Hi everyone. Relatively new to WhatIf, but a long-time baseball fan and sabermetrics afficionado. I've been toying around with the available stats in Excel, and put together a multiple regression analysis using $/IP as the dependent variable and ERA, ERC, OAV, WHIP, HR/9, BB/9 and K/9 as independent variables (all using the normalized stats as given). The analysis yielded the following:

R-square = .871412 (high, these factors unsurprisingly strongly predict the player's price)

Coefficients:
Intercept = 74522.56 (p = 0)
ERA = -91.6313 (p = 9.06E-06)
ERC = 2598.706 (p = 4.2E-283)
OAV = -195551 (p = 0)
WHIP = 82.41085 (p = .871955)
HR/9 = -7724.62 (p = 0)
BB/9 = -2747.95 (p = 0)
K/9 = 371.6703 (p = 0)

Because the variables have different scales, the coefficients themselves aren't useful, but the signs are meaningful (as are the p scores). The conclusion that WHIP isn't very useful when controlling for OAV and BB/9 isn't very suprising.

What I can't figure out, though, is why the coefficient for ERC would be positive. This suggests that when these other variables are controlled for, a higher ERC makes a player more valuable (whereas common sense would suggest a lower component ERA would be more valuable).

If WhatIf uses Bill James' definition of ERC, then:
ERC = (((H + BB + HBP)×PTB)/(BFP×IP))×9 - 0.56
where H is hits, BB is bases on balls (walks), HBP is hit by pitch, BFP is batters faced by pitcher, IP is innings pitched, and PTB is defined as:
PTB = 0.89×(1.255×(H - HR) + 4×HR) + 0.56×(BB + HBP - IBB)
where HR is home runs, IBB is intentional walks, and others are as above.

Do people like pitchers who hit batters more? This would be my conclusion, since HBP is the one component of ERC not controlled for in this regression.

Anyone know why this might be the case?

It is known that the SIM has an inordinately high amount of HBP in general. In the SIM, an HBP is actually a preferable result to a BB because less pitches are (on average) used to do so, and therefore, fatigue is impacted less.

Welcome to WIS.

If you fit a model with all the predictors above except for ERC, then fit a model with everything including ERC, is there a big change in the R^2 value? I'm guessing it's akin to WHIP, in that all those predictors are already a part of ERC, so there is redundancy. However, I would think that the p-value would be much higher, as it is for WHIP...

Posted by AKlopp on 1/17/2012 7:38:00 AM (view original):
If you fit a model with all the predictors above except for ERC, then fit a model with everything including ERC, is there a big change in the R^2 value? I'm guessing it's akin to WHIP, in that all those predictors are already a part of ERC, so there is redundancy. However, I would think that the p-value would be much higher, as it is for WHIP...

My thoughts are the same.  I also think it has to do with how normalized (#) stats and regular stats interact in the pricing scheme.

To answer some of your questions, this is the full regression results with all of the aforementioned categories:

SUMMARY OUTPUT
Regression Statistics
Multiple R	0.933495
R Square	0.871412
Adjusted R Square	0.871385
Standard Error	2250.252
Observations	32758
ANOVA
	df	SS	MS	F	Significance F
Regression	7	1.12E+12	1.61E+11	31705.74	0
Residual	32750	1.66E+11	5063633
Total	32757	1.29E+12
	Coefficients	Standard Error	t Stat	P-value	Lower 95%	Upper 95%	Lower 95.0%	Upper 95.0%
Intercept	74522.56	352.4922	211.4162	0	73831.67	75213.46	73831.67	75213.46
ERC_Norm	2598.706	71.56305	36.31352	4.2E-283	2458.44	2738.972	2458.44	2738.972
ERA_Norm	-91.6313	20.64174	-4.43912	9.06E-06	-132.09	-51.1727	-132.09	-51.1727
OAV_Norm	-195551	2167.393	-90.2242	0	-199799	-191303	-199799	-191303
WHIP_Norm	82.41085	511.3046	0.161178	0.871955	-919.765	1084.587	-919.765	1084.587
HR_Norm	-7724.62	67.71866	-114.069	0	-7857.35	-7591.89	-7857.35	-7591.89
BB_Norm	-2747.95	46.20412	-59.4741	0	-2838.51	-2657.39	-2838.51	-2657.39
K9_Norm	371.6703	9.827309	37.82015	0	352.4084	390.9322	352.4084	390.9322

These are the results with ERC removed from the model:

SUMMARY OUTPUT
Regression Statistics
Multiple R	0.930717
R Square	0.866235
Adjusted R Square	0.86621
Standard Error	2295.073
Observations	32758
ANOVA
	df	SS	MS	F	Significance F
Regression	6	1.12E+12	1.86E+11	35348.1	0
Residual	32751	1.73E+11	5267358
Total	32757	1.29E+12
	Coefficients	Standard Error	t Stat	P-value	Lower 95%	Upper 95%	Lower 95.0%	Upper 95.0%
Intercept	63754.02	194.3543	328.0299	0	63373.08	64134.96	63373.08	64134.96
ERA_Norm	31.17509	20.76842	1.501082	0.133344	-9.53176	71.88194	-9.53176	71.88194
OAV_Norm	-174726	2131.775	-81.9626	0	-178904	-170547	-178904	-170547
WHIP_Norm	10202.96	437.2097	23.33653	1.8E-119	9346.013	11059.91	9346.013	11059.91
HR_Norm	-5697.33	39.09259	-145.739	0	-5773.96	-5620.71	-5773.96	-5620.71
BB_Norm	-2781.35	47.11508	-59.0332	0	-2873.7	-2689.01	-2873.7	-2689.01
K9_Norm	366.2474	10.02189	36.54473	1.3E-286	346.6041	385.8907	346.6041	385.8907

These are the correlations of the variables included:

	ERC_Norm	ERA_Norm	OAV_Norm	WHIP_Norm	HR_Norm	BB_Norm	K9_Norm
ERC_Norm	1
ERA_Norm	0.870553	1
OAV_Norm	0.848078	0.731859	1
WHIP_Norm	0.944289	0.811412	0.788259	1
HR_Norm	0.500885	0.490973	0.262321	0.264511	1
BB_Norm	0.475847	0.38691	0.059736	0.64534	0.071221	1
K9_Norm	-0.22782	-0.19368	-0.4264	-0.21361	0.089248	0.155949	1

As you can see, including ERC in the model does raise the model's precision (R-squared), but not by a huge amount. ERC and WHIP are highly correlated, so without ERC in the model, WHIP "stands in" for ERC in the model, and takes it's positive coefficient. This would presumeably dis-prove my previous theory that hit batsmen might be causing this effect (since WHIP does not account for HBP).

Yeah, I think the problem with your analysis is that you have assumed pricing is based exclusively on the # stats, which I don't believe to be the case.  I feel fairly certain there is some component based on raw stats and then an adjustment based on normalization, but not necessarily a direct correlation to the # stats.  It could be that the directions of normalization are what are causing the apparent reverse ERC dependence.  If the real prices are somewhere between what you would expect based on the RL stats and the # stats, as I would suspect is the case given that good normalizers have generally been seen as slightly better "values," particularly in lower caps, it would make sense that any statistic that is also normalized but not actually included in the calculation of price would have a positive coefficient.  This would indicate it was driving prices back into the median range between standard and normalized stats.  And I notice that WHIP, which is essentially in the same boat, did in fact do this very thing, albeit with a low significance.

If we remove 1990 Eckersley, 2006 Papelbon, 1981 Gossage, 2002 Hammond, and 2003 Gagne from the data then square the normalized ERA and ERC (these five pitchers have an ERA or ERC < 1, so they would throw off the model when squaring), you get the following results:

SUMMARY OUTPUT
Regression Statistics
Multiple R	0.94302
R Square	0.889287
Adjusted R Square	0.889263
Standard Error	2083.327
Observations	32753
ANOVA
	df	SS	MS	F	Significance F
Regression	7	1.14E+12	1.63E+11	37574.26	0
Residual	32745	1.42E+11	4340253
Total	32752	1.28E+12
	Coefficients	Standard Error	t Stat	P-value	Lower 95%	Upper 95%	Lower 95.0%	Upper 95.0%
Intercept	76694.87	234.5722	326.9564	0	76235.1	77154.64	76235.1	77154.64
ERCsquared	256.9048	3.3044	77.74628	0	250.4281	263.3816	250.4281	263.3816
ERAsquared	-8.4717	1.943675	-4.3586	1.31E-05	-12.2814	-4.66203	-12.2814	-4.66203
OAV_Norm	-161500	1956.795	-82.5329	0	-165335	-157665	-165335	-157665
WHIP_Norm	-5440.58	432.0181	-12.5934	2.79E-36	-6287.35	-4593.81	-6287.35	-4593.81
HR_Norm	-7591.43	39.75399	-190.96	0	-7669.35	-7513.51	-7669.35	-7513.51
BB_Norm	-2061.09	43.73234	-47.1296	0	-2146.81	-1975.37	-2146.81	-1975.37
K9_Norm	334.5729	9.104127	36.74959	9.3E-290	316.7285	352.4174	316.7285	352.4174

As you can see, this increases the model fit (R-square increases to .889287), and WHIP now becomes significant in the model. ERC still has a positive coefficient.

Returning to the original (full) data set, and removing ERC and WHIP from the model gives the following results:

SUMMARY OUTPUT
Regression Statistics
Multiple R	0.929522
R Square	0.864011
Adjusted R Square	0.86399
Standard Error	2314.04
Observations	32758
ANOVA
	df	SS	MS	F	Significance F
Regression	5	1.11E+12	2.23E+11	41618.05	0
Residual	32752	1.75E+11	5354782
Total	32757	1.29E+12
	Coefficients	Standard Error	t Stat	P-value	Lower 95%	Upper 95%	Lower 95.0%	Upper 95.0%
Intercept	61200.02	161.9387	377.9209	0	60882.62	61517.43	60882.62	61517.43
ERA_Norm	177.4876	19.96308	8.890789	6.37E-19	138.3592	216.6159	138.3592	216.6159
OAV_Norm	-127325	652.4689	-195.144	0	-128604	-126046	-128604	-126046
HR_Norm	-5700.06	39.4155	-144.615	0	-5777.31	-5622.8	-5777.31	-5622.8
BB_Norm	-1721.47	12.63809	-136.213	0	-1746.24	-1696.7	-1746.24	-1696.7
K9_Norm	384.2618	10.0747	38.14126	0	364.515	404.0086	364.515	404.0086

ERA takes a positive coefficient in this model, again seemingly "standing in" for the effect of ERC. Going further and removing ERA from the model, we get the following results:

SUMMARY OUTPUT
Regression Statistics
Multiple R	0.929345
R Square	0.863682
Adjusted R Square	0.863666
Standard Error	2316.796
Observations	32758
ANOVA
	df	SS	MS	F	Significance F
Regression	4	1.11E+12	2.78E+11	51879.18	0
Residual	32753	1.76E+11	5367541
Total	32757	1.29E+12
	Coefficients	Standard Error	t Stat	P-value	Lower 95%	Upper 95%	Lower 95.0%	Upper 95.0%
Intercept	60485.69	140.7688	429.6812	0	60209.77	60761.6	60209.77	60761.6
OAV_Norm	-123135	451.7266	-272.587	0	-124020	-122249	-124020	-122249
HR_Norm	-5530.18	34.51578	-160.222	0	-5597.83	-5462.53	-5597.83	-5462.53
BB_Norm	-1661.8	10.72109	-155.003	0	-1682.81	-1640.78	-1682.81	-1640.78
K9_Norm	383.9182	10.08662	38.06212	0	364.1481	403.6884	364.1481	403.6884

As the r-squared value changes very little, I think it's safe to say that ERA, ERC and WHIP just aren't strong predictors of $/IP. The others are much more important factors (particularly OAV).

Ah, takes me back to my multiple regression grad class...my prof was a big advocate of the "build from the ground up" method...start with the predictor with the highest correlation and run a simple regression. Then run a bunch of multiple regressions (2 predictors) with that var and each of the other possible predictors. Choose the best model if a) that predictor is significant and b) there is a big enough increase in R^2 to warrent the additional variable (I kinda think that's where looking at the adjusted R^2 comes in, but it's too long ago and I forget). Continue until you have the "best" model. Even if a new var is significant, if the increase isn R^2 isn't big, why go with a more complicated model. I think your last is the one to go with. So, the big question then is what is the other 14% of variability unaccounted for by the model? The pitcher's hitting stats, fielding(?),... 

But thanks, jimkelley! Interesting stuff!

As a potentially interesting side note, if you take the model a step further and say that OAV#, HR/9#, BB/9#, and K/9# SHOULD predict player value (this is dubious, and I have no information about how these players perform in the sim), you can "rank" players by the residual given by the model equation. So using this ranking mechanism, here are the top twenty "most undervalued" players, first by $/IP (using the raw residuals) and second by absolute salary (Residual*IP).

Season	LName	FName	Throws	IP_162	True_Salary	True_$/IP	Expected_$/IP	Residual	Expected_Salary	Residual*IP
1914	Black	Dave	R	27	$300,693.00	$11,136.78	$20,387.04	-$9,250.27	$550,450.19	-$249,757.19
2005	Osoria	Franquelis	R	30	$394,029.00	$13,134.30	$22,061.09	-$8,926.79	$661,832.73	-$267,803.73
1915	George	Lefty	L	30	$441,636.00	$14,721.20	$23,612.33	-$8,891.13	$708,370.04	-$266,734.04
1906	Moroney	Jim	L	29	$284,900.00	$9,824.14	$18,483.37	-$8,659.23	$536,017.64	-$251,117.64
2008	Castillo	Alberto	R	26	$321,847.00	$12,378.73	$20,860.47	-$8,481.74	$542,372.14	-$220,525.14
1905	Caldwell	Ralph	L	37	$371,084.00	$10,029.30	$18,496.30	-$8,467.00	$684,362.96	-$313,278.96
1915	Vance	Dazzy	R	30	$409,368.00	$13,645.60	$22,067.30	-$8,421.70	$662,018.95	-$252,650.95
1909	Ables	Harry	L	32	$552,644.00	$17,270.13	$25,668.70	-$8,398.58	$821,398.46	-$268,754.46
1903	Pearson	Alex	R	36	$529,463.00	$14,707.31	$22,890.53	-$8,183.22	$824,059.09	-$294,596.09
1915	Russell	Allan	R	29	$345,264.00	$11,905.66	$20,030.11	-$8,124.46	$580,873.23	-$235,609.23
1908	Glaze	Ralph	R	37	$588,033.00	$15,892.78	$24,000.99	-$8,108.21	$888,036.75	-$300,003.75
1910	White	Kirby	R	28	$474,927.00	$16,961.68	$25,069.12	-$8,107.44	$701,935.46	-$227,008.46
2006	Sauerbeck	Scott	L	26	$292,496.00	$11,249.85	$19,274.66	-$8,024.81	$501,141.10	-$208,645.10
2006	Sauerbeck	Scott	L	26	$292,496.00	$11,249.85	$19,274.66	-$8,024.81	$501,141.10	-$208,645.10
1917	Monroe	Ed	R	31	$271,535.00	$8,759.19	$16,736.79	-$7,977.60	$518,840.59	-$247,305.59
1902	Heismann	Crese	L	39	$525,642.00	$13,478.00	$21,451.56	-$7,973.56	$836,610.78	-$310,968.78
1920	Zinn	Jimmy	R	33	$534,240.00	$16,189.09	$24,113.95	-$7,924.86	$795,760.23	-$261,520.23
1886	White	Will	R	31	$324,300.00	$10,461.29	$18,370.67	-$7,909.38	$569,490.69	-$245,190.69
1989	Gardner	Mark	R	27	$359,687.00	$13,321.74	$21,098.85	-$7,777.11	$569,668.85	-$209,981.85
1916	Smith	Pop-boy	R	27	$312,004.00	$11,555.70	$19,319.70	-$7,764.00	$521,631.89	-$209,627.89

Season	LName	FName	Throws	IP_162	True_Salary	True_$/IP	Expected_$/IP	Residual	Expected_Salary	Residual*IP
1906	Young	Irv	L	385	$7,759,350.00	$20,154.16	$22,899.28	-$2,745.13	$8,816,223.39	-$1,056,873.39
1908	Summers	Ed	R	319	$7,349,185.00	$23,038.20	$26,182.49	-$3,144.29	$8,352,212.90	-$1,003,027.90
1902	Pittinger	Togie	R	461	$10,323,428.00	$22,393.55	$24,506.69	-$2,113.14	$11,297,584.45	-$974,156.45
1909	Rucker	Nap	L	328	$7,884,868.00	$24,039.23	$26,970.76	-$2,931.53	$8,846,410.17	-$961,542.17
1885	Galvin	Pud	R	539	$10,228,958.00	$18,977.66	$20,740.98	-$1,763.32	$11,179,386.95	-$950,428.95
1885	Galvin	Pud	R	539	$10,196,387.00	$18,917.23	$20,654.38	-$1,737.15	$11,132,712.12	-$936,325.12
1902	Willis	Vic	R	485	$11,878,746.00	$24,492.26	$26,418.60	-$1,926.34	$12,813,022.62	-$934,276.62
1902	Taylor	Dummy	R	280	$5,600,012.00	$20,000.04	$23,235.25	-$3,235.20	$6,505,869.04	-$905,857.04
1907	McGinnity	Joe	R	329	$6,814,295.00	$20,712.14	$23,413.35	-$2,701.20	$7,702,990.68	-$888,695.68
1908	Dygert	Jimmy	R	253	$5,764,115.00	$22,783.06	$26,295.40	-$3,512.34	$6,652,736.28	-$888,621.28
1907	Waddell	Rube	L	319	$8,543,596.00	$26,782.43	$29,530.93	-$2,748.49	$9,420,365.40	-$876,769.40
1904	Pelty	Barney	R	321	$6,878,280.00	$21,427.66	$24,143.19	-$2,715.52	$7,749,962.77	-$871,682.77
1909	Warhop	Jack	R	262	$5,776,599.00	$22,048.09	$25,372.83	-$3,324.75	$6,647,682.45	-$871,083.45
1903	Flaherty	Patsy	L	348	$6,083,710.00	$17,481.93	$19,984.36	-$2,502.43	$6,954,557.07	-$870,847.07
1908	Sparks	Tully	R	278	$5,774,382.00	$20,771.16	$23,876.28	-$3,105.12	$6,637,604.60	-$863,222.60
1907	Dorner	Gus	R	297	$5,818,124.00	$19,589.64	$22,422.63	-$2,832.98	$6,659,519.71	-$841,395.71
1906	Dorner	Gus	R	310	$5,661,325.00	$18,262.34	$20,931.31	-$2,668.97	$6,488,705.93	-$827,380.93
1906	Dorner	Gus	R	310	$5,661,325.00	$18,262.34	$20,931.31	-$2,668.97	$6,488,705.93	-$827,380.93
1902	Taylor	Dummy	R	240	$4,812,249.00	$20,051.04	$23,452.98	-$3,401.94	$5,628,715.56	-$816,466.56
1903	Patterson	Roy	R	347	$7,908,680.00	$22,791.59	$25,122.90	-$2,331.32	$8,717,647.56	-$808,967.56

Looking at the "most over-valued" players might help more in determining what the remaining 14% accounts for. Taking out pitchers who the model predicts negative salaries for, here are the top twenties on the other side of the spectrum (highest residual and highest residual*IP):

MasterPlayerID	Season	LName	FName	Throws	IP_162	True_Salary	True_$/IP	Expected_$/IP	Residual	Expected_Salary	Residual*IP
16934	2009	Adams	Mike	R	37	$2,569,430.00	$69,444.05	$45,272.79	$24,171.27	$1,675,093.13	$894,336.87
4817	2003	Gagne	Eric	R	83	$5,463,221.00	$65,821.94	$43,248.00	$22,573.94	$3,589,584.18	$1,873,636.82
10281	2006	Nathan	Joe	R	69	$3,906,519.00	$56,616.22	$39,902.66	$16,713.56	$2,753,283.38	$1,153,235.62
14684	1999	Wagner	Billy	L	75	$4,314,571.00	$57,527.61	$40,964.53	$16,563.08	$3,072,339.96	$1,242,231.04
3979	1990	Eckersley	Dennis	R	74	$4,293,010.00	$58,013.65	$41,511.79	$16,501.86	$3,071,872.70	$1,221,137.30
17371	2010	Kuo	Hong-Chih	L	60	$3,398,795.00	$56,646.58	$40,480.16	$16,166.42	$2,428,809.59	$969,985.41
17418	2006	Saito	Takashi	R	79	$4,085,629.00	$51,716.82	$36,267.01	$15,449.81	$2,865,093.85	$1,220,535.15
17221	2006	Papelbon	Jonathan	R	69	$3,731,155.00	$54,074.71	$39,071.55	$15,003.16	$2,695,936.79	$1,035,218.21
17676	2008	Devine	Joey	R	45	$2,467,906.00	$54,842.36	$40,093.61	$14,748.75	$1,804,212.32	$663,693.68
8905	2000	Martinez	Pedro	R	217	$11,707,138.00	$53,949.94	$39,223.66	$14,726.29	$8,511,533.30	$3,195,604.70
12000	2008	Rivera	Mariano	R	70.667	$3,784,493.00	$53,553.89	$39,777.50	$13,776.40	$2,810,956.28	$973,536.72
12000	1996	Rivera	Mariano	R	108	$5,393,081.00	$49,935.94	$36,558.00	$13,377.93	$3,948,264.14	$1,444,816.86
15965	2004	Rodriguez	Francisco	R	84	$4,192,760.00	$49,913.81	$36,582.11	$13,331.70	$3,072,897.04	$1,119,862.96
12585	1951	Scheib	Carl	R	151	$5,287,607.00	$35,017.26	$21,898.52	$13,118.75	$3,306,675.85	$1,980,931.15
2209	1886	Caruthers	Bob	R	452	$19,253,474.00	$42,596.18	$29,752.81	$12,843.37	$13,448,270.88	$5,805,203.12
17066	2007	Putz	J.J.	R	71	$3,666,818.00	$51,645.32	$38,964.74	$12,680.58	$2,766,496.60	$900,321.40
13722	1894	Stratton	Scott	R	147	$3,434,808.00	$23,366.04	$10,744.07	$12,621.97	$1,579,378.70	$1,855,429.30
2358	1995	Charlton	Norm	L	54	$2,858,396.00	$52,933.26	$40,350.07	$12,583.19	$2,178,903.73	$679,492.27
10356	1955	Newcombe	Don	R	248	$8,940,624.00	$36,050.90	$23,473.26	$12,577.64	$5,821,368.62	$3,119,255.38
17759	2011	Romo	Sergio	R	48	$2,444,230.00	$50,921.46	$38,346.24	$12,575.22	$1,840,619.40	$603,610.60

MasterPlayerID	Season	LName	FName	Throws	IP_162	True_Salary	True_$/IP	Expected_$/IP	Residual	Expected_Salary	Residual*IP
12333	1890	Rusie	Amos	R	679	$26,668,329.00	$39,275.89	$29,990.30	$9,285.59	$20,363,415.01	$6,304,913.99
2209	1886	Caruthers	Bob	R	452	$19,253,474.00	$42,596.18	$29,752.81	$12,843.37	$13,448,270.88	$5,805,203.12
7532	1888	King	Silver	R	703	$29,119,784.00	$41,422.17	$33,178.10	$8,244.07	$23,324,206.25	$5,795,577.75
13722	1890	Stratton	Scott	R	529	$20,971,368.00	$39,643.42	$29,853.33	$9,790.09	$15,792,410.70	$5,178,957.30
6001	1895	Hawley	Pink	R	546	$18,937,942.00	$34,684.88	$27,152.79	$7,532.08	$14,825,424.49	$4,112,517.51
11642	1886	Ramsey	Toad	L	702	$26,183,261.00	$37,298.09	$31,501.70	$5,796.39	$22,114,195.71	$4,069,065.29
14069	1890	Terry	Adonis	R	465	$16,300,407.00	$35,054.64	$26,371.62	$8,683.02	$12,262,801.84	$4,037,605.16
4338	1885	Ferguson	Charlie	R	597	$21,254,213.00	$35,601.70	$29,118.47	$6,483.23	$17,383,725.60	$3,870,487.40
6058	1886	Hecker	Guy	R	502	$16,603,893.00	$33,075.48	$25,626.69	$7,448.79	$12,864,598.43	$3,739,294.57
14982	1885	Welch	Mickey	R	712	$25,567,428.00	$35,909.31	$30,897.06	$5,012.25	$21,998,708.25	$3,568,719.75
12825	1888	Seward	Ed	R	632	$23,545,198.00	$37,255.06	$31,678.88	$5,576.18	$20,021,050.44	$3,524,147.56
2530	1885	Clarkson	John	R	902	$30,279,672.00	$33,569.48	$29,678.38	$3,891.10	$26,769,897.50	$3,509,774.50
7068	1913	Johnson	Walter	R	364	$16,372,767.00	$44,980.13	$35,353.65	$9,626.48	$12,868,729.53	$3,504,037.47
13664	1894	Stivetts	Jack	R	415	$10,211,098.00	$24,605.06	$16,192.35	$8,412.71	$6,719,825.10	$3,491,272.90
6414	1896	Hoffer	Bill	R	389	$13,544,640.00	$34,819.13	$25,917.32	$8,901.81	$10,081,837.02	$3,462,802.98
8642	1994	Maddux	Greg	R	288	$13,307,875.00	$46,207.90	$34,420.37	$11,787.53	$9,913,067.70	$3,394,807.30
7068	1912	Johnson	Walter	R	394	$17,525,356.00	$44,480.60	$35,881.22	$8,599.38	$14,137,200.36	$3,388,155.64
13664	1890	Stivetts	Jack	R	504	$15,380,556.00	$30,516.98	$23,932.08	$6,584.89	$12,061,769.08	$3,318,786.92
4338	1886	Ferguson	Charlie	R	563	$20,553,002.00	$36,506.22	$30,615.90	$5,890.32	$17,236,749.25	$3,316,252.75
7498	1893	Killen	Frank	L	522	$16,683,860.00	$31,961.42	$25,775.06	$6,186.36	$13,454,580.85	$3,229,279.15

If anyone has played with the above pitchers, maybe they can shed some light on what makes them more valuable.

Finally, the truly dubious honor of the "worst" pitcher in WIS goes to 2004 Denny Stark (by a landslide, with an expected $/IP of -$16889.55). His stats:

LName

FName

Franchise

Throws

IP_162

OAV_Norm

HR_Norm

BB_Norm

K9_Norm

FieldingGrades_P

FirstSeason

LastSeason

True_Salary

Stark

Denny

Colorado Rockies

R

26

0.427383

2.849271

6.129083

3.106784

B/A+

0

1

$200,000.00

Just from a cursory observation: For the "over"-priced guys--Caruthers is there due to his bat. A lot of the others are some of the best relievers in the sim. It seems since a lot of the biggest residuals are guys on the high end of $/IP, then some of the predictors might not be included in the pricing model in a linear fashion.

Yup, makes sense. Is there a way to pull the pitchers' batting actuals from draft mode?