2011 Projection-Based Rosters: Methods, Formulas, & Discussion ...

[TripleThreat1973]

Here is some information on some rosters I am workin on. It should give an overview of how they are put together, as well as, keeping my thoughts straight, and ensuring I'm on the right track.

This will be for MLB10, although if I buy MLB11, it could easily be done there too.

I'll be making about 10 consecutive posts. Give me a minute.

2011 Projection-Based Ratings System
[1] Projections
[2] Batting Splits
[3] Average Starter Ratings – OSFM v2
[4] Ratings Formulas
[5] Non-Formula Ratings
[6] Minor League Players
[7] Entering Ratings
[8] Lineups & Rotations
[9] Remaining, Absent, and Retired Players
[10] Testing
[11] Global Changes or Individual Tweaks

[TripleThreat1973]

Re: 2011 Projection-Based Rosters: Methods, Formulas, & Discussion ...

[1] Why use 2011 Projections instead of 2010 Stats or 3-year Weighted Averages?

- My preference – I would prefer to play the “2011” season with ratings that projected players are calculated to have, versus “replaying the 2010 season”.

- Projections include 3-yar weighted averages already, plus MLE’s for minor leaguers, and aging curves, and league/park effects, career paths, etc.

Why ZiPS projections?

- I usually average 3 projection systems … CHONE, ZiPS, and Guru. The maker of CHONE projections has a job with a ML club, and can no longer produce projections for release to the public. CHONE recommends ZiPS.

Guru and ZiPS are already similar. I purchased THT (The Hardball Times), but they lack batting hand, games pitches, and games started, which makes it more difficult to calculate batting splits, pitcher stamina, and durability.

- ZiPS provides projections for about 65 players per team. Basically anyone that has a chance at playing at the ML level during the 2011 season is included.

[TripleThreat1973]

Re: 2011 Projection-Based Rosters: Methods, Formulas, & Discussion ...

2] Batting Splits

Batting Splits were obtained from Scoresheet. The list suggests how many points should be subtracted from BA, OBP, and SLG based on pitcher handedness.

www.scoresheet.com/FOR_WWW/LR_diffs.txt

I then regressed the splits toward the league average, using 70% players splits and 30% league average splits as outlined by Mitchell Lichtman (co-author of ‘The Book’) and Dave Cameron (Fangraphs).

[TripleThreat1973]

Re: 2011 Projection-Based Rosters: Methods, Formulas, & Discussion ...

[3] Average Starter Ratings


My MLB10 plays very well with my sliders and the OSFMv2 rosters. So, in order to come up with a system that is congruent with the ratings in the roster, I needed to find the average rating for each attribute for the average ML starter and work from there.

So, what I did was load the rosters and start a franchise with a fantasy draft. I recorded the top 30 position players at each position and the top 150 starting pitchers, top 30 relievers, and top 30 closer (Basically a starting team if the talent were distributed evenly throughout the league). The averages are displayed below.

POS.....CR.....CL.....PR.....PL.....BT.....DB.....VIS.....DIS.....CLT.....DUR.....SPD.....AST.....AAC.....RCT.....FLD.....BLK.....BAB.....BAG

C..........60.....63....55.....58.....22.....13.....57......65.......64.......71........32.......80......65........61.......66......76.......16........8

1B........73.....65....75.....67.....17......8......57......84.......76.......88........35.......64......64........62.......66........3........19.......8

2B........66.....71....47.....50.....42.....30.....69.....64........67.......83.......66........65......63........67.......71........0........50.......29

3B........68.....72....62.....63.....24.....14.....61.....64........72.......88.......50........75......68........67.......70........7........33.......15

SS........65.....73.....42.....44.....48.....39.....76.....56........70.......82.......71........71......65........71.......72........0........51.......33

LF........77......79....73.....74.....26......15.....67.....79.......81.......93........61........72......68.......66........69.......2.........44.......27

CF........67......64....51....54.....40.....34......56.....65.......65........85........75.......68......61........74......76.........0........65.......45

RF.......72......66.....64.....63.....23.....14....58......70.......71........83........60.......75.......65.......66.........64......0........37.......22

ALL.....68.......68.....58.....58.....30.....21......62.....67......70........83........55.......39.......23.......

I formed my ratings formulas to work from this information. BTW, thanks Knight & Co for the great work.

[TripleThreat1973]

Re: 2011 Projection-Based Rosters: Methods, Formulas, & Discussion ...

[4] Ratings Formulas

I had all my formulas made, and then I stumbled upon Bahnzo’s spreadsheets. In these spreadsheets you can view the formulas he uses. Since this was released to the public, I figured I could use the formulas. I noticed that the results that his formulas come up with were about 10 points less in contact, power, vision, and discipline, when compared to knight’s ratings. So, I made my formulas a “little high” and averaged them with Bahnzo’s. The result is a system that can be applied to all players in a uniform and consistent manner … and be congruent with Knight’s ratings and my slider. *grin*

Here are the formulas …

Durability
-----------
(GP/162)*100

Contact
----------
Bahnzo
=IF((BA/1000)<0.28,(((BA/1000)*254.315428326364)-(((0.28-(BA /1000))*254.315428326364)/0.6)),(( BA /1000)*254.315428326364)+(((( BA /1000)-0.28)*254.315428326364)*1))

Me
=68+(((Y80-0.269)*1000)/2)

It’s basically “average starter rating + the difference in players BA and league average BA, with the difference divided by 2 in order to keep it to scale.

As I mentioned before, BA/4 works rather well too.

Power
---------
Bahnzo
=IF(HR550<18,(( HR550*3.0256)+((18- HR550)*0.65)),( HR550*3.0256)-((( HR550-18)*3.0256)/1.9))
HR550 = Home runs per 550 AB’s. It can be calculated with (550/AB)*HR

Me
=(58+(((HR550-15)*2)))*0.95

58=average starter power rating; 18-average HR550

Vision
---------
Bahnzo
=(((((650/PA)*SO)*0.590223)+((108-((650/ PA)* SO)))*1.05)-(10-(( PA / SO)/1.5)))

Me
=62+((100-K550)/2)

K550 = strikeouts per 550 At Bats

Discipline
----------------
Bahnzo
=IF(((650/PA)*BB)>60,((((650/ PA)* BB)*1.205393)-((((650/ PA)* BB)-60)*0.52)),((((650/ PA)* BB)*1.205393)+((60-((650/ PA)* BB))*0.15)))

Me
=67+((BB550 - 55)*0.9)
67=average DIS rating; 55=average BB550

Speed
---------------
Modified James Speed Score

5 formulas, averaged to get a “speed score” and multiplied by a modifier to get a speed rating.

[TripleThreat1973]

Re: 2011 Projection-Based Rosters: Methods, Formulas, &amp; Discussion ...

Clutch
-------------------
Here’s where I diverge from common practice. Traditionally clutch uses BA on RISP or a clutch rating from a sabermetric website. However, since we’re dealing with projections, and clutch factor does not correlate one year to the next (i.e., ARod has been very clutch and very not clutch, as has “Mr. Clutch” David Ortiz. They can be among the league leaders one year and among the league’s lowest the next. In short, there is no clutch skill. There is overall player talent, and BA w/ RISP is victim to a whole host of small sample issues. It is my preference that Ramon Hernandez not have a 95 clutch rating, nor would McGehee be a 99.

Rather than have a clutch rating for everyone, such as making everyone “league average” or something like 65, I used a formulas that would measure their “true talent” (amount above and below league average on the key batting ratings), and scaled it to be essentially between 55 and 75.

The stat I used was “Runs Created” which factors in things like BB and SB that improve WPA (Win Probability Added). Sometimes a walk is “very clutch” and that won’t be encapsulated in Batting Average with Runners in Scoring position, or Close and Late batting average.

=70+((RC-68)/5)

70=average clutch ratings for OSFM; 68=average runs created … dividing by 5 causes the difference to fall with the 55-75 range. What this means is that players with higher talent are more likely to be clutch than those with lesser talent.

This is what research tells us. When runners are on 1st and 2nd with 2 outs, you want Pujols at the plate, even though Rasmus may have a higher BA with RISP. It’s because he has more talent, and Rasmus’s BARISP could just be the result of a small sample. Just because Brandon Inge is 4-for10 against Sabathia does not mean I want him at the plate instead of Miguel Cabrera. Furthermore, this is how it’s done in projections, so it remains uniform.

Likewise, pitching clutch is the difference between each pitchers 4 main ratings and the league average added to the league average clutch ratings. Same deal, the better pitchers will tend to have higher clutch ratings, also in the 55-75 range.

[TripleThreat1973]

Re: 2011 Projection-Based Rosters: Methods, Formulas, &amp; Discussion ...

Pitcher Durability
----------------------------
Starting Pitchers
(GS/35)*100

Relievers
(G/81)*100

Mixed Pitchers
((GS/35)+((G-GS)/81))*100

Pitcher Stamina
------------------
Bahnzo
=IF(GS>0,IF((((IP/G)*13.5)+10)<50,(((( IP / G)*13.5)+30)),IF(((IP / G)*13.5)>99,99,( IP / G)*13.5)),((( IP / G)*13.5)+10))

Me
=IF(POS="SP",(85+((K2*10)-(58))),IF(POS="RP",(25*K2)))

Again using league averages as the basis. For starters the difference between their IP/GS is added to the league average stamina rating of 85. The lower the IP/GS, the more is subtracted (adding a negative number) from 85.

For relievers, the same thing is done to 25.

Hits Per 9 IP
--------------------
Bahnzo
=((HR/9*8.160999)+((8.5- HR/9)*15))

Me
=64+((9.3-R2)*10)

64 = average rating; 9.3 = average H/9 stat for top 150 starting pitchers

Home Runs Per 9 IP
-----------------------------
Bahnzo
=IF(((HR/9*71.11)+((0.9- HR/9)*125))>99,99,(( HR/9*71.11)+((0.9- HR/9)*125)))

Me
=62+((1-HR/9)*100)

Strikeouts Per 9 IP
-----------------------------
Bahnzo
=IF(K9<7.6,(( K9*10)+(7.6- K9)),IF(((K9*10)+(7.6- K9)+((7.6- K9)*0.5))>99,99,(( K9*10)+(7.6- K9)+((7.6- K9)*0.5))))

Me
=IF(K9>9.9,((68+(K9-6.6)*10)*0.95),((68+((K9-6.6)*10)*0.9)))

Walks Per 9 IP
-------------------------
Bahnzo
=IF(L2<3.4,IF(((L2*18.21593)+((3.4- L2)*35))>99,99,(( L2*18.21593)+((3.4- L2)*35))),( L2*18.21593)+((3.4- L2)*30))

Me
=(65+((3.1-BB9)*10))

[TripleThreat1973]

Re: 2011 Projection-Based Rosters: Methods, Formulas, &amp; Discussion ...

[5] Non Formula Ratings

Sac Bunts, bunt singles, specific fielding stats are not projected stats for various reasons.

I left bunting and drag bunting as is. If a player saw a huge jump in contact, but not power, I increased his bunt rating some (not much).

Essentially all minor league players can field at a replacement level. The difference between a major leaguer and a minor leaguer is basically “all hitting”. This means minor league players will have fielding ratings in the 50s instead of the 40s.
I have fielding value projections, and at the end I’ll go through and note the fielders that are exceptionally good and bad, and tweak their ratings to be congruent with the average starter fielding ratings from OSFM.

[TripleThreat1973]

Re: 2011 Projection-Based Rosters: Methods, Formulas, &amp; Discussion ...

[6] Minor League Players

Using projections that include MLE (Major League Equivalents) is a BIG bonus for those that enjoy minor league players and roster flexibility.

Minor league players are essentially replacement level players. They can often be just as good as the “4th outfielder” or the “utility guy”. Some players in the minors are there, even though they are better than current starters. They are kept in the minors due to age, emotional reasons, and most likely “contract reasons” … specifically to avoid being awarded “Super 2 Arbitration” privileges. In other words, keeping them in the minors, allows teams to pay them much less over their team controlled years.

So what you’ll see in these rosters is the gap between poor major leaguers and good minor leaguers is razor thin, as it should be. They are both replacement level. Some minor leaguers like Mike Trout will be starters for many teams, as they should be. Jason Heyward is a good example of a minor leaguer that was ready for the ML’s. Mike Stanton is another.

Your minor league players will be adequate replacements for injured players.

The aspect I like the best, is that it dramatically decreases “trade rape” where a user can trade for talented prospects easily due to their ratings being so low.

The traditionally of handling minor leaguers is to base their ratings off minor league stats and then reduce a certain percentage or points. So basically it makes all minor league players well below replacement level. Not fun, nor realistic, IMO.
This should also open the game wide up for those that like to play RTTS and MiLB games. This method also rewards player (Yes you, Brandon Wood) for playing MiLB games in hitter firendly leagues. MLEs incorporate park and league factors, so that MiLB stats are "neutralized" before they are scaled to ML projections.

For teams like MIN, that has 4 quality, speedy, decent hitting CF’s … this opens up a ton of roster flexibility in terms of trades, etc. For the Craig Counsell’s of the world, this means that MIL has 2 2nd basemen in the minors that are better than him, and the Doug Davis’s of the world are in heavy competition for a 5th rotation spot.

[TripleThreat1973]

Re: 2011 Projection-Based Rosters: Methods, Formulas, &amp; Discussion ...

[7] Entering Ratings

Ah, the fun part. This is basically going through player to player, row by row, entering the ratings from the spreadsheet. Since I am undecided as to what team I’ll use in a 2011 franchise (Seriously considering Kansas City), I use this as “scouting”, and that’s the only thing that makes it tolerable.

Everything up to this point is fun. I now hear the MLB10 music everywhere I go.

A few things to mention …

Bunting rating is left alone unless the batter’s contact rating goes way up.

Durability is sometimes left alone if the player is viewed to be a part-time ML player, but full time MiLB player. In other words, if he is projected at 52 games in the majors, I’m not giving him a 38 durability, when his most likely role is full time minor league starter.

Relievers are the ones that will see the most jump in their ratings. The reason for this is simple. Middle relievers are the least important job in baseball. They are often replacement players themselves. So the difference between them and the minor league pitchers is thin.

Since MiLB relievers in OSFM have their ratings all down in the 40s and 50s because of the traditional way of handling them, their rating will go up to 60s and higher on occasion … especially K9. When this happens I make small additions to their pitch ratings. If a reliever projects to have a K9 rating of 85, then having his pitch ratings in the low 40s is not desirable. So, I boost them to an average level of someone with an 85 K9 rating, by increasing their break rating. I use break for K’s, and BB9 for walks. It is rare, but it does happen … specifically with a young flamethrower that strikes out a lot of guys, with poor control.

[TripleThreat1973]

Re: 2011 Projection-Based Rosters: Methods, Formulas, &amp; Discussion ...

[8] Lineups and Rotations

Once everyone is entered and trades are up to date, lineups and rotations will be set. There are a few good wbsites that already have likely lineups, batting orders, and rotations available … and change with every roster move.

[9] Non-roster players

What to do with the remaining scrubs on the rosters, retired players, and missing players.
A part of me says “forget em, they aren’t integral to the game anyway.” There will be about 5 guys on each roster. They stick out because their rating are in the low 40s, and everyone else is 50s and up. I may just leave them as is.

Retired players will have each rating reduced by 50. If you wish to keep them active or bring them back at a reduced rating, you’ll know how much to add back.

Missing players that will be important will be created. I haven’t come across anyone yet. Great job OSFM guys.

[10] Testing

10 individual seasons will be simmed. The top 10 leaders for each stat recorded, and averaged, and then compared to ML leaders in those same categories. I am not anticipating any big differences since Bahnzo already test his ratings thoroughly, as did the OSFM guys.

[11] Global changes or individual tweaks

I am curious as to how the durability rating works. Relievers see far too many appearances. I recall a middle reliever in one of my franchises that had 69 appearances by the all-star break (AI controlled).
I also notice that starter stamina is too low, not enough pitchers get enough wins, and some middle reliever is the CYA running with a 15-4 record. I also am curious about the relievers K9 ratings.

Relievers will always have better k9 stats because they come in for one inning and face batters that they have a high leverage against and/or have not seen them previously (and they don;t have to worry about stamina or pitch counts, etc .... just attack). Starters on the other hand have to face the same batters 3+ times per game. I just want to make sure all strikeouts relievers don’t rack up Marmol strikeout rates. I don’t anticipate any problems.

[12] Progress.

I have entered 15 teams, have 9 more to go, and then am waiting for the remaining 6 to have their projections released. If I get down to the end and am ready to finish, I can use the THT projections in their place … they are very similar anyway. I think Dan releases 2 teams per week, so it should not be an issue.

[gtm]

Re: 2011 Projection-Based Rosters: Methods, Formulas, &amp; Discussion ...

Very interesting and nice. How do you figure pitcher's pitch speeds with there pitch types?

[gtm]

Re: 2011 Projection-Based Rosters: Methods, Formulas, &amp; Discussion ...

I'm taking it that these are for historical player's also, correct?

[TripleThreat1973]

Re: 2011 Projection-Based Rosters: Methods, Formulas, &amp; Discussion ...

I am leaving individual pitch ratings alone, unless the player sees a significant change in their "per 9" ratings. I know the guys put a lot of effort into those, and I am not interested in going pitcher-by-pitcher and looking at their pitch value ratings, although if needed, it could be done with fangraphs and exporting data into excel.

For historical rosters, I would figure what the "average stat" would be and what the average rating should be and then work on a formula from there.

I'm guessing Bahnzo's formulas are designed to use for anything. Likewise my formulas would work as well, but I have not tested them for that.

[Rwings832]

Re: 2011 Projection-Based Rosters: Methods, Formulas, &amp; Discussion ...

I am leaving individual pitch ratings alone, unless the player sees a significant change in their "per 9" ratings. I know the guys put a lot of effort into those, and I am not interested in going pitcher-by-pitcher and looking at their pitch value ratings, although if needed, it could be done with fangraphs and exporting data into excel.

For historical rosters, I would figure what the "average stat" would be and what the average rating should be and then work on a formula from there.

I'm guessing Bahnzo's formulas are designed to use for anything. Likewise my formulas would work as well, but I have not tested them for that.

Nice work, interesting, are you punching them into a spreadsheet then?