I will put up a game thread tomorrow discussing how much one ought to worry about any baseball series played before The Masters is completed. In the meantime, a day without Braves baseball calls for my first Retrosheet Retrospective for the season.

Today’s rambling discourse is inspired by a comment from long-time Braves Journal stalwart Roger:

Just want to remind everyone that this is NOT our team yet.  What will we be like when we have Acuna, Strider, Murphy, Verdugo, and Kimbrel?  We have already lost Profar and Lopez and Jimenez for large parts of the season so, overall, this team could be and will be better than it is now.

Roger

I agree with that, but it got to me to thinking, which is of course never a good thing.  Teams at the beginning of a season are always different from teams at the end of seasons.  There are two reasons for this: injuries and failure.  But they work in opposite directions.  Players who become injured must be replaced on rosters; on average, this must make teams worse if they have chosen wisely in giving the injured player a job.  Sure, there are Wally Pipp/Lou Gehrig counterexamples, but they are celebrated because they are counterexamples of what everyone knows: when Ronald Acuña, Jr. gets hurt, your team gets worse, and that is true even if you end up winning the World Series with him out.

Failure works the other way: when you have a personnel problem and you address it, the team gets better.  There are counterexamples here as well, of course, as just about every bad trade attests.  But the alternative to responding to failure by hoping does not have a long track record of success.  Mostly, it has a track record of getting managers fired.  (Replacing failures with people who are worse failures has a record of getting GM’s fired.)

Roger’s comment already encapsulates some of this: Acuña, Strider and Murphy are returning from injury and should be expected to make the team better.  Verdugo and Kimbrel represent experienced parts to be substituted in when the initial constituents fail.  Conditional on their failure, Guys like Verdugo and Kimbrel represent an expected improvement, on average.

So you’ve got one effect that runs one way, and one that runs the other way,  What do the data say about which effect predominates?  Do teams that make personnel changes do better or worse?

I decided to take a Bill James-like look at this.  I want to state emphatically that this study is not intended to prove anything.  If there’s one thing anyone should take away from reading Bill James, it’s that trying prove something this way is a bad idea – it leads you to a selectivity in gathering evidence that is the death of insight. And the conclusion it does reach is pretty weak: changing personnel turns out to be just about neutral, with maybe a slight ability to improve a team through churn. But the real payoff is just giving myself something to do on an off day.

So first, I need a metric of team change.  There is definitely no right way to do this.  And in many ways the data dictate what you can do.  If I had data on absences explicitly due to injury, for example, I’m sure I would have done something more nuanced.  But I don’t.

So here’s what I did:

I took the years 2014-2024, omitting 2020, as 10 recent years to study.  And I looked at the personnel for every team in the first ten and last ten games of their season.  On the assumption that every team tries to put its best foot forward, the first ten games’ personnel are therefore a statement of the team you meant to go with.  For playoff teams, those last ten games will include some or all of their playoff games, when personnel is presumably at its most critical.  For bad team, those last 10 games will often focus on new personnel for tryouts for the next season.

So from these sets of personnel, I adopt a very simple metric: what is the overlap between the personnel who appeared in the first 10 games and the last ten games?  (I shoud note that by appearance, I count only batters and pitchers, pinch runners and late-inning defensive replacements who don’t bat don’t count,)

So I have 300 team-years, and a metric: the number of players who were there in both the first 10 and last 10 games of the season.  Before looking at the data, think about it a little bit?  Has any team fortuitously ended up with the same players they started with? (Under this method, they might have made any number of changes during the season.)

So the first chart just shows the distribution. 

The interesting thing here is that MLB teams all have a lot of churn. While very few teams lose half of their opening roster, it’s normal to lose 8 or 9 players. Those guys you broke camp with and who were there in the trenches in the opening week will be elsewhere at crunch time.  Now it probably makes a lot of difference who those people are: losing your whole bullpen is a lot different than losing seven position players.  But for this exercise I’m not going to look at that.

Now to the question that animated the discussion:  does churn make your team better or worse?  For this you need to add another dimension: winning percentage,  This creates the next chart.

 I have added a third dimension by coloring in the playoff teams.  (I have also colored the Braves teams in this period in blue.)  The results here are clear: keeping your team together, even with a measure as crude as this first 10/last 10 game personnel overlap, is a boon.  No team that opened and closed the season with 23 players or more has ever finished below 0.500, and only one team which finished a season with 13 or fewer players they started with has ever even made the playoffs – the 2015 Texas Rangers.

We’ve talked a lot about the indifferent play of the 2021 Atlanta Braves for ¾ of the season, but in the current circumstances, the 2015 Rangers, who started 8-16 and were still 42-46 at the All Star break.  And their improvement came both from players added (Cole Hamels) and the return of injured players (Derek Holland). 

What this chart clearly shows is that it’s better to keep your team together, at least if you have a good enough team that can win. it also shows that teams that have a lot of churn turn out to be not very good teams. But what this chart can’t do is attribute improvement to moves or collapse to injury, at least not directly. Does churn cause losing, or do losing teams try to churn to recover?

So let’s graph the continuity measure against not the final winning percentage, but against a measure of improvement: winning percentage in the final three quarters of the season minus winning percentage in the first quarter. Now the result is far more nuanced, which is fancy way of saying muddled.

 

While I think it’s easy to look at this blob of points and see “well, that just looks like a blob of points” I think there are some things can be gleaned. First, this chart has to be centered around 0 in the difference axis, because the differences actually sum to zero: somebody’s win is somebody else’s loss each season. But there is a slight upward tilt to the data: it is slightly easier to improve a team that has less turnover, though it is a very small effect. Even so, the largest upward gains come from teams in the middle of the distribution… teams with a more average amount of turnover. I’d summarize this by saying that it’s very hard to improve a team that doesn’t have a lot of turnover, but (a) it’s not impossible; and (b) a ton of turnover almost insures you’re going to do worse, at least in that season.

If you were looking for grand conclusions, you’ve come to the wrong place. But the next time an announcer in September says: “There are only 18 remaining players from the 26 we started with on opening day” check back to this page and say “Only? that’s actually above average.”

Enjoy the offday.