If you read our primer on the expected play values for the basic list of lacrosse plays, you, like we did, might have wondered how those calculated values differ between teams. Obviously, a 30 second penalty isn’t going to result in as many expected goals against a team with a great penalty kill. At least that is the working assumption.

But we don’t traffic in assumptions here at Lacrosse Reference. Once we posted that initial article, it was time to dig into the team by team splits. Not surprisingly, there was a good bit of variation within the D1 mens teams.

Before we dive into the numbers, I’ll summarize what we wanted to investigate:

• Which teams were best/worse relative to a specific play type? For example, are there teams that are best at turning unforced turnovers into goals? Conversely, which teams can weather an unforced turnover best?
• Are there relationships between metrics? Put another way, do teams sacrifice an ability to prevent goals off a saved shot when they are especially good at turning missed shots into goals?
• Lastly, do these trends hold up year to year or within halves of a season or does a team’s performance in a single metric fluctuate?

Note: the data used here was all games from 2015 and 2016. We limited the regressions to teams and play types to those where the teams experienced the plays enough in a given time period to have reasonable sample size. For the various analyses, this meant that they would need to have 15 to 30 instances of a given play in the time period of interest.

Editor’s Note: To answer the questions posed in this post, we resorted to z-scores or standard scores, the basic measure of how far a given team falls outside of what we might think of as average.  Z-scores allow us to define a great team (in a given metric) as one that far above the pack. For those familiar with z-scores, feel free to skip this section.

The z-score is defined as the number of standard deviations a particular observation falls above the mean.  A standard deviation is a measure of how distributed the data is in a population.  In a normally distributed population, you would generally expect 68% of the data points to be within one standard deviation from the mean.  Put another way, if you expand in both directions out from the mean, when you’ve moved far enough to encompass 68% of the data points, then you’ve moved one standard deviation.

If you’d like a more in-depth explanation, Khan Academy does a great job.

Question 1: Which teams were best/worse relative to a specific play type?

 So who stands out, when we look at the z-score of every team relative to each of the play types?

For 2015:

For 2016:

So Albany, am I right? If you think back to the year that the Great Danes had in 2015, some of these numbers really come into focus. They ranked among the top 10 teams in individual metric performance for unassisted goals, faceoff wins, saved shots, good clears, and ground balls. In other words, when they initiated any of those plays, they scored in the following sequence far more than their opponents. Certainly, that is largely driven by the fact that that offense (a.k.a. the Thompsons) was fairly unstoppable, but I thought it was interesting that they fared so well in both positive plays (faceoff wins, good clears, unassisted goals, and ground balls) and negative plays (saved shots).

Also, if you look at the goals for and against, you can see that their high net values were not necessarily because they held opponents without scoring (indeed, when they turned the ball over, their opponents scored 45% of the time). It was because of how often they themselves scored. Contrast that with Yale, who both showed up on the list because they allowed a goal(within 60 seconds) only 2.7% of the time after a successful clear. On one level, it’s surprising that 2.7% is a leading value on that metric, but again, this is based on net goals, not goals allowed.

Just one minute on the Faceoff win stat for Albany ’15. You’ll notice that no team registered a dominant z-score value in Faceoffs in 2016. Given the specialist FOGOs in the game today, it’s pretty amazing that Albany’s performance on won faceoffs were so much more dominant than any other teams. When they won a faceoff, they scored a goal within 60 seconds 55% of the time. Just a testament to the efficiency of that offense, whenever they got their hands on the ball.

2016 was much more distributed, with Duke showing up several times, but no team coming close to the statistical domination of Albany in those select metrics. One tidbit that did stand out is how several teams scored highly in the two “goal” categories, where only Albany registered in 2015. Now is not the time to dive into why that may have been the case, but it’s interesting. (Something about the units on these teams that allowed them to take advantage of a goal situation to produce another? Maybe they were especially good at detecting some sort of weakness in the defense?)

Question 2: Are there strong statistical relationships between any individual metrics?

To answer this question, we used a simple Pearson correlation to look at the z-scores of all qualifying teams for every unique pair of metrics. The relationships we were looking for could be positive (Team A is good in both metrics 1 & 2, where Team B is bad in both metrics 1 & 2) or negative (Team A being good in metric 1 means they are not good in metric 2). At the core of this question is whether or not excellence in one category means having to accept a negative trade-off elsewhere.

Quick answer: no

Instead, what we find is that for the preponderance of metrics, there is either a strong positive relationship or there is no statistically strong relationship either way. Side note: given that talent is not evenly distributed within college sports, this makes sense; I wonder if there would be a different set of relationships for pro lacrosse?

Below are the combinations of metrics that had a correlation coefficient above .5:

For 2015:

For 2016:

Safe to say that there are strong relationships. To recap, a strong correlation here means that teams either have high z-scores for both metrics, middle z-scores for both metrics, or low z-scores for both metrics. It doesn’t say anything about performance, just about the relationship between the performance of the population on both metrics. And of course, there are many more combinations where there is no observed relationship at all.

So why do we see the ones we see? You could dig into this with a ton of statistics, but a few surface observations first.

First of all, the strongest relationship, across both years is between “Good Clear” and “Ground Ball” and given it’s significantly higher than the other pairs, it seems to be an almost especially noteworthy relationship. And on the surface, we might as why? But remember, this is not to say that the teams are good at getting ground balls or at clears; it’s about how often they score after these events. And since both of these are in effect two halves of a change of possession, it makes sense. It’s not impossible, but most ground balls are going to present a clear opportunity.

As for some more unrelated events, “Ground Ball” and “Saved Shot” are strongly correlated in both years. And this is a bit more counter-intuitive. Teams that score a lot after picking up a ground ball can be said to be strong in transition offense. Teams that have a high net goals average after giving up possession on a shot saved by the goalie do so because they have a solid transition defense. If there were a trade-off to be made (at least in terms of z-score), you would not expect these two to be correlated at all (or if they were, a negative relationship). This is evidence that at least when it comes to z-scores, the good teams are going to have high ones, across metrics, and the bad ones are going to have low ones across metrics. In reality, this probably means that these analyses should either A) be done within groups of more comparable teams talent-wise or B) be done using the raw net values, rather than the z-scores. Another post perhaps.

Question 3: Do these trends hold up year to year?

I won’t spend a ton of time on this one, some metrics are correlated, and some are not. There could be many reasons for this, namely, the fact that teams lose 25% of their roster every year, so the player that made you great at goal differential after a saved shot last year graduates, and suddenly, you aren’t as good at that anymore.

Here is the data:

Make of this what you will. You could argue that the impact of FOGOs shows up here in the goal year to year correlations. Goals are only scored when you have possession, so strong or weak performance after a goal is dependent on your face-off guy. It could be that teams’ performance after a goal is not correlated from year to year because that faceoff percentage is so dependent on one person.

On the flip side, it’s interesting that so many of the unsettled situations score highly. Missed shots, ground balls, good clears, faceoff wins are all followed by a somewhat chaotic situation: transition. And since transition situations require something of every player on the field, it makes sense that the impact of one player is diminished and something akin to coaching/talent takes over. This analysis probably suffers from the similar problem as that which we identified in the previous analysis: good teams are going to score highly relative to bad teams across all metrics when you are using the entirety of D1 Men’s teams as your population.

Up next, we’ll probably want to look at these analyses within a single conference (2015 ACC anyone?).