Roster Turnover across Major League Soccer, 2013 – 2015

While Toronto and Seattle are preparing to face off in MLS Cup, the rest of the teams in Major League Soccer are turning their attention towards next season. Most immediately, teams are making decisions about which players should return and which should be let go.

In Columbus that process has already claimed two high-profile victims: goalkeeper Steve Clark and defender Michael Parkhurst, who were the two most-used players in league action in 2016, are both on the outside looking in. This is an unprecedented step for Columbus. It is only the seventh time in 21 off seasons that either of the two most-used players has not returned, and now both such players have been released.

Plotting departed players for each year in Columbus Crew SC history, ranked by playing time.

How does this turnover compare to the whole of MLS, however? Are Gregg Berhalter’s actions so far this offseason in line with what other teams have done?

Turnover across Major League Soccer

In order to put these steps in context, I prepared a dataset that details roster turnover for each team in MLS over the last three offseasons. This leaves us with instances of team offseasons, spread across 21 teams. This dataset, and the R code to generate the plots in this article, have been posted to GitHub.

For the purposes of this analysis, turnover is measured along two scales.

The first and simplest to count the number of players who don’t return the next year. In doing so, I made no effort to distinguish between players who left during a season and those who left during the offseason. Michael Parkhurst is treated the same as Kei Kamara in that regard; each played during 2016 and neither will apparently return for 2017, even though Kamara left via a midseason trade and Parkhurst’s contract expired after the season.

The second scale gives players different weights according to their playing time. This allows us to distinguish between bit players such as Amro Tarek and stalwarts like Clark or Parkhurst.

Using these approaches, Crew SC has thus discarded 34% of last year’s players (10 out of 29 who made appearances). Measuring by playing time, the club has discarded 29% of last year’s playing time – the departed players accounted for 10,481 minutes out of a total of 35,565 minutes in league play.

The following two plots show the range of values seen across MLS in the past three years.

Figure 1 - Histogram of roster turnover, measured by players

Measuring by player, teams have turned over between 22% and 71% of their roster. Most commonly the figure rests somewhere around 40% (22 times the figure has been between 37.5% and 42.5%). The data doesn’t appear to be normally distributed – the histogram shows a longer tail on the right side, but falls off pretty steeply on the left.


Measuring by playing time, the range of numbers gets lower – extending from 10% of minutes up to 58%. Usually teams bid adieu to just over 30% of their playing minutes from year to year.

Taken together, these two measurements are not particularly surprising. Players who don’t see the field regularly are more likely to move on, which should skew turnover towards slightly more-than-average players, and slightly less-than-average playing time.

The comparison of these two approaches can be seen in Figure 3.


Team-specific trends

Within this history, certain teams stand out from their peers.


Some teams were not active enough during the period from 2013-2015 to make much impression. Such is the case for Chivas USA (which registered only one offseason, as the team folded following 2014) as well as expansion clubs Orlando and New York City.

Others, like the New York Red Bulls, exhibited a variety of approaches during this period. Following the 2014 season the team shed more than half of its players and minutes, but held firm the next year with the lowest turnover of any team during these three years.

A third group of teams, spearheaded by the New England Revolution, have been remarkably consistent from year to year. The Revolution have largely kept the faith with their roster, making minimal changes each of the last three years. In opposition some teams, such as the Chicago Fire or Toronto FC, have exhibited higher-than-average change in each year.

Turnover by performance

One easy explanation for postseason turnover, of course, is to look at how successful the team was the previous year. Clubs that miss the playoffs, for example, are likely to make more significant changes to their roster than clubs that performed well. This becomes very clear in Figure 5, comparing team success (in terms of Points Per Game) with the following offseason’s turnover (in terms of minutes played).


Of the seven teams that earned less than one point per game from 2013 to 2015, each changed more than a third of their roster. Five of the seven changed roughly half of their roster. This stands in stark contrast to the eleven teams that earned more than 1.6 points per game, who each changed less than a third of their roster – and usually changed less than 20% of their roster during the next offseason.

This dynamic does not explain the whole picture, however. A trend line fit through these data points confirms that a relationship exists, but only tells half the story.

> model <- lm(data$Pctg.Minutes ~ data$PPG)
> summary(model)

lm(formula = data$Pctg.Minutes ~ data$PPG)

 Min 1Q Median 3Q Max 
-0.180299 -0.065318 -0.006573 0.065164 0.265937

 Estimate Std. Error t value Pr(>|t|) 
(Intercept) 0.71258 0.06112 11.659 < 2e-16 ***
data$PPG -0.29151 0.04322 -6.745 9.85e-09 ***
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.09478 on 55 degrees of freedom
Multiple R-squared: 0.4527, Adjusted R-squared: 0.4428 
F-statistic: 45.5 on 1 and 55 DF, p-value: 9.854e-09

Previous-year success only explains about 45% of the variability in offseason changes in this dataset. After controlling for league success, some teams still change their roster more thoroughly than others.

Team-specific data

Following are a series of team-specific charts that illustrate how each club has responded to their varying levels of success. Some, like the Chicago Fire, have changed a great deal without – so far – anything to show for their efforts. Others, such as Toronto FC, have had much more success.

Teams with only one season in this dataset (Chivas USA, New York City, and Orlando) are not included in this list.


Chicago had one semi-successful season, when they narrowly missed out on the playoffs in 2014 despite earning more than 1.4 points per game. Each year, however, they’ve changed a significant portion of their roster.



Colorado also had  a decent year in 2014, and perhaps as a result changed only a quarter of their roster. Each of the next two years, however, saw much lower success – yet only once (after 2015) did they make drastic changes. It will be very interesting to see how they react to a much better year in 2016.



Columbus had two decent years in 2014 and 2015, culminating in an MLS Cup appearance after 2015. Unsurprisingly, they made few changes – but coach Berhalter speculated during a disastrous 2016 that perhaps they changed too little after MLS Cup and paid the price this year.



Dallas has enjoyed a range of success during this period, ranging from 1.2 to nearly 1.8 points per game. After each year, however, they have predictably changed about 30% of their roster.



DC may be the clearest example in this dataset for the phenomenon of making drastic changes after a bad year (2013, when they earned less than half a point per game – and then blew up half their roster) and standing pat after good years (2014, when they rebounded to earn 1.74 points per game – and then changed under 20% of their roster).



Houston has stayed relatively stable during this period, as evidenced by their turnover-success plot all showing dots toward the bottom of the observed band. This is consistent with a team that, whatever their success, changes less than other teams in similar conditions.


Kansas City

In contrast to Houston, Kansas City’s data points tend to be in the upper edge of the “turnover by success” band. This is consistent with a team that is likely to make more changes to their roster, whatever their level of success.


Los Angeles

Los Angeles, led by Bruce Arena, has achieved a high level of success – never failing to earn at least 1.5 points per game. Despite this consistent success, however, their turnover has ranged from 20% to over 40% of their playing minutes each year.



Montreal, like DC United, shows a predictable variability in turnover based on that year’s success.


New England

New England is the poster child for stability in this data. While they have been consistently successful it is also notable that, like Houston, they have changed their roster less often than many other teams with similar PPG values.


New York

New York is the anti-New England. Their worst year (2014) saw them earn almost 1.5 points per game, yet they still changed over half their roster. Much of that turnover can be attributed to the arrival of Ali Curtis, who jettisoned coach Mike Petke in favor of Jesse Marsch – as well as the departures of Thierry Henry and Tim Cahill.



Philadelphia, like several teams such as DC and Montreal, shows variability in turnover that coincides with competitive success.



Portland’s consistent success in terms of PPG has resulted in relatively low turnover. Interestingly, their MLS Cup-winning team in 2015 was not followed by their lowest turnover of the measured period – last year they still changed 28% of their roster.


Salt Lake

Salt Lake may be unique among MLS for changing relatively few players, but those that do leave were more used than most. The band of dots in the left-hand plot below shows that RSL’s departures, measured by headcount (horizontal axis) are at the upper side of the band – meaning those players saw more-than-average playing time.


San Jose

The Earthquakes changed about 40% of their headcount after every year, but in some years that group has been more central than others. The 2014 group that earned 0.88 points per game, for example, saw departures that totaled 40% of the playing minutes – while after 2015 a roughly similar number of departures accounted for only 13% of playing minutes.



The Sounders, like Los Angeles and Kansas City, have been more likely to change a bigger proportion of their roster.



For so long the poster child for MLS ineptitude, Toronto appears to have finally assembled a winning group. In doing so, they have progressively stabilized their roster – from a significant overhaul after 2013 (when they earned 0.85 points per game and then changed over half their roster), to an improved 2014 and even better 2015 that saw them earn their first-ever playoff berth. 2016 finds them hosting MLS Cup.



The Whitecaps’ relative stability in terms of PPG is not reflected in their roster from year-to-year. Yet the trajectory they have followed has been similar to Toronto’s – improving from 1.41 PPG in 2013 (and missing the playoffs, after which they changed 45% of their roster) to 1.56 PPG in 2015 (followed by a turnover of only 22% of their roster). Unfortunately, the comparisons to Toronto end there – while the Reds are in MLS Cup this weekend, Vancouver regressed to only earn 1.15 points per game in 2016.


I hope you found this exploration of roster turnover interesting. If you want to explore this data yourself, I’ve posted the data – and the R code that generated the plots, to GitHub.

Leave a comment

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