Introduction to using Expected Goals as a way to evaluate goalkeepers

Wes Foderingham, courtesy of Thomas Gadd

written by – Rangers Report

The application of advanced stats in sports is a fluid process & is forever evolving as more & more ideas are applied to analytics.  The way we evaluate players has changed over the years as more information becomes available.  However, the analysis of certain positions has advanced much quicker than others.  For example, it is far easier to use analytics to evaluate attacking players then defending players.

When it comes to goalkeepers, the application of basic stats commonly used in ice hockey are limited in how much they can really tell us about how effective a goalie is.  Goals against averages & save percentages have been proven to be ineffective in evaluating keepers.  As Bill Renowho uses advanced stats to research goalkeepers, pointed out in a post for American Soccer Analysis – goals against average is really a team stat & does little to help analyze the performance of the individual player.  Save percentage does begin to individualize a goalie’s performance, but it is still limited.  Reno wrote, “Save percentage only looks at shots on goal. It is not impacted by the difficulty of shot, but by the goalkeeper’s ability.”

What Reno & others at American Soccer Analysis have spearheaded is factoring in Expected Goals, which is explained here, to enhance the evaluation of individual goalkeepers.  The metric that they have developed is called Goals Minus Expected Goals.

Reno’s explanation:

The GmxG looks at where shots are taken, calculating the likelihood of a shot going in from that distance and angle to goal, which ends up telling us if a goalkeeper is giving up too many goals given the circumstances.

This is great for a couple of reasons, the main one being that we have a more accurate reading on shots on goal than the ol’ shot percentage. If one goalkeeper consistently sees shots from distance while another is routinely left out to dry by his defense, the GmxG will let us know how many goals each goalkeeper should be conceding even if their SOG and goals are similar.

There are limitations to this statistic, which Reno outlines here, mainly that it only measures a keeper’s ability to stop shots & does not measure his ability to handle crosses or distribute the ball.  Also it is very important to note that GmxG is designed for long-term analysis & that small sample sizes are too easily influenced by fluky plays, deflections, or impossible saves.  But over the long haul, you should see the better goalkeepers rise to the top & the below average goalkeepers struggle.

Cammy Bell, courtesy of SNS

The statistic is very simple in its application.  You take the goals allowed & subtract the Expected Goals Allowed.  In this case  you would prefer the results to be going in a negative direction, meaning the goalie is allowing fewer goals then the projections say he should.  Conversely, if the resulting data is in the positive, the keeper is allowing more goals then the data says he should.

It is important to note that unlike the Expected Goals numbers used to evaluate teams & attacking players, you are only looking at Shots on Target when applying it to assessing a goalkeeper.  You can’t examine a goalie’s performance on shots that go sailing above the bar or are blocked by a defender.

Let’s apply the GmxG metric to Rangers’ match against Peterhead.  

Looking at only Shots on Target, Wes Foderingam’s Expected Goals Against was only 0.09, while Peterhead’s Graeme Smith had an xGA of 2.60.  As you know, the final score was 3-0.  So Foderingam’s GmxG was -0.09 & Smith’s was 0.40.

Again, this metric is not designed to evaluate goalies on a short-term, isolated basis & it should be applied over a long stretch of games.  The purpose of the example above was simply to point out how the statistic is calculated.

Reno concludes his introduction of this advanced stat by stating, “Goals Minus Expected Goals is an intense look on one dimension of goalkeeping. Although it is obviously an important one, and one we can learn a lot from, it does have it’s limitations.”

“Is it better than the GAA? Absolutely. Is it better than the Save Percentage? Yes, significantly. Should it be the only thing taken into account in gauging goalkeepers this season? Definitely not. At least for now. The more the stat is smoothed out and missing pieces are marked down, the more confidence we can place in it.”

As part of using advanced stats to enhance our coverage of  Rangers & the rest of the Scottish Championship, a separate page will be created that displays the table of goalkeepers in the league & how they are ranked by the Goals Minus Expected Goals data.  Look for it once the season begins.

You can follow Rangers Report on Twitter @TheGersReport

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s