Jump to content


Yo! You're not logged in. Why am I seeing this ad?

Photo

Draft Slot Probabilities and Values


This topic has been archived. This means that you cannot reply to this topic.
17 replies to this topic

#1 philly sox fan


  • SoSH Member


  • 9,748 posts

Posted 04 March 2006 - 11:42 AM

Introduction

I’ve spent a good chunk of time this winter updating and reworking my draft research. There’s a lot of stuff to get out and I’ve just been slow getting over the hump and really writing it up in reasonably final form. A couple weeks ago I participated in my strat league draft and of course a lot of players from the June 2005 draft were selected. Someone asked about the probabilities of these players actually panning out and I thought that was a good opening to at least knock out a quick summary of some of my research just based on those players that were selected in our draft. I was originally just going to send out it to the league, but I really liked the idea of putting specific names to the generic draft slots. Instead of looking at draft slot #4 we can look at draft slot #4, Ryan Zimmerman. I think that adds a nice put of context both for the draft slots and for the players many of whom are popping up on prospect lists this winter.

I started to think of this as a nice summary of some of the work I’ve done, so I’ve added additional sections to that framework and it became much longer than I intended, but that’s life.

Part I: What does an X WARP3 player look like?

This work is designed to show the frequency of different types of players in the draft. The most important “type” of player distinction that I make is quantified by the player’s career WARP3 total. I’ve looked at these players so much that it’s very easy for me to refer to an X WARP3 player and know exactly what type of career that player has had, but they may be just numbers for other people. In order to familiarize people with what I mean by certain WARP3 types of careers I’ve pulled a number of players with Red Sox connections from my dataset. After running through these examples you should have a pretty good idea of what a 20-40 WARP3 player looks like.

60+ WARP3: the very good to HoF great
Rd     Pick     Team       Player           Pos      School       WARP3
  4      109      Bos       Jeff Bagwell*     3B         C         125.3
  3       79      Tor       John Olerud*    1B-LHP       C         111.6
  1       13      Cle       Manny Ramirez*    OF        HS         100.2
  1       12      Bos       N Garciaparra*    SS         C          63.6
  8      200      Pit       Tim Wakefield*    1B         C          61.7
  1s      35       KC       Johnny Damon*     OF        HS          61.4
I actually have this range split into three smaller intervals, but there are so few players in each one that it makes it easier to just collapse them into one. Wakefield, naturally, is a bit of an oddball member of this group, but basically a 60+ WARP3 player is a very significant contributor to his team over a period of many years. A 60+ WARP3 player is in the top ~5% of expected value for every draft slot except #1 overall.

40-60 WARP3: good players
Rd     Pick     Team       Player           Pos      School       WARP3
  1       23      Bos       Mo Vaughn         1B         C          56.4
  5      121      Bos       John Valentin     SS         C          56.1
  5      127      Tor       Mike Timlin*     RHP         C          50.2
  8      215      Sea       Derek Lowe*      RHP        HS          49.9
  1       23      Bos       Aaron Sele*      RHP         C          47.8
  5      127      NYY       JT Snow*          1B         C          45.5
  7      207      Mil       Mark Loretta*     SS         C          44.7
  6      161      Bos       Paul Quantrill*  RHP         C          43.4
  9      256       SF       Keith Foulke*    RHP         C          43.2
15      414       SF       Bill Mueller*     SS         C          42.3
  1       10      NYY       Carl Everett*     OF        HS          41.9
I think 40 WARP3 works reasonably well as a cutoff line for “good” players. Vaughn and Valentin were better than merely “good”, but injuries cut short their ability to accumulate enough WARP3 to get up into the next category. Wherever you chose to draw these kinds of lines there are going to be a few players on the edges of each interval that probably belong in the adjacent one. To me players like Mueller and Lowe typify good, solid players that may win a batting title (Mueler in 03) or contend for the Cy Young (Lowe in 02) in a fluke year but mostly contribute solid performances for several years.

One issue to always keep in the back of your mind is that if two players have similar career WARP3 totals, then the one with the shorter career was most likely much more valuable because most of his production occurred prior to free agency. For example, Valentin and Timlin are going to end up with very similar career totals, but over 90% of Valentin’s production was concentrated in his first six years. Timlin is a WARP3 accumulator who will end up with ~70% of his production in his post free agency years. Valentin was much more valuable to the team that drafted him, then Timlin.

Foulke is arguably better than a “good” player, but his role makes it difficult to accumulate very high WARP3 totals. That’s something to keep in mind in the discussion about Hansen.

20-40 WARP3: useful players
Rd     Pick     Team       Player           Pos      School       WARP3
  2       49      Bos       Jeff Suppan*     RHP        HS          37.7
  1        2      Det       Tony Clark*       OF        HS          35.1
  1        3      Atl       Steve Avery      LHP        HS          33.7
  1        7      Bos       Trot Nixon*       OF        HS          33.2
  1        8      Min       Todd Walker*      2B         C          31.5
  6      140      Cub       Frank Castillo*  RHP        HS          29.5
13      330      Mil       Troy O’Leary      OF        HS          29.3
  6      158      Stl       Rheal Cormier*   LHP        CC          29.1
18      463      Oak       Darren Lewis      OF         C          27.7
15      386       SF       Pat Rapp         RHP         C          27.5
  3       86       SD       Matt Clement*    RHP        HS          27.3
  1s      29      NYM       Jay Payton*       OF         C          24.7
  1       20      Cin       Pokey Reese*      SS        HS          24.4
13      355      Bos       Carl Pavano*     RHP        HS          23.5
  1       26      Det       Rico Brogna       1B        HS          22.0
  2       56      NYM       Pete Schourek    LHP        HS          21.8
  1s      43      Bos       Scott Hatteberg*   C         C          21.7
  3       85      Sea       Jim Mecir*       RHP         C          21.1
10      256      Atl       T Graffinino*     SS        HS          21.1
30      765      Tex       Jeff Frye         2B         C          20.9
  8      199      Bos       Tim Naehring      SS         C          20.8
  5      144       SF       Bobby Howry*     RHP         C          20.3
I generally use the 20 WARP3 line to denote players who’ve had “useful” careers. The best players in this group – Suppan, Nixon, Clement – should end up in the 40-60 WARP3 group. Any time you’re labeling career totals of active players you’re going to run into some that are characterized a little lower than they should be. For the most part this isn’t a big problem with the players in the study. Players like O’Leary, Lewis and Brogna typify the “useful” player type. They do represent good value on their draft position and they are good contributors to their teams for a few years, but they’re generally just complementary players on good teams.

<20 WARP3: not quite useful players
Rd     Pick     Team       Player           Pos      School       WARP3
25      641      Bos       John Flaherty*     C         C          19.6
23      640      Hou       John Halama*     LHP         C          19.1
11      291      Bos       Phil Plantier     3B        HS          18.6
25      657      Sea       Matt Mantei*     RHP        HS          15.6
  4      120       SF       Mike Myers*      LHP         C          15.5
17      454      NYM       Brian Daubach*    1B        HS          15.3
  5      123      Cle       Alan Embree*     LHP        HS          14.4
28      781      Det       Dave Roberts*     OF         C          14.0
  5      131       SF       Doug Mirabelli*    C         C          13.1
Just to make my life simpler I lumped every pick that doesn’t produce at least a 20 WARP3 player into one catchall category. Although the vast majority of picks in this category never make the majors at all, some are certainly positive contributors. You find career backups like Flaherty – who was over 20 WARP last year, boy I’m glad the Sox signed a 35 yr old moving backward down the career ladder! – and Mirabelli. There are decent niche relievers like Myers and Embree and fringe swingmen like Halama. The most interesting players are the one-year wonders like Plantier. However, most of the MLB players in this group or more like Tim VanEgmond or Donnie Sadler. Generally, these players just aren’t quite good or durable enough to hold down regular jobs long enough to clear 20 WARP3.


Part II: A general look at draft slot probabilities for picks #1-100

These are actually just the slots whose players were picked in my recent strat league draft. It includes every 1st rd pick, a decent sampling of the supplemental and second rounds and a handful of third round picks.

I only have eight drafts in my dataset which is much too small of a sample to generate draft slot data based solely on the players drafted in each particular slot. With a sample of eight each successful player is going ot have much too big an impact on the final average or probabilities. For example, Bagwell was drafted #109 and has produced over 120 WARP3 in his career. I can’t say that there is a 12.5% chance to draft a Bagwell-like player at #109. or that the average WARP3 return is 15 (plus whatever contribution from the other seven players drafted at #109).

A simple way to boost the sample is to use a rolling average. I’m now defining the value of the 109th pick not by players drafted 109th exclusively, but by the group of players drafted just before and after slot #109. This process really smoothes out the slot to slot variability a great deal and because there is very little slot to slot differentiation in the draft I don’t think I’m loosing any significant granularity in the data.


So the next technical question is what size group should I use and should that group size change. For the vast majority of the draft there is so little differentiation that I could have picked very large group sizes for the rolling average. For example, I don’t think pick #50 is much different than pick #70. I do think that there’s enough differentiation at the very top of the draft that it’s important to choose as small a group as is reasonably possible.

I decided on a group size of seven for most of the first round. That means the probabilities for slot #10 are actually based on picks from slots #7-13 from the eight drafts in my study. That yields a fairly large sample of 56 players per draft slot. I considered using a group size of 5 for the first half of the draft, but I wanted to minimize the number of arbitrary sample size changes as much as possible and from looking at the players in these slots I don’t think I’m missing anything meaningful differences by using a slightly larger group size.

The first overall pick is so different from the rest of the first round that I couldn’t include it in any of the rolling averages. That means the sample size for slot #1 is just the eight #1 overall picks and slots #2-4 are groped together creating a sample size of 24. I also increased the group size twice after the first round. A full description of the group and sample sizes is in the next table.
   Picks      Group Size     Sample
        1           1            8 
      2-4           3           24
     5-30           7           56
   31-100           9           72
  101-260          11           88

I’m not thrilled with these different sample sizes, but the reasoning behind the decisions is sound and it does improve the meaningfulness of the data.

Let’s take a quick look back at slot #109 to see how the number of drafts in a study and/or the methodology dramatically changes the expected probability of finding a Bagwell-like player.
#Drafts      Source              Probability
    8        me-straight ave        12.5%
   16        BP (2005)               6.3
  ~25        James (early 80s)       4.0
   41        all drafts              2.4
    8        me-rolling ave          1.1        

Obviously as you increase the number of drafts in a given study the probability of finding a Bagwell-like player goes down. I think the “true” probability of finding a Bagwell-like player at slot #109 should be very close to zero. By doing a rolling average I can get pretty close to zero with just eight drafts of data.

With that way too long pre-amble let’s move onto the data.

First round
Pick   Player       WARP3 Ave   %<20    %20    %40    %60+
   1    Upton          57.9        25     25     13     38 
   2    Gordon         19.4        54     33     13      0
   3    Clement        19.4        54     33     13      0
   4    Zimmerman      19.4        54     33     13      0  
   5    Braun          16.2        66     20     11      4
   6    Romero         16.3        68     16     11      5
   7    Tulowitzki     16.8        70     13     11      7
   8    Townsend       14.6        75     11      7      7
   9    Pelfrey        17.4        70     13      9      9
  10    Maybin         19.4        64     18      9      9
  11    McCutchen      20.5        63     16     13      9
  12    Bruce          20.5        64     14     13      9
  13    Snyder         20.1        64     14     13      9
  14    Crowe          19.6        64     14     14      7
  15    Broadway       19.6        64     14     14      7
  16    Volstad        17.4        68     14     13      5
  17    Henry          16.6        70     13     13      5
  18    Carillo        12.7        75     14      7      4
  19    Mayberry       14.5        71     18      5      5
  20    Pawelek        16.2        70     16      9      5
  21    Pennington     14.3        73     14      7      5
  22    Thompson       15.5        71     14      7      7
  23    Ellsbury       15.1        73     14      5      7
  24    Bogusevic      14.0        77     11      7      5
  25    Garza          15.4        71     14      9      5
  26    Hansen         12.3        79      9      9      4
  27    Devine         11.2        79     11      7      4
  28    Rasmus         10.4        80     11      5      4
  29    Marceaux        9.3        82     11      5      2
  30    Greene          8.9        82      9      7      2

I’ve never liked draft studies that state simple draft slot averages like these WARP3 averages because the average at every slot after #1 is a barely useful player. In fact, these data show only three slots (#11-13) with WARP3 averages over 20. That means a player like Rico Brogna with a 22.0 WARP3 career represents an above average return on every draft slot. That’s technically true, but it doesn’t really tell us anything useful. Teams don’t pay their 1st rd picks seven figure bonuses hoping to get someone nearly as good as Brogna.

I think it’s much more useful to establish some reasonable levels of production and then try to determine the probability that a draft slot will produce those types of careers. That draft slot #10 has an average return of 19.4 WARP3 isn’t meaningful to me. I’d much rather know that draft slot #10 has a 64% chance of not producing anything “useful”, an 18% chance of producing a “useful” player, a 9% chance of producing a “good” player and a 9% chance of producing a very good to great player. That may average down to 19.4 WARP3, but it’s a fuller more meaningful picture of what draft slot #10 can produce.

The first thing that jumps out from this table is the huge difference between the first pick and the second pick. At least in this specific time period, the first overall pick includes some once in a generation talents (Griffey, ARod) and the 2nd pick is really just another pick in the top half of the first round.

There are two interesting observations from the %<20 column. One, even premium top 5 picks are less than 50/50 bets to reach 20 WARP3 in their careers. And I think this is a nice example of why having specific names attached to these slots provides some very useful perspective. Alex Gordon and Ryan Zimmerman are already consensus top 20 prospects in the game. Clement and Tulowitski have also received considerable hype in their first prospect season as professionals. Based on their draft pedigree they are perhaps much riskier than the new prospect buzz they’ve received. Of course, that assumes prospects with longer track records have higher success rates. They may not. The second thing to note is how quickly the percentage rises. In just 29 slots from #2 to #30 the percentage of “failures” rises from 54% to 82% and as we’ll see we’re not far from 90+%.

The %20 (“useful” players) column isn’t very revealing. The top 5 slots are at least 20%, but the next 20 or so spots are fairly consistent bouncing around 15%. There is a second drop to ~10% for the last five picks of the first round. All of these outcomes represent positive return on investment, but there isn’t much differentiation.

The %40 (“good” players) column is pretty consistent at 13% right down to pick #17. There’s a little blip at #8-10, but that’s somewhat offset by an increase in %60+ and just in general with these low success rates all of the numbers are going to be prone to bouncing around a bit. Obviously, there’s no good reason for their to be such a significant difference between pick #17 and pick #18, but the drop that happens to occur there is substantial and it persists for the rest of the draft. This is the beginning of a steady decline in the percentage of “good” players.

There are a couple significant blips in the %60+ column. I don’t think these changes are very meaningful. It just happens that in this eight year period there weren’t any very good players drafted in the #2-4 slots. There happened to have been a slight concentration of very good players in the #7-15 range. I think you can mentally smooth out the blips and reasonably think about it as a ~7% rate for the first half of the round and ~5% rate for most of the back half of the round. The difference between the first and second halves of the draft is actually bigger in the %40 column (11.5 vs 7.7%) than in the %60+ column (5.9 vs 4.6%).

These probabilities are generated from whatever types of player happened to be drafted in the appropriate slots during this eight-year period. However, we wouldn’t really expect a HS pitcher drafted 17th to have the same success probabilities as a C hitter drafted 17th. I don’t have a big enough sample to break things down quite that finely, but I can look at the overall first round success rates for HS pitchers and hitters and C pitchers and hitters. Please note that this table assumes that each of these classes are evenly distributed in the first round, but that may not be the case.

Type        #players    %20    %40    %60+    (%40+)
HS pos          58       17     14     12       26
  C pos          47       19      9     13       21
  C pit          67       12     10      2       12
HS pit          34       12      3      0        3

The two groups of position players were much more successful than the two groups of pitchers. The group of HS position players produced the highest percentage of “good” players. C pitchers were the most frequent type of player drafted, but they succeeded at a rate far less than that of both groups of position players. HS pitchers barely registered any 40+ WARP3 successes.

Also note that position players have much greater career production upside than pitchers. There were a total of 101 1st round pitchers and only one (Mike Mussina) exceeded 60 WARP3. In contrast, roughly the same number of drafted position players (105) produced thirteen 60+ WARP3 players. (This is true looking at the entire draft as well. There are six 60+ WARP3 pitchers from this period and thirty-five 60+ WARP3 position players.)

With that information in the background we can look more closely at some of the specific players from the last June draft. There were three HS pitchers selected – Volstad, Pawelek and Thompson. It’s likely that their specific success probabilities are lower then their generic rates. We can also make a reasonable mental adjustment to reduce the %60+ probabilities for all pitchers. And conversely, we can give slight adjustments up for the position players. I don’t think you can quantify these little adjustments, but it’s good to keep them in the back of your head.

A closer look at the Sox two first round picks brings up some interesting points. Prior to the draft every perceived college oriented team was rumored to be interested in Trevor Crowe and Jacoby Ellsbury. Both are polished college centerfielders that project to be top of the order hitters. BA ranked them very similarly at #25 and #29. We know that the Indians valued Crowe over Ellsbury, but it’s quite possible that the Sox or some other team would have taken Ellsbury over Crowe if given the opportunity. If we had access to all 30 team’s draft board I doubt we’d see much of a consensus opinion about these two similar players. And yet the historical record will show that Crowe was taken a seemingly significant nine spots ahead of Ellsbury. That fairly large gap – even in the middle of the 1st round where we might expect slot to slot differentiation - doesn’t really reflect the relative merit of the two players. (And it’s examples like this that convince me that large rolling averages even in the first round don’t cover up meaningful distinctions.)

However, because Crowe was picked ahead of the #17/18 breakpoint and Ellsbury was picked after it, it appears that Crowe has a much greater probability of becoming a good player. I don’t think it makes sense to think that this type of draft slot analysis should outweigh the known pre-draft similarity of two players. Don’t take the generally small differences in these numbers too seriously. These two players are closer than the following numbers suggest. In fact, based on Boston hype and a couple hundred AB many people probably think of Ellsbury as the better prospect.

Pick   Player       WARP3 Ave   %<20    %20    %40    %60+
  14    Crowe          19.6        64     14     14      7
  23    Ellsbury       15.1        73     14      5      7

Hansen is interesting because his signability dramatically changed his draft position. I can establish reasonable success probabilities based on the guaranteed money he received (4th in the draft), by his pre-draft ranking (8th by BA excluding Drew and Weaver) or by his actual draft slot. For 99% of the players in the draft we can use the player’s actual draft slot because it is a good proxy for the other two methods. Unfortunately, players that slip for signability reasons are a major pain because it is reasonable to look at them in all three ways. For Hansen that looks like this:
Pick   Method       WARP3 Ave   %<20    %20    %40    %60+
   4    cash           21.9        54     33     13      0  
   8    pre-draft      14.6        75     11      7      7
  26    slot           12.3        79      9      9      4

Each of those lines is obviously quite a bit different. The cash method suggests a much higher probability to at least exceed the 20 WARP3 threshold and the slot method suggests a much lower probability to become a %60+ player (note: I’m treating the 0% for pick #4 as random noise). For large studies I think you have to just leave these players in their slots and note when you can that they may not really be representative of that slot. When you try to look at an individual player though, you have to consider all of the possibilities and subjectively squeeze that player somewhere into that range.

As it turns out the pick 4/cash line is a pretty reasonable way to look at a college reliever like Hansen. The Sox are banking on a very high probability (and in the draft that means ~50%) that he will be at least “useful”, but the chances of any college reliever exceeding 60 WARP3 are pretty slim. That line actually does a nice job representing the pluses and minuses of drafting Hansen.

First round – supplemental

It was a large supplemental round last year with eighteen picks and my league drafted a little more than half including all three Sox players.
Pick   Player       WARP3 Ave   %<20    %20    %40    %60+
  31    Torra           8.9        82     10      6      3
  33    Drennen         6.2        88      7      3      3
  36    Buck            4.5        90      7      1      1
  38    Iorg            3.5        93      6      0      1
  39    Sanchez         3.1        93      6      0      1
  41    Jones           2.4        96      3      1      0
  42    Buchholz        3.6        94      3      1      1
  43    McCormick       4.8        93      3      1      3
  45    Lowrie          5.7        92      4      1      3
  47    Bowden          5.4        92      4      1      3

The WARP3 average continues to plummet into the low single digits. For the rest of the draft it’ll bounce around between 0-5 WARP3 with the baseline steadily decreasing. On average, no pick after the first round is worth much of anything in terms of career production. The %<20 jumps over 90% by the mid-30s and it gradually increases over time. In terms of “good” players there’s a trough between #36 and #42 (happens to be Buchholz), but that’s just randomness. If a player isn’t picked in the 1st rd his chances of being a “useful” or “good” player are never higher than 10% and 5% respectively.

Lowrie is an interesting player because there’s a bit of a perception that he was underdrafted and that perhaps he has already answered some of the questions about him. He reportedly played a surprisingly solid shortstop and even if he still projects as a 2B that represents a nice improvement in his toolbox. And more importantly, he hit well with wooden bats perhaps quieting some of those “aluminum bat hitter” concerns. How much should 200 AB in the NYPL change our perceptions about a very recent draft pick? In my opinion not very much at all, but it’s possible that with an improved defensive profile and some solid wood bat production under his belt he would have been a late 1st rd pick. Just for the hell of it how would his success probability profile change if he went at #25 instead of #45?

Pick   Player       WARP3 Ave   %<20    %20    %40    %60+
  25    Garza          15.4        71     14      9      5
  45    Lowrie          5.7        92      4      1      3

The dropoff in this range is so steep that a twenty spot improvement makes a huge difference. I don’t think you can let 200 good NYPL ABs change your perception of a player that much, but I suppose you can start to build a case that a player is better (or worse) than his draft position suggests with each professional game.

Ideally it would be possible to find a large group of college position players whose draft status gives them very similar probabilities of long-term success. That group would include some high and low performers in their first abbreviated professional season. Would the probability of success for the two groups differ from each other and/or from their baseline draft slot status? Beats me. I would guess that the signal would be so small that it would be quite difficult to pick up a meaningful difference. If you could though, it would probably be more likely for college position players. I think that the unpredictability for pitchers (both HS and C) is so great that it would be very unlikely that a half season of performance data would make any difference at all. That’s even more true now that teams like the Sox severely limit the innings for most high round pitchers. I know of enough anecdotal examples of very good HS position players that stunk in the GCL that I doubt GCL performance of a teenage position player would change their success probabilities very much either.

From eyeballing my data I consider the 1st supplemental round through the 3rd round to be the next tier of draft picks. Here’s a table with the success rates of the various types of players from these rounds. I added the %40+ data from the first round table just to make the steep decline clear.

Type        #players    %20    %40    %60+    (%40+)   (1st rd %40+)
  C pos         101        6      1      5        6          21
HS pos         131        2      5      0        5          26
  C pit         163        4      0      2        2          12
HS pit          92        7      0      0        0           3

Again, the two groups of position players are at the top of the list although in these rounds the C hitters edge the HS hitters. HS pitchers are again the least successful group although I expect a couple of the %20 pitchers to clear 40 WARP3 and that will let them catch up a bit to the C pitcher group. Nevertheless, a HS pitcher like Bowden has some very long odds to overcome. There aren’t enough JC picks at the top of the draft to say much of anything about the historical chances of a player like Buchholz.

Second Round

We drafted about a third of the 2nd round though not Egan who slots in at #57. Reimold and Escobar received the most buzz from this group of players, but most of these names will be recognizable to people who follow the minors or the draft closely.
Pick   Player       WARP3 Ave   %<20    %20    %40    %60+
  50    Bianchi         4.7        93      4      0      3
  52    Carte           2.8        94      6      0      0
  53    Italiano        2.4        94      6      0      0
  56    Mason           2.9        96      3      0      1
  60    Wood            4.6        92      6      1      1
  61    Reimold         4.4        93      4      1      1
  62    Head            4.4        93      4      1      1
  65    Costanzo        4.0        93      4      1      1
  66    Headley         3.8        94      4      0      1
  67    Whittleman      4.3        94      4      0      1
  75    Escobar         6.1        92      3      1      4

The most obvious characteristic of this round to me is its flatness. The WARP3 averages are between 2.5-4.5, the %<20 bounces between 92-94, the %20 is around 4% and the combined %40+ is around 2%. There aren’t many data points that fall outside of those boundaries. Slot #75 is an exception, but a lot of that is because John Olerud, a top of the draft talent that fell to the 3rd rd due to health/signability issues, happens to be included in the slot 75 range. If I dropped him the slot 75 data would fall right in line with the rest of the second round.

By the time we get to the 2nd rd the chance of any player becoming a “good” 40+ WARP3 player is roughly 1 in 50.

BA just released its Top 100 list and it includes a pull quote for each player. The quote for Reimold is quite timely.

"He's got five plus tools and he has a little bit more ability to use them than maybe we knew. He was a good first-round talent that we got in the second round. We're fortunate--really fortunate."
--Orioles scouting director Joe Jordan


He’s making the case for Reimold that I suggested might be possible for Lowrie. While Reimold may have profiled as a solid second round pick on draft day, now that Jordan has seen him play professionally for a few months he believes that Reimold’s tools play more like a good 1st round pick. A comparison of Reimold’s draft slot and what Jordan now thinks he profiles as would look something like the following set of numbers.
Pick   Player       WARP3 Ave   %<20    %20    %40    %60+
  20    Pawelek        16.2        70     16      9      5
  61    Reimold         4.4        93      4      1      1

In Jordan’s opinion, Reimold may have been a player with a 2% chance to exceed 40 WARP3 in June, but he’s now better characterized as a player with a 14% chance to exceed that level of productivity. And it seems like BA, which ranked him above ~40 players that were drafted above him, agrees.

A brief BA Top 100 sidebar smack dab in the middle of the post

I’ve never looked at June draft picks in the BA Top 100, but since the list came out while I was writing this I thought it might be interesting to take a quick look at the June draft picks that are considered Top 100 prospects prior to their first full professional season.

Player          BA   Slot    %40   %60+    %40+
Upton            2     1      13    38      51
Gordon          13     2      13     0      13
Zimmerman       15     4      13     0      13
Tulowitski      25     7      11     7      18
Maybin          31    10       9     9      18
Clement         33     3      13     0      13
Pelfrey         36     9       9     9      18
Braun           49     5      11     4      15
McCutchen       50    11      13     9      22
Hansen          54    “8”      7     7      14
Bruce           76    12      13     9      22
Pennington      83    21       7     5      12
Pawelek         85    20       9     5      14
Romero          87     6      11     5      16
Carillo         88    18       7     4      11
Volstad         97    16      13     5      18
Reimold         99    61       1     1       2

There are seventeen June picks on the list. I have no idea if that’s high, low or average in comparison to other years. The players that made it match pretty well with their draft position. Sixteen of the players were taken within the first 21 picks (I listed Hansen as slot “8” to better reflect his pre-draft status). The lone notable exception is the previously mentioned Reimold.

This suggests both that the brief on field performances didn’t change their status very much and that BA believes that the top prospects in the June draft have very comparable success probabilities to more experienced minor leaguers with similar rankings. For example, Upton with zero professional experience ranks right between Delmon Young and Brandon Wood with two plus years of professional experience. Alex Gordon (no professional experience outside of the AFL) and Ryan Zimmerman (308 pro AB including a great MLB trial) are ranked alongside fellow 3B Andy Marte with 4+ years of professional experience. My probabilities for Gordon’s and Zimmerman’s draft slots are unfortunately low, but if we give them a more realistic 20% chance of being 40+ WARP3 players based solely on their draft status (and for Gordon that’s all we have), then that suggests that Marte’s chance to be that good should be about the same if BA did a good job of making their list. I’ve never looked closely enough out old BA lists to have a feel for how accurate that might be, but it would be an interesting way to judge how successful BA (or any group with a list) is at merging inexperienced players in with established prospects with professional track records.

Also note that BA seemed to do a pretty good job reflecting the greater risk for recently drafted pitchers. There are ten June draft picks within the top 54, but just two are pitchers. And they were the highest rated C pitchers to sign. The other four June pitchers (two HS, two C) that made the list are clustered between #85-97.


Back to the Third Round to wrap this part up

We only drafted a handful of players from the third round.
Pick   Player       WARP3 Ave   %<20    %20    %40    %60+ 
  81    Neighborgall    4.9        92      6      1      1
  83    Owings          5.5        90      7      1      1
  85    Inman           3.6        93      6      1      0
  93    Erbe            2.3        99      0      0      1
  99    Teagarden       1.3        97      3      0      0

The probabilities in the 3rd round are pretty similar to the second round with a slight downward trend. The jump to 97-99% in the %<20 column for slots #93 and 97 is just a random fluctuation. That percentage mostly stays in the 93-94% range until around pick #225 when it jumps into the high 90s for good.

#2 philly sox fan


  • SoSH Member


  • 9,748 posts

Posted 04 March 2006 - 12:00 PM

Part III – Who cares about generalities what about the Sox?

Using this data I can look at the success probabilities of a single team’s draft class through about the 250th pick (that’s the eighth round these days). I’ve added two additional financial columns to the next set of tables. The first is Cost and that’s simply the signing bonus each player received. The second is Value and that’s a little more complicated. You might recall that last summer Nate Silver wrote an article on valuing draft picks. However, he made a couple significant (imo) mistakes. The first was that the formula he used to convert WARP production into a valuation overvalued very low WARP players. He subsequently corrected that formula in an article about the free agent market and I then re-applied that change to his method on valuing draft picks. That article (with the links to the original Silver article) can be found here:

http://sonsofsamhorn...?showtopic=1443

The second mistake that Silver made is that he based his valuations on the average return for each draft slot. As I’ve stated I don’t think those averages are very meaningful so to me any valuation based on an average return just isn’t very useful. Instead I based these values on the probabilities of achieving different levels of career production. Using the method that Silver outlined in his original article, I determined a value for five different levels of career production. I then used the specific draft slot probabilities to achieve that level of production to find a value for each potential production level at that specific probability. The value of each draft slot is then just the sum of the individual values for each level of production. That’s probably not entirely clear so let me go through a couple examples.

Value determination for slot 10
Potential Production   Val of Prod     Probability     Value
     20-40 WARP3           7.6M           17.9%         1.360M
     40-60                14.2             8.9          1.264
     60-80                31.7             1.8          0.571
     80-100               34.0             3.6          1.224
     100+                 55.2             1.8          1.987
                                               Total:   6.406  vs 3.590

For each Potential Production I determined the average WARP production during the six pre-free agency years. I used Silver’s valuation method in the Valuing Draft Picks article to determine the Value of Production for each level. A “useful” 20-40 WARP3 player will produce about 7.5M of value on average and the values increase up to a great 100+ WARP3 player that will produce roughly 55M of value. I then multiply the Value of Production by the specific probabilities for each draft slot. An 18% chance of drafting a “useful” player that is worth 7.6M has a Value of 1.360M, a 9% chance of drafting a 40-60 WARP3 player is worth 1.264M, etc. I then sum the five individual values for a total value, which in this case is just over 6.4M.

If I had simply taken the average production for slot #10 and applied Silver’s method with his revised formula the calculated value would be just under 3.6M. That’s quite a large difference. Although I think the probability method makes more sense than the average production method, when I first looked at the two sets of very different numbers for the early part of the first round I was a bit troubled because I didn’t have an intuitive feel for which set of numbers made more sense. If anything, the 3.6M value seemed more accurate than 6.4M for slot #10. And yes, that was pretty annoying. But let’s look a little further down the draft.

Value determination for slot 30
Potential Production   Val of Prod     Probability     Value
     20-40 WARP3           8.5M            8.9%         0.757
     40-60                15.1             7.1          1.072
     60-80                32.6             1.8          0.587
     80-100               34.9             0.0          0.0
     100+                 56.1             0.0          0.0
                                               Total:   2.415  vs –0.256

One quick note – the Values of Production in this table are slightly different than the Values from the previous table. That’s because Silver’s valuation method includes the cost of the signing bonus. In the 2005 draft, slot #10 cost 2M and slot #30 cost 1.1M. Each set of values is customized for the specific draft slot.

At slot #30 the success probabilities are lower across the board including two 0% probabilities for the most valuable careers. As a result the total value of 2.415M is much lower than the calculated value for slot #10. As you might expect, the 80-100 and 100+ WARP3 categories really drive the total value. A lot of the slot to slot variability comes directly from those probabilities bouncing between 0-3%. Silver’s method produces a value of negative 0.256M. In fact, nearly every draft pick after the first round yields a negative value. The reason that happens is that the average yearly WARP production for these slots is in the 0-1.5 WARP range and those seasons basically have no value. The return on average production method attempts to put a six year value on players who, in the real world, would be quickly replaced for lack of production. Their WARP based valuation is below the minimum MLB salary. As a result you get nonsensical results like the 40th overall pick in the draft is “worth” –1.5M. Teams that lose free agents certainly don’t treat supplemental picks like money pits.

These negative valuations for fairly high draft picks is a result that only Brian Sabean could love. In the past I’ve mildly defended Sabean’s decision to forfeit low 1st rd draft picks for MLB talent so I can definitely buy the idea that picks after the first round have very little value (and the probability values that I’ve generated really don’t disagree), but I know that these draft picks have some probability of producing players worth 7-55M and that has to have some positive value. And that’s the beauty of looking at probabilities. As long as there is some demonstrable chance to draft a good to great player, then there is going to be some quantifiable net positive value. If not, then the 60+M MLB as an industry spends on signing bonuses past the 1st round is completely wasted. The negative valuations based on average draft slot production suggest that teams should pocket the money and stop drafting after round one. That just doesn’t make sense.

These valuations are based on the cost of WARP production in the free agent market. I think it’s important to keep that context in mind. What does 5-6M (the actual 6.4M value for the 10th slot is a little high) or 2-3M actually buy in the free agent market? Here are the players that signed this winter for contracts with AAVs in those ranges.

Players with AAVs roughly comparable to pick from the top of the first round
Player             AAV
Kyle Farnsworth    5.7M
Todd Jones         5.5 
Jaime Moyer        5.5
Jacque Jones       5.3
Juan Encarnacion   5.0
Bob Wickman        5.0
Reggie Sanders     5.0

Outside of Moyer these are remarkably similar players – second tier closers and decent corner outfielders. This reduced my initial qualms about the size of my first round valuations quite a bit. Five million dollars *is* a huge sum of money, but in the post free agency service time baseball talent market its purchasing power isn’t all that impressive. And as the valuations quickly fall off that point becomes even stronger.

Players with AAVs roughly comparable to very late 1st round picks
Player             AAV
Roberto Hernandez  2.75
Jose Mesa          2.52
Josh Byrnes        2.5
Neifi Perez        2.5
Ramon Ortiz        2.5

In any other context these are large sums of money, but in this context it’s merely buying role players on one year contracts in the free agent market. My second round valuations average 1M and the players who actually sign for 1M are afterthought players like John Mabry, Joe Mays and Sidney Ponson. After putting the valuations back into the free agent market context, then they don’t seem quite as large as they did (to me at least).

That’s my case for valuations based on draft slot probabilities of success. Before I write up the final report I’d love a little feedback if you think that I’ve completely missed the boat somehow.

One last note - I went through each of the steps in Silver’s valuation method more explicitly in a post a couple months back titled Valuing Marte. BTF poster Darren pointed out that some of the expected yearly salaries where much too high. That is an issue for the high end career production levels. I made a correction and re-calculated the Values for slots #10 and #30. The corrected Value for slot #10 would be ~450k higher and for slot #30 it would be ~50k higher. At least for slots with measured probabilities to reach 80-100 or 100+ WARP3 the difference is big enough that I will go back and re-do all of these calclulations.

Anyway, let’s look at the Sox 2005 draft class and their expected 2006 draft class through the 8th rd to get a sense of how these probabilities and valuations really work on a practical team level.

2005 Draft class
Rd    Pick   Player       Cost   Value   WARP3 Ave   %<20   %20    %40   %60+
  1     23    Ellsbury     1.4      5.1     15.1        73    14      5     7
  1     26    Hansen       1.325    3.2     12.3        79     9      9     4
  1s    42    Buchholz     0.800    0.9      3.6        94     3      1     1
  1s    45    Lowrie       0.763    1.5      5.7        92     4      1     3
  1s    47    Bowden       0.730    1.5      5.4        92     4      1     3
  2     57    Egan         0.625    1.0      3.4        94     3      1     1
  3    108    no pick               1.2      3.8        93     5      1     1
  4    138    Blue         0.075    0.6      2.1        95     2      2     0
  5    168    Engel        0.154    0.8      2.8        94     5      0     1
  6    198    Corsaletti   0.050    0.7      2.1        95     3      0     1
  7    228    Yema         0.090    0.1      1.0        97     3      0     0
  8    258    Zink         0.075    0.2      1.3        97     1      2     0 
Tot                       6.087   15.6     54.8              

Due to the loss and gain of various free agents the Sox added a late 1st rd pick, three 1supp picks and upgraded their 2nd rd pick and they lost their 3rd rd pick. They spent a little over 6M on signing bonuses (this doesn’t count ~2.7M in guaranteed salary to Hansen). By my method of calculating Value these picks are worth 15.6M with half of that concentrated just in the two late 1st rd picks.

The individual average WARP3 production for each slot is quite low. I’ve seen the argument made that as long as a team exceeds the average return for it’s top draft picks, then that team had an above average draft. That’s not true in any meaningful way. You can add one WARP3 to each of these slots and you wouldn’t have eleven above average picks. You’d have a couple backups and a bunch of players that can tell their kids about their cups of coffee in the big leagues. That’s pretty good for the drafted players (fulfilling their dreams to be in the majors and all that), but teams don’t want to draft eleven players that add up to 50 WARP3. Teams want to draft one 50 WARP3 player. That would be a good draft. As with so much in baseball the distribution of the production is more important than just the straight sum of individual components.

You can see little fluctuations in both the Value and the probabilities. When I write up the final version I think I’m going to have to include round averages after the first round. From looking at the full data set, pick #42 is low and pick #228 is low (though it really doesn’t matter that far down). But overall, this is a pretty good template of what a draft looks like through the 8th rd. At the end of the first round there are relatively high valuations and a greater than 10% chance to become a good player. By the supplemental round and picks in the 40s the valuation drops to 1.5M and the chance to draft a “good” 40+ WARP3 player dips below 5%. After that the valuations hold steady for a few rounds at roughly 700k-1M with a 2% chance to draft a “good” player and by the 8th round the picks have negligible value and a very low probability to become a “good” player.

It can be fun to look at the specific names and speculate a little about whether the player has materially improved or decreased his probability to succeed. We don’t have to guess about Hansen because we know that his draft position doesn’t reflect his perceived potential. On the downside we also don’t have to guess about how post-draft issues affected Scott Blue’s perceived potential. His expected slot bonus was ~225k, but the shoulder injury that was diagnosed in his physical dropped his bonus to 75k. In theory, a HS pitcher with a shoulder injury should have an even lower probability to succeed than whatever slot he was drafted in.

2006 Draft Pick Compensation

Now that all of the free agent compensation picks have been settled I can set up the same table for the Sox 2006 draft. First, let’s look at the value of the compensation picks the Sox received.

Both Damon and Mueller were Type A free agents. The Sox received a late 1st rd pick (#28) and a supplemental pick (#40) for Damon. The Sox received a supplemental pick (#44) and a 3rd rd pick (#83) for Mueller because the Dodgers 1st rd pick was protected and Mueller was the second highest rated free agent the Dodgers signed. At the time Mueller signed I mentioned that the Sox lost a tremendous amount of value dropping from a late 1st rd pick to a 3rd rd pick. Now we can do a more specific comparison.

At the start of the free agent signing period the Twins were rumored to be interested in Mueller. They were an ideal team to sign him because their first round pick (#20) was not protected and they were not likely to sign anyone better (ie more expensive) than Mueller. The following table is a comparison between pick #20 and pick #83.
Pick   Player       Cost   Value   WARP3 Ave   %<20   %20    %40   %60+
  20    Pawelek      1.75     5.2     16.2        70    16      9     5
  83    Owings       0.440    1.6      5.5        90     7      1     1

The Sox will save over 1M in signing bonus, but by my calculation they will have lost over 3.5M in Value (note: the 1.6M Value for slot #83 should be closer to 1M). More importantly, the probability of getting a useful player decreases from 30% to 10% and the probability of getting a “good” 40+ WARP3 player decreases from 14% to 2%. There are bigger problems with the free agent compensation system, but the fact that the amount of compensation can change so much on the basis of which team signs your free agent is definitely a source of unfairness.

The next table is the expected return and value of the picks the Sox actually received for Damon and Mueller.
Damon
Pick   Player       Cost   Value   WARP3 Ave   %<20   %20    %40   %60+
  28    Rasmus       1.0      2.9     10.4        80    11      5     4
  40    Hochevar     0.850    0.6      2.5        94     4      1     0

Mueller
Pick   Player       Cost   Value   WARP3 Ave   %<20   %20    %40   %60+
  44    West         0.775    1.3      4.7        94     1      1     3
  83    Owings       0.440    1.6      5.5        90     7      1     1

As compensation for losing Damon the Sox received a pick with a 9% chance to be a “good” player and a pick with a 1% chance to be a “good” player. The Value listed for pick #40 is quite a bit lower than it should be. I’d put the value of these two picks closer to 4.5M.

As compensation for losing Mueller the Sox received a pick with a 4% chance to be a “good” player and a pick with a 2% chance to be a “good” player. Since the Value listed for pick #83 is higher than it probably should be, I’d put the value of the two picks closer to 2.5M.

On one level it makes sense that the Sox should receive more compensation for the player with a market value of 52M than for the player with a market value of 9.5M, but as long as MLB considers them equivalent it would be nice if their respective compensation was more equitable.

2006 Draft Class

Rd    Pick   Player       Cost   Value   WARP3 Ave   %<20   %20    %40   %60+
  1     27    Devine       1.3      3.1     11.2        79    11      7     4
  1     28    Rasmus       1.0      2.9     10.4        80    11      5     4
  1s    40    Hochevar     0.850    0.6      2.5        94     4      1     0
  1s    44    West         0.775    1.3      4.7        94     1      1     3
  2     71    Phillips     0.505    1.4      5.1        93     3      1     3
  3     83    Owings       0.440    1.6      5.5        90     7      1     1
  3    103    O’Sullivan   0.375    0.2      0.9        98     2      0     0
  4    133    Matusz       0.220    0.8      2.3        98     0      0     2
  5    163    Mendoza      0.159    0.8      2.9        92     7      1     0
  6    193    Moore        0.100    0.1      0.9        99     1      0     0
  7    223    Romero       0.110    0.1      0.7        99     1      0     0
  8    253    Hall         0.075    0.2      1.0        98     1      1     0
Tot                       5.909   13.1     48.1              
The expected 2006 draft class looks pretty similar to the 2005 draft class. The Sox have one fewer supplemental pick and two additional 3rd rd picks. There’s enough clustering of good players at the top of the draft that that works out to a slightly lower expected value for 2006, but the presence of Hochevar in the supplemental round is a clear sign that a large revenue team like the Sox can always make more of these picks than a simple slot based analysis suggests.

The Sox have really done an excellent job of sloughing off free agents to reap extra draft picks while maintaining a very good major league roster. In the four years that the current ownership group has managed this process the Sox will have had extra high round picks in three drafts (2003, 2005, 2006). The one year they kept the roster together and actually had a net loss of top picks was 2004 the year they won the championship. And though it’s barely worth mentioning in comparison to the WS win, even that year they spent their 1st rd slot money on Mike Rozier.

Fans and players alike have complained about losing this or that free agent, but that’s basically a perfect overall track record. The Sox have either shed free agents to accumulate extra draft picks while maintaining a playoff caliber roster (hopefully still true about the 2006 team) or they’ve kept the roster intact and won the WS. It’s hard to complain too much about that.

OK, you can say they woulda, coulda, shoulda kept the roster together and won the WS every year. But who would do a thing like that?

That’s a pretty good overview of my draft research at this point. Hopefully, actually finishing this will kick me in the butt and get to me start knocking out some of the specific draft reviews that will fill out the details behind these findings.

#3 vasoxfan

  • 27 posts

Posted 04 March 2006 - 01:52 PM

I decided on a group size of seven for most of the first round.  That means the probabilities for slot #10 are actually based on picks from slots #7-13 from the eight drafts in my study.  That yields a fairly large sample of 56 players per draft slot. 

<{POST_SNAPBACK}>


This is way too much to fully digest on a Saturday . . .

In any event, could you better explain how you performed your groupings? I understand pick #10, and I also understand pick #1 and picks 2-4 (these are essentially one group as the data is exactly the same). But I am confused over the construction of groups for slots 5-8 and 27-30. In particular, #10 was constructed as the median pick within its group (7-13). If you are using data only within that section of picks (5-30) you can't use the median method with pick 5 because it is at an endpoint. So is it the case that pick 5 is constructed using picks 5-11? What about pick 6? The WARP3 average for pick 6 differs from pick 5, so the pick 6 group cannot reflect picks 5-11. Rather, does slot #6 reflect data from picks 6-12? Similarly, could you explain the construction of picks at the end of the 1st round?

#4 OCD SS


  • SoSH Member


  • 6,902 posts

Posted 04 March 2006 - 02:25 PM

I have a question about your methodology:

What is the benefit of using Warp3 vs Warp1? Since all of the players in your study are in the same era, doesn't adjusting "for all time" apply an unnesecary distortion to the player's values (in the same way that Manny and Miggy's Warp1 totals are seperated by less than 1 warp, but this expands to over 6 in Miggy's favor in Warp3)? Or does this additional adjustment help seperate the discussed player's value from within very close sets of values within the data?

#5 DSG

  • 229 posts

Posted 04 March 2006 - 03:28 PM

Philly,

I'll be honest; I haven't had time to digest this all. But IMO you're on the right track. You should definitely value each player seperately, and use a non-linear formula. But here's what I would do: first off, find each player's "per season" production, i.e., if they have 50 WARP in 6,480 PAs, that's 5 WARP/season (648 PAs). Secondly, would you mind re-doing this analysis with my player value formula? The first, which I introduced a few months before Silver (and if you want my opinion, he saw it before developing his, so yes, I am a little bitter since he didn't even acknowledge previous work on this in his article), is Wins^2/6.25*1,000,000. Wins is simply Win Shares divided by 3 or WARP + 2 or pretty much WARP3, I think. You'd have to check that. The second is a bit more complex, and this is the first time I've presented it: Wins^2/28.481*1,000,000+1,750,000*(Wins - 3). Anything under 3 wins needs to simply be replaced with a 0, as it is lower than my replacement level, and therefore worth nothing. Here's a comparison of my two systems, Silver's, and a $2,000,000/marginal win system: http://www.hardballt...Value_Graph.GIF

Edited by DSG, 04 March 2006 - 03:28 PM.


#6 PedroSpecialK


  • Comes at you like a tornado of hair and the NHL salary cap


  • 17,080 posts

Posted 05 March 2006 - 03:40 PM

Superb work, Philly - I learned a lot from that post. It was really interesting to me to see how you evaluated players such as Hansen who should have been drafted a good amount higher but dropped due to signability issues (Mike Rozier is another, as well as Pedro Alvarez at Vanderbilt), and how this will/has affected the future data for drafts. It was also interesting to see the sharp decline in WARP3 between the 17th and 18th picks in the draft - I never knew that there was such a concentrated area of dropoff of talent.

Keep it up - a great read.

Edited by PedroSpecialK, 05 March 2006 - 03:41 PM.


#7 singaporesoxfan

  • 3,843 posts

Posted 06 March 2006 - 06:09 AM

Random question: given the clear difference between a #1 pick and the next few picks, does this mean teams who're doing badly have a very strong incentive to be absolutely horrific? (They don't alternate first picks between the AL and NL still, do they?) Or is that a function of the small sample?

#8 philly sox fan


  • SoSH Member


  • 9,748 posts

Posted 06 March 2006 - 07:38 AM

This is way too much to fully digest on a Saturday . . .

In any event, could you better explain how you performed your groupings? I understand pick #10, and I also understand pick #1 and picks 2-4 (these are essentially one group as the data is exactly the same). But I am confused over the construction of groups for slots 5-8 and 27-30. In particular, #10 was constructed as the median pick within its group (7-13). If you are using data only within that section of picks (5-30) you can't use the median method with pick 5 because it is at an endpoint. So is it the case that pick 5 is constructed using picks 5-11? What about pick 6? The WARP3 average for pick 6 differs from pick 5, so the pick 6 group cannot reflect picks 5-11. Rather, does slot #6 reflect data from picks 6-12? Similarly, could you explain the construction of picks at the end of the 1st round?

<{POST_SNAPBACK}>


Aside from the first 4 picks the slot is always in the middle of the group. PIck #5 is #2-9. Pick #6 is #3-10, etc. And that continues the whole draft. The end points for the first rd or after pick 100 don't make a difference. PIck #30 is #27-33 and pick 31 is #26-34.

It's maybe a bit odd that pick 31 includes a slot that precedes the slots that roll into pick #30, but the differences are all so small it's not a big deal.

#9 philly sox fan


  • SoSH Member


  • 9,748 posts

Posted 06 March 2006 - 07:41 AM

I have a question about your methodology:

What is the benefit of using Warp3 vs Warp1?  Since all of the players in your study are in the same era, doesn't adjusting "for all time" apply an unnesecary distortion to the player's values (in the same way that Manny and Miggy's Warp1 totals are seperated by less than 1 warp, but this expands to over 6 in Miggy's favor in Warp3)?  Or does this additional adjustment help seperate the discussed player's value from within very close sets of values within the data?

<{POST_SNAPBACK}>


Unfortuantely I chose WARP3 mostly out of laziness two years ago and I've just stuck with it. WARP3 corrects for playing time in the strike seasons of 94/95 and I didn't want my values to be a little low just because of the strike.

However, as I keep incorporating more and more recent drafts into the data set that effect gets much smaller. I should have picked WARP1 though. For the next iteration I'm going to have give some serious thought to re-doing everything in WARP1 or perhaps even going back to WS.

#10 philly sox fan


  • SoSH Member


  • 9,748 posts

Posted 06 March 2006 - 07:45 AM

Superb work, Philly - I learned a lot from that post. It was really interesting to me to see how you evaluated players such as Hansen who should have been drafted a good amount higher but dropped due to signability issues (Mike Rozier is another, as well as Pedro Alvarez at Vanderbilt), and how this will/has affected the future data for drafts.

<{POST_SNAPBACK}>


Rozier is actually a better example than Hansen because the cash method and pre-draft rank method were pretty close for Hansen, but they would be very far apart for Rozier.

Rozier got the 17th highest bonus in the draft, but he was rated as more of a 3rd-5th rd pick (say slot #90) and of course he was picked a couple hundred picks after my study ends. There's a huge difference between what the says paid him at #17 and what otehr teams seemed to think about him at ~#90. (actually that's pretty similar to the Mueller example I gave).

#11 philly sox fan


  • SoSH Member


  • 9,748 posts

Posted 06 March 2006 - 07:51 AM

Random question: given the clear difference between a #1 pick and the next few picks, does this mean teams who're doing badly have a very strong incentive to be absolutely horrific? (They don't alternate first picks between the AL and NL still, do they?) Or is that a function of the small sample?

<{POST_SNAPBACK}>


There is an element of my small sample. Two of my 8 #1 overall picks are Griffey and ARod (as opposed to Mark MErchant and Darren Dreifort at #2) and those kinds of players aren't 25% of #1 picks.

BP's draft study which covered 16 years also found a very large effect though so there is probably something to it.

The last couple of #2 picks - Alex Gordon and Rickie Weeks - have had almost as much buzz as the guys drafted at #1 in those years. And a couple years earlier the Pirates took the "safe" C pitcher Bullington over Upton. So it's possible that the #2 pick in these recent drafts will either close the gap with #1 or at least assert some difference from picks #3+.

DSG

The next round of posts aren't going to contain any of the Value stuff, but I will dig around through your methodology to see what difference it makes. I'm not sure I like the course flip flop from WARP to WS although I know from previously doing it that there isn't all that much difference.

#12 lonborgski

  • 250 posts

Posted 06 March 2006 - 03:46 PM

Josh Byrnes 2.5?

#13 bowiac


  • I've been living a lie.


  • 9,766 posts

Posted 06 March 2006 - 06:23 PM

The last couple of #2 picks - Alex Gordon and Rickie Weeks - have had almost as much buzz as the guys drafted at #1 in those years. 

<{POST_SNAPBACK}>


Not a big thing, but Alex Gordon's buzz is almost nonexistent compared to Justin Upton's.

Upton's buzz is A-Rod/Griffeyesque, as evidenced by his likely rating as the #2 prospect in baseball by BA, without even playing a game.

Alex Gordon might have #1 hype in other years, but Justin Upton was just on a whole other level.

#14 Eddie Jurak


  • Go Leafs Go


  • 9,025 posts

Posted 08 March 2006 - 10:47 AM

Excellent work, Philly.

I think you've laid the groundwork here to start answering some very interesting questions. Most interesting, to me, is that you can start to measure the importance of different factors that affect a team's draft.

It is obvious that each of the following factors will have an affect, but less obvious (without your work) how much of an effect:

1. Draft position. The keys here are who has the early picks - especially the #1 - and who has gained or lost picks by signing or losing free agents. The work you've done here will allow you to rate each team's draft position, and measure how much better one team's position is than another, before the draft even happens. That's essential if you want to find out which teams draft better than other teams.

2. Money. Of course, it's not all about draft position. The financial approach teams take to the draft is also going to be crucial. If a team offered 'slot money' to each of its picks, then the value of its draft would basically be a function of draft position only. A team that offered below slot money would wind up with a lesser draft than one would expect based on draft position (e.g. the Moneyball draft, after Swisher and Blanton). And a team that offered more would be able to do better. Also, I suppose there are different strategies (go above slot for the first couple of picks and below for the rest, or vice versa) that could be used. Again, this is something you can now look at without even considering the actual players who were drafted.

3. Once you have a handle on the value of draft position and money, you can then measure each team's 'drafting skill' (or 'drafting luck', I suppose). Does a team consistently get a better crop of players than would be predicted based on their draft position and money spent? You could also evaluate different drafting strategies... for example, high school vs. college, pitcher vs. hitter, etc.

Again, really terrific work, Philly. If you aren't working for the Bosox, you should be.

#15 DSG

  • 229 posts

Posted 08 March 2006 - 04:48 PM

DSG

The next round of posts aren't going to contain any of the Value stuff, but I will dig around through your methodology to see what difference it makes.  I'm not sure I like the course flip flop from WARP to WS although I know from previously doing it that there isn't all that much difference.

<{POST_SNAPBACK}>


Thanks. Silver's formula underrates players from replacement level to above average, and overrates superstars, which is why I wouldn't use it. My old formula is better, but still overrates superstars as well as replacement level (and below) players. My new formula is essentially perfect. The marginal wins method has problems with superstars but is good beyond that. BTW, if you can do this, it's better to evaluate value in each season seperately, I think, than on a "per season" basis as I earlier suggested.

#16 dnramo

  • 3,669 posts

Posted 09 March 2006 - 03:29 AM

Philly,

Why use career WARP3 rather than the first 6 years of MLB service time? Particularly with the high-value players, this would seem to make a huge difference.

#17 philly sox fan


  • SoSH Member


  • 9,748 posts

Posted 09 March 2006 - 08:03 AM

1. Draft position.  The keys here are who has the early picks - especially the #1 - and who has gained or lost picks by signing or losing free agents.  The work you've done here will allow you to rate each team's draft position, and measure how much better one team's position is than another, before the draft even happens.  That's essential if you want to find out which teams draft better than other teams.


3. Once you have a handle on the value of draft position and money, you can then measure each team's 'drafting skill' (or 'drafting luck', I suppose).  Does a team consistently get a better crop of players than would be predicted based on their draft position and money spent?  You could also evaluate different drafting strategies... for example, high school vs. college, pitcher vs. hitter, etc.

<{POST_SNAPBACK}>


Those kinds of things are definitely on the drawing board, but they may be a few iteratons down the road. The next round of stuff that I'm writing is going to include some broad GM/scouting director specific information, but the real in depth stuff may end up waiting until next winter.

Why use career WARP3 rather than the first 6 years of MLB service time? Particularly with the high-value players, this would seem to make a huge difference.


The next set of posts are going to be year by year draft reviews and those will include a section separating players by pre-FA service time. It can make a big difference although when I look at a team level it generally tracks pretty well that teams that drafted well in terms of career production will rank about the same in terms of pre-FA production.

The last summary that I have sorta done will be the slot by slot rate of return and that will include a career average and pre-FA average.

Oh, and the Values are actually derived just from the pre-FA piece of the players career production so in that respect it is already built in to the economic piece.

#18 dnramo

  • 3,669 posts

Posted 09 March 2006 - 08:15 AM

The last summary that I have sorta done will be the slot by slot rate of return and that will include a career average and pre-FA average.

Oh, and the Values are actually derived just from the pre-FA piece of the players career production so in that respect it is already built in to the economic piece.

<{POST_SNAPBACK}>


Cool. I'm looking forward to reading more.