clock menu more-arrow no yes

Filed under:

How computer models predict MLS will shake out

New, 2 comments

Nerds like sports too, and they’re better at predictions than you think

If you buy something from an SB Nation link, Vox Media may earn a commission. See our ethics statement.

MLS: MLS Cup Geoff Burke-USA TODAY Sports

It’s the ultimate division in American culture: jocks and nerds. One group does sports; the other group does their homework. Ne’er the twain shall meet, unless their names are Leonard and Penny. That’s not really true, of course. There are plenty of athletes with advanced degrees, such as this PhD holder, and plenty of others who are really smart, including this guy.

As of right now, the US’ ubernerd is probably one Nate Silver. He is best known for the website FiveThirtyEight.com, which over the past few cycles has proven to be one of the more accurate election predictors. It is not a polling site, but uses polling and other data dumped into complex mathematical and statistical models to come up with its predictions. However, Silver started life as a baseball analyst, and is the inventor of PECOTA, a hitting and pitching performance analysis system. Think Moneyball for fantasy players. FiveThirtyEight.com itself, is in fact owned by ESPN.

538 itself looks at plenty of non-political stuff, including, obviously, sports. Not just baseball, but also football, basketball, and, yes, soccer. Up until this season they had not looked at MLS, but added the league to their coverage this year. They have what they refer to as the Soccer Power Index (SPI), which they use as the basis for their predictions. The detailed predictions for MLS can be found here. They are, to say the least, interesting. Just in the past few days, Paul Carr, ESPN’s senior soccer analyst, tweeted out this 538 prediction:

The prediction shows Toronto FC as the likely Supporters’ Shield winner. That’s not too surprising. It also shows Atlanta United having a 4% of winning, and the top five candidates all being in the Eastern Conference. Again, that seems very probable, although I personally think the Atlanta chances are rather better.

But here’s where it gets interesting. 538 makes several other MLS predictions. Here’s how they pan out for Atlanta United:

  • Overall SPI: 34.8, 8th overall, 5th in the Eastern Conference
  • Total points for the season: 54, tied for 5th overall with Seattle, and also 5th in the East
  • W-D-L for the season: 15.5-7.9-10.6. Reset as W-L-T and adjusted to match the points, that’s 15-10-9
  • End-of-season goal difference: +17, 3rd both overall and in the East
  • Chance of making the playoffs: 89%
  • Chance of first-round bye in the playoffs: 15%
  • Chance of winning MLS Cup: 6%

That deserves some examination. First, United currently are 10-7-5 with 35 points. That means 538 think we are going 5-3-4 for the remaining 12 games for an additional 19 points. Given that 8 of those games are at home, and looking at the strength of the remaining schedule (which is low), that looks very pessimistic. Also odd is the goal difference, which would be only +4 for the remaining games. Again, we should do way better than that.

The predictions also give Atlanta a much better chance of a first-round bye (15%) than of winning the Supporters’ Shield. That makes sense, since that only requires second place in the conference. On the other hand, we also get better odds, 6%, of winning MLS Cup. At first glance, that seems contradictory. In fact, it’s not.

The way 538 generates its soccer predictions sounds like a lot of fun. What they do is to run an entire simulated season lots of times over, repeating the process after every round of games. These are called Monte Carlo simulations. How many times do they run the simulation? Ten thousand every week. Obviously, they can’t do that by playing FIFA 24/7, so it really isn’t much fun after all, and Atlanta United isn’t even in FIFA yet anyway. Instead, they feed four data points into an algorithm and let the computers have at it. The data points are: goals, adjusted goals, shot-based expected goals and non-shot expected goals. If you know what “expected goals” are (or xG to acronym lovers), congratulations, you too are a soccer nerd. If not, well, you’re probably better off not knowing. If you need some sleep, you can geek out and learn about them here.

However, this is also why Atlanta United gets a better chance of winning MLS Cup than it does of winning the Supporters’ Shield. In the simulations, the playoffs have to be treated as a separate season. Which is what they really are when you think about it. So the two chances are not directly correlated.

So: what do you think of 538’s predictions? Are you going to rush off to Vegas to lay down your life savings based on Silver’s mathematical wizardry, or are you going to rest comfortably knowing that this proves nerds should just stick to the sciency stuff and leave sports to the rest of us?