K or 4-3

[Member]

The Lahman's Datbase, of course.

 

The definition of outs I use is:

 

AB - H + SH + SF + CS + GIDP.

 

That's slightly different than the AB+HB+BB+SF denominator of OBP.

 

So, outs per plate appearance is actually:

 

 AB ? H + SH + SF + CS + GIDP ---------------------------- AB + SH + SF + HB + BB

[rluzinski]

The Lahman's Datbase, of course.

 

I meant the actual spreadsheet with your calculations.

[Member]

Yep. Easy to do when you have MS Excel.

 

Taking the team totals from the last 50 seasons (1956-2005), I simply ran the CORREL function, comparing the team-by-team statistics with each team's winning percentage.

 

EDIT:

Inbox me if you want a copy.

[rluzinski]

I ran across the following post in the Scout's Sabermetric forum. It was posted by Mitchel Lichtman, aka "MGL", who contributed to "The Book" and currently works for the Cardinals:

 

Quote:

---

Why people use regression equations when we know the exact win and run value of almost everything in baseball, I don't know. It can only get you in trouble! There are all kinds of cross correlations and hidden variables when you regress w% on selected events in baseball. For example, you will find a positive correlation between K's and run scoring or w% (I think) because K's correlate positively (fairly strongly) with HR's. Etc.

 

You will probably find a negative correlation with both SB's and CS's (I guess you did) allowed because more CS's mean more SB attempts which means more speed which means more run scoring (by advancing the extra base more often) by your opponents. Do you need to run a regression analysis to tell you the *sign* (negative or positive) of the influence of catching your opponents stealing more often, everything else being equal?

---

 

Maybe throw this thread in the stats forum?

[rluzinski]

A post from MGL in this thread about strikes:

 

Quote:

---

This is not a complicated matter. Based on empirical data (the average change in RE state after a K versus the average change in RE state after a non-K out, including a ROE), the average value of the K out was -.280 in the NL in 93-02. The average value of the non-K out, including ROE's, was -.261. Therefore a non-k out was better by about .019 runs.

 

In the AL, the results were similar. K out= -.305 and non-kout= -.287.

 

So a non-K out is around .0185, or .02 runs better than a K out. It will vary slightly, depending on the slot in the batting order and the profile of the other players, as will all the lwt values. As well, it vares depending upon whether the player is RH or LH and what his G/F ratio is.

 

It is that simple. Everything else being equal between two players, you would prefer the one who had more non-K outs, but it don't make a whole lot of difference, as 100 extra K's as compared to non-K outs is worth a grand total of 2 runs. Now, if we are taking a batter's sample (historical) stats and crafting a projection from them, it is a different story and a different question, as K's regress quite a bit less than non-k outs. But that is something else altogether...

---

 

This is based on actual in game data, not simulations or permutations. It's what has actually happened.

[mothershipconnection]

So a non-K out is around .0185, or .02 runs better than a K out.

Correct me if I'm wrong here, but he's saying that 50 extra K's would equal 1 run. Thus, a team would need to convert 500 (50k/run X 10 runs/win) non-k outs into k's to lose 1 additional game, which could never happen.

[rluzinski]

Thus, a team would need to convert 500 (50k/run X 10 runs/win) non-k outs into k's to lose 1 additional game

 

Give or take some, yes. Let's look at the Crew from last year:

 

SO: 1162

SF: 38

SH: 66

GIDP: 137

 

The above estimation says that the Brewers would have won about 2 more games if they hadn't struck out once.

 

The flip side is if every out the Brewers made last year was via a K (about 3,000 extra Ks). Let's try and use a common sense approach to estimate the run/win impact of that:

 

* They would have forgone 38 sac flies. Some percentage of those runners would still have eventually scored. We'll say they lose the majority of those runs (-30).

 

* They would have zero sac hits, instead of the 66 sac hits. There would have also been other siutions where a runner wouldn't have advanced as well. This is hard one to estimate... maybe 45 runs?).

 

* They would have had 137 less outs from having no GIDP. Linear weights gives an out a value of about -.25 runs. -.25 run/out x 137 outs = about 34 runs saved.

 

-30 runs - 45 runs + 34 runs = -41 runs = ~ 4 wins

 

That's in the ballpark of the original estimation of .04 runs/K (.02 runs/K x 3,000 Ks = 60 runs = 6 wins))

 

On a team level, K's barely hurts a team. It's why most run estimators don't even bother using K numbers at all.