I think the real lesson - and the premise behind Moneyball - is that when you have a complex objective, you can't evaluate people in a vacuum. You have to optimize at the team level. You can try your damnedest to come up with heuristics and metrics to evaluate individuals, but if you don't allow yourself to consider the context that they'll be working in, you'll miss a lot of potential.
I don't think this is the central thesis of Moneyball. I'm a huge baseball nerd, including on-field and statistical side. The idea behind the WAR (Wins Above Replacement) stat is that you should adjust for context, then evaluate players in that vacuum.
Moneyball was about identifying and exploiting market inefficiencies. It actually rejects the team level optimization ("he's a team player" is explicitly rejected as a valuation of a player).
Instead, Billy Beane identified what components led to team success, then found which of those the market doesn't pay for. Then he got those guys.
Used to be thought that a good baseball team had a speedy leadoff hitter. Teams would play objectively worse players because they thought their style better "fit" their role.
Moneyball was the start of the revolution that you just want to get the best players. Then accept that playoffs don't really mean anything other than randomness with small sample sizes.
Agreed. It's also worth observing for those who might not really be baseball fans that baseball is really less of a team sport than most major sports. Yes, team dynamics can reach toxic levels and sometimes teams seem to outperform their individual roster members. But other than some degree of matching players with positions (pitchers in particular), you mostly want players who would be best/best value on key stats under any coach or any team.
The entire offensive side of the game pretty clearly violates the notion of independence. In particular, the value of a batter with particular stats is dependent on the batters that come before and after him. This is the reasoning behind stacking people with high on base percentages in front of sluggers. If it were not a team sport, you'd see no clear patterns in line-up. This applies at the team selection level as well. For instance, a player who can draw a lot of walks is worth different amounts to different teams depending on their ability to convert those into runs.
The fact that saving a run is nearly equivalent to scoring a run means it bleeds over into the defensive side as well. A phenomenal batter with a high error rate may be of enormous value to one team since their pitching staff strikes out more batters than average while being a terrible addition to another team.
If you're still unconvinced, I could throw together some simulations later that might make this clearer.
"In particular, the value of a batter with particular stats is dependent on the batters that come before and after him. This is the reasoning behind stacking people with high on base percentages in front of sluggers."
This is the older, common wisdom that has essentially been refuted by modern analytics (SABRmetrics, if you will).
Team-dependent statistics like R + RBI (and pitcher ERA!) are dependent on the performance of the rest of the team. They also correlate much less with a team winning games than other stats, like OBP.
That's one of the central theses of Moneyball and the SABR movement: it's better to ignore stats like (pitcher) Wins, RBI, and R, because they measure the teammates contribution more than the player you're tryign to measure.
One popular advanced pitching statistic is literally named Fielding Independent Pitching, and is calculated using only BB, K, and HR. This tracks more with contributions to a team winning than stats that include fielding skill, randomness, sequencing, etc.
What you described for a good hit / bad field team does work around the edges, but these effects are not very strong. The latest research suggests the biggest impact might be matching your pitching staff tendencies with fielders; if you have a flyball heavy staff, invest in better OF defense (Mariners, 2017). If you have worm burners (Houston, esp w/ Keuchel), invest in your IF defense.
The goal of the modern stat movement is to evaluate a player independent of team context using statistics that do not depend on their teammates. This is because estimating a players "true talent level" is much more useful than evaluating RBI.
It seems like we're more or less in agreement...Team performance and individual performance are interdependent, particularly in naive statistics like the ones you've mentioned are traditionally used. You're talking about making team-independent estimations of player value. That's completely valid. What I'm also saying is that in a similar vein, when you are considering how a player's addition will impact a team, you need to correct for his interaction with the team as well.
Yes, agreement! Thank you for nudging me to recognize that.
However, the studies I've seen (regular Fangraphs and BP reader) suggest these interactions between player skill sets is minimal, if such an effect can be shown.
Take lineup position. It's been studied ad nauseam, by some of the brightest in the field. Turns out, lineup order, over the course of a season, doesn't really matter that much. An intentionally suboptimal lineup underperforms an "optimized" one by maybe a couple wins per year. Almost all lineups actually implemented are more like fractions of a win, which is generally within error bars.
By and large, baseball is a game where you just assemble the best talent and they will win. (This is in regard to on field talent; I do not believe "clubhouse culture fit" is as silly as in the tech world, and is usually retroactively defined.)
This kind of reasoning go only so far. While you are technically correct, the difference that something like batting order makes is marginal in the long run. The advantage of a fully optimized vs. a totally random batting order would equate to approximately 1-2 additional win(s) over the span of a full 162 game season.
That being said... one win can be the difference between a playoff spot or not, so it clearly does matter. I'm just trying to emphasize that the teamwork aspect of the offense is minimal, as a lineup of hitters with solid statistics is more important than the order in which they bat.
However, the defensive side of things is much more tricky to measure. Baseball statistics still have a difficult time calculating the value (for better or worse) of a player's defense and nobody is arguing that it doesn't significantly affect a team's performance.
That's fair enough. Although I'd argue that it's more a case of assembling an appropriate portfolio than it is a "team" in the way that people normally use that term.
(I'd also argue that overall offensive ability trumps a lot of optimization around base stealing or the type of hits.)
ADDED: i.e. I'd argue that if populating a roster with about a half dozen All Star sluggers were financially viable, I'd argue that would probably be a pretty effective team however unbalanced.
As covered in Moneyball (the book), the "sluggers" tend to swing at more. So although they may hit home runs, they also tend to strike out a lot too.
The actual approach was to get a bunch of people who worked to extend the inning by not getting out.. aka getting on base safely. You don't need many big hits when you can get singles and doubles consistently.
A couple years ago, I met a small scale angel investment group that complained "we haven't had a unicorn" but when we chatted more, I found out 60% of their investments were acquired in under five years with them making ~5x each time.
Not as sexy but sounds like a good strategy to me.
True, only if you have a great team already and you don't desire messing it up. If that is the case, you wouldn't need that method at the first place.
On the other hand, teams are pretty dynamic these days given the interfaces and responsibilities are well defined. Or at least are largely defined well. So replacing team members is never really a problem.
If you have a team of 6, sometimes the best individual 6 people are not the same group of 6 that you put out there to solve a problem.
Let's continue to use hockey as an example, because I love ill-fitting sports analogies and I also love hockey. For sure, we already know we need 3 forwards, 2 defensemen and a goalie, so for the sake of argument, even if your 6th best skater is Wayne Gretzky, he's going to sit because you need to put Neuvirth in goal (to pick on my own team for a second). But let's think about forward lines only for just a second. Pittsburgh has Crosby and Malkin centering different lines, even though they're the best two guys on the entire team - they do the same job, they're playmakers first, goal scorers second (even though Crosby is the league's top goal scorer). Washington has Ovechkin playing the wing with Backstrom, because even though he's their best player, his job is to take a pass and fill the net.
That's a really long-winded way for me to say that "best" has a situational component, and teams have compositional elements that often involve roles. The 10 best people at that level in the company may all fill the same role, but on a team of 6, maybe there's only 3 of that role, and the other 3 are complementary parts. Best is situational.
Gretzky is great example because he was not the best player because he scored the most. He was the best because just him being on the ice lead to others scoring as well. Just being out there he was able to help the rest of the team score as well. Contrast this with a player like Kobe Bryant who was a great individual player but horrible team player. Kobe did not get along with his team mates and other than the points he put up on the board Kobe did not really contribute much to the Lakers. Kobe was such a ball hog that they refer to a missed shot that is rebounded and scored by the offense as a kobe assist because it was only way he was going to give up the ball
But he played ball for 20 years. I think average number of assists per game is a better stat here, where he was ranked 136th according to this link [1].
There have been over 3000 people to play in the NBA. Ranking 136th still makes him better than the vast majority of people to have ever played the game. Especially as a two guard on a team that primarily ran the triangle, an offensive system that devalues traditional pick/roll dribble creation and generates most of it points on secondary actions (eg hockey assists) off high post play.
When people either can't or don't want to measure the performance of who they picked and who they didn't, they resort to stories that sound good. Even if you have "rigor" around how you score candidates, if you don't connect those scores to subsequent performance it's all still just a narrative.
If the "right people" aren't the same as the "best people" are the criteria we're using for "best" correct?