else to take a player’s numbers and explain a system-wide study.
It could very well be a simple notation that Holliday just happens to be an exception to an overwhelmingly popular idea and justification of home/road splits and playing at coors vs road.
And it’s equally unreliable to listen to the quote of …”When projecting future performance, you’re better off using one of the well tested systems, such as ZIPS or CHONE, which include park adjustments, rather than relying on that player’s previous home/road history.”
–i.e. Zips is projecting ISO to be .176 on out, when he’s never hit below .198 in a year. Yeah yeah, I understand the regression and mean reversion aspect, but just think about that for a second.
Not everything can be explained. And everything, at one time or another, is incredulous.
If you don’t get why a player could be projected to perform worse than ever before, then you don’t understand regression to the mean as well as you think you do.
Dave, this seems a little bit more than normal regression to the mean. ZIPS projected a .196 ISO before the season started. Now, he puts up a .214 mark this year, and that lowers his projection by .20 points? That doesn’t make much sense at all Dave.
I see Zips projecting .191 ISO now. It would seem Dave got a little carried away and trusted the projections to be correct a little too much.
ZiPS is projecting a .205 ISO for the rest of the year. Don’t know if there was some kind of error beforehand or what.
You are isolating a non-discussion point. Further, you misunderstood the hints of sarcasm. For almost every player, one could make a valid argument as to why/why not they will play a certain way (much better/worse) based on their home/road splits, and in this particular case, their specific home splits given the hitter’s park. To argue against that is ludicrous. Yes a few hundred ABs is a small sample size, but the argument that Holliday should have been crappy this year was valid. So the last 3 paragraphs of your article of “pointing the finger and saying you shouldn’t have made the data sets unreliable to explain how Holliday should play this year” are frankly erroneous. Maybe I misunderstood you, but that’s how I see it.
Your article goes like this:
Guy plays at a hitters ballpark, gets traded, and now shouldn’t play as well
Oh crap he is still playing at the same level
Now now, let’s not forget this could be an exception
We’re idiots for extrapolating with splits and making data unreliable with small sample
We should have just used ZIPS/CHONE after seeing Holliday’s great season
Let’s not use splits to explain something (data set small); Holliday isn’t an exception
And that would be wrong.
Todd Helton must be psyched to have a reference point in the case that Coors Field made him look way better than he really is.
THE WAY I READ IT, HITTING BEHIND MR. ALBERT PUJOLS IS SIMILAR TO HITTING IN COORS FIELD???!!!
TOLAXOR! Awesome.
If you define home as the NL, and away as the AL, Holliday certainly prefers home cooking. I doubt any AL teams will be looking at Holliday when he becomes a FA.
Pretty much every argument against using just his road stats to determine his neutral park value applies here as well.
Question: If it takes Rockies a while to get acclimated on the road, wouldn’t away teams also be un-acclimated when they show up at Coors Field? Does anyone have stats that show home team batting statistics vs. road team batting statistics at Coors Field? Then compare that to the stats overall (or overall xCoors). If that argument – a big part of the reason the Rockies hit worse outside Coors has to do with adjusting to the altitude or what have you – holds water, I’d expect to see the Rockies’ Coors vs. non-Coors numbers are much better than the other teams’ Coors vs. non-Coors numbers. I don’t know if this is the case, but I’m hoping someone has the numbers handy…
Per Retrosheet gamelogs:
Typical HFA is about .012 points wOBA (from 2000-2008, at least). The Rockies hit about .066 points wOBA higher in Coors than on the road (also 2000-2008). Their opponents hit about .019 points wOBA higher in Coors than against the Rockies at their own home park. So the difference between the boost the Rockies get from hitting in Coors and the boost their opponents get from playing in Coors is significantly greater than the typical .012 point advantage in wOBA the home team enjoys. A small amount of that is probably just from the fact that Coors wOBA is on a higher scale than at other parks, so the absolute difference should be a little higher, but most of it is probably the adjustment factor the road teams face.
Another stellar piece of Dave Cameron statistical analysis.
Matt Holliday 2008 road stats – .308/.405/.486
Matt Holliday 2009 stats (Oct 2) – .311/.390/.514
Matt Holliday 2009 ZIPS Proj – .289/.367/.486
Matt Holliday 2009 CHONE proj – .286/.357/.479