MLS Shootout (part 2) - One Table to Rule Them All

The MLS shootout was introduced in 1996 to prevent American sports fans from suffering through a tied soccer game.  The assumption was that we would prefer an NHL hockey-style shootout to a game without a winner and that we wouldn't mind becoming the laughing stock of the rest of the soccer world.

Assuming a competitively balanced league where any team has an equal probability of winning, losing, or tying any game, I showed in the last post that teams have less incentive to fight for a win in a late tied game now that there is no shootout.  That is, teams are more likely to settle for a tie now that it guarantees a point.  However, when looking at the percent of tied games each season, it became clear that the jump in the percent of tied games didn't occur immediately after the shootout era, which ended after the 1999 season (see the reproduced graph below).  If we take a closer look at the data, we can see that the percent of games ending in a tie has more to do with conferences and the rules of playoff entry than shootouts.

The reason for this is simple: since the top teams in each conference gain entry to the post-season, teams are less averse to sharing points when playing non-conference opponents.  For example, when Columbus Crew SC tied the Vancouver Whitecaps this past April, the Crew didn't mind sharing points with a Western team. The point given away to Vancouver had no bearing on the Crew's playoff chances since Vancouver competes for spots in the Western Conference.  However, the 3-3 draw Columbus had with Toronto was less beneficial as it gave away a point to Toronto, which remained 3 points behind Columbus in the Eastern standings.  So on net, Columbus gained no ground on Toronto whereas the tie against Vancouver gained Columbus a point against all Eastern teams.

It stands to reason then that seasons with more inter-conference games should have less ties as teams fight harder to net 3 points (3 to them - 0 to opponent = 3 net points) instead of 0 (1 to them - 1 to opponent = 0 net points) against conference opponents.  This has already been shown to be the case in NHL by Shmanske and Lowenthal (2007).  But what about MLS?

Taking a look at the percent of tied games plotted against the percent of games played between opponents of the same conference for all non-shootout seasons, it seems that this is indeed the case.  There is a clear downward trend indicating that the greater the percent of inter-conference games, the less the percent of tied games there will be.

Use the data and R script to reproduce this graph and all others below.

What is up with those outlier years with 100% of games being between same-conference teams you ask?  Although MLS actually had 3 "conferences" (East, Central, and West) for the 2000-2001 seasons, entry into the playoffs was granted to the top two-thirds of all teams regardless of conference.   This was also true in 2002 although the league reverted back to 2 conferences (East and West).  So in practice the conferences only served as lists of teams that were geographically close and played each other more often.  This is why I'm treating those years as if 100% of the games played were inter-conference; for the sake of playoffs, they were.  

You have to appreciate the irony here.  In the year where MLS seemed to rebel once more against European football and create 4 team-conferences a la the NFL, what it really had was a single table with an unbalanced schedule.  Single-table purists should add this fact to their arguments; a single table should decrease the number of tied games in MLS.

Back to the shootout era.  Adding these years to the plot, we see that outside of that crazy 1999 outlier season, the pattern still holds.  Additionally, we see the effect the shootout had on tied games independent of conference effects.  Again, outside of 1999, the shootout seasons saw a lower percentage of ties compared to other seasons with the same percent of inter-conference games.

So what can MLS do to minimize the amount of tied games we see? There are two changes it could implement: a single table for playoffs and/or reintroduce the shootout.  Although the latter would explicitly eliminate ties as well as reduce the number of games that ended regulation in a tie, I don't actually want to return to the shootout days. Neither does MLS and neither do you.  But a single table?  When asked about the possibility of a single table back in 2010, Don Garber replied

...every year we do deeply analyze whether or not it makes sense for us to have a single table and no playoffs. We also evaluate whether it makes sense to have a single table and playoffs, or whether it makes sense to have conferences and playoffs.

...We're going to change it if we believe we could have a more compelling format and one that might be perhaps more balanced. ...It should be a format where there is more media coverage, more television ratings and more attendance as we get down to an event that is a single, stand-alone event, our championship game, the MLS Cup.

...All of those can happen within a single table format. The single table discussion is whether there is a single table or whether there are conferences?

I have no idea what goes on when the top brass of MLS meet to deeply analyze using a single table.  What I do know now is that the evidence above strongly suggests that doing so would reduce the number of ties.  It would also appease the Eurosnobs out there who use words like "cappo" and refer to games as "matches" and fields as "pitches" and....ok I'm getting nauseous, enough.

And if a single table is not in the works, we could  theoretically baby-step our way to one by returning to the days when last playoff positions were given to the best teams regardless of conference (2007-2011).  For example, in 2008 the top 3 teams in each conference were granted access to the playoffs along with the remaining top 2 teams regardless of conference.  Although this yielded a strange situation where a team from New Jersey, the "New York" Red Bulls (NYRB), entered the playoff bracket in the West and was crowned the Western Conference Champions, it should have provided incentive for NYRB to give more effort against Western Conference teams during the regular season since they were competing for that last playoff spot with Western Conference teams.

2008 MLS Standings

Eastern Conference		Western Conference
team	points	team	points
Columbus Crew	57	Houston Dynamo	51
Chicago Fire	46	Chivas USA	43
New England Revolution	43	Real Salt Lake	40
Sporting Kansas City	42	Colorado Rapids	38
New York Red Bulls	39	FC Dallas	36
D.C. United	37	San Jose Earthquakes	33
Toronto FC	35	LA Galaxy	33

                                                               *teams in bold admitted to playoffs

Both Sporting Kansas City and NYRB  had to earn more points than the 4 lowest ranked teams in the Western Conference to make the playoffs in 2008.  Thus, their intra-conference games should have been given the same importance as their inter-conference games.  The same should be true for all teams ranked around the last playoff spots and we should see more teams push for wins in closely fought games.  Although this incentive should exist in theory, it's not blatantly obvious at first glance.

The years for which some teams are ranked on a single table to gain entry into the playoffs are represented by blue circles.  The areas of the circles represent the proportion of teams that will make it to the playoffs via a single table comparison.  The absolute number has been as few as 2 teams and as many as 10 teams.  If the single table comparison caused more teams to push for wins in late tied games, we would see less ties when the circles are bigger and the circle years should be below the dots representing no single table comparisons.  Unfortunately, the years that used a single table for some teams are not close to many comparison years in the plot for us to draw a conclusion.

Moving forward, I think things are complex enough that it's time to carry out a logistic regression on game level data so that we can simultaneously control for all the moving parts influencing tied games in MLS.  These include but are not limited to, overtime structure, conference structure, playoff structure, and of course competitive balance within the league.  Maybe then we can find out just what was going on in that 1999 season.

This project started out in my mind as a simple investigation of the incentive effects of the MLS shootout, a rare gem in soccer history.  Upon diving into the data, it has become so much more.  That is how research usually goes.  And for all of my projects, you can replace the word "usually" with "always".  But this is why you should choose topics you love.  Then you don't mind so much when it takes 10 times as long as you thought it would.

The MLS Shootout (Part I) - Fast Kicking, Low Scoring, and Ties? You Bet!

American sports fans aren't known for their love of tied games.  Other than MLS, the NFL is the only league to currently allow games to end in a tie and we usually go an entire season without even one tied game.  Fearing that Americans couldn't support a soccer league with games that frequently ended in a draw, MLS initially decided to end all tied games with a shootout in its inaugural season.  Not the penalty shootouts you're use to seeing in the World Cup or UEFA mind you, but more like a NHL-style shootout where players charge at the keeper and try to get a shot off before the 5-second clock runs out. Because that would be more American.  For those of you who didn't have the pleasure to see one of these in person, by the magic of Youtube, you still can.

 Now this is clearly a sub-optimal way to decide a winner. MLS did at least try to dissuade games from ending this way by awarding 3 points for a win in regulation and only 1 point for a shootout victory.  So were Americans more entertained by a NHL-style shootout? Did less games end in a tie?  Should we have expected them to?

The answer to the first question is of course subjective but my guess is that the answer is a resounding "no".  Otherwise, we'd still have the shootout.  To answer if less game ended in a tie and whether or not we should expect them to, we can turn to the data and probability theory.  Let's first see what we should expect to happen with a shootout.

Consider, as we did in the last post, a league where all teams are of equal strength and the likelihood of any team winning, losing, or tying a game is equal to 1/3.  Now with shootouts, when a game in regulation is tied each team will have a 50% chance to win the shootout.  So we have 4 possible outcomes for any given team, they can win, lose, win in a shootout, or lose in a shootout.  The probabilities for the outcomes are PW=PL=1/3 and PSOW=PSOL=1/6.  So we can model the outcomes of the games as draws from a multinomial distribution with four outcomes.  Of course these probabilities are somewhat arbitrary.  We could just as easily assume any numbers where PW=PL and PSOW=PSOL and still have an ideal competitive league. The probabilities chosen are simply for ease of illustration.  

To be specific, the probability density function (pdf) of points is

Before any game, a team can expect

with a variance of

so that

 

If a game is tied in regulation, then each team has a 50% chance of winning or losing and the conditional expectation becomes

with a variance of

 so that 

The expected points earned in a shootout is much less than in regulation but does have a lower variance.  So although the expected payoff in a shootout is lower, it is more certain.  However, given the large difference in expected outcomes, a team would have to be severely risk averse to prefer to end a game in a shootout. Or, abstracting from both teams being of equal strength, one team would have to believe it had a very low chance of winning in regulation to prefer the shootout.  There may be evidence that this is what the San Jose Clash were thinking in 1999 with their record 13 shootouts (more on that later).

So did this setup give teams incentive to try to win in regulation and avoid shootouts altogether?  Let's look at what the distributions for expected points are for games are now without the shootout.   We can repeat the process above for the situation where teams can win, lose, or tie with equal probability (1/3).  Or we could just plug in a value of 1 game played (gp=1) in the calculation we did in the last post.  Either way, we'll get

and of course if a game ends in a tie, each team gets a guaranteed 1 point with 0 variance. Now let's look at this all in a convenient table.

	Regulation (SO)	Shootout	Difference (SO)	Regulation (Ties)	Tie	Difference (Ties)
Expected Value	1.167	0.50	-0.667	1.333	1.000	-0.333
Variance	1.806	0.250	-1.556	1.556	0	-1.556

As we can see from the first 3 columns of data, the expected points earned when going from regulation to a shootout decreases by two-thirds (-0.667) and the outcome is much more certain.  In the last 3 columns, we see that the decrease in expected points after regulation is half that of the shootout era (-0.333 vs. -0.666)!  Furthermore, it's a guaranteed outcome of 1 point each.  So having a tied game in regulation loses less points in expectation without the shootout.  Saying it another way, teams have less incentive to avoid tied games when there is no shootout.  So it seems the answer to the question of whether we should expect less ties in the shootout era is, yes.  The payoff structure did suit the game well for avoiding games that were tied in regulation.  But did it actually work?

You can recreate the graphs using this data and this R script.

While the graph above indicates that games were less likely in the initial years to end in a tie after 90 minutes, the switch doesn't seem to perfectly coincide with the end of the shootout era, which lasted from 1996 to 1999.  So although the answer to our question of whether there was actually a higher chance of ending a game in regulation during the shootout era is "yes", we now have new questions that need to be answered.  What is going on in 1999 that there were so many games going to shootout? And more importantly, why does the jump in games tied in regulation occur in 2003 and not immediately after the shootout era?

For a hint, below is the same graph with lines drawn around the period of time for which MLS essentially had a single table.  I say "essentially" because there were actually multiple conferences during this time but entry into the playoffs was granted to the top 8 teams regardless of conference. This led to a somewhat odd situation where the entire Western Conference made the playoffs in 2002.

We'll dig into why divisions matter when it comes to whether or not games end in a tie next time.  Until then, I'll leave you with pop culture's initial reaction to MLS a la The Simpsons.