Using Forecasting to Explore How Players Impact Goal Differentials

Goal differential is the ultimate team stat. It’s beautiful in its simplicity. It’s easy to understand, but still revealing.  We can learn quite a bit about a team by simply looking at how many goals it’s scored and how many it’s allowed.

But even in its minimalist beauty, the weight of the messages hiding behind a team’s goal differential can’t be overstated. What looks like an elementary level stat actually can teach us quite a bit in predicting the future. 

While goal differential has two distinct components—goals scored and goals allowed—it’s entirely possible to evaluate a team’s future trajectory by looking only at its scoring. Sure, defense is important, but we still can glean quite a bit of information about a team’s direction by focusing purely on its scoring. Perhaps more accurately, we can do so by using math to quantify how much a club is over-performing or under-performing on offense and how that impacts its goal differential.

In this exploration, we’re going to compare each team’s actual scoring output to a scoring forecast churned out by a pair of LaxMetrics models. Using two different multi-variable regression models, we will come up with an average of “Expected Goals” and “Goals Over/Under Expectations” for each forward in the NLL, and then we’ll aggregate their values by team affiliation to create team forecasts. For the purposes of this exploration, we must hold each team’s defensive performance constant, assuming roughly the same production on that side of the ball from each squad in the second half of the season.

Here’s how it’ll all look. Let’s start by taking Albany’s Joe Resetarits for example. Resetarits is by all accounts having a phenomenal season for the Firewolves, perhaps even his best as a professional. The two regression models we are using each have one unique variable input paired with a constant variable input. In this case, the constant variable will be “shots on goal” and the unique variables will be “usage rate” and “shooting %+”. You can see more information on those stats on their respective pages. We’ll then take the average of the two forecast outputs to arrive at an average “Goals Forecasted” number for Resetarits. The result is shown below in Figure 1.

Figure 1

As the graph illustrates, Resetarits’s real-world total of  25 goals slightly exceeds both LaxMetrics forecasts and, therefore, also the forecast average. We can subtract the forecast average from his actual scoring total, and would describe the ensuing result as Resetarits having scored “+1.88 Goals Over Expectations”. This is a number that reflects well on Resetarits’s offensive performance, specifically his efficiency as a scorer. A negative value would indicate a lack of efficiency as a scorer with room for improvement.

Now with this concept of “Goals Over/Under Expectations” in hand, we can apply it to all of the forwards in the NLL and then aggregate their forecast averages to create a team forecast. That team forecast can be compared against each team’s actual goal total. For the purposes of this exploration, the goals scored total we are talking about will not match the overall goal totals that you might see on the league’s standings page or in a stat pack. Instead, these goal totals are strictly limited to goals scored by forwards, specifically the forwards for whom we are creating forecasts.

The below graph, Figure 2, demonstrates these team goal forecasts versus their actual goal totals.

Figure 2

We can compare each team’s goal forecast next to its actual goals scored total. Most teams fail to meet their forecasts, but a handful including Buffalo, Colorado, New York, and San Diego actually exceed their forecasts. Doing so illustrates a high degree of offensive efficiency. Unsurprisingly, three of those four teams are in playoff position entering Week 13.

Now let’s take the data from Figure 2, combine it with a little bit of basic math, and lay it out in a table that gives us a clear visual for some of the things we are about to discuss. Below in Figure 3, there is a whole bunch of information that might seem like too much at first. But hang with us, it’ll make sense in just a moment.

Figure 3

Let’s break the table down moving from left to right. In addition to the number of games played and total goals scored by a team’s forwards, we’ve included each club’s overall goal differential. This goal differential is the same value that you’d find on the league’s standings page. It includes all goals scored and allowed by a team, not just those contributed by forwards. In the next column, we’ve taken the goal differential of each team and divided it by that team’s number of games played to create a “Goal Differential Per Game” value.

Then in the lime green column dead center in the chart, you’ll see the total “Goal Forecast” for each team. As mentioned before, this “Goal Forecast” is the sum of each forward’s individual average goal forecast.

To the right of that column in the bright blue is each team’s “Goals Over/Under Expectations”. Like the Joe Resetarits example, this number tells us how many more or fewer goals a team has scored than would be expected by the LaxMetrics forecast average. A positive value indicates an efficient offense, while a negative value demonstrates the opposite.

Where the fun really begins is in the final two columns on the right side of the table in Figure 3. First, we have the average “Goals Over/Under Forecasted Per Game”. This is just the number in the bright blue column divided by the team’s number of games played. These values are demonstrated in the graph below in Figure 4.

Figure 4

The final column on the right side is the one that’s going to be the most interesting to us in just a moment. In percentage form, it’s the quotient we get from dividing each team’s “Goals Over/Under Forecast Per Game” by its “Goal Differential Per Game”. These percent values tell us what proportion of a team’s goal differential can theoretically be directly attributed to the team’s over-performance or under-performance of their goal forecast. Remember, for the purposes of this article, we are holding all defensive performances as constant so we can isolate the impact of offensive efficiency on wins and losses. We can see the percentages represented visually in the graph below in Figure 5.

Figure 5

Even just a cursory glance at Figure 5 reveals a trio of teams with particularly high values. Colorado, Philadelphia, and Saskatchewan each have “Goals Over/Under Expectations Per Game” values well over 100% of their “Goal Differential Per Game”. This is interesting because it means that each of the three could easily see their goal differentials swing either way in the season’s second half.

For Colorado, their overachievement at +1.227 goals per game is somewhat fragile, while Philadelphia’s razor-thin goal differential could easily be flipped in either direction. If the Wings improve upon their fairly unremarkable -0.341 Goals Under Expectations per game, they could position themselves to steal a few wins that they would have struggled to earn in the first half. The inverse is also true should Philly regress from that mark. Given how many overtime games they seem to play, a few added goals could make all the difference in the world.

Of the three, Saskatchewan is clearly the most interesting. It’s “Percent of Goal Differential Per Game” value tells quite a story.

With a “Percent of Goal Differential Per Game” value of 173%, the Rush, who are 2-6 on the season, could easily flip their goal differential from negative to positive with an improvement on their nearly league-worst first-half performance. The fact that only Panther City has a greater magnitude of underperformance is an indictment on Saskatchewan’s first-half offensive efficiency. Other than a metric measuring Heart and Hustle, no team with playoff aspirations should feel comfortable being closely associated with Panther City on a statistical basis.

While the three teams mentioned above stand out, most of the rest of the league floats in the 30-40% range of “Percent of Goal Differential Per Game”. Falling in that group is largely unremarkable and closely aligned with the league average. 

Similar to the three most visible outliers, Rochester breaks the mold somewhat at around 61%. Their “Goals Under Expectations Per Game” accounts for more than half of their overall goal differential. We’ll dive into what that means a little bit later in this exploration.

The two other main outliers in Figure 5 are New York and Toronto. The Riptide have wildly over-performed their goals forecast, while the Rock have substantially under-performed theirs. Whereas New York is fighting to stay out of last place, Toronto has managed to produce a quality goal differential and win-loss record through nine games. Interestingly, most of the Rock’s under-performance can be traced to Rob Hellyer, who has produced a league-worst average of -8.55 “Goals Under Expectations”. We’ll have more on that a little bit later.

But other than who has over-performed and who has under-performed the LaxMetrics projections on offense, what do these numbers tell us?

First, they offer an insight into which teams in pursuit of the wildcard spot may be in position to improve upon their records in the second half. More importantly, these figures create an opportunity for us to examine a collection of key players, who carry outsized responsibilities in impacting their teams’ second-half performances. Five players stand out as carrying a significant load, and they’re illustrated below in the table in Figure 6.

Figure 6

Beginning with Saskatchewan, no player could potentially swing his team’s fortunes more than Ryan Keenan of Saskatchewan. Sask has struggled through a series of close games, subsequently sliding out of playoff position at the season’s midpoint. The good news for Rush fans, however, is that a reversal of fortunes is entirely possible. If Sask can get Keenan to produce at a rate commensurate with the LaxMetrics forecasts, their circumstances could change quickly.

To date, Keenan has posted the league’s second-worst goal scoring performance relative to the LaxMetrics forecasts. Only Toronto’s Rob Hellyer, who we mentioned earlier and will dig into more in a moment, has faired worse. Despite having taken 65 shots on goal and carrying a usage rate of 12%, Keenan has managed only 6 goals on the season. For context, the LaxMetrics forecast average projects that Keenan should have between 14 and 15 goals at this point. 

For a Saskatchewan team with a goal differential of -6 and four losses by two goals or less, an improvement from Keenan could easily provide the mathematical tipping point toward winning more games. Also, given that the rest of Sask’s offense has been roughly in line with expectations, the weight we’re placing on Keenan’s performance isn’t overblown. If he can cut his “Goals Under Expectations” in half from -7.8 to -3 in the second half, the Rush could be in position to steal two to four wins in the second half that they likely would have failed to secure in the first half. Their pending goaltending change notwithstanding, the Rush are only a couple goals here and there from potentially being a wildcard team.

Another player in a similar situation to Keenan’s in Saskatchewan is Rochester’s Shawn Evans. While Evans has had a phenomenal first half as a passer, he’s put together an underwhelming 8-game start as a scorer. As Figure 6 demonstrates, Evans’s average of -6.13 “Goals Under Expectations” could cover over 55% of the Kighthawks’ negative goal differential. For a team that’s given up a ton of goals at times this season, it’s not reasonable to expect Evans to make up the full difference. But given how far off the forecast he’s been as one of the NLL’s three biggest scoring underachievers, a marked improvement from Evans would bolster a Rochester offense that desperately needs help.

While he can’t do it all himself, Evans—like Keenan—could potentially be the lynchpin in leading a strong second-half push in Rochester. Considering that he’s one of the league’s all-time best players, it’s entirely reasonable to suggest that Evans is likely to improve over the season’s final 10 games. Such a step up in production could help the Knighthawks steal one to three close games, similar to Keenan’s situation in Saskatchewan.

After Evans comes an interesting pair from Colorado—one of the league’s two biggest overachievers and one of its five worst underachievers.

Connor Robinson’s +8.66 “Goals Over Expectations” trails only New York’s Connor Kearnan for best in the NLL. On its face, Robinson’s performance reflects a remarkably efficient first half of the season. He’s been staggeringly efficient. While, of course, nothing definitively says that Robinson can’t maintain his pace in the season’s second half, it wouldn’t be surprising to see him regress to the mean, at least somewhat. If Robinson were to simply break even at the LaxMetrics forecast the rest of the way, he would finish in the top-35 of players over the last six seasons, which includes 555 player entries. Doing so would place him in the 95th percentile. For context, Dhane Smith’s record-setting 72-goal season holds the highest Goals Over Expectations mark at just over 19. Robinson is at 8.66.

In contrast to Robinson is his teammate Zed Williams, who has struggled to one of the league’s five worst “Goals Under Expectations” marks. At more than six “Goals Under Expectations” (-6.095), Williams has effectively been a black hole of bad shots (52) and excessive usage (13.6%). Fortunately for the Mammoth, Robinson’s phenomenal first half paired with a quality first half from Ryan Lee (+3.71) has out-weighed the negative impact of Williams’s struggles.

But here’s the thing, Colorado already has a relatively narrow goal differential (+9) respective to their record (6-2). This means that Williams clearly needs to be better. Should Robinson falter in the second half—which is statistically likely—there isn’t room in the Mammoth’s goal production for Williams to continue to underperform at this clip. Dillon Ward might be challenging for status as the league’s top goaltender, but Colorado can only rely on him so much. The Mammoth are currently in great position to make the playoffs, but should Robinson cool off without Williams or another teammate stepping up, their margin for error will shrink substantially.

Lastly, Toronto’s Rob Hellyer presents a unique, but interesting study in a player that has quite a bit more to offer a team that is already on the ascent. Toronto has performed well through the season’s first half despite Hellyer’s struggles. 

But unlike the top three teams in the NLL—Buffalo, Halifax, and San Diego—the Rock have won four straight games and built a solid goal differential in spite of one of their leading threats producing the least efficient, most underwhelming scoring performance of any forward in the NLL, relative to the LaxMetrics forecast. To some extent, the Rock are winning in spite of Hellyer.

Hellyer has the highest usage rate on his team (16.1%) and is tied with Tom Schreiber for the squad’s most shots on goal (89). But despite the volume of his load, Hellyer has produced only 10 goals in the eight games he’s played. Based on the LaxMetrics forecast average, Hellyer should have somewhere between 18 and 19 goals on the season. He’s produced just over half of that in actuality.

But unlike the players discussed before him, Hellyer’s performance isn’t likely to be the difference between Toronto making and missing the postseason. Whereas Keenan and Evans have the opportunity to help lead turnarounds, Hellyer has the chance to bolster a club already on the climb. Entering Week 13, Toronto is firmly planted in third place in the East. If Hellyer were able to improve his “Goals Under Expectations” mark from -8.55 to an even 0.00 in the season’s second half, Toronto could reasonably see an additional goal per game added to their scoring average. An improvement of that magnitude could easily swing the Rock from a middle-of-the-road offense to one of the league’s top four.

Essentially, an improvement from Hellyer paired with a continuation of Toronto’s stellar defense could reasonably be the difference between the Rock hosting a playoff game or not. And as is always the case in sports, home floor advantage is a tremendously valuable commodity in big games. Surely he’d love to help his team score home floor. An improvement in his goal scoring efficiency would go a long way toward making that happen.

While these are just five of the most interesting cases, there are countless other goal scorers around the league primed for either major improvements or significant regressions in the second half. How that impacts their teams ultimately remains to be seen, but each one is its own story waiting to unfold. For now, using multi-variable forecasting as we did in this exploration can give us one item of note: a clear sense of where to focus our attention in the season’s second half. What unfolds from here is just the story developing the way it’s supposed to.

If you’re interested in reviewing the collection of raw data used to compose this article, you can do so at the link below.

2 thoughts on “Using Forecasting to Explore How Players Impact Goal Differentials”

Leave a Reply