Using CS to determine what combo of players (given a franchise) would be the best in terms of fWAR over the last 25 years with one player taken from each year, complete team filling every position with no duplicates. Using the Blue Jays because Go Jays!
The prompt is quite open ended, but what I was interested in particularly is how to maximize the value of a team based on the given criteria. In order to do this, there were a few choices to make in terms of how to design the team. WAR was an easy choice for maximization criteria because it attempts to represent a value for a player, so using that you'd assemble the team that could potentially be the most valuable and thus in my mind most productive overall. Another choice was how to partition the roster.
The roster will be partitioned as follows:
- 5 Starting Pitchers
- 7 Relief Pitchers
- 4 Bench Players (Catcher, 1 Infielder (a shortstop for versatility), 2 Outfielders)
- 8 Starting Position Players
- 1 DH
This is only valid for an AL team so maybe for adaptability this will be changed going forward. The only real choice here was between 7/8 man bullpen and 3/4 man bench, but I figured that because WAR for relievers tends to be pretty low and the system may waste a good position player year to accommodate a reliever, but in reality I just felt like another position player would be more interesting than an extra reliever. The Baseball Prospectus article (see below) had a 5 man bench so I figure may as well go with a more normal 4 and have some basis for comparison between the author's list and the one generated from this, though she used bWAR for her WAR versus this system using fWAR so theoretically there could be some differences in the concept of value on top of her list not being programmatically generated.
Another thing of note is that I chose to not adjust for the fact that relievers may be undervalued in terms contributing to wins in WAR, so the team that is produced will likely not put an emphasis on taking strong reliever seasons where it can find better position player seasons because it just wants to maximize WAR, not create a balanced team. The result of this may be that the relievers are relatively lacking, though I'm somewhat curious as to what the bullpen will look like. Maybe the WAR lost by maximizing reliever seasons will be harder to make up by poorer relievers rather than the difference between one player-season to another. That being said, relievers are not treated specially because they have a fundamentally different role than other players.
Data was compiled in a fairly simple manner. First, I went to the Fangraphs leaderboard section, set multiple seasons to be from 1992 (25 years from now) to 2016. I then split the seasons and created a custom leaderboard that just included WAR (because season and name are automatically included). I then selected the team to be Blue Jays.Then, I went through each position and exported the data. The results of this can be seen in data/jays_raw. I then compiled that into a final CSV which included the positions of the players so that it could be easily read into the program. There are a few things of note here.
- Players can be listed with same season statistically for multiple positions for a particular year. An example would be Edwin Encarnación in 2015, where he was listed as both a first baseman and DH. This isn't really a problem because players can't be repeated so really it just adds a dimension of positional versatility. This is also intersting in the case of backup outfielders, which can be anyone from the OF list, which is composed of all outfielders in all years. This allows positional versatility for the backups, but it could create a situation where one of the two OF positions (which should be bench players) have a higher WAR value than that of a starter in the outfield. This is actually likely in the case of left field, where left fielders probably won't end up being the backups, but if there is a situation where manually I can switch a supposed backup with a higher WAR than a starter where the backup actually players the starting position I will manually do so at the end.
- We switched the order of the characters for the position "1B", "2B", and "3B" because we thought it would make it easier to parse. We then changed the parser and but there is no compelling reason to change it back.
- The data contains Fangraphs player IDs and they are used to make sure no players are repeated.
After all of that the data was ready to use in our program!
This Bluebird Banter article...
Which led to the following tweet:
<script async src="//platform.twitter.com/widgets.js" charset="utf-8"></script>Pick one franchise. One guy from each of the last 25 years. No repeated players. Must form a coherent 25-man. Don't cheat on positions. Go.
— Russell A. Carleton (@pizzacutter4) January 27, 2017