clock menu more-arrow no yes

Filed under:

Orlando City’s Underlying Stats Show Significant Improvement

New, 1 comment

The Lions’ early-season performances are providing hope as the numbers trend away from last year’s decline.

MLS: Orlando City SC at New York Red Bulls Vincent Carchietta-USA TODAY Sports

We’re only four games into the MLS season and the Lions are still finding their feet. Orlando City permanently parted with 15 players from last season and brought in 14 new faces as James O’Connor took to rebuilding the squad in his first off-season in Central Florida, a recruitment drive that currently accounts for 52% of the roster.

After such a bad season, changes were obviously needed, but at such a drastic level, the outcomes can often be unpredictable. It can take time for the team to gel, for new players to form an understanding of their new teammates, and of a new coach’s style and philosophy. Likewise, coaches can spend the early part of a season figuring out how to get the best from their new players on the field as well as man management off the field.

But even in these early stages, the underlying stats suggest Orlando City has come out fighting in 2019 with total improvement across the board using one key model: expected goals. Expected goals, more commonly noted as xG, is the trendy new stat in soccer analysis that boomed in 2017 after it was picked up by Opta and has continued to increase in popularity ever since. For those unfamiliar with xG, it is a metric that grades the quality of a chance using various statistical models to assess the likelihood a shot will become a goal.

The figure is a percentage shown as a decimal (e.g., a shot valued 0.6 has a 60% chance of becoming a goal). The total from all the shots are added at the end of the game, giving the expected total for each team. It’s an underlying stat that can often tell us more about a game than the actual score line does: who’s taking chances, who’s squandering them, who is giving up the most, etc.

How Can We Use xG?

As an example, take a look at last season’s six-game win streak under Jason Kreis. In hindsight, you could have predicted the unsustainability of such a streak by using xG modelling. Orlando’s expected goals for (xGF) total was 11.46, 5.54 — less than the 17 that was actually scored. On the defensive side, the team’s expected goals against (xGA) was 8.9, only 0.1 short of their actual total of nine. In short, the attack was over performing and the defense had simply played to its level. More telling was the game by game split. Whereas Orlando took all 18 points in real life, xG modelling suggested the team would ordinarily have only walked away with half that, winning two of the six games with a further three draws and one loss. The expected loss was Orlando’s away trip against the Philadelphia Union, a game that in reality finished 2–0 to the Lions but had an expected scoreline of 1.53–1.26 in favor of Philadelphia, which rounds to a 2–1. The 2.27 difference between the expected score and the actual outcome was the biggest in Orlando’s favor across last season.

This Year’s Numbers

Hopefully that gives you a better understanding of just what xG can mean as I turn your attention to this year’s numbers. The data set is split into three parts: Jason Kreis’ 15 games in charge of the 2018 season, the 17 games after James O’Connor took over in 2018, and finally, the opening four games of the 2019 season. For ease of comparison, the goals for (GF), goals against (GA), goal difference (GD), and points (P), as well as their expected (x) variants are converted into per game (/90) numbers. The final value, here noted as eta (η), is efficiency, or the difference between the real outcomes and expected values.

On a per game basis, the team has an xGF of 1.48, similar to the level of Jason Kreis’ 1.47 and a vast improvement of O’Connor’s own spell last season. However, unlike last season, the team is actually in debt to the tune of -0.23 per game, essentially meaning across the opening four games the team would have been expected to score one goal, and when translated across a 34-game season it could mean the team has under-performed by 7.82 goals. Across the league, the 1.48 xGF per game is good enough to rank in the middle of the pack, at 11th.

The team has also improved defensively, conceding an average of below two goals (1.50) per game for the first time since June 2, 2018. Last year under O’Connor the team conceded 2.18 goals per game. Likewise, Orlando’s xGA for this season is the lowest of the data set, at 1.32. The efficiency, although highlighting that the team is conceding more goals (0.18) than the xG model would suggest, is under performing at a much lower rate than before which, when combined with the improved xGA figure, is an early indication that 2018’s defensive record could become a distant memory.

Using the xGF and xGA figures we can see that the goal difference once again shows improvement. In essence, Orlando City is playing closer games than the blowouts it was last season. While the real life average goal difference remains a negative, the -0.25 deficit is an incredible turnaround from O’Connor’s 2018 numbers, which in reality stood at over a goal per game (-1.06) which, across all games, meant the Lions were outscored by 18.

It’s early stages, of course, but despite Orlando City’s real life negative GD, the Lions’ current xGD is the only one from the data sets that is positive, suggesting the team has actually gotten the better of the opposition across the opening four matches. With the -0.41 η rating between reality and expected goal difference, we can cite a combination of minor inefficiencies on either side of the ball in comparison to the offensive over performance and defensive ineptitude that still led to the mass deficits in last season’s numbers. For reference, across all 34 MLS games in 2018, including the two Bobby Murphy was in charge for, the Lions outperformed its xGF by only +0.9 goals (total, not per game) but conceded 14.09 more goals than xGA modelling suggested they should have.

Finally, extrapolating the game by game data we can work out that Orlando City is currently matching its expected points total of five (1.25 per game), an η score of 0. The xP is a huge 0.72 more per game compared to O’Connor’s 2018 expected per game total of 0.53 points. Across a full 34-game season, that per game difference equates to a staggering additional 24.48 points. Curiously, despite having the exact same 1-1-2 real world record that the xG data suggests Orlando should have, the model actually predicted City to win the opening weekend 2–2 draw by a score of 2–1, while on paper the Lions should have lost to Chicago on the road 1–0 in week two instead of the 1–1 draw they actually got. The team arguably did well to get anything from that game as the 0.28 xGF total for Orlando is the fourth lowest single-game xGF in Orlando’s entire MLS history. The other team games were expected to end in a draw: 2–2 against Montreal instead of the real life 3–1 loss, and a 1–1 draw in last weekend’s 1–0 win over the Red Bulls.

Orlando City has not only improved compared to its own standards, but it stacks up pretty well compared to the rest of the league, ranking 11th in xGF, 11th in xGA and ninth in xGD, consistently good to middling in every area. xG isn’t the be all and end all, and with such a small sample size, as well as the team’s historical inconsistencies, it is perhaps difficult to read too far into stats. But in short, the start of the 2019 season has seen Orlando City improve in every measurable xG factor and that can only be promising: the team has managed to keep up its offensive chance creation albeit with a negative η rating for a change and has seen a significant uptick in defensive form, both in terms of reducing the chances given up and increasing the efficiency with which it defends them.