[ChrisE]

I'm interested in looking at how rosters are changing in MLS from year to year; it's not really a question of whether or not teams win games (I doubt we'd see much correlation), but of how the league is changing. However, I'm having some doubts about what would be the best way to do this, so I thought I'd solicit the opinion of the board.

I can think of two ways to look at how rosters are changing, though I'd imagine there are more. One would be comparing the rosters (or, because I don't have that information, the players who have played at least one minute) of the teams from different years.

To illustrate, here's how Columbus's 2002 and 2003 rosters would look:

Player Name	2002	2003	Both
Dunseth, Brian	1	1	2
M'neuve, Brian	1	1	2
Torres, Daniel	1	1	2
C'ham, Jeff	1	1	2
West, Brian	1	1	2
Denton, Eric	1	1	2
Oughton, Duncan	1	1	2
Clark, Mike	1	1	2
Martino, Kyle	1	1	2
Presthus, Tom	1	1	2
Buddle, Edson	1	1	2
Perez, John	1	0	1
Mcbride, Brian	1	1	2
Busch, Jon......	1	1	2
McCarty, Chad	1	1	2
W'ton, Dante	1	0	1
Leitch, Chris	1	0	1
Harkes, John	1	0	1
Warzycha, Robert	1	0	1
Garcia, Freddy	1	1	2
Lapper, Mike	1	0	1
Yeagley, Todd	1	0	1
Matteo, Jeff	1	1	2
Paule, Ross	0	1	1
Hejduk, Frankie	0	1	1
Walsh, Diego	0	1	1
Akwari, Nelson	0	1	1
Williams, Mark	0	1	1
Chacon, Alex	0	1	1
Perea, Trevor	0	1	1
Ritch, Michael	0	1	1
Traeger, Jake	0	1	1
Number playing	23	25	16

So, if you look it this as from 2002 to 2003, you have 16 of 23 players playing both years, 7 new players. However, if you look at it from 2003 to 2002, you get 16 of 25 players playing both years, 9 new players. I'd imagine it would be obvious that you want to use 9, but I have no idea how to quantify this.

Things get significantly harder when you use minutes played instead of rostered players:

Columbus	 team     2002	2003	2002-2003	2003-2002
Dunseth, Brian	2436	1589	847	0
M'neuve, Brian	2238	1730	508	0
Torres, Daniel	2085	562	1523	0
C'ham, Jeff	1962	1405	557	0
West, Brian	1944	1543	401	0
Denton, Eric	1875	2336	0	461
Oughton, Duncan	1608	1717	0	109
Clark, Mike	1599	2673	0	1074
Martino, Kyle	1455	1765	0	310
Presthus, Tom	1365	613	752	0
Buddle, Edson	1304	1509	0	205
W. Perez, John	1256	0	1256	0
Mcbride, Brian	1239	2183	0	944
Busch, Jon......	1236	2194	0	958
McCarty, Chad	1176	1226	0	50
W'ngton, Dante	1122	0	1122	0
Leitch, Chris	989	0	989	0
Harkes, John	739	0	739	0
Warzycha, Robert	495	0	495	0
Garcia, Freddy	215	1278	0	1063
Lapper, Mike	141	0	141	0
Yeagley, Todd	74	0	74	0
Matteo, Jeff	14	286	0	272
Paule, Ross	0	2212	0	2212
Hejduk, Frankie	0	2128	0	2128
Walsh, Diego	0	612	0	612
Akwari, Nelson	0	488	0	488
Williams, Mark	0	391	0	391
P. Chacon, Alex	0	325	0	325
Perea, Trevor	0	34	0	34
Ritch, Michael	0	21	0	21
Traeger, Jake	0	12	0	12
	28567	30832	9404	11669

Hopefully this is clear. The first two rows are the minutes played by player x in 2002 and 2003. The third row is their 2002 minutes - their 2003 minutes; if the number is less than 0, though, it simply becomes 0; otherwise, you end up with just +/- 28567-30832. I'm not sure that is the right way to do this at all, however (in fact, I really doubt it).

A better way might be to take whichever is the lower number between the two, which would obviously be the minutes that the player played in both years, summing them up, and dividing by one (or both??) year's total minutes. In this case, we get 19163, which, divided by 28576+30832 gives us 32.2%.

I assume it's pretty clear that I'm completely at a loss here, any suggestions would be appreciated.

[mpruitt]

Doing it by minutes would obviously be a lot more precise, but I'm not sure how you would group that in terms of minutes played from one year to the next in any kind of understandable way. I think however that trying to get that percise may be muddying the issue. Maybe try and look at it this way... Any player on a roster has the potential of being used, that is pretty signifigant. Any player not on the roster obviously does not have the potential of being used. Whether or not they're being used I think is somewhat insignifigant. However, of course if ou do it that way then your numbers are goign to get muddied by guy swho sit on the roster as a developmental player. Perhaps an easier way to do this would be looking at Games Started. That way you have some idea of who's made a signifigant contrabution to the team, but don't get bogged down by minutes played. Games started to might get you around the problem of those extra guys too that use up DV spots etc.

[ChrisE]

Originally posted by maxim-1
Doing it by minutes would obviously be a lot more precise, but I'm not sure how you would group that in terms of minutes played from one year to the next in any kind of understandable way. I think however that trying to get that percise may be muddying the issue. Maybe try and look at it this way... Any player on a roster has the potential of being used, that is pretty signifigant. Any player not on the roster obviously does not have the potential of being used. Whether or not they're being used I think is somewhat insignifigant.

I really disagree here, Maxim. There's no particularly good example here, but for Dallas (e.g.) D.J. Countess went from playing 100 minutes in 2002 to playing over 2000 in 2003, while Chivas Martinez went from 1700 to 560. If you just use whether these guys are on the roster or not, you're ignoring the fact that Martinez got replaced by a superior player and Dallas had a significant change at goalkeeper. I admit, it may be possible that you're going to run into other problems here (nats call-ups, injuries) but I'd at least like to try something more complicated than simply counting the rostered players.

Perhaps an easier way to do this would be looking at Games Started. That way you have some idea of who's made a signifigant contrabution to the team, but don't get bogged down by minutes played. Games started to might get you around the problem of those extra guys too that use up DV spots etc. [/B]

This is a good idea, I feel sort of silly for having thought of it, but it runs into the same kinds of problems as minutes played. Teams haven't generally played the same number of games from year to year, and even if they had, I don't run into this problem of not really knowing what I'm supposed to be adding up.

(an additional problem with just counting players is that I don't have info for these developmental players like Andres Murriagui or Johanes Maliza or even Brian Carroll who didn't get into any games this year. Some developmental players may get credited for a season because they got 5 minutes, while guys who didn't play at all will necessarily be ignored)

[mpruitt]

Yeah the counting of it has me a bit stumped too but I'm kinda on overdrive with this stuff right now. Your point about minutes being important is valid, just would make it infinately more complicated. There has to be some kind of plus/minus percentage change and then lump 'em all together but if anything my little project has told it's that I should've studied harder in math.

[beineke]

Regardless of which numbers you plug into the two columns, I think you'd do pretty well by taking their correlation. I think the excel command is correl().

BTW, this is an interesting question ... I'd also like to see comparisons across multiple seasons. Quite a few of the current Quakes are entering their fourth year together. Not sure if that's a good thing, though -- last year, quite a few Crew players had been together for a long time, too.

[ChrisE]

moving on...

I decided to use the minimum of minutes for any player between the two years. That, I suppose, is the minutes they played in common those two years. I don't really think the subtraction thing makes sense after all, though I'm not wholly convinced of this method either. Additionally, I averaged the total minutes available in the two years being averaged. For turnover, I just divided the minimum minutes sum by the averaged available minutes. I'll post those numbers (or share the file) if anybody likes, but they don't really seem necessary, so here I've only included the ratios.

The numbers ended up being a great deal higher than I had expected, considering that this seems to be a relatively stringent way of determining turnover.

Turnover		1997	1998	1999	2000	2001	2002	2003	Club
Chicago		------	------	0.759	0.622	0.679	0.627	0.565	0.652
Colorado		0.392	0.657	0.521	0.419	0.412	0.333	0.623	0.480
Columbus		0.472	0.473	0.698	0.594	0.525	0.562	0.645	0.567
D.C		0.667	0.568	0.664	0.675	0.389	0.349	0.479	0.542
Dallas		0.436	0.633	0.671	0.695	0.687	0.646	0.543	0.616
K. City		0.548	0.677	0.398	0.441	0.604	0.561	0.676	0.558
L.Angeles		0.648	0.597	0.744	0.685	0.670	0.639	0.712	0.671
M'stars		0.444	0.392	0.385	0.431	0.599	0.441	0.257	0.421
N. Eng		0.384	0.317	0.433	0.360	0.532	0.308	0.582	0.417
San Jose		0.511	0.456	0.406	0.418	0.290	0.693	0.480	0.465
Miami		------	------	0.408	0.580	0.451	------	------	0.480
Tampa Bay		0.581	0.385	0.388	0.521	0.526	------	-------	0.480
Yearly		0.508	0.515	0.540	0.537	0.531	0.516	0.557	0.530
Min		0.384	0.317	0.385	0.360	0.290	0.308	0.257	0.257
Max		0.667	0.677	0.759	0.695	0.687	0.693	0.712	0.759

The all-time most stable MLS teams have been, unsurprisingly, the two best, Chicago and Los Angeles. The teams with the all-time highest turnover have been, unsurprisingly, the two worst (I believe), New England and NY/NJ Metrostars. However, there have been several teams that seriously made over the roster and found great success, most notably 2001's Earthquakes and 2002's New England. The Metrostars of this year were the all-time leaders in turnover, while Chicago's 1999 championship successor was the least turned-over. Perhaps the most surprising result is that MLS turnover ratios appear to change very little by year.

edit: 1997, e.g., refers to the combination of 1996 and 1997.

[ChrisE]

Furthermore, I did the second part of this investigation, looking at the number of players who carried over from year to year. I divided that number by the average number of players who played for the team in those two years (I suspect I should have used the max...). The numbers hew quite closely to each other, with a .862 correlation (Beineke or somebody will have to explain exactly what that number means). Here's the table:

Ratio		1997	1998	1999	2000	2001	2002	2003	Club
Chicago		------	------	0.680	0.549	0.627	0.642	0.549	0.609
Colorado		0.409	0.591	0.558	0.458	0.453	0.490	0.609	0.508
Columbus		0.556	0.560	0.756	0.636	0.609	0.609	0.667	0.625
D.C.		0.692	0.600	0.720	0.692	0.471	0.440	0.449	0.582
Dallas		0.468	0.706	0.653	0.723	0.578	0.545	0.640	0.619
K. City		0.465	0.667	0.583	0.500	0.622	0.558	0.727	0.589
L.Angeles		0.625	0.654	0.694	0.706	0.625	0.667	0.612	0.655
Metros		0.424	0.441	0.464	0.481	0.577	0.364	0.250	0.427
N. Eng		0.436	0.340	0.566	0.481	0.566	0.415	0.471	0.468
San Jose		0.553	0.490	0.360	0.500	0.426	0.750	0.600	0.517
Miami		------	------	0.426	0.607	0.471	------	------	0.500
Tampa Bay		0.622	0.453	0.526	0.571	0.542	------	------	0.540
Yearly		0.523	0.545	0.576	0.574	0.546	0.539	0.550	0.552
Min		0.409	0.340	0.360	0.458	0.426	0.364	0.250	0.427
Max		0.692	0.706	0.756	0.723	0.627	0.750	0.727	0.655

[numerista]

Originally Posted by ChrisE

The numbers hew quite closely to each other, with a .862 correlation (Beineke or somebody will have to explain exactly what that number means).
[/b][/size][/code]

Resurrecting an old thread ...
I'm not sure which numbers you're referring to, but I expect that it means that some teams have consistently high (and low) rates of turnover. In any case, that's certainly true.

Looking at pre-season rosters, my impression is that 2004 is shaping up to be a very low turnover season ... for instance, I believe that every one of the top 20 2003 draft picks is still in the league, and 18 of them are with their original teams. Is that impression easy for you to confirm/refute?

[ChrisE]

Originally Posted by numerista

Resurrecting an old thread ...
I'm not sure which numbers you're referring to, but I expect that it means that some teams have consistently high (and low) rates of turnover. In any case, that's certainly true.

That's not what I meant. I was comparing turnover-by-players to turnover-by-minutes, which unsurprisingly are very strongly related to each other.

Although I'm sure the Metros have a higher turnover than the Galaxy (e.g.), I'm confident it wouldn't be anywhere near that high (doing a quick calculation, I got around .38 for both).

Looking at pre-season rosters, my impression is that 2004 is shaping up to be a very low turnover season ... for instance, I believe that every one of the top 20 2003 draft picks is still in the league, and 18 of them are with their original teams. Is that impression easy for you to confirm/refute?

I get the same impression, but I really have no idea, I don't have any other preseason rosters to compare to (if anybody has any old media guides/season previews...). I'll see what I can do to look at offseason turnover, but I'd rather wait until the season starts. The reason being, I'm not really confident in the rosters that MLS has given us yet (I'm pretty sure you're right about last year's rookies, but MLS's preview doesn't list Scot Thompson on LA's roster), and several teams still have developmental slots available.

[ChrisE]

If we want to predict turnover, we've got to know what causes it. A lot of things result in new players being brought in, but probably the easiest to track is trades. In MLS's history, there have been 151 players who have played for more than one team in a season (two who have played for three, Orlando Perez and Winston Griffiths). Nobody needs to see a chart, but I like em, so here it is:

Year	Teams	Players	Ratio
1996	10	7	0.70
1997	10	20	2.00
1998	12	23	1.92
1999	12	27	2.25
2000	12	20	1.67
2001	12	26	2.17
2002	10	14	1.40
2003	10	14	1.40
Total	88	151	1.72

If beineke were still around I'd leave the interpreting to the more qualified, but since he's absolutely nowhere to be found, I'll just assume that we ought to expect something around 1.7 players added to each team's roster via mid-season trade.