Daily/Weekly Projections: Updated Factor Opposing Pitcher and New Factor Left/Right Matchup

ken_admin  4/15/2013 6:13 AM

You can now optionally turn on these two factors for daily and weekly projections:

Opposing Pitcher:

Hitter projections are based on their batting average, HR-rate, 3B-rate, 2B-rate, K-rate, Walk-rate, batting order, strength of players batting before/after them, etc.  This factor looks at what the opposing pitcher allows for batting average, HR-rate, K-rate, Walk-rate compared to the average pitcher, and then the hitters projected stats are adjusted accordingly.

L/R Matchups:

We've looked at historic data for the effects of hitters vs starters based on L/R.  Based on these results, we adjust a hitters batting average, Hr-rate, 3B-rate, 2B-rate, K-rate, Walk-rate. Since pitchers change during a game, the adjustment does not assume all at-bats for the hitter will be against the L/R of the pitcher.  We simply look at the effects on hitters based on the starter of the game being L/R.

#28675  4/17/2013 6:13 PM

Regarding opposing pitcher quality. Does it use opposing pitcher projections or actual YTD stats?

ken_admin  4/17/2013 7:07 PM

Currently, we use our pitcher projections to determine the opposing pitcher difficulty.

#28675  4/19/2013 10:18 AM

What happened to the Park Factor option in the preview site?

ken_admin  4/19/2013 11:01 AM

There's an updated version there which will be deployed to the real site tomorrow along with an explanation of how they work.

awinters  4/20/2013 1:42 PM

Are the pitcher's projections also adjusted for quality of the opposing team's batters and the park?

bcavers  4/23/2013 10:27 AM

Would like the above question answered also

ken_admin  4/23/2013 11:19 AM

Currently, it's only hitters, but we're in the process of adding park and opponent strength for pitchers too.

awinters  4/26/2013 6:37 AM

Thanks for filling us in on this.  I contrived a model for factoring in the park and opposing team for pitcher's but I suspect yours will be much more accurate. Let us know when you get it done. 

QPiz25  4/28/2013 4:00 PM

Nice work with the hitter factors! I've noticed a change for the better in all of my teams. Is there any timetable for the updated pitcher factors? 

ken_admin  4/28/2013 8:10 PM

Thanks. We'll work on these early this week and should have them by Wednesday. Sorry for the delay.

ehrenberggreg  5/2/2013 1:12 AM

Hey I have been playing around the pitchers parks feature and I was wondering why the point projectons for almost all pitchers go down with it on versus it off.  For example Kershaw is projected for about a full point less on Friday pitching in San Francisco with the pitchers park on, which is a known pitchers park.  What is the cause of this?

Rapierman   5/2/2013 7:00 AM

I'll take a wild guess:  It depends on the type of pitcher and what type of out that pitcher goes for.  If the pitcher goes for fly balls, he'll have a harder time in parks with a short distance to the fence (i.e., Minute Maid Park, Fenway Park, etc.) as opposed to pitchers that go for ground ball outs.

ken_admin  5/2/2013 9:56 AM

Looking into Kershaw specifically, here's the factors we are adjusting:

HR, 3B, 2B, BB, SO

Many pitchers benefit at SFO because HRs are down.  In Kershaw's case, he doesn't allow many HRs so his benefit is less than others.  The major downgrade is SO.  I wasn't sure if SO should be adjusted for park, but cases like COL have a -20% for SO so it certainly needs to be used for some.

The adjustments also are based on your scoring system, and not all reward good pitchers as much as others.  I assume you are using FanDuel so the loss in value occurs because they value strikeouts so much.

dk maestro  5/2/2013 10:55 AM

I think that what youa re doing is great. It is really interesting to look at how the different factors affect hitters and pitchers, and especially useful for weekly H2H matchups, where a players sample size is a bit bigger than just limited to 4 PA's. My only concern is hwo often the projections for the players are updated? It seems that some hitters (Josh Hamilton) is a shining example of someone needing an adjustment - yes - he will not be this bad all year, but on daily and weekly projections, he consistently rank in the top 20 or so, and he is producing like a 7th outfielder.... Same goes for a couple of the pitchers. I know it is a lot of work adjusting projections on all the players based on one month of the season, but some need it. Also - any way to take into consideration, if a pitcher is going up against a hot hitting TEAM? CLE have been on fire recently, and I am inclined to sit my pitchers against them, until they cool off.

mbuser   5/2/2013 11:57 AM

There is no set schedule for adjusting players, it's an ongoing process. A player like Hamilton (.313/.370/.583 over the past three seasons) is going to get a longer leash than others. I'll adjust him down a bit now, I certainly can't argue that he hasn't looked really bad to this point.

QPiz25  5/2/2013 1:30 PM

I look forward to implementing the new park factors.  Also, I noticed there is a 'HomeAway' column for the daily projections. This column is always blank on my screen. Am I not using this correctly, or has this feature not yet been fully included in the daily/weekly projections? 

Thanks again.

bcavers  5/3/2013 12:05 AM

Question: When you adjust for the opposing pitching, are you adjusting the right amount for the starter? What I am asking is do you adjust hitter stats up and down based on how many innings you project the starter to go + league-average quality relief pitching for the remainder, or do you just do everything off of the starter?

ken_admin  5/3/2013 1:12 PM

Currently, it's only considering the opposing starter but adding the bullpen effect is a good idea.

bcavers  5/3/2013 5:22 PM

That would be great. I'd encourage you to use league-average bullpen quality rather than trying to estimate it as bullpens are inherently unstable from year to year and I've seen some research indicating that assuming league-average rather than prior year is better for projections.

ken_admin  5/3/2013 9:56 PM

Using league-average bullpen to project all players would have the same effect on all players, right?  If so, it may affect overall projected stats, but it wouldn't help/hurt players relative to others.

bfyoung  5/4/2013 11:21 PM

On average, do the additional factors improve the utility of the regression model (e.g., classification accuracy, H-L goodness of fit, R^2)?

If so, to what extent? 

ken_admin  5/5/2013 4:54 PM

We have not done an analysis with those techniques, but we'll do some research and get back.  If there's anything you'd like to add in the meantime, feel free. Thanks

awinters  5/5/2013 7:41 PM

I just saw that you added park and opposing team factors for pitchers which is awesome.  One question I have is - are pitchers stats also adjused for opposing pitcher?  I assume that would involve only the probability of a win for a given game.  That is, a pitcher's chances of a win if his team is hitting against a good pitcher will be lower than if his team is hitting against a below average pitcher.  Thanks!

bcavers  5/6/2013 6:38 PM

"Using league-average bullpen to project all players would have the same effect on all players, right?  If so, it may affect overall projected stats, but it wouldn't help/hurt players relative to others."

Players would converge, though. Also value might change relative to pitchers. Both of these things could change what players you select on daily sites since they have different salaries so an "optimizer" might select different players in this case.

dk maestro  5/7/2013 12:53 PM

Kslight/Buser, Possible to adjust "Opposing Pitcher" factors to implement their stats YTD? Some pitchers are clearly much worse than their current projections and some better, and I wonder, if the projections could be adjusted to show a combination of their YTD stats and yearly projections. Thanks

ken_admin  5/7/2013 1:05 PM

We'll be adding a new Analysis Source for current season and past # of days shortly.

robjc99  5/8/2013 10:37 AM

Great work with the addition of these new factors as it seems like the daily projections have improved noticably since I've been using them.

Is it possible to get the value(s) used to adjust the players stats based on the inclusion of a factor as an output in the projection table? Ideally I would like to see whether the factor represent a positive or negative adjustment to the players stats and with some back testing see which of the factors are most predictive.   

dk maestro  5/15/2013 12:14 PM

Is it possible to add one more factor - groundball/flyball hitters/pitchers. Reasoning is that GB pitchers will do better against a GB hitter and FB pitchers against FB hitters. But a FB hitter against a GB pitcher gains an advantage. Let me know, if it is worth looking at.

ken_admin  5/15/2013 12:21 PM

We may not have those stats, but I'll see what I can find.

descender  7/9/2013 5:39 PM

Do the park factors apply straight to the value column?  Or does it only apply to park factor related columns like HR?

mbuser   7/9/2013 6:11 PM

Both - you should notice both overall Value and categorical values change when you apply/remove a factor.

descender  7/21/2013 12:05 PM

Fenway has a pitcher friendly park factor?  Is that right?  

mbuser   7/21/2013 12:15 PM

It has a rather large -HR%.

http://baseballmonster.com/MLBParkFactors.aspx

descender  7/21/2013 12:21 PM

Yeah I was looking at that, but are we sure about the total result?  ESPN still has the park as +runs.  Not by much, but certainly not negative like this is suggesting.

I'm not even considering pitching CC there.  Minor @ CHW scares me less (and has a negative run factor on ESPN...).

What is the discrepency between the two?  I'm see that the ESPN data is solely for this season, and they still show the depressed HR rate.

descender  7/21/2013 12:28 PM

Just to expand on that, Fenway has had a +runs factor on ESPN every year except 2002.

mbuser   7/21/2013 1:17 PM

We apply factors a bit more specifically than up/down expected runs, we concentrate more on the events that would produce expected runs (2B, 3B, HR, BB, K, etc). In Sabathia's case, he's pitched fairly well but has a poor HR/9 this season (his xFIP is half a run lower than his ERA), so scaling back his expected HR/9 improves his outlook.

"With that said" we do need to review our inputs, outputs, and weights and values assigned to it all to make sure it's where it should be. We'll keep you posted on the results there.

descender  7/21/2013 1:40 PM

My curiosity continues.  Park factors probably should be applied seperately for LHP/RHP results.  Some parks heavily favor certain pitchers with their lopsided designs (yankee stadium comes to mind).  If that sort of information is available it could give users of this website a significant advantage.

mbuser   7/21/2013 2:36 PM

Can you point to some data that shows how Yankee Stadium treats certain pitchers versus others?

descender  7/21/2013 9:00 PM

If i could point to the data myself I wouldn't be paying you guys any money.  :)

LHP pitch better in places like Safeco because of that deep left/left-centerfield, which is death to right-handed power.  They constantly try to build their staff around that fact.  I assume the opposite should be true as well.  RHP generally face lefty power hitters and lefty heavy lineups, and somewhere like Yankee stadium that would be the short side of the park.

It probably correlates better with the batted ball profiles of the batters than the pitcher's handedness, but it still seems worth looking at (or show me if you know where it has been looked at already).

mymaus  7/21/2013 10:36 PM

You ask and you shall receive. Descender is correct on both Yankee Stadium and Safeco. Go to the following site and click on the team who plays at the park you want to see. Stats for LHB and RHB are on the right along with some other interesting and very useful stats. I agree that you should look into these for your adjustments with the obvious caveat to beware of Safeco, Petco and Citi field changes.

http://www.statcorner.com/index.php

Do I get a free subscription next year for pointing this valuable piece of info out? :)

Tell you what....I'd trade that free subscription for that Ease adjusted projections I've been talking about.  :)

mbuser   7/21/2013 10:38 PM

mymaus, that data (and everything else you are likely to find) is for batters and not pitchers.

mymaus  7/22/2013 11:02 AM

Good point but I wonder if this data could still be useful as Descender pointed out if I am a RHP pitching in Yankee stadium which favors LHB I think I would be in for a rougher game if the circumstances were for the opposite. Maybe these factors can be applied to that day'sOpposing Lineup projections before it is then applied to that day's pitcher. Just a thought.

mbuser   7/22/2013 12:22 PM

Oh, batter data is certainly useful - that's why we're already using it

descender  8/2/2013 11:14 AM

Have you guys gone over the park factors and weights at all yet?

When I see things like this "-1.18 @ Pit" I still think it is far too extreme of an adjustment.  When it takes a player from rank 30 to rank 700+ there is no bloody way that is a useful adjustment.

ken_admin  8/2/2013 9:07 PM

It looks like the @PIT example is for COL so the factor difference is going to be more dramatic because it's relative to their normal home hitting.  What's the specific example you're citing?

descender  8/4/2013 11:16 AM

It was Cuddyer that day, but he ended up not playing anyway.  It just seems a bit extreme, by about double what would be a useful adjustment.  He only has a 3 HR difference in his home/road splits this season.

descender  8/4/2013 11:23 AM

Today is another good example.  It has Quentin 26th, in SD (where he doesn't hit at all) against Hughes (who pitches better on the road).  Cuddyer is 768th with fairly neutral home/road splits (and yes I understand Burnett at home is really good).

I guess my real question is how far back does it pull/utilize splits data?  Quentin is hitting .208/.209/.350 at home this season, but it obviously doesn't seem to care about that with the -0.2 park adjustment.

mbuser   8/4/2013 11:50 AM

For standard 5x5, Quentin's unqualified projection for today is 0.14, and it drops to -0.01 when you add park. We are not taking individual player home/away splits from the current season and using them to move player's dramatically up/down. We are using expected production on a per-game basis and applying the factors as they apply to all hitters. In other words, we don't say "Quentin has been terrible at home, so his projection is lowered by X" but rather "We expect Quentin to do this on average but lower him by X based on his home park playing to this level."

For Cuddyer, PIT is a pitchers park (and Burnett has been excellent). Again, it's not "Cuddyer has been solid on the road so we'll only lower his projection by X" but rather "Cuddyer is playing at a pitchers park so we'll lower him by X based on PNC Park playing to this level."

descender  8/4/2013 12:09 PM

That certainly makes more sense, but now I really think they are too severe if they don't include any of the players own home/away split data.  

I also guess my next question is... can we get home/away splits added? :)  Maybe they would balance out the severe swings in value?

Unless... you don't see an issue with the fact the daily projections recommend starting Quentin today.

mbuser   8/4/2013 12:34 PM

Today comes down to Phil Hughes being pretty bad. We'll be adding more layers of data next season, no question, although I can't speak to specifics at this point.

descender  8/4/2013 12:39 PM

... but that statement scares me because Phil Hughes is NOT actually bad.  He's only bad in Yankee Stadium.
 

Hughes has a 3.02ERA and a 1.21 WHIP on the road this season.

This is the kind of important data I think we are missing right here.  Quentin hits .200 at home, Hughes pitches like an Ace on the road.  The park factors aren't cutting it to bring that data forward.

mbuser   8/4/2013 11:53 PM

So Hughes got destroyed today (2.2 IP, 5 ER, 3 BB, 1 K) and was much worse on the road last season than at home. We need to be careful about relatively small samples for baseball, and that's where saying "Phil Hughes is only bad at home because of his splits in 2013" can becomes dangerous. Quentin's OPS at home last season was 160 points higher than it is this season (and his road OPS was 90 points lower).

descender  8/5/2013 1:03 AM

The depth of the splits is entirely important, I agree.  Hughes getting "destroyed" by singles today is hardly relevant, he shouldn't have and the numbers leading up to today would support that.  Looking at just last year, you would think he was worse on the road.  And you would be wrong again.  Over his entire career he is still a run better on the road.  Is that 750ip enough of a sample for you yet? :)

Another question springs to mind... Do you split your park factor data for left/right-handed batters?  Fangraphs has this data i believe, but I know some sites like ESPN don't show the split data.

Back to what started this... IMO the park factors are simply changing the player values too much in the daily projections.  A -1 or +1 adjustment is way, way too much when most players are being projected between 1 and -1 in the value column.  You could half the park factors and they might still be too strong, but at least they would keep good players playing in big parks out of the 700's.

It's all about narrowing down the factors, to make single day decisions in the best light possible.  When it suggests to play guys in situations they are not currently performing in (or vice versa), it makes me ask all of these questions.

mbuser   8/5/2013 10:34 AM

We appreciate all the feedback, thanks.

Yes, we do split park factors. Check the page here: http://baseballmonster.com/MLBParkFactors.aspx

descender  8/6/2013 1:45 PM

Here are the ranges for today's factors for all hitters.

ken_admin  8/7/2013 5:43 AM

Applying a weight to the factors would be simple to do.  Before trying that, maybe it's good for me to explain the underlying process for hitters with "Apply Opposing Pitcher" applied:

1. Find the league average for pitchers (avg against, %hr allowed, bb/9, k/9, etc.)

2. Find the projected stats for the opposing pitcher (avg against, %hr allowed, bb/9, k/9, etc.)

3. Find the % difference between average stats and the opposing pitcher.

3. Adjust all of the hitters offensive stats based on this adjustment (batting average, %hr, pa/k, pa/bb, etc.)

4. The hitter stats are also adjusted to remove the factor of their home field since this is built into their projection.

5. The resulting hitter inputs are then used to generate their projection.

If we applied a weight to this process, it would mean that we're over adjusting in some area.  Instead, we'd like to analyze the process, and adjust it if needed.

Did I explain this process clearly?

Thanks

descender  8/8/2013 11:33 AM

I can certainly follow that process.  It looks reasonable enough :)

I'm just not convinced that the amount of adjustment it makes is entirely useful.  I find myself switching the factors on/off repeatedly to figure out how much each is effecting the overall value, rather than it just being "on".

bcavers  2/20/2014 12:19 PM

Can you explain how you adjust for this: "strength of players batting before/after them"?

ken_admin  2/20/2014 8:18 PM

For example, if someone has Cabrera batting after them, their projected runs would be increased. If someone has hitters who get on base before them, their projected RBIs would be increased too.

bcavers  2/21/2014 8:10 AM

Right I was wondering about the nitty-gritty on how that is calculated.