Daily/Weekly Projections: New Factor MLB Park Factors

ken_admin  4/20/2013 7:18 AM

Under Daily and Weekly projections, you can now include MLB Park Factors in the results.  There is also a new display column Factors which shows the change in player Total Value based on all of your selected Factors to Apply.

A table of the MLB Park Factors can he found here: http://baseballmonster.com/MLBParkFactors.aspx

Here's how players are adjusted:

Home Players: Add 1/2 of the park adjustment to their projections.  We do this because our projection already assumes the player is playing half their games at home so we don't want to apply a full adjustment and repeat part of what's already been done.

Away Players: Remove 1/2 of their home adjustment (built into their projection), and add a full adjustment for the away game park.

The adjustments go back to 2000 when applicable for a team's ball park (newer parks use less years). Here are some comments about specific parks:

NYM - We only use last year and this year since they made significant changes to their park for 2012.

DET - They moved their fences in for the 2003 season so we're using data starting then.

SD/SEA (shortened fences this year) - These will affect the results, but we're still using past data for these parks until more data is available.  Keep in mind that these effects are likely over penalizing hitters at the moment.  We thought it would be better to use some data rather than none.  If you have any thoughts on this, let is know.

bcavers  4/20/2013 9:04 AM

Very nice. Some of the triple numbers are very high. My guess is this is partly being driven by lack of sample size - thoughts?

ken_admin  4/20/2013 9:08 AM

Yes, the higher numbers are for the stadiums with less data.  Maybe creating a weighted system where the results are applied more for each year of data.  For example, if we apply a weight of 25% for each year of data.  This would reduce the current dramatic effects of MIN and MIA.

Thoughts?

liam4  4/20/2013 10:19 AM

Yes, I think some manual adjustments and weighting would help, vs the purity of just using the data itself.  Enhances the usefulness of the adjustment.

if using data from both teams playing at a given park, is the data somewhat skewed by quality of home team pitching?  

Very smart to only apply half the adjustment.

mymaus  4/20/2013 11:25 AM

I like the idea of park factors and I think your general approach is good and I am sure will get better (like the idea of weighting based on how many years are available). The one exception is Safeco and Petco. Obviously the changes WILL have an effect, but I agree that it is very hard to predict the effect with such a small sample size, I wonder if a web site out there somewhere is tracking this (or maybe they have done a study based on last year's fly ball data) - even with the small sample size. I definitely think an adjustment to those parks should be made. I think I'd feel better if you assumed that they were half way between neutral and being a pitchers park, although, based on the changes, it looks like Safeco will be better for RHB and may not make a difference for LHB. I think Petco changes might yield benefits for everyone.

mbuser   4/20/2013 11:27 AM

The sample size is going to be the same for all stats, based on the number of games played. But with triples, there is less volume, which helps increase the variance - 8261 2B, 927 3B, 4934 HR last season.

mbuser   4/20/2013 11:39 AM

Assuming anything is dangerous, because it's not going to help anyone if we assume wrong. Small sample size is indeed a sample size, but the advantage is still that it's actual results. With Safeco and Petco, can we really do anything other than simply acknowledge that there is less data to rely on than with other parks?

liam4  4/20/2013 12:12 PM

Makes sense on triples, far greater variability.  Found it interesting to compare findings here to billjames.com article, as well as espn's park factors (smaller sample size)

http://www.billjamesonline.com/the_best_and_worst_hitters_parks/

Clear agreement on Colorado and Texas as hitters parks.  Data leads to some different conclusions too.  One that surprises me here is the findings on chw, us cellular; seems a little extreme on home runs?

mymaus  4/20/2013 1:32 PM

I agree sample size for Petco/Safeco is extremely small, but you must agree that the changes will have some positive effect on hitters. Yeah, maybe going halfway to neutral is too much without data, but keeping the park factor the same as it's always been is not repsonsible either. I am sure someone has done a study of fly ball distances in each of those parks over the last few years and remapped the results -- especially the homeruns -- with the new dimensions. I know websites have flyball distances and angles data -- seems like, with a little bit of work,  the data park factors for these 2 parks could be better estimate. Yeah, it's not perfect, but it's better than what we have now. No -- I'm not expecting you guys to do this, but it would be great if you could find someone who has. I would be willing to bet the teams in those parks have done it, as well as MLB -- when they approved the dimensions.

ken_admin  4/20/2013 11:19 PM

We've added the concept of weight to the Park Factors where the more ABs, the higher the weight up to 1.00 which is achieved in about 4 seasons.  This weight is shown in the MLB Park Factors page.

We've also changed SD and SEA to only use data since their park changes for this season.  Obviously, their weight is currently very low.

bcavers  4/21/2013 8:57 PM

Just to be clear - you are applying each factor to event (double, triple, HR), and differently for left and right handed batters rather than just across the board right?

mymaus  4/21/2013 9:16 PM

Love the dedication to make your tool the best!!!

I'm not sure about this approach to the park factor. I THINK what you are assuming is that a park starts out is neutral and adjusts from that point as time goes on up until 4 years. If that is correct players stats at parks like SD, SEA and NYM COULD be off by a good measure. For example, based on the changes in SEA I would expect RHB to see a benefit over years past, but LHB may see no benefit. If I use park factor, a guy like Seager could be way over valued (because you are giving him a virtually neutral park this year).  Unfortunately, we don't have enough data to tell for sure. The only useful adjustment you could make at this point is to find someone who remaped hit data like I suggested above.

It seems like a better (but still not completely accurate) approach maybe to start with the historical data for the parks with changes and then heavily weight  more recent data.

Also, I find it really hard to beleive Coors Field's HR factor is not higher. Every study I have seen rates them always in the top 3 and most of the time, #1. Not sure why your stats are different.

Here are some interesting data to look at in comparision (you may have already seen all this, so sorry if it's a repeat):

http://www.statcorner.com/index.php - click on team and then see 3 year park factors on lower right. It's pretty comprehensive. It might be nice to use their adjustments, though obviously their 3 year rolling average still isn't going to help with the NYM, SEA and SD problem.

http://www.billjamesonline.com/the_best_and_worst_hitters_parks/

http://www.parkfactors.com/

http://espn.go.com/mlb/stats/parkfactor/_/year/2012/sort/HRFactor

ken_admin  4/22/2013 8:57 AM

Yes, for now SEA and SD are considered nuetral parks, but we'll think about other solutions including your recommended approach.

For COL and HRs, it's likely different because of what we're measuring.  In our case, we care about the % of hits that are HR, 3B, 2B, along with batting average.  For COL, we don't have a very high % of hits as HRs, but we do have a big boost for overall average for COL.  So the end result is more HRs because we project more hits for the players.  Hopefully that made sense.

mymaus  4/22/2013 9:37 AM

I understand your statement above but I don't understand the method. Maybe a real example would help. Let's say I'm looking at Allen Craig. He's projected for 21 HR's the rest of the year. Let's say we were only applying the park factor. What is the math that you go through and the results to figure out his daily projection for a game at COL and a game at CHW?

ken_admin  4/22/2013 10:28 AM

Here's Craig vs COL:

• Default batting average: 0.300
• Default % of hits HR: 16%
• STL Factors: -20% HR, +4% AVG
• COL: Factors: +9% HR, +26% AVG
• Road adjustments: -1/2 Home = [+10% HR, -2% AVG] + Away = [+9% HR, +26% AVG]
• HR Adjustment: 16% (+10% +9%) = 16 x 1.19 = 19.04% of hits HR
• AVG Adjustment: 0.300 (-2% + 26%) = 0.300 x 1.24) = 0.372
• He's now getting a boost in HR% (percent of hits that are HR) plus his average gets a boost increasing his # of hits (and then HR).

mymaus  4/22/2013 4:20 PM

Thanks for the info. It is helpful but not exactly what I was wondering about. How about this -- with all else equal on a particular day, how many HR's will Craig hit at CHW, COL and DET (I threw Detroit in there because they look close to neutral according to other sources). 

Rapierman   4/22/2013 11:28 PM

I dunno if it's possible, but I've seen some instances where, on TV, the producers will have someone show the actual track of a ball that was hit in one particular park and where it landed, and then compare them to other parks and show you where it would end up being a homer in a particular park as opposed to a double or triple in other parks, and it mostly has to do with home run fence distance and height and the exact shape of the park, given that many parks are asymetrical (Minute Maid Park in Houston, Fenway Park in Boston, etc.)