[Updated] tRA Player Cards - Now Taking Requests

by Graham on Apr 10, 2008 1:57 PM PDT Comment 149 comments

I've just set up my spreadsheet to generate player cards from 2007 with a bunch of useful information on them. However, I don't really have a sensible way of actually producing them all for easy viewing, so I thought I'd throw up a thread and ask who everyone wanted to take a look at. I'll start with someone dear to our hearts: King Felix.

(click for a clearer picture)

So, if anyone wants to see a particular player, go for it in the comments and I'll put a player card up. I'm bored of revision.

Note: Although I've done my best to make tRA as accurate as possible, all information has been collected by hand and thus typos are inevitable. If you see a mistake somewhere, please point it out.

Updates: Fixed a little bug in the regression algorithm that seems to be hitting the small sample size-types pretty hard, hasn't affected anyone interesting very much as far as I can tell. I've replaced Excel's automatic percentile ranking with a z-score-->% system which doesn't fit the data -quite- as well but puts the mean on the 50% mark, which I like. Also, BIP% column now in the table, pie chart's a bit bigger, and a sample size indicator is to the right of the first graph. I'm still taking requests, too.

12 recs | Comment 149 comments

Story-email Email Printer Print

Comments

Here's a question

Are all the percentiles based on descending totals? That is, the 99th percentile pitcher in OFB% is the pitcher who allows the most flyballs right?

What do you think about making a judgment on each stat whether more or less is better and doing percentiles that way?

by Matthew on Apr 10, 2008 2:03 PM PDT reply actions 0 recs

Here are some answers

the 99th percentile pitcher in OFB% is the pitcher who allows the most flyballs right?

Yep

What do you think about making a judgment on each stat whether more or less is better and doing percentiles that way?

I did consider this.

But then I ran into the batted ball types, and I honestly couldn't decide which way GB and OFB should go. Obviously, GB/FB ratio is important, but it's important for limiting HR more than anything else; straight outfield flies are better for pitchers in terms of runs and outs than ground balls.

After puzzling that over for a while I just decided that a straight descending percentile would be the best way of doing it, and I changed the rest back to be consistent across the board

by Graham on Apr 10, 2008 2:09 PM PDT up reply actions 0 recs

Better colors.

What about R. Soriano. Let's see what we missed out on.

...and now I'm here

by Librocrat on Apr 10, 2008 2:08 PM PDT reply actions 0 recs

What colours would you prefer?

Rafael Soriano

by Graham on Apr 10, 2008 2:20 PM PDT up reply actions 0 recs

Huh...

Not as good as I thought, though clearly still awesome. The regression looks like it will kill his stats, though, unless I'm reading it incorrectly. LD% from 29 to 45? BB% from 12 to 21?

As far as colors go, I'm just not a fan of earth toned/dulled colors, especially in the graph. But they are fine for now.

...and now I'm here

by Librocrat on Apr 10, 2008 2:25 PM PDT up reply actions 0 recs

That's % rank in the league

He actually gets better from regression due to a huuuge drop in home runs.

by Graham on Apr 10, 2008 2:27 PM PDT up reply actions 0 recs

Ah, you're right. I didn't look that far over.

Though I prefer my way of thinking, since he's not on our team.

...and now I'm here

by Librocrat on Apr 10, 2008 2:29 PM PDT up reply actions 0 recs

He is on the 15 day DL at the moment BTW.

And to think, I could have chosen to support the Yankees or Red Sox...

by EnglishMariner on Apr 10, 2008 2:25 PM PDT up reply actions 0 recs

Brent Lillibr....oh, never mind. Not a pitcher.

How about GS52?

Nice Guys Finish Third - Hopelessly lost, but makin' good time.

by pdb on Apr 10, 2008 2:14 PM PDT reply actions 0 recs

Here's GS52

Link

by Graham on Apr 10, 2008 2:24 PM PDT up reply actions 0 recs

Hahaha

His GB% is hilariously low.

I reject your reality and substitute my own!

by Phildopip on Apr 10, 2008 2:27 PM PDT up reply actions 0 recs

Yeah

I wanted to look at Sherrill and Green side by side.

Yin/Yang of set-up men.

by marc w on Apr 10, 2008 4:12 PM PDT up reply actions 0 recs

Tim Lincecum, Andrew Miller

and next to them, Brandon Morrow

by seattlebruin on Apr 10, 2008 2:16 PM PDT reply actions 0 recs

Ok

Lincecum
Miller
Morrow

by Graham on Apr 10, 2008 2:26 PM PDT up reply actions 0 recs

So the lesson from this excercise

is that Tim Lincecum is really fucking good.

by seattlebruin on Apr 10, 2008 2:39 PM PDT up reply actions 0 recs

Our friend, Jered Weaver

plus:
Cupcakes
Zito
Sean Green
Pat Neshek

by marc w on Apr 10, 2008 2:21 PM PDT reply actions 0 recs

Ok

Weaver actually looks pretty good, although he's in line for a regression.
Cupcakes
Zito
Neshek
Green

by Graham on Apr 10, 2008 2:35 PM PDT up reply actions 0 recs

I rec'd this diary.

Everybody else should do the same.

I reject your reality and substitute my own!

by Phildopip on Apr 10, 2008 2:28 PM PDT reply actions 0 recs

I am recommending this FanPost

I support Graham's quest to find more efficient ways to measure pitchers. As long as they prove that Jon Garland sucks.

by seattlebruin on Apr 10, 2008 2:34 PM PDT up reply actions 0 recs

Garland, you say?

Here he is. That one makes me smile.

by Graham on Apr 10, 2008 2:38 PM PDT up reply actions 0 recs

Holy fuck

How is he successful? He doesn't strike anyone out and is in the bottom 10% of LD% even WITH REGRESSION?? Wow...

Do HoRam. That should be amusing

by seattlebruin on Apr 10, 2008 2:42 PM PDT up reply actions 0 recs

Look at the HR allowed

He isn't really successful, either. Gave up 4.92 R/9 last year.

Here's Ho.

by Graham on Apr 10, 2008 2:46 PM PDT up reply actions 0 recs

It looks like this is saying that Ho wouldn't be that terrible

if he would strike a few people out

by seattlebruin on Apr 10, 2008 2:47 PM PDT up reply actions 0 recs

And he gave up a lot of HRs for a guy who didn't give up that many OFBs

or were his HR allowed more of the 390 ft screaming line drive types?

by seattlebruin on Apr 10, 2008 2:48 PM PDT up reply actions 0 recs

I have no idea

Information of that nature was not readily available.

by Graham on Apr 10, 2008 2:50 PM PDT up reply actions 0 recs

Graham

I checked the Hit Tracker database (and taught myself how to import spreadsheets... you would think I would have figured that out by now) and anyway, HoRam was no more likely or unlikely to give up line drive home runs, so it would appear to me that he simply got unlucky in his HR%

After running it through by Hit Tracker's criterion (Just Enough + Lucky, Just Enough, Plenty, No Doubter), *WARNING SMALL SAMPLE SIZE ALERT* and over last year, Ho Ram gave up 13 HRs - 1 JE/L, 5 JE, 6 PL, 1 ND, so by %s, he gave up more cheap homers than the average Mariner pitcher would have been expected to give up (by % of HRs allowed)

by seattlebruin on Apr 10, 2008 4:30 PM PDT up reply actions 0 recs

And stop giving up line drives

by Graham on Apr 10, 2008 2:49 PM PDT up reply actions 0 recs

Well it looks like if you regress him and he falls all the way to the 75th percentile in LD%

combined with a relatively low OFB% and a high GB%, all he would need is a few K's and all of a sudden he goes from giant albatross to eh #5 guy.

by seattlebruin on Apr 10, 2008 2:51 PM PDT up reply actions 0 recs

Where'd you get the batted ball types?

I just checked out HoRam's page on fangraphs and it lists very different percentages for LD, GB, etc.
I know people classify things differently, so it may not matter - just thought I'd ask.

by marc w on Apr 10, 2008 4:16 PM PDT up reply actions 0 recs

They're not per ball in play

They're true percentages per plate appearance. Fangraphs uses the former

by Graham on Apr 10, 2008 4:19 PM PDT up reply actions 0 recs

got it, cheers!

by marc w on Apr 11, 2008 10:32 AM PDT up reply actions 0 recs

Oh, and to answer your question, the raw data comes from Fangraphs

by Graham on Apr 11, 2008 3:47 AM PDT up reply actions 0 recs

Whoops. I called it a diary instead of a fanpost.

I reject your reality and substitute my own!

by Phildopip on Apr 10, 2008 2:42 PM PDT up reply actions 0 recs

Rec'd

I have no requests but only want to say that Graham's work fucking rules.

by Gomez on Apr 10, 2008 7:24 PM PDT up reply actions 0 recs

I see that you're bored of revision...

...but FWIW, I would maybe have bars of zero length for 50th percentile so that a 0th percentile performance shows up with a bar the same length as a 100th percentile performance. Also, that would kinda sorta help with the problem of whether a 100th percentile performance is a large OFB% or a small OFB%, since both an extremely large or small OFB% would show up as an extreme.

It also seems kind of wrong to have the HBP% bars be as long as the K% bars, but it doesn't immediately strike me what the appropriate way to scale those would be.

by ubelmann on Apr 10, 2008 2:40 PM PDT reply actions 0 recs

Hmm

Why would it be better to just denote extreme values rather than saying which extreme it is?

The scaling problem I tried to resolve with that pie chart and the actual % per plate appearance, but you're right, it's still awkward.

by Graham on Apr 10, 2008 2:43 PM PDT up reply actions 0 recs

I think what I said wasn't very clear...

You should still say which extreme it is, but instead of having no bar for a 0% performance, have a "negative bar" from midway through the graph to the bottom, and for a 100% performance, have a "positive bar" from midway to the top.

I'm thinking that, at a glance, that would make it easier to pick out what makes a pitcher unique. If a pitcher was median at everything except that he had a big GB% and small OFB%, there would be a bar extending upwards for his GB% and a bar extending downwards for his OFB%.

by ubelmann on Apr 10, 2008 2:51 PM PDT up reply actions 0 recs

Ah yes, that would be a good idea

That's definitely something to consider for the new version, but I'll leave it for now as it would require quite a lot of spreadsheet poking.

Thanks for that.

by Graham on Apr 10, 2008 2:53 PM PDT up reply actions 0 recs

So nevermind about my leaving it

Do you guys like this version better?

(That's Jake Peavy. He's really good)

by Graham on Apr 10, 2008 3:56 PM PDT up reply actions 0 recs

Apart from forgetting to change AL SP to NL SP.

Stupid copy-paste.

by Graham on Apr 10, 2008 3:59 PM PDT up reply actions 0 recs

And forgetting to hide those numbers at the bottom.

It's getting late...

by Graham on Apr 10, 2008 4:02 PM PDT up reply actions 0 recs

I like it

That's exactly what I was talking about.

by ubelmann on Apr 10, 2008 5:21 PM PDT up reply actions 0 recs

Because I'm curious

and because it will complete the trifecta of upcoming Angels pitchers, Joe Saunders.

Also since no one asked for them yet, Bedard and Silva.

*Visiting Angels fan* Never give up, never surrender!

by TheOptimist on Apr 10, 2008 2:51 PM PDT reply actions 0 recs

I'm excited for Silva

I second this one

by seattlebruin on Apr 10, 2008 2:52 PM PDT up reply actions 0 recs

Silva

that line drive rate is RIDICULOUS

by CKel on Apr 10, 2008 4:39 PM PDT up reply actions 0 recs

It's not really that bad

He gives up a lot of line drives because he allows a lot of balls in play. His LD/BIP is only in the 52nd percentile.

by Graham on Apr 11, 2008 2:25 AM PDT up reply actions 0 recs

A random question

what kind of sample size would you consider reasonably significant for this type of work? x > 250 batters faced? More?

by seattlebruin on Apr 10, 2008 2:52 PM PDT reply actions 0 recs

The regression algorithm takes sample size into account automatically

It might not do it very well, though. When I came up with the correlation coefficients and r^2 values I was playing with ~350ish BF, but I went and dropped that number down a bit for relievers and raised it for starters because all of the relievers were just getting pushed back to average really hard.

by Graham on Apr 10, 2008 2:56 PM PDT up reply actions 0 recs

The Book Blog is picking up on this

by Jeff on Apr 10, 2008 3:22 PM PDT reply actions 0 recs

Oh holy crap

everybody act nice. And tidy up - we'll probably have guests coming over soon. I hate being on my best behavior.

Nice Guys Finish Third - Hopelessly lost, but makin' good time.

by pdb on Apr 10, 2008 3:32 PM PDT up reply actions 0 recs

Now we just need to find a metaphorical closet to hide Coach and Corco in

they can come out when our guests leave

by seattlebruin on Apr 10, 2008 3:33 PM PDT up reply actions 0 recs

I either want this to be a gay joke or an R Kelly joke

but, alas, it is neither.

Free Barry Bonds

by JI on Apr 10, 2008 8:05 PM PDT up reply actions 0 recs

Thats when I opened the closet, closet, closet

I fucking hate you Mariners

by kentroyals5 on Apr 10, 2008 9:40 PM PDT up reply actions 0 recs

a man... a man... a man...

Free Barry Bonds

by JI on Apr 10, 2008 9:42 PM PDT up reply actions 0 recs

Oh my God, a rubber.....

I fucking hate you Mariners

by kentroyals5 on Apr 10, 2008 10:08 PM PDT up reply actions 0 recs

I TRIED MY BEST TO QUICKLY PUT IT ON VIIIIIIIIBRAAAATEEEE

Free Barry Bonds

by JI on Apr 10, 2008 10:39 PM PDT up reply actions 0 recs

AND THEN I PULL OUT MY GUN

by Jeff on Apr 10, 2008 10:10 PM PDT up reply actions 0 recs

"I pull out my Baretta"

I fucking hate you Mariners

by kentroyals5 on Apr 10, 2008 10:27 PM PDT up reply actions 0 recs

A long long time ago

Me and the people at work would do Trapped In The Closet freestyles just to kill time. The point being that the lyrics were so tuneless and devoid of art you could just riff on what you were doing at that moment, fade out, and then BAM you had written a hit R Kelly song.

Free Barry Bonds

by JI on Apr 10, 2008 10:37 PM PDT up reply actions 0 recs

Thats very true

And its only really the first time you hear that song that you actually laugh..the other times you just think to hard and analyze. Its the "WTF factor" that makes it so great.

I fucking hate you Mariners

by kentroyals5 on Apr 10, 2008 10:59 PM PDT up reply actions 0 recs

I like how this subthread went from tangotiger to R Kelly

by Jeff on Apr 11, 2008 9:03 AM PDT up reply actions 0 recs

I like how you just looped it

Cunning.

by Graham on Apr 11, 2008 9:03 AM PDT up reply actions 0 recs

I got descibed by Dave Cameron as 'kind of like a European mgl'?

Words cannot express the pride I feel right now.

by Graham on Apr 11, 2008 3:44 AM PDT up reply actions 0 recs

Nitpick...

any chance that you can get your speadsheet to reflect the percentiles as being the more positive of the outcomes?

EX. #1 in K% = 99%
#1 (lowest) in BB% = 99%

Right now Silva is in a very low percential in BB rate, which seems counter intuitive to me.

by PLU Tim on Apr 10, 2008 5:38 PM PDT reply actions 0 recs

I'm still thinking about whether I want to do this

I almost prefer having absolutes on there just to be more consistent, but there's an argument to be made for highlighting the 'good' stats over the 'bad'.

by Graham on Apr 10, 2008 5:46 PM PDT up reply actions 0 recs

Well....

the way I see it is that when someone looks at Bedard's card they see he's in the 99 %ile for K% and say "Wow. He's good!" Then they look at Silva's BB% and see he's in the 6-7 %ile and think "WTF?"

by PLU Tim on Apr 10, 2008 5:54 PM PDT up reply actions 0 recs

So would everyone prefer it if I did something like...

K positive
BB negative
HBP negative
LD/BIP ???
GB/BIP ???
FB/BIP ???
IF/FB positive
HR/FB negative

I'm not at all sure what to do with the middle ones though...

by Graham on Apr 11, 2008 2:19 AM PDT up reply actions 0 recs

Soooo

Do y'all prefer this version (Sonnanstine)? Is it clear which categories are inverted and which aren't? How's the labelling?

by Graham on Apr 11, 2008 6:25 AM PDT up reply actions 0 recs

Oh. And I've flipped line drives now too.

by Graham on Apr 11, 2008 9:32 AM PDT up reply actions 0 recs

So at a quick Glance.....

I see that Sonnanstine is a pitcher who:
Get his K's
Doesn't Walk many
A flyball pitcher who's profile suggests that he should give up fewer linedrives than he does and has been extremely fortunate to have given up so few homers.

Does that fit Sonnanstine?

by PLU Tim on Apr 11, 2008 10:41 AM PDT up reply actions 0 recs

Am confused about the HR thing

He's ~avg for HR/FB

by Graham on Apr 11, 2008 2:25 PM PDT up reply actions 0 recs

That's perfect

I like this version. The %/batted ball avoids the weirdness discussed above re: Silva's LD rate. You'd still get a sense that he should allow more total LDs by looking at the K/BB%.

This is awesome.

by marc w on Apr 11, 2008 10:37 AM PDT up reply actions 0 recs

I like it...

It seems fairly intuitive to me to have the X/BIP stats in that direction. Somehow it was weird to have BB% in ascending order, but it's not weird for LD/FB. I'm not sure what the deal is with that.

It's starting to get a bit busy, but would it be worth it to have a BIP% column in the per batter faced cluster? I know it's implied from the first three rates, but it might be good to have it in there explicitly? I'm not sure, just throwing that out there.

by ubelmann on Apr 11, 2008 11:13 AM PDT up reply actions 0 recs

I am on the fence RE: line drive ascending or descending

Should probably give it some more thought at some point, but it's quite easy to switch between the two - the hard part was trying to figure out a way of indicating which measures were ascending and descending without too much clutter.

As for the BIP% column... I don't think it's really necessary, mainly because the pie chart gives you a pretty good idea of the rough BIP% anyway. What would the merits of having a an numeric value there? I'm not really seeing it..

by Graham on Apr 11, 2008 2:31 PM PDT up reply actions 0 recs

re: BIP% column

I would include it for the same reason you include the FB/BIP column, essentially. The FB/BIP column isn't strictly necessary, but it's nice to know at a glance whether or not a pitcher gives up a lot of fly balls. Similarly, a BIP% column isn't strictly necessary, but it might be nice to know at a glance whether or not a pitcher could be considered a contact pitcher.

At this point, I can see the advantages of not adding anything more to the cards as there is already a lot of information there. (I wouldn't mind seeing the actual and park-adjusted stats dropped entirely in favor of just presenting the regressed statistics...but that probably wouldn't be a popular viewpoint.)

by ubelmann on Apr 11, 2008 2:46 PM PDT up reply actions 0 recs

How's this?

Josh Beckett.

by Graham on Apr 12, 2008 11:15 AM PDT up reply actions 0 recs

Brandon Webb
Micah Owings
Danny Haren
Chad Qualls
Brandon Lyon
Jose Valverde

Felix Hernandez may be The King, but Justin Upton is a GOD.

by Goose on Apr 10, 2008 7:29 PM PDT reply actions 0 recs

I sense a commonality

by Jeff on Apr 10, 2008 7:32 PM PDT up reply actions 0 recs

I'm mostly interested in the Chad Qualls/Brandon Lyon comparisions

because I've been leading the "Chad Qualls for closer" bandwagon and Brandon Lyon is terrible.

Felix Hernandez may be The King, but Justin Upton is a GOD.

by Goose on Apr 10, 2008 8:07 PM PDT up reply actions 0 recs

Lyon was better last year but Qualls regresses much better

by Graham on Apr 11, 2008 2:09 AM PDT up reply actions 0 recs

You'll love RJ's card.

Yesterday's Pants
A blog-thingy about the Mariners and stuff.

by BrettJMiller on Apr 10, 2008 9:20 PM PDT up reply actions 0 recs

Yep

Webb
Owings
Haren
Qualls
Lyon
Valverde

by Graham on Apr 11, 2008 2:09 AM PDT up reply actions 0 recs

Sweet, thanks.

Ah Michah :(

QUALLS FOR CLOSER!

Felix Hernandez may be The King, but Justin Upton is a GOD.

by Goose on Apr 11, 2008 4:48 AM PDT up reply actions 0 recs

err Micah

Felix Hernandez may be The King, but Justin Upton is a GOD.

by Goose on Apr 11, 2008 4:48 AM PDT up reply actions 0 recs

Probably worth pointing out

That this doesn't have an ageing curve built in. Micah Owings will get better.

by Graham on Apr 11, 2008 5:04 AM PDT up reply actions 0 recs

Adam Wainwright plz

I can't think of anyone else on the Cardinals even worth looking up.

...and that's even with Player A throwing a full shutout inning last year.

Free Barry Bonds

by JI on Apr 10, 2008 8:07 PM PDT reply actions 0 recs

Anthony Reyes could also be interesting.

Free Barry Bonds

by JI on Apr 10, 2008 8:08 PM PDT up reply actions 0 recs

Ok

Wainwright
Reyes

by Graham on Apr 11, 2008 2:16 AM PDT up reply actions 0 recs

neato

Free Barry Bonds

by JI on Apr 11, 2008 8:29 AM PDT up reply actions 0 recs

I'll play too

Verlander
R. Hill
Billingsley
B. Wilson
Broxton
Neshek
H. Bell

Time to get excited about baseball again!

by wwbaker3 on Apr 11, 2008 2:40 AM PDT reply actions 0 recs

That looks eeriely like someone's fantasy team ;)

I'll get round to this and any new requests late this afternoon. I need to study bioinformatics :(

by Graham on Apr 11, 2008 3:31 AM PDT up reply actions 0 recs

Ok

Neshek
Hill
Verlander

by Graham on Apr 11, 2008 6:29 AM PDT up reply actions 0 recs

Part 2

Billingsley
Broxton
Bell

by Graham on Apr 11, 2008 6:29 AM PDT up reply actions 0 recs

Pat Neshek is the man.

I wonder how long he can stay effective v. LHB...

Free Barry Bonds

by JI on Apr 11, 2008 9:29 AM PDT up reply actions 0 recs

Thank you, Graham!

Time to get excited about baseball again!

by wwbaker3 on Apr 11, 2008 11:03 AM PDT up reply actions 0 recs

Looks really nifty.

Ran accross this at Tango's blog. As far as hand-collating the data, assuming I correctly identified what you're using, I went ahead and parsed it out of the 2004-2007 Retrosheet events logs. Hope this helps.

I has a blog.

by cwyers on Apr 11, 2008 10:44 AM PDT reply actions 0 recs

I was literally just trying to post a link to that spreadsheet

nice timing.

Graham, this looks great and should really help you gauge the predictive power of tRA. Should be better than FIP/xFIP, but it'll be cool to know how MUCH better.

by marc w on Apr 11, 2008 10:47 AM PDT up reply actions 0 recs

he had to long collect it the first time

because he foolishly didn't ask me. I already went through and grabbed the retrosheet data for him and data from MLB for going forward. The hope is to have daily tRA updates housed somewhere relatively soon.

by Matthew on Apr 11, 2008 1:09 PM PDT up reply actions 0 recs

I think you talked about this before

But this is not necessarily meant for projections yeah? Although regression implies projections, you cannot or are currently not using these player cards in order to estimate a tRA for 2008, for example, correct?

...and now I'm here

by Librocrat on Apr 11, 2008 1:50 PM PDT reply actions 0 recs

Well,

RA in year 0 correlates better with ERA in year 1 than ERA in year 0.
Similarly, FIP in year 0 correlates better with ERA in year 1 than RA in year 0.
xFIP in year 0 is a bit better.
tRA should be better still.

If you have a better handle on a pitcher's skills, you should, over time, see that pitcher's results fall in line. In that sense, it's predictive.

Predicting tRA isn't all that interesting; skills are skills. Felix isn't suddenly going to become a flyballer who doesn't strike anyone out. His tRA profile should vary, but not by much.

by marc w on Apr 11, 2008 2:04 PM PDT up reply actions 0 recs

Marc's got it spot on

And I fully plan on using this to predict RA. Otherwise there wouldn't be any point.

by Graham on Apr 11, 2008 2:08 PM PDT up reply actions 0 recs

Predicting tRA

Predicting tRA isn't all that interesting; skills are skills.

What are we left to predict then? If results are the fuction of skills, teammates, opponents, park, and luck in some combination, and we already have a good handle on a player's skills, his teammates' skills, his opponents' skills, and the park factors, then all we have left to predict is luck--and no one is going to be good at that.

I would assert that what we want to do is measure everyone's skills accurately and then figure out for players of a given skill level, what the variance around their expected results is.

(Also, there should be some expected change in skills as players age, so I don't think that forecasting tRA is a completely trivial task. Or at least it remains to be shown that it's a trivial task.)

by ubelmann on Apr 11, 2008 2:35 PM PDT up reply actions 0 recs

So the key thing here is that I'm not writing a projection system

So I'm ignoring player development curves, etc, etc, etc. However, one of the goals of the project is to eventually get around to the player skill projections, which is why I've done stuff like include age and handedness, so I can write a PECOTA-style clustering algorithm and do my best to match players (this will be unfeasible for a while - AFAIK we don't have good historic BIP data, which makes matching pitcher types difficult).

tRA, as it stands, is an attempt to evaluate pitcher performance by removing park effects and team defensive performance. It will hopefully (well, I'll use park and defense adjusted tRA*) correlate well with y+1 RA.

I think player projection and a performance/skill indicator are actually very different things, despite the relation.

by Graham on Apr 11, 2008 2:43 PM PDT up reply actions 0 recs

I mostly agree

I definitely think that it would be really great to see the lists of comparable pitchers based on the categories that you are tracking.

I guess I don't understand why we shouldn't at least park and defense adjust the y+1 RA. Predicting a pitcher's y+1 park factor and y+1 defense are distinct issues from predicting a pitcher's contribution to winning, and by not park and defense adjusting RA, things seem to get a bit murky.

by ubelmann on Apr 11, 2008 2:54 PM PDT up reply actions 0 recs

Part adjusting tRA does the same thing as adjusting RA, but makes it easier on users reading the output

It's essentially just a user-friendliness thing. If people look at a player card and see an xRA of 5.00, I'm not sure I can count on them to take into account park and expected defence. I'd much rather do the processing on the inside so I can deliver better predictions.

by Graham on Apr 11, 2008 3:02 PM PDT up reply actions 0 recs

I think I see what you are saying now

Instead of taking a park and defense neutral stat in year y and a park and defense neutral stat in year y+1 and determining the correlation, you take the park and defense neutral stat in year y, then in effect do a park and defense un-adjustment based on your best guess of the year y+1 park and defense factors, and then do the correlation to the y+1 RA.

I kind of don't like that approach because I really like to drive home the point that RA is a measure of team run prevention, not pitcher effectiveness, but I suppose that you need to make a compromise somewhere, and Rome wasn't built in a day, etc., etc.

by ubelmann on Apr 11, 2008 3:22 PM PDT up reply actions 0 recs

I'm with you

But I think if you're going to make predictions, you have to have all the information on hand. It's not really anything to worry about, though - I'll still have tRA in its original form when I start predicting stuff with xRA* (or whatever I'll call it), and tRA will still be the focus.

by Graham on Apr 12, 2008 11:25 AM PDT up reply actions 0 recs

It might be cool if you threw up cards

in Matthew's series previews he has been doing.

by Edgar for Pres on Apr 11, 2008 8:35 PM PDT reply actions 1 recs

Lenny DiNardo?

The A's colors are green and gold.

by mikeA on Apr 12, 2008 4:22 PM PDT reply actions 0 recs

Since I know y'all are big Lenny fans....

Today: 5 innings, 3 K, 0 BB, 12 GO, 0 FO.

The A's colors are green and gold.

by mikeA on Apr 12, 2008 5:48 PM PDT up reply actions 0 recs

I am!

I said he'd be better this year

by Matthew on Apr 12, 2008 8:09 PM PDT up reply actions 0 recs

hmmm

I read that when you posted it, but I was in a nocompetewithoutHarden mindset (actually I still am), and Lenny was the #7 starter.

So if I understand correctly, what you're saying there re: Lenny is:
1) He should have a better than average LOB% due to all the GBs, whereas it was somewhat below average last year; and
2) His swinging strike % and ball % were significantly out of line with his K and BB %s? That is pretty difficult to check on any of the stat sites... Your and Graham's work over here mostly just makes me frustrated with bref and even tht, since baselining everything on PAs makes a million times more sense than baselining it on IP. Would it kill them to have K% and BB%?

Answers/thoughts much appreciated as I'm considering a "Lenny doesn't suck, he's kinda decentish" fAnPoSt in the southerly blogosphere.

The A's colors are green and gold.

by mikeA on Apr 12, 2008 9:50 PM PDT up reply actions 0 recs

Close, but not quite

Based on his 2007 pitch results, I expected DiNardo's:
-K rate to go up
-BB rate to go down
-LOB% to go up (not above average)
-HR/FB rate to go down

Granted, that's if his skill level doesn't change and given his limited sample size, it could but he's certainly decent-ish.

by Matthew on Apr 12, 2008 10:05 PM PDT up reply actions 0 recs

thanks

The A's colors are green and gold.

by mikeA on Apr 13, 2008 12:41 AM PDT up reply actions 0 recs

Yep

DiNardo.

Would AN like to take a look at this stuff, by the way?

by Graham on Apr 12, 2008 11:35 PM PDT up reply actions 0 recs

Thanks

Yeah, some of us would, but we're not a real analytically-inclined community on the whole. This is great stuff though (along with all the preceding diaries), and I think even the most analytically-inclined among us don't much know what to make of pitching. Sal Baxamusa of THT (and of course AN) has a stat primer every few weeks on the front page, and we could certainly use one dealing with pitching. I'll email him, and if and when he does a post on this sort of stuff, if you posted/reposted a diary I think it would be well received, but I think as things stand now it would be mostly ignored, as no one knows who you are.

But anyway, this is great stuff, and it's sort of shocking that it is so hard to find reliable pitching metrics (I don't know how much I trust your tRA, but I trust it more than the other crap out there...)

The A's colors are green and gold.

by mikeA on Apr 13, 2008 1:21 AM PDT up reply actions 0 recs

In terms of trust

Without inputs for total runs scored or total innings pitched, tRA is within 7 runs and 3 IP for the entire season, for all of MLB. I'm inclined to trust tRA, tRA+, tROA, but less so for the regression (tRA*), since I'm still trying to find the best thresholds and exponents.

And the people not knowing who I am thing is strange. I've gotten used to it here, and I find the anonymity at AN a little intimidating.

by Graham on Apr 13, 2008 1:42 AM PDT up reply actions 0 recs

RE: First sentence

Wow.

by Fett42 on Apr 14, 2008 6:42 PM PDT up reply actions 0 recs

It's also apparently a lot more accurate than my brain

I meant 5IP, not 3.

by Graham on Apr 15, 2008 2:48 AM PDT up reply actions 0 recs

Wower

by Fett42 on Apr 15, 2008 9:00 AM PDT up reply actions 0 recs

That's a bigger error! It's like 0.1% rather than 0.06%!

(Yeah I’m pretty pleased with that :D)

by Graham on Apr 15, 2008 9:06 AM PDT up reply actions 0 recs

You could go great places with such error expansion abilities

by Fett42 on Apr 15, 2008 9:41 AM PDT up reply actions 0 recs

How about....

Smoltz
Hudson
Glavine

Two Teams, Two Coasts - One Obsession

by RainDelay on Apr 13, 2008 6:52 PM PDT reply actions 0 recs

I was hoping that tRA would look more favourably on Glavine than current stats.

Nope. But it's fond of the other two.

Smoltz
Hudson
Glavine

by Graham on Apr 14, 2008 9:56 AM PDT up reply actions 0 recs

Yeah...

..doesn’t like Glavine at all..heh

Two Teams, Two Coasts - One Obsession

by RainDelay on Apr 16, 2008 8:42 AM PDT up reply actions 0 recs

Jonathan Sanchez plz

by Fett42 on Apr 14, 2008 6:40 PM PDT reply actions 0 recs

Ok

Here

by Graham on Apr 15, 2008 1:15 AM PDT up reply actions 0 recs

As long as you've got most of the Angel staff anyway

How ‘bout Lackey and Escobar?

~Till the Halo burns out...

by Zu Long on Apr 15, 2008 7:18 AM PDT reply actions 0 recs

As long as you've got the other Angel Starters

How about Lackey and Escobar, just so we know what we’re missing if nothing else.

~Till the Halo burns out...

by Zu Long on Apr 15, 2008 7:25 AM PDT reply actions 0 recs

Sure thing

Escobar
Lackey

by Graham on Apr 15, 2008 7:29 AM PDT up reply actions 0 recs

Got a few.

Scott Kazmir
James Shields
Zack Grienke
Joakim Soria
Randy Johnson
Yovani Gallardo
Cole Hamels

Yesterday's Pants
A blog-thingy about the Mariners and stuff.

by BrettJMiller on Apr 15, 2008 10:21 AM PDT reply actions 0 recs

Sorry for the delay

I’m a lazy bastard sometimes. In no particular order:

Hamels
Soria
Grienke
Gallardo

by Graham on Apr 15, 2008 12:05 PM PDT up reply actions 0 recs

cont...

Kazmir
Shields
Johnson

by Graham on Apr 15, 2008 12:06 PM PDT up reply actions 0 recs

my turn my turn

Miguel Batista
Moyer
Buehrle

also

Putz, Jenks, and F-rod?

by Jamelah on Apr 16, 2008 4:18 AM PDT reply actions 0 recs

Ok

Putz
Jenks
F-Rod

by Graham on Apr 17, 2008 1:31 PM PDT up reply actions 0 recs

cont

Beluga Tits
Moyer
Buerhle

by Graham on Apr 17, 2008 1:32 PM PDT up reply actions 0 recs

How about Brian Bannister while you're at it?

by seattlebruin on Apr 16, 2008 8:24 AM PDT reply actions 0 recs

K

Bannister

by Graham on Apr 17, 2008 1:34 PM PDT up reply actions 0 recs

So Brian Bannister is decidedly average

by seattlebruin on Apr 17, 2008 2:23 PM PDT up reply actions 0 recs

Yep

by Graham on Apr 17, 2008 2:38 PM PDT up reply actions 0 recs

Is there data to do this for earlier years

Like Pedro, Randy, Clemens, Maddux around their prime? Especially Pedro.

by Fett42 on Apr 16, 2008 2:01 PM PDT reply actions 0 recs

2003 forward I'm afraid

by Graham on Apr 17, 2008 1:34 PM PDT up reply actions 0 recs

Since this seems to have slowede down a bit

One more I was interested in: Ervin Santana

If possible though, I’d like two versions- one normal one, and one that only uses home games (if that makes the sample size too small, throw in the 2006 numbers as well). I just want to see how much difference there is between the two sets of number, if any.

~Till the Halo burns out...

by Zu Long on Apr 21, 2008 10:15 AM PDT reply actions 0 recs