from Amazon
Wednesday, April 28, 2010
Do players who sign with their own teams age better than those who don’t?
About 3 weeks ago, Matt Swartz wrote an article on BP that concluded fairly emphatically that players (pitchers and hitters) who sign multi-year deals with their own teams aged and performed significantly better than those who did not. To be honest, I did not re-read the article carefully and it is not real clear what the criteria were for the various groups of players.
Here is what he wrote at the end of the article:
What appears to be happening is that teams seem to have some sense of the aging curve of individual players, especially if they are already in their organization. There are probably a variety of reasons that this subset of players aged well, but the team knowledge about the player’s medical and scouting information appears to contribute to the decision to give a player two-year deals. This is important to keep in mind when we hear of a player signing a new contract and look to a projection system to figure out how smart the deal was. Chances are that there is additional information—especially about aging—which teams have that we may not.
Today, he wrote a long article telling us why the Ryan Howard contract might not be as bad as some analysts are making it out to be. One of the reasons he cites that it may not be so bad is that because Howard is signing with his own team, he may age better than we think (using comparable players, the traditional aging curves, etc.).
Now, when his first article came out, I must say that I thought that the results were fascinating, if not ground breaking. In fact, it should have gotten much more attention than it did. The idea that one, teams may know a lot more about players who have been around in the majors for a while than do analysts, and that two, our projections can be vastly improved using this new found knowledge, is, well, fascinating and groundbreaking. One could even say that it is one of the most interesting and significant things to come out of sabermetrics since DIPS.
I must also say that I was highly skeptical of the results. I forgot who said it - maybe no one in particular - but in science, when one gets surprising results from an experiment or from research, although in this case they are not necessarily counter-intuitive, one should be very careful before one draws any conclusions from the data. I don’t think Matt was nearly careful enough. There are so many things that need to be looked at before you can declare his thesis to be true. For example, sample size (I realize that he said that his results were very significant), ages of the players, do teams tend to sign their own players after a particularly good or bad season. What about other teams? What about park effects? Obviously for players that sign with other teams, park effects have to be considered. What about the fact that players who switch leagues might automatically do worse until they get used to their new opposition (the opposite is actually true for pitchers)? Maybe a team is better at determining whether a bad or good year was a fluke or not, so we are just seeing other teams tending to sign players after a flukey good year and teams signing their own players after a fluky bad year (and thus, it would APPEAR that when teams sign their own players, they get “better” the next year)?
As you can see, there are all kinds of things to look at before you declare that teams are much better at projecting their own players because they know their character, makeup, work habits, and health more so than anyone else. Of course, what he proposes is not necessarily surprising. What surprised me was the magnitude of the results. I would have expected little or nothing to “show through” in the data. Keep in mind that I am not in any way shape or form saying that Matt’s thesis is not true or if it is, to what extent. I’m just saying that I am skeptical of his results, and I think that there is much more to look into before you can draw conclusions with any kind of certainty.
Not being satisfied with just being skeptical, I decided to do a little quick research myself. Keep in mind that I did some quick and dirty things, and that my research is probably less comprehensive than Matt’s, although I took a much different approach.
First I looked at all player years since 1980. I broke players down into 2 groups: Both groups has at least 7 years in the majors, with no min playing time in any year. One group played at least 3 years straight with the same team after their 6th year, and they were with that team in at least the 5th and 6th years. That was my proxy for players who “signed multi-year deals with their teams during their FA years.” The other players were those who switched teams in their 7th year. Those are my “players who signed with other teams”. Obviously it is not the same as using a contract database, but it is close enough for government and sloppy sabermetric work.
The first thing I looked at was the overall aging curves of both groups of players using all of their major league years, not just those after the 6th year. I used my usual delta method corrected for survivor bias (by filling in the “year x+1” for players who dropped out after year x).
Both curves were fairly typical of the average player, shape-wise, however there were some significant differences. The players who stayed with their teams peaked about a year later (age 28) than those who did not, and the rise from age 20 or 21 was much steeper for the same-team players. In fact, the same-team group gained around 27.5 runs (lwts per 500 PA) from age 20 or 21 to their peak at age 28, while the switched-team, group only gained around 17 runs to their peak at age 27. After that, both groups declined at about the same rate, although the same-team group declined a little steeper. So there seems to be something going on there, but I am not quite sure what it is. As I said, there are so many variables and factors that one needs to look at before drawing any cause-effect relationships, at least in my opinion.
Next I looked at the same aging curves, but I only started each player at year 7. So the first year of aging deltas for all players was from year 6 to year 7. For the same-team players, they were obviously with the same teams, and for the new-team players, the 7th year was the first year with their new team.
Again, the overall curves were similar. Again, the first group (same-team) peaked a year later than the second group. Both groups peaked early though - 26 and 25. The first curve was very smooth and looked exactly as you would expect from a typical player, without the normal steep rise from age 21 to 27 or so. Obviously not many players in either group were less than age 26 or 27 since we were starting in their 6-7 year interval, so for the two curves, most of the curve is in the decline phase. For the first group, there is no suggestion that the players aged particularly well or got better after age 26. There is a smooth decline from age 26 onward for both groups. For some reason, the second curve, the players who switched teams, is very choppy even though overall it is a normal decline-phase curve. Maybe that has to do with park effects, which I am not controlling for. I don’t know.
There is one interesting difference between these curves for the two groups. Remember these are aging curves starting with every player’s 6-7 season interval. Both curves start at around age 24-25, although there are only a few players in that interval of course, so any inferences made from the first part of these curves is suspect at best. Anyway, there is a sharp drop after the age 25 peak for the switch teams group, while the same-team group’s curve shows somewhat of a plateau from 24 to 28 (a small decline). After age 28, both groups show the same rate of decline until about age 34, at which time the switch-team group drops sharply again. This could also be a small sample size fluke, but perhaps the inference here is that teams know much more about the future development of their young players than do teams that sign them after they are let go. Maybe.
Finally, I looked at the same curves, but this time I restricted them to players who were at least 29 in their 7th year when they either stayed with their old teams or moved on to another team. These are most of the players of course, and certainly players like Howard fit into this group. With these players, we are obviously not talking about teams knowing how well a young player is going to develop. If anything we are talking about whether teams know how well a seasoned player will age, which is pretty much what Matt is talking about in both of his articles.
Here is where it gets interesting. A comparison of the two curves is exactly the opposite of that of the younger players. For the same-team group, the players decline from the get go (we are starting at age 28-29 for these curves), as expected. For the switch-team group, they actually improve from age 28 to age 29 and then decline. After age 30 or so, the declines in both groups are very similar, although the curve of the switch-team group is once again choppy while the same-team group’s is smooth.
Basically, I don’t see a whole lot of evidence that the aging patterns of the two groups are significantly different although one can make some inference that for younger players, the ones that stay with their teams do much better than the ones who leave, at least for the first few years after their 6th MLB season. For older players, which are most of the sample, either they appear to age the same after their 6th season or the players who switch teams do better, again, at least for a few years into free agency.
As I said, I think there is a lot more work that needs to be done before we draw the sort of conclusions that Matt did.
(That is one of those gratuitous apologies we were talking about on the other thread.)
"Extraordinary claims require extraordinary evidence.” - Carl Sagan (though he wasn’t the first to say it)
Interesting results, and as you say, hard to draw any definitive conclusions. A related question is whether traded prospects do worse than we’d otherwise expect. I always think of Andy Marte, but of course that’s just one data point.