Now with Fangraphs Park Factors

Steamer has officially adopted Fangraphs

Parks Factors, the differences are considerably larger than I might have
expected and I think we can all feel good about this change. Park factors were
never something we had great enthusiasm for, some of our park factors were out
of date and, personally, I am quite pleased to pass the buck and rely on Fangraphs here. Go here to
read up on how Fangraphs creates its park factors.

So, how big are the changes and whose projections changed the
most?

I’ll start with the batters.

The following are root mean square
differences (in wOBA points) between batter
projections with the new park factors and with the old park factors 
and one version I ran
with NO park factors at all:

Steamer NEW pf
& Steamer OLD pf: 
5.0

Steamer OLD pf & Steamer NO pf: 5.2

Steamer NEW pf
& Steamer NO pf: 
6.2

So, the typically difference between the old
park factors and new park factors is 5 wOBA points
which is just as large as the difference between using the old park factors and
using NO park factors. This kind of blows my mind. To put it in perspective,
here are the differences between different projection systems:

Oliver & ZiPS: 13.1

Steamer NO pf
& ZiPS: 
11.3

Steamer OLD pf
& ZiPS: 
11.1

Steamer NEW pf
& ZiPS: 
10.8

Steamer NO pf
& Oliver: 
14.4

Steamer OLD pf
& Oliver: 
14.2

Steamer NEW pf
& Oliver 
13.9

So, Steamer is more similar to both ZiPS and Oliver with the new Fangraphs
park factors than it was with our old park factors. This is a good sign. When
we make changes that generally move us away from the consensus for we have
to be doubly sure that we’re doing something clever and not something horribly
foolish. Two other things of note: Steamer looks more similar to
ZiPS than to Oliver, and the Fangraphs
park factors are further from having no park factors at all than our old park factors were.

Of course, there will always be winners and
losers.

Three biggest winners (min 300 projected
PA):

1.
Will Venable: +13 wOBA

According to Fangraphs
Guts, what had been an 81 lefty HR factor in 2012 became a 109
lefty
HR factor in 2013 after the right-field wall was moved in from 360
feet to 349 feet. Here’s a good example of Fangraphs
being on the ball while we were asleep at the wheel.

2.
David Lough: +12 wOBA

Lough was traded to the Orioles. Fangraphs
has Camden increasing lefty HR’s by 14%, we were only
using a 4% increase before.

3. Justin Morneau: + 12 wOBA

Another lefty, this time just signed by the Rockies. We had Coors helping
righties more than lefties. Fangraphs has it the
other way.

Three biggest losers (again min 300
projected PA):

1.
Michael Morse: -23 wOBA

Wow. 23 points. Our AT&T park factor was
several years old and looks like the Fangraphs
version from a few years back. It’s played worse for righties of late. The same
could be said for our Safeco park factor (it was behind the time)– but it has moved in the other
direction. This combo it a tough one for Morse.

2.
Curtis Granderson: -14 wOBA

He got more help from Yankee Stadium that we
were previously accounting for.

3.
Dexter Fowler: -13 wOBA

Fangraphs has a significantly
bigger penalty for going from Coors to Minute Maid that our old park factors
did.

Now, on to the
pitchers.
Here are the root mean square differences (in ERA):

Steamer NEW pf
& Steamer OLD pf: 
0.06

Oliver & ZiPS: 0.41

Steamer OLD pf & ZiPS: 0.50

Steamer NEW pf & ZiPS: 0.50

Steamer OLD pf
& Oliver: 
0.50

Steamer NEW pf
& Oliver 
0.50

Here the changes due to the park factor
change are much smaller compared to the differences between systems. ZiPS and Oliver are more like each other than we are like
either of them and the park factors don’t get us any closer.

The winners (min 30 IP) :

1. Wandy Rodriguez: -0.18 ERA

2. Joe
Thatcher: -0.14 ERA

3.
Trevor Bauer -0.12 ERA

The losers (min 30 IP):

1. Ian
Kennedy +0.22 ERA

2.
Huston Street +0.14 ERA

3.
Burch Smith +0.14 ERA

 

 

6 thoughts on “Now with Fangraphs Park Factors”

  1. Really interesting. Especially the change caused by the Coors factor. Thanks!

    (Unfortunately the pitcher csv at Fangraphs only has about 630 pitchers in it this morning. Is that intentional?)

  2. A couple comments:
    1) When you say, “So, Steamer is more similar to both ZiPS and Oliver with the new Fangraphs park factors than it was with our old park factors. This is a good sign”, I wouldn’t really agree. You guys have beaten the other projection systems in avg. accuracy over the past few years. Couldn’t one of the reasons for your edge be the others’ use of overly aggressive park factors?

    2) Handedness component park factors (e.g., LHB HR PF) are extremely noisy overall. As you know from being a projections guru, HR/FB rates are very fickle from year to year. So to base future projected PFs off of these rates — even an average taken over a few years is not enough — is very dangerous in terms of predictive accuracy. In your example, there is almost no way the “true” LHB HR PF would go from -19% (2012) to +9% (2013) after a wall move of 11 feet in San Diego, yet your projections are now assuming that it did. Have you (or anyone at FG) done work on the predictive accuracy of these LH/RH component PFs. I suspect the y2y correl. is not strong and that they would have to be massively regressed to optimize accuracy. Even when using them historically, you are essentially attributing a (perhaps large) portion of a player’s success (or lack thereof) to playing conditions, when in fact the causality (park–>performance) may not be nearly as strong as the raw data suggests, due to random chance overwhelming a relatively small sample.

    3) To determine if the old Steamer park factors have been a source of consistent error over the past couple seasons, couldn’t you take a look at your average wOBA/ERA projections by team and see which teams, if any, were consistently over/under-projected?

    Sorry to be negative here. I recognize this is a free resource and it’s your prerogative to make any change you wish. It’s just that Steamer is my favorite projection system (by far), and this change strikes me a bit reckless, unless there is some behind-the-scenes research supporting it that I am unaware of?

    1. Thanks for the thoughts. I’ll have to think about this more about each of your points but here’s my first reaction.

      Regarding #1, it is true that these are somewhat more aggressive than our older factors but, and I think more than that, they’re also just different and I think that the fact that they’re different in a way that makes them more similar to other system is encouraging. We’re not claiming that other system have this pegged perfectly, of course, I’m just thinking that their errors in this case would be largely independent of our errors — if they’re not, and we’re all being overly aggressive in our park factors and in the same direction we could get more similar by being worse. Certainly worth considering. If we were all building park factors roughly the same way and all not regressing enough, this would be the case.

      Re #2, Fangraphs is regressing these park factors and is doing so heavily (hopefully to the correct extend). When the ballpark changes (fences are moved in or out or whatever), as I understand it, Fangraphs is starting fresh and with one year’s worth of data, they’d be regressing quite a bit. I do think that it would be better to model how the park factors would be expected to change based on the park changes made and then regress (even more heavily) to that estimate instead. So this isn’t perfect — but I’d argue that it’s bad.

      Re #3, it’s a good idea but we haven’t done it. Unless we’re missing by quite a bit though, I think we’d need several years worth of data to feel confident that we had a team level bias.

      Maybe the answer (as usual, unfortunately) is that we need to put more time and thought into this.

  3. First of all thank you for the great work you guys are doing! I’ve used your projections for a few years now and I am really satisfied with them. Just to be sure, if I want to park neutralize the projections, should I correct the numbers with Fangraphs Park Factors?

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>