Derby Curse? Nah

by R.J. Anderson on Jul 13, 2008 10:01 PM EDT Comment 17 comments

Another reader requested piece today: testing the myth (or is it) of homerun derby participants changing their approach and suffering second half regresses.

Taking the combatants from 1998 through 2007 (84 total) I looked up their first half and second half OPS (with help from user SaberToothedPie) to see if the players really did suffer a let down post derby. Why not homeruns? Because for one the games are disproportioned and 15 homers in the second half is actually more impressive than 15 in the first half on a rate scale. Hitters also can have their ratios like HR/FB% regress or progress, meaning homeruns just even out over the 162 game season. Instead we're looking at an outline of their entire offensive performance, or at least a summed up version.

Year	Player	PreASBOPS	PostASBOPS	NET	W
2007	Morneau	0.944	0.702	-0.242
2007	Ordonez	1.05	1.004	-0.046
2007	Rios	0.87	0.831	-0.039
2007	Guerrero	0.962	0.935	-0.027	W
2007	Fielder	0.996	1.034	0.038
2007	Howard	0.933	1.016	0.083
2007	Holliday	0.964	1.073	0.109
2007	Pujols	0.886	1.097	0.211
2006	Wright	0.961	0.844	-0.117
2006	Dye	1.043	0.965	-0.078
2006	Cabrera	0.998	0.998	0
2006	Tejada	0.871	0.886	0.015
2006	Glaus	0.859	0.879	0.02
2006	Berkman	1.011	1.078	0.067
2006	Ortiz	0.996	1.121	0.125
2006	Howard	0.932	1.259	0.327	W
2005	Abreu	0.955	0.787	-0.168	W
2005	C. Lee	0.864	0.747	-0.117
2005	I. Rodriguez	0.761	0.697	-0.064
2005	A. Jones	0.93	0.911	-0.019
2005	Choi	0.777	0.814	0.037
2005	Ortiz	0.982	1.024	0.042
2005	Teixeira	0.93	0.982	0.052
2005	Bay	0.93	0.998	0.068
2004	Blalock	0.941	0.743	-0.198
2004	Thome	1.059	0.868	-0.191
2004	Sosa	0.939	0.777	-0.162
2004	Bonds	1.421	1.421	0
2004	Tejada	0.887	0.902	0.015	W
2004	Berkman	1.008	1.024	0.016
2004	Palmeiro	0.771	0.825	0.054
2004	Ortiz	0.954	1.021	0.067
2003	Edmonds	1.066	0.864	-0.202
2003	Boone	0.963	0.815	-0.148
2003	Anderson	0.943	0.807	-0.136	W
2003	Delgado	1.053	0.965	-0.088
2003	Giambi	0.966	0.898	-0.068
2003	Pujols	1.121	1.084	-0.037
2003	Sheffield	1.019	1.028	0.009
2003	Sexson	0.889	0.98	0.091
2002	Konerko	0.949	0.743	-0.206
2002	Berkman	1.053	0.901	-0.152
2002	Sosa	1.059	0.911	-0.148
2002	Hunter	0.911	0.78	-0.131
2002	Sexson	0.882	0.847	-0.035
2002	Giambi	1.032	1.035	0.003	W
2002	A. Rodriguez	1.008	1.023	0.015
2002	Bonds	1.342	1.432	0.09
2001	L. Gonzalez	1.189	1.032	-0.157	W
2001	Boone	0.945	0.956	0.011
2001	Glaus	0.877	0.922	0.045
2001	Helton	1.09	1.148	0.058
2001	A. Rodriguez	0.993	1.052	0.059
2001	Sosa	1.126	1.225	0.099
2001	Giambi	1.082	1.202	0.12
2001	Bonds	1.314	1.455	0.141
2000	E. Martinez	1.114	0.883	-0.231
2000	Everett	1.05	0.831	-0.219
2000	Guerrero	1.131	1.004	-0.127
2000	Delgado	1.185	1.069	-0.116
2000	C. Jones	1.023	0.913	-0.11
2000	Griffey	0.935	0.952	0.017
2000	Sosa	0.962	1.138	0.176	W
1999	Surhoff	0.923	0.737	-0.186
1999	S. Green	1.047	0.889	-0.158
1999	Griffey	1.024	0.882	-0.142	W
1999	Bagwell	1.109	0.968	-0.141
1999	Burnitz	0.998	0.886	-0.112
1999	Garciparra	1.054	0.988	-0.066
1999	Walker	1.188	1.135	-0.053
1999	Jaha	0.981	0.956	-0.025
1999	Sosa	0.972	1.036	0.064
1999	McGwire	1.027	1.236	0.209
1998	Thome	1.09	0.792	-0.298
1998	Griffey	1.061	0.876	-0.185	W
1998	Lopez	0.937	0.789	-0.148
1998	Easley	0.876	0.731	-0.145
1998	McGwire	1.252	1.189	-0.063
1998	A. Rodriguez	0.94	0.894	-0.046
1998	Palmeiro	0.959	0.928	-0.031
1998	C. Jones	0.955	0.946	-0.009
1998	Alou	0.975	0.988	0.013
1998	Castilla	0.926	0.98	0.054

Above you see the 843 players, their first and second half OPS, a "net" column (second half OPS - first half OPS) and a "W" column - only the players who won that derby have a W listed. 50 of the 84 players either saw no post-derby gain or suffered an OPS drop, meaning only 34 had positive post-derby experiences.It's important to note that Ivan Rodriguez played in all of 10 games following the derby, hence his absence.

So can we predict a fall off for some of the contestants in tomorrow's derby? It seems quite likely that at least a few will have lower second half OPSes, but correlation doesn't always stem from causation, so while people may think the derby affected a player, there's nothing too conclusive to be drawn, despite only a .443 R2 between first half and second half OPSes which I actually figured would be a bit higher.

To see if this is regular or not we took the top 50 homerun hitters from 2007 and compared their first and second half OPSes - since in the end we are talking about some of their power hitting brethren who, again in theory, didn't "mess up their swings" or anything of the sort. The R2 of the first and second half OPSes: .344.

That means your average HRD participant actually has a better chance at maintaining his first half OPS than your top 50 homerun hitters, at least in 2007.

0 recs | Comment 17 comments

Story-email Email Printer Print

Comments

by P Brady on Jul 13, 2008 10:04 PM EDT reply actions 0 recs

improvement of graph

Interesting work. Might I suggest to make the graph more readable you scale down both the axes so only numeric coordinates that contain data are shown. Second, if you were to plot a line where the x and y values were exactly equal across the graph it would be easier to visualize the results as one could directly see how many players improved/declines and the rates of those declines.

I assume that you took out individuals who may have gotten injured early in the second half, whereas this could have led to a unrealistic OPS split?

by RaysTheRoof on Jul 13, 2008 11:08 PM EDT reply actions 0 recs

On your last point, yes.

You definitely have a point about the scales, I should’ve done that originally.

by R.J. Anderson on Jul 13, 2008 11:13 PM EDT up reply actions 0 recs

sorry

about the x=y line suggestion. Now that I think about it that doesn’t make too much sense.

by RaysTheRoof on Jul 13, 2008 11:14 PM EDT up reply actions 0 recs

It makes sense

I’m just not sure how helpful it would be visually though.

by R.J. Anderson on Jul 13, 2008 11:21 PM EDT up reply actions 0 recs

database?

Is there a specific database where you get access to the statistics you use? I tried using mlb.com to compile some data and it was brutal.

by RaysTheRoof on Jul 13, 2008 11:44 PM EDT up reply actions 0 recs

No database

I hand collected it through BR’s splits pages.

by R.J. Anderson on Jul 13, 2008 11:47 PM EDT up reply actions 0 recs

BR?

by RaysTheRoof on Jul 13, 2008 11:50 PM EDT reply actions 0 recs

Baseball-Reference.

by R.J. Anderson on Jul 13, 2008 11:57 PM EDT up reply actions 0 recs

I have faith

Evan will be just fine after the Derby and ASG. I’d be more worried about the pressure put on him by the offensive struggles of the team in general.

TAMPA BAY RAYS: 2008 CITRUS SERIES CHAMPIONS

by JMB on Jul 14, 2008 9:04 AM EDT reply actions 0 recs

Are there any statistics

for rookies in the HRD who also happen to be the most handsome contestant? Because I am predicting that he will get an OPS of 1.500 after the HRD, and then we will elect him president of the world for life.

B Rad the Ray Fan
9 = 8

by B Ray on Jul 14, 2008 9:54 AM EDT reply actions 0 recs

the jinx is a sampling bias

Which players tend to be picked for derby? It tends to be players having breakthrough power years or breakthrough years in general. Older, established power threats tend not to participate. (I realize this is not a hard and fast rule, just a trend.)

Ok, so if we’re selective players who have had breakout seasons in the first half, what can we expect from them, whether or not they participate in the derby? We would expect that part of their breakout is due to luck. Which is more likely, that someone who hit 25 homeruns actually has 20 homerun talent but got lucky or that they actually have 30 homerun talent and got unlucky? The first one, obviously. That’s the whole point of regression. Performances far away from the mean most likely include random variation.

Therefore, since breakout performances should have future expectations regressed, it’s not the derby’s fault for bringing their stats down.

It’s like the SI cover jinx. They only put people on the cover who do awesome shit. People who do awesome shit have nowhere to go but down.

by Sky Kalkman on Jul 14, 2008 11:54 AM EDT reply actions 1 recs

Very good point

by R.J. Anderson on Jul 14, 2008 12:06 PM EDT up reply actions 0 recs

IIRC your coefficient of .9076

would indicate an overall difference of .0924 between a perfectly, positive correlated sample. Would this indicate a roughly 9% decrease in first half to second half OPS across the board?

Be who you are and say what you feel, because those who mind don't matter, and those who matter don't mind.

by Sandy Kazmir on Jul 14, 2008 2:12 PM EDT reply actions 0 recs

Yeah, I can make up statistical sounding stough too!

The problem is that the leading coefficient of the reverse normalizer is well over 1.034, the mean that we rely on for most normalized calibrations with sample sizes over 15. The truthometer also has a suspiciously low standard deviation, which is liable to discount the overall contingency barometer to -0.502 or lower.

Is it fun to read things that don’t make sense!? That’s what I thought.

B Rad the Ray Fan
9 = 8

by B Ray on Jul 14, 2008 2:55 PM EDT up reply actions 0 recs