THE BOOK--Playing The Percentages In Baseball

Very cool. Would be interesting to compare sports too, to see where the biggest disparities are in pick value by round. Obvious guess is Basketball is at the top of the “top heavy” list by far, with football and hockey in the middle of the pack, and baseball way behind.

Agreed, that’s really cool. Would love to see what you find for other sports too.

Seems I already did it for baseball a while back.

The equation there was:
1/sqrt(pick)

Though, I didn’t try different schemes of:
1/(a+pick)^b

I just forced a=0, so b came in by default at 0.5. 

Anyway, presuming that is correct, then a #1 pick is 100%, a #2 pick is 71%, a #4 pick is 50%, and a #16 pick is 25%.

That sure doesn’t seem right, does it?  I would have expected a much flatter curve than the hockey one.

The baseball one was based on WAR, while the hockey based on playing time.  So, that probably explains it, as you can’t really have a super career based only on playing time.  There’s a cap essentially.  Using Awad’s GVT or Ryder’s PC would have been much better.

Hmmm… looks like I have draft data for baseball and I included WAR through age 29.

My best fit is:
7/(pick + 6)

So, a #1 pick is at 100%, a #8 pick is at 50%.  A #3/#4 pick is at 75%.  A #22 pick is at 25%.

I think that the hockey one is not good.  It’s based on playing time instead of performance, and it makes the curve way too flat.  I think it has to be steeper than the baseball one.

In baseball there’s a huge premium for the #1 overall pick, but after a steep drop to #2 it’s pretty flat in the mid-teens before dropping off again.  At least that’s been true ay time I’ve looked at drafts of the late 1980s and 1990s with various metrics.

Here are the historical results for the 1987-1999 drafts.  Formatting will be terrible, but this is first 30 slots, pre-FA WAR, career WAR.  (oh and rWAR)

1 15.9 30.8
2 5.6 7.5
3 5.6 7.5
4 4.9 7.3
5 5.1 8.0
6 5.0 7.9
7 5.5 8.5
8 6.2 9.6
9 5.9 8.7
10 6.3 9.0
11 5.9 9.1
12 5.3 9.0
13 3.8 6.8
14 4.6 8.2
15 3.9 7.8
16 3.2 6.1
17 2.8 4.9
18 4.0 7.1
19 3.2 5.8
20 3.6 6.2
21 4.3 7.4
22 4.1 7.2
23 3.1 5.1
24 3.0 4.9
25 2.0 3.1
26 1.5 2.3
27 1.5 2.1
28 1.7 2.5
29 1.9 2.8
30 1.9 2.6

Mine from 1965-1999 drafts, using rWAR, with WAR through age 29:

1 13.4
2 9.0
3 7.6
4 8.3
5 4.9
6 6.5
7 3.5
8 4.0
9 4.6
10 7.4
11 1.9
12 4.1
13 4.2
14 4.0
15 2.9
16 4.5
17 4.0
18 2.3
19 4.4
20 4.9
21 2.1
22 4.3
23 1.9
24 1.7
25 1.7
26 1.9
27 1.9
28 1.7
29 3.3
30 4.5

Philly: for your #11 pick, you have an average of 5.9 WAR for 13 players, or 77 WAR.

The only players of note that I have for 1987-1999 are Estes and Eaton.  When you do your average, are you doing sum(WAR) / 13, or sum(WAR) / someSmallerNumber?

And before that, I just have Luzinski and Mack and Weiss.

"The only players of note that I have for 1987-1999 are Estes and Eaton.  When you do your average, are you doing sum(WAR) / 13, or sum(WAR) / someSmallerNumber? “

I should have mentioned that I do a simple rolling average.  Pick #11 is really everyone who went #9-#13 from those years.  So instead of having a really bumpy line based on 13 players I have a much smoother one based on 65 players.

Just looking at straight average (career rWAR):

#10 -14.1
#11 - 1.0
#12 - 11.1

For whatever reason that slot was a black hole in those years.

With the rolling average it becomes:

#10 -9.0
#11 - 9.1
#12 - 9.0

That seems a lot more reasonable when trying to talk about the intrinsic value of a slot.

Sure, I agree.  Or just best-fit it.

The playing time is a first order approximation. I wanted to use Tom Awad’s GVT, but had trouble finding the data. I have it now and will do it in a future update.

But based on your formula, I still think it fits pretty well. Hockey seems to have a good top-3 or so.

Also, Notice i coupled sample size in groups of no smaller than 3. I only have sample size of 11 years available (97-06) so going by individual draft pick would leave too small a sample size.