The rise of Expected Goals (xG) in Hockey and other innovative metrics motivated me to try and model shot quality in a similar but slightly different lens. Expected Goal models assign a probability to each unblocked shot attempt of being a goal, and then summing every attempt’s expected goal probability is a player’s or team’s total xG per game or per season. So at the end of a game, if a team had 3.4 xG but scored 4, they over performed or their opponent let in a goal that probably shouldn’t have gone in. Similarly, if that only scored 2 or 3, their opponent did a good job of stopping a high quality chances. The Shot Quality Index (or SQI) attempts to solve one potential missing piece to xG:
- Rather than goal or not goal, weight each shot based on the probability of being one of a goal, shot saved, or shot missed and therefore assign high quality, mid quality and low quality shots.
What is the SQI?
The Shot Quality Index (SQI) is a shot rating system that assigns a rating on a 1-100 scale to every shot attempt taken. Based on unblocked shots taken in the last 5 seasons, excluding shootout and empty net shot attempts, the SQI model was trained to predict if a shot is one of a goal, save, miss. The probability of each event happening to a shot is then weighted, and based on the sum of the probability multiplied by its respective weight, the shot is given a rating score.
Based on the final SQI, a shot is assigned a Quality, High, Mid, or Low.
- High: SQI Greater than or Equal to 50
- Mid: SQI between 35 and 50
- Low: SQI Less than 35
The model uses many variables to predict the outcome and probabilities, including position on the ice, situation (even, powerplay, etc.), type of shot (slap, wrist, snap, etc.) and others. To be clear, the SQI scores cannot be added to see how many goals a team or player should have scored, like xG. However, the SQI model also provides an xG model using the traditional method by summing the goal probabilities.
How Does this Help?
First of all, we can see if players are able to get more shots on net or even score more goals than the average based on their probabilities of either getting saved or missed. That insight I am particularly excited to utilize for defensemen who are skilled at generating chances from the point. Additionally, we can see how many goals are let in that have high probabilities of being saved, and how many saves a goalie based on expectations of goal vs save, similar to an xG model. Lastly, we are able to reward shots that aren’t goals for being on net vs misses completely.
Similar to xG, we can also understand how teams played against one another based on the quality assignment of shots, as well as see which goalies are stopping high quality chances.
I will write on the data used, variables used and how we finalized some of the technical components, like quality boundaries. For now, I am very excited to use this model for future player and game analyses and am excited to see how we can use it to evolve our other metrics going forward.