Statistical Scouting: The 2017 MLS SuperDraft

Last year my Superdraft model did fairly well, given that it only uses college and PDL team name, position, GA status, and USYNT history. However, this year it didn’t seem to be as accurate with three North Carolina players(Colton Storm, Walker Hume, and Tucker Hume) projected to go over 2000 minutes with none seeming to be particularly close, leading me to try a different method. You can find that model under Raw Computer Model Exp. Mins in the posted spreadsheet with all the raw data used for my new method.

For the new method, I first created a polynomial model for minutes played in the first two years and then used mock drafts and big boards to simulate where the player was picked and run the player’s position in the mock draft/big board into the polynomial model to create Expected Minutes played numbers.

Y1+Y2 Minutes Projections(sorted by Median)

You can view the full viz here, it takes a while for it to load sometimes.

A couple of notes:

Since Abu Danladi was only selected first, second, or third and was most often selected second, his box plot is just a straight line. However, this doesn’t mean that Danladi has a very high chance to play exactly 2690 minutes, the box plot appears this way because he is selected second in more than half the mock drafts and 2690 is the polynomial model’s prediction for minutes for the second pick.

Also this has a slight bias towards Soccer By Ives’ draft material since it has published 5 mock drafts and big boards compared to TopDrawerSoccer’s 3 and’s 1. This reflects especially in Reagan Dunk’s boxplot, as SBI has ranked him anywhere between 21 and 31 compared to the other two sources’ rankings between 5 to 11.

Finally, I used a clustering algorithm to cluster the results to form more concrete “rankings,” that describe the difference between two players that ranking does not provide. The difference between Jeremy Ebobisse and Abu Danladi is one spot if you rank the players, the same difference between Jacori Hayes and Eric Klenofsky. However, there is really no difference between Ebobisse’s and Danladi’s expected minutes (0 difference in median exp. mins) and there is a much bigger difference between Hayes’ and Klenofsky’s expected minutes(472 difference in median exp. mins).

Positions Cluster# Players(Median Exp. Mins)
1-3 1 Abu Danladi(2690)
Jeremy Ebobisse(2690)
Miles Robinson(2507)
4-8 2 Jackson Yueill(2176)
Brandon Aubrey(2026)
Jonathan Lewis(2026)
Chris Odoi-Atsem(1886)
Jacori Hayes(1886)
9-13 3 Eric Klenofsky(1414)
Nick DePuy(1414)
Niko Hansen(1316)
Shamit Shome(1280)
Lalas Abubakar(1140)
14-20 4 Justin Schmidt(1061)
Zeiko Lewis(989)
Colton Storm(922)
Brian Wright(804)
Julian Gressel(804)
Francis DeVries(752)
Jordan Wilson(752)
21-28 5 Reagan Dunk(705)
Christian Thierjung(661)
Adonijah Reid(652)
Walker Hume(585)
Brian Nana-Sinknam(521)
Guillermo Delgado(469)
Nazeem Bartman(469)
29-46 6 18 players
47-69 7 23 players

I tried clustering by third quartile out of curiosity and out of the players within the top 5 clusters(28 players both times) only one player moved out, Brian Nana-Sinknam, replaced by Napo Matsoso. Dunk shot up two clusters while Francis DeVries and Jordan Wilson dropped down one cluster but really there was not any notable change. Interestingly enough FC Dallas whittled down the combine invitees to a 29 player list, it would be interesting to see the overlap. San Jose have the 6th and 28th picks so are guaranteed to get a second cluster and fifth cluster player.

Pick #6: A player that shone through the models but is not receiving much hype is Akron’s Jonathan Lewis. He is only 19 years-old, and finished tied for first place in Division 1 in assists(12) and plays on the wing, where the Quakes are thin as of now. He is also a Generation Adidas prospect so he does not count against the cap and will be protected automatically in next year’s Expansion Draft. Jacori Hayes and Brandon Aubrey also look like enticing options. Jackson Yueill probably will not fall down to 6 and Chris Odoi-Atsem plays fullback, where the Quakes are actually young and have good depth. Shamit Shome’s GA status also brings him into the picture.

Pick #28: Just pick the whatever player is left. Probably players from cluster four will not drop so far but if one does it should be close to a no-brainer. A couple of players are shortchanged by using median like David Goldsmith and Kwame Awuah, who were highly rated in early TopDrawerSoccer rankings but fell off and thus have medians of 0.

*I only used Soccer By Ives, TopDrawerSoccer, and for their big boards and mock drafts since I trust their credibility. Other mock drafts are can be found by NBCSport’s ProSoccerTalkBleacherReport, and SBNation.

**I used the second version of the MLS Combine invite list. This excludes players like Trevor Haberkorn and Alex Neff who appeared on multiple mock drafts and bog boards. Haberkorn would be an interesting pick  because he played with Quakes Homegrown Nick Lima for four years and partnered with Homegrown prospect Josh Morton for three years on the Cal backline.

You may also like...

Leave a Reply

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