Reconsidering My Player Valuation Method

Prior to the season, as part of the FanGraphs+ offering, I went over my methodology for creating ottoneu player valuations, which included my decision to use Points/PA rather than total points or points per game evaluate players in points leagues. In fact, my preference in almost all formats is to break players down to their per PA values and rank them based on how much production above replacement level they provide in that increment.

This led to a debate on the relative merits of Pts/PA vs. Pts/G, in which I basically stood by the fact that the per PA method better accounted for platoon players – guys who pinch hit a lot and therefore have a lot of low-point games thanks to getting only 1-2 PA. And while I still contend that Pts/PA is better than Pts/G, I have been wondering if there is not a third way.

In ottoneu leagues, owners are limited to a certain number of games played by position. This means that if you had started, for example, A’s 1B/OF Brandon Moss everyday, you would get 145 games played. If you only played Moss when he started, you lost 30 of those games, but you only missed out on 28 of his plate appearances. Contrast this with Cincinnati OF Jay Bruce, who played in 160 games and started them all.

Over the 2013 season, both the A’s and Reds has .327 team OBP, and Moss and Bruce both hit in (roughly) the same lineup spot (both hit 5th most often; Moss’s average lineup spot was 4.7, Bruce’s was 4.8). Based on this, you would expect both players to get roughly the same number of PA per game.

As is evident by the 28 PA in 30 games Moss added as a sub, that is not the case. In fact, Bruce managed nearly an extra PA per game – 4.36 to Moss’s 3.48. Assuming they put up the same stats per PA (they didn’t), Bruce would have been worth about 25% more per game than Moss.

Instead, by using Pts/PA, I effectively assumed these two – on similar teams in similar lineup spots – would put up roughly the same number of PA per game started. This, however, was also not accurate.

Moss started 115 games and in those 115 came to the plate 477 times, for 4.15 PA/Start. Bruce, as noted, started every game he played, so his 4.36 PA/G holds. Despite the fact that Moss hit slightly earlier in the lineup on average, he put up more than .2 fewer PA per game. This may not seem like a lot, but over the 115 games that Moss started, that works out to about 23 fewer plate appearances, or roughly five games lost. This means that Pts/PA actually overrates Moss compared to Bruce.

The reason for this is the same as the reason why Pts/G overrates Bruce – Moss’s platoon split means that he often gets pulled for pinch hitters late in the game. The same problem can occur for weak defensive players, who may lose PA to defensive replacements, or even to National League players who get pulled for a double switch.

At the end of the day, if you look only at total points (904.7 for Bruce, 728.2 for Moss) or points per game (5.65 to 5.02), Bruce looks much better. On a per PA basis (1.44 for Moss to 1.30 for Bruce), Moss appears significantly more valuable. But what really matters is what you can expect to get from each player if you use your limited resource (one game played) on them. And on a per game started basis, Moss (6.20) outshines Bruce (5.65), but by a smaller margin than on a per PA basis (note that Moss actually had a slightly better Pts/PA split as a starter than overall).

Let’s look at two fictional players, based (roughly) on Moss and Bruce. Players A and B both hit in the same spot in similar lineups. Player A plays in and starts all 162 games with 700 PA and 1000 points scored. Player B plays in all 162 games, but starts only 120, accruing 522 PA 480 as a starter) and 805 points (740.2 as a starter).

Here are their stats on per game, per PA, and per start basis:

  A B
Pts/G 6.17 4.97
Pts/PA 1.43 1.54
Pts/GS 6.17 6.17

Per game, A seems like the better bet, and if you build values based on Pts/G, you will bid more on him. On a per PA basis, B looks like the better value. In reality, you get exactly the same contribution from them for each game they start. In this case, you would bid more on Player A, as you would get his contribution in 162 games, vs. 120 games of that contribution plus 42 of a replacement level player.

Regardless, what both this example and the Bruce/Moss comparison show is the best method for valuation is likely to look at start vs. sub splits and only consider stats as a starter. My goal for this off-season is to build a spreadsheet that can consider the following equation for each player:

Players Projected Value = [(Projected Points per PA)*(PA per Game in which Said Player is a Starter)*(Games Said Player is Projected to Play)+(Replacement Level Points per PA)*(Average PA per Game Started)*(162-GAmes Said Player is Projected to Play]*(Some per Point Dollar Value)

Some of these variables are relatively easy to come across, and are closely related to what I did for 2013. But since no projection system is going to project starter vs. sub splits, I need to a) find a spreadsheet of past starter split statistics and b) figure out if it is safe to assume that a guys playing time split between will be constant moving forward.

For now, suffice it to say, neither Pts/G or Pts/PA are ideal and if you can figure out a players production on a per start basis, you are one step closer to perfecting your auction values.




Print This Post



13 Responses to “Reconsidering My Player Valuation Method”

You can follow any responses to this entry through the RSS 2.0 feed.
  1. SurprMan says:

    Good stuff, Chad.

    I also prefer to value players by Pts/PA but have found that one folly of doing so is the tendency to over-value injury-prone players. I got burned last year having Allen Craig, Jake Peavy, Josh Johnson, and Carlos Quentin (among others) on my roster, all of whom lost significant chunks of time (which could have been predicted based on injury history) despite top notch past/projected Pts/PA[IP].

    Vote -1 Vote +1

    • Chad Young says:

      I think the best way to account for this is to assume a number of games played for those types of players, add replacement level stats to their stats to fill out the rest of the year, and value that combined player. Technically, the “starter” in that case is worth the value of that combined player minus $1 (needed to add the replacement player).

      The key wtih guys like that is too look for a discount when other owners write them off completely. Too many owners will avoid Quentin like the plague for fear of the injuries, but if you value him appropriately and buy him at a discount, you can add in an above-replacement-level partner for him and make it work.

      Vote -1 Vote +1

  2. Keep Moss? says:

    Is Moss a keeper worthy player (Keep 5) in a 16-team league w/ larger benches and daily moves? Considering his HR potential in limited ABs? According to Yahoo! he was within the Top 80 players, which mathematically/technically makes him keeper worthy on paper in this league.

    Vote -1 Vote +1

    • Chad Young says:

      This is a really tough call. Yes, he is top 80, but depending on positional eligibility and other rules in your league, that may not make him a top 80 in terms of value. He should qualify in the OF in more formats, which makes a big difference.

      The real opportunity with platoon guys is the ability to acquire them for a price below their value because other owners miss out on what those guys produce in even part-time duty. The word is likely out on Moss at this point, though, which may make that hard.

      With only five keepers, I guess it just depends who else you have. I would hesitate to keep a guy who you know you have to draft a handcuff for, but if he is one of your top five, then you do it.

      Vote -1 Vote +1

      • Steve says:

        I have Moss in a somewhat similar, yet even deeper League. 14 teams, 10 keepers, auction format, 3 year contract maximum. Moss just finished the first year of his contract, which is $1!. I think that tilts the table toward “keeper.”

        Vote -1 Vote +1

  3. LuckyStrikes says:

    It’s a good point to consider GS with platoon players, which can be critical to Ottoneu. Baseball-Reference seems to be the best place I can find to find the right splits. Here’s a link to a spreadsheet you can use to begin the process (http://goo.gl/al5T2F) you’re describing above.

    Vote -1 Vote +1

  4. Chad Young says:

    Like the spreadsheet. I build something quite similar. I try to start by compiling different projection systems so that I can have a master stats spreadsheet that can be used to create values for all the scoring formats.

    B-R seems to have good start/sub splits on their player pages, but I have been unable to find a leaderboard/spreadsheet style dataset which is what I really need. The splits int he article for Moss and Bruce came from B-R. I also looked at B-R to compare Trout and Joyce (both hit top 2-3 most often, Angels and Rays had same OBP, so similar experiment to Bruce and Moss).

    Vote -1 Vote +1

  5. bryce99 says:

    how would you put a dollar value on a player?

    Vote -1 Vote +1

    • Chad Young says:

      If you have FG+, you can check out my FG+ piece from last year for this. If not, the short version is that once I know how many points each player will earn above replacement level, I can convert that into a dollar figure by comparing how many points are up for grabs with how many dollars are available to spend.

      Vote -1 Vote +1

  6. bryce99 says:

    Would this also work if you were to take there projected games played and projected plate appearances.

    Vote -1 Vote +1

    • Chad Young says:

      This is actually the exact problem I am trying to solve. If a player is projected for 480 PA in 120 GP, you don’t know if that is 4 PA per G or if it is actually 5 PA per start over 80 starts and 2 PA per game in which he enters as a sub.

      The problem is that no one projects this stat and I was unable to find any good way to predict it.

      Vote -1 Vote +1

  7. Jim says:

    I’ve been considering the same problem, but from the angle of stolen bases. I use per plate appearance rate stats in my valuation model for my traditional Ottoneu 5×5 roto league. For steals, I use two factors for projection – ratio of SB/CS and attempts per time on first (SB+CS)/(1B+BB+HBP).

    Using this standard, Rajai Davis is elite. Crazy elite, up there with Miggy and Chris Davis. But he is used as a pinch runner, which throws off the model, because he has been at first base more times than his batting statistics would suggest. Do you know of any place I can find a “times used as pinch runner” stat? Adding that to the attempts denominator should balance out the effect.

    Vote -1 Vote +1

    • Jim says:

      I should clarify that I’m referring to past year’s stats for players, not projections of future pinch running appearances.

      I’ve found that using projected stats from the available models somewhat biases your valuation results. Each projected stat has it’s own variance, but in all cases that I’ve seen, you are only given the average result for each player.

      Some rate stats, like SB/PA, are very stable from season to season, whereas batting average sees relatively large year to year fluctuations. A base stealer is more reliable than a high average hitter, and therefore should be valued more. My model projects expected fantasy value per plate appearance directly from past season stats, solving this problem. The model gives a big boost to Starling Marte, but hates Joe Mauer.

      Once you have value per PA, you get into the muck of playing time projections. If someone didn’t play a full season last year, were they a) a call-up from the minors, b) injured for part of the year, or c) in a platoon role? Each of these types would need very different projection models to assess future PA/season. Some good work has been done on player injuries, but call-ups and platoons are often at the mercy of team prerogative. I just fudge these numbers, I have no idea how to approach that problem.

      Vote -1 Vote +1

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>