Park factors for RAA data

[jinaz]

Hi,

Could someone either explain or point me to a link that explains specifically how to use park factors on Runs Above Average offensive data.

I'm comfortable using park factors on RC-style (i.e. absolute runs) data, as explained here.

But you can't just divide by a park factor if you're using RAA data, as that would just move everyone back toward zero instead of giving a consistent across-the-board boost or penalty to all hitters on a team.

Not understanding how to handle this is the main reason that I've been so reluctant to move to using PA-based linear weights instead of the outs-based linear weights I've been using up until now. And I just haven't seen a good explanation (which means I've probably missed it).

Thanks,
Justin

[SABR Matt]

Any particular reason you want to move to the average-centric framework? There's no inherent advantage to doing so, and frankly, I consider it a waste of time to build things in an average-centric frame of reference. To compare players, you're going to need to "un-average" everything anyway because when you're in an average frame, you lose playing time considerations and a 100 AB defensive replacement can look just as good as a 700 AB stud.

To answer your question, ratio factors are poorly designed anyway, but the way to do it would be to...un-average your stat, and then re-average it after you make the adjustment. And no, there's really no other way.

[jinaz]

Matt,

Thanks. So what you're saying is that I'd need to basically convert my RAA measure for a player to an absolute runs measure using league average runs per PA or whatever, apply the park factor, and then convert back to RAA?

That helps a lot...I think now I'm starting to understand what I saw Colin and Tango talking about where they somehow subtract or add runs per PA to apply the park factor. That's probably a shorthand way of doing the same thing, and I think the math works out the same as what you described, ultimately. I'll have to try it. I couldn't figure out how to estimate how many runs to adjust based on a ratio park factor, but I think you just gave me a framework to do it. Maybe it was obvious to other people, but it wasn't to me for some reason.

The reason I'm considering moving to RAA with my linear weights calculations is basically just that "everyone else is doing it." I base a lot of what I do on Tango's work, and use FanGraphs a lot as a quick 'n easy data source, and so it would be more convenient for me to also start with RAA.

There was also some discussion at a point some months back that I frankly never could get my head around indicating that calculating rate stats as Absolute Runs per Out and then converting to RAA won't necessarily give you the same answer as RAA per PA. And that this made using an RAA framework a bit more desirable. Colin and Tango seem to understand it, but I never quite figured it out as to whether I'm doing something wrong or not.

Cheers,
Justin

[SABR Matt]

I await Tom's response to this, I anticipate he'll show up here and make a case for starting with an average-centric frame of reference...even I find it necessary to use average-centric starting points for some parts of my analysis (especially anything requiring matrix math because when you solve a system of equations, it's simpler (logically) to resolve things around zero = average)...but I see no real value in using average-centric linear weights to rate players. It won't hurt you, other than making the steps in the contextual analysis more annoying...and if it makes it easier to compare your own work to theirs, then by all means...

But yes...what you will have to do to use ratio park factors is convert average = 0 to average = average RS/G, apply the ratio factor, and then figure out how many runs above or below average results from the ratio (in the given playing time)...and that will be your average-centric park adjustment.

[Tango Tiger]

I prefer simply making it additive. If a park suppresses .01 HR per PA, then add that to everyone. I don't see why you would necessarily want to make it a multiplier, since that presumes that a great HR hitter will have been more impacted by a great HR park.

When THAT is proven, then I will change my mind. Until then, I prefer the safe (and much much easier) way of just giving every player the same boost, on a per PA basis.

(I know, you end up getting negative numbers for Juan Pierre at Coors or something. I can live with that.)

[AstrosFan]

I prefer simply making it additive. If a park suppresses .01 HR per PA, then add that to everyone. I don't see why you would necessarily want to make it a multiplier, since that presumes that a great HR hitter will have been more impacted by a great HR park.

When THAT is proven, then I will change my mind. Until then, I prefer the safe (and much much easier) way of just giving every player the same boost, on a per PA basis.

(I know, you end up getting negative numbers for Juan Pierre at Coors or something. I can live with that.)

Matt, isn't this one of the things you discussed in your explanation of the FSIA-Matrix?

[SABR Matt]

Yes...I prefer making the park factors additive as well. Tom knows this...but the original poster asked specifically about rehabilitating ratio park factors so I assumed he was using prepackaged park factor data, rather than "rolling his own".

If he wants to calculate additive factors, that is by far the better approach.

[jinaz]

Right, I'm just using Patriot's park factors. I like 'em because they're 5-year regressed park factors, and because they just operate on runs rather than components (which is relevant for value, though less good if you're trying to make a forecasting system...which I'm not).

But I think I should be able to convert Patriot's park factors to an additive park adjustment for an average player, and then use those numbers to adjust RAA numbers.

Thanks, folks.
-j