THE BOOK--Playing The Percentages In Baseball

Didn’t we just have this conversation?

Are there actually major views on how to do park factors?  It seems like the only issue is sample size but shouldn’t we have big enough sample sizes to do all the breakdowns we want and get answers that are reasonable approximations.

Why don’t we have agreed upon park factors?  Why don’t we have LH/RH and GB/FB park factors?  What is the barrier preventing us from making a protocol which people believe in to calculate these things?

Bryan/2: only effort and desire is stopping us.  This one is strictly 1% inspiration and 99% perspiration.

So far, we’ve put 0.5% inspiration, 5% perspiration, and 94.5% yapping about not doing anything about the rest.

"Why don’t we have LH/RH and GB/FB park factors? “

I do calculate these for the Hardball Times Forecasts.

I have Safeco HR’s at 0.93 for RHB, 0.89 for LHB - not much of a difference.

Miller Park has a 1.09 factor for ground ball hits to the outfield, while at Petco it’s 0.82.

When doing my projections, I count how many PA’s by each batter as a RBH and LHB at each park, and how many batters of each hand were faced by each pitcher at each park, then created a weighted mean custom factor for each player.

I have Safeco HR’s at 0.93 for RHB, 0.89 for LHB - not much of a difference.

Is your pool of RHH as representative as the pool of LHH?  What is the career HR/PA of each pool?

Right, “we” have component park factors - if by we you mean pretty much anyone involved in doing projections significantly more complicated than the Marcels.

Looking at HR per contacted ball, I have Safeco from ‘06 to ‘09 as +0.002 for LHB, -0.004 for RHB. (I’m using additive components, which are then used to derive your more traditional park factors.)

Before anyone asks - yes, I’m adjusting for “that,” where “that” is pretty much anything I can think of. What I’m doing is:

* Breaking everything out into independent binomial components
* Taking batters and pitchers and producing mostly-road binomial lines (the park of interest is downweighted to 1/15th of the batter’s overall line if it’s his home park)
* Regressing those batting lines to the mean
* Using the odds ratio method, calculating expected batting lines from the regressed lines
* Finding and regressing observed batting lines for that park
* Finding the difference between them to create component factors
* Using the component factors to derive adjusted batting lines
* Running the adjusted batting lines through linear weights to produce R/PA and R/O park factors.

So I tend to think I’m being pretty thorough here. (For those curious - roughly 60 hours to do, 1,780 lines of code. Should have a full writeup at BP next week.)

Colin, that’s a .006 HR per contacted ball gap, or 3 HR per 162 games.

Brian’s gap, (93/89 - 1) x 17, where 17 is the average number of HR is a gap of 0.8 HR.

So, fairly big difference.  Your numbers, in PF-speak, would be something like 107/87.  Much more believable.

Over at THT, I used matched pairs of teams playing home/road. What the Mariners and Rangers did in Safeco, matched with what they did in Arlington. Repeat for every combination of ballparks. I do not break it down by individual player, but by teams each season.

Each component for each park starts with a factor of 1.0. I then do an iterative process where at each step the road totals for the matched pairs are adjusted by the factors at that point, then recalculated.

Knowing how many PAs of each hand at each ballpark for each batter and pitcher, I get a custom weighted mean factor for each player for each component. Use the odds ratio to adjust each player’s component rates and then use flow charting to create a new adjusted batting line.

Derek Jeter would be a good example of a RHH who likely would experience no problem with SAFECO as his home park, assuming the weaker lineup did not get him.  The park might even help him.  Just like Tony Gwynn was not hurt playing in SDG.  So you can’t just look at handedness or assume everyone has an average hit chart.

“All the reason that you need to increase the sample, and do a better job of adjusting for the characteristics of players”

Well, doing a better job on the latter means decreasing the sample. Catch 22.

You have teams who play as few as 3-4 games at some parks, and at most 9 games per season (30-35 AB).  It’s kind of like looking at catchers ability to call a game by comparing them with his backup catcher who may catch as few as 30 games per year, and only the same starting pitcher 5-6 games for 30-35 IP.

There is no control group, and a very SSS, so there is no way to take macro PF’ers and apply them to individuals with any great certainty, at least not without looking at their hit charts, and even then this assumes a hitter or pitcher can not adjust to his park, and does not take into account the quality of the OPP Pitcher faced.  Of course, adjusting can sometimes have negative consequences.

Adrian Beltre and Jason Bay might be a good polar examples.  Beltre went to being a CF-RF FB guy with Seattle, to a guy hitting bombs to LF at Fenway and on the road this year.  Virtually every HR he hit this year would have been a HR at SAFECO, even those hit to LF. 

Jason Bay looked like he made some adjustments as a result of moving to CITI, despite the fact most of his HR last year to LF and CF were on the road and would have been HR at CITI.  This hurt him both at CITI and on the road (who the heck is the Mets hitting coach).

Of course, the big unknown with both is the effect of hitting in a good lineup which wears down opposing pitchers, vs a a weak lineup which allows pitchers to be more efficient and thus pitch around the better hitters.  I can not say for sure, but it seems plausible, and would explain some teams going into team hitting streaks and deep slumps where everyone is hot or cold at the same time.  Once you hit some critical point (number of hitters) of hot cold it may act like a switch going on (hot) or off (cold).  A team whose lineup projects to be at that critical point (hot or cold) will have an amplifying effect or dampening effect for most of the season, at least for some hitters. 

In any event, the simplistic view will be that Jason Bay suffered from a change in park, and Beltre benefited from the change, despite the fact that Jason Bay hit better at CITI on the than on the road, and Beltre hit better on the road than at Fenway.