Match States and what happened at PSV? (part deux)

(Follow @RowZReport )

Previously, I had looked at the Eredivisie 16-17 season and seen that low shot conversion rates (CR%) hurt PSV. Since the StrataData contains very detailed information for each shot taken, I now continue with other team stats. Let’s go!

Match States Match states have been used by many excellent analysts in expected goals context, however its impact was debatable.

Shortly, we may define match state as a team’s score-based status at a given point during the match. Intuition suggests that teams that are behind will find it difficult to create opportunities (and score) since the opponent would have incentive to defend stronger, in order to keep their advantage. Conversely, leading team could have better, easier chances since the opposition would be attacking full-force and leaving defense more open.

I calculated the match states at the time of each shot taken in three leagues. Below chart shows the shot conversion rates with this regard, among three leagues: Italy, Netherlands and Turkey in 16-17. (1)

First, let me remind the casual reader that conversion rate is the average goal scoring ratio over shots. (Conversion Rate (CR%) = Goals/Shots).

Relation between matchstate and conversion rates does exist to a certain extent, yet it’s not too strong nor always linear. From a modelling perspective, this would essentially mean that matchstate would not turn out to be a significant factor in determining a shot’s outcome.

Leading teams, who presumably face lighter defense while they attack, do have significantly higher conversion rates. Compared to a ‘tie’ state, there is 5 percentage points more likelihood: 11.2% vs. 16.4%. This relation holds in all three leagues, as seen in the above graph.

• Being behind on the scoreboard, on the other hand, is a little contradictory. Conversion ratio (CR%) are practically equal to ‘tie’ states 10.8% vs 11.2%. • CR% is actually higher for Dutch and Italian teams when they are trailing. (With some hope, I actually looked at the sample sizes for each these groups, but they were large enough - 500+ chances minimum) More surprisingly, Turkish Super league teams do show a clear linear relationship.

• As suggested by 11tegen11, I broke down chance types and looked only at Open Play situations. Relation is still not linear: CR% in matchstate -2 or -1 are at least as good as 0 (i.e. tie).

PSV a (short) re-visit Focusing on team level for Eredivisie only, we see that PSV actually had the best CR% when they were behind: a striking 10 goals in 49 shots (20.4%). Where (when?) they were hurt was during ‘tie’: 26 goals in 311 shots (8.4%). Feyenoord, the champions, had scored 28 times in 179 tries (15.6%).

To be clear; the nominal shot/goal numbers highly depend on number of minutes a team was in a certain match state. Good teams spend more time in positive match states, which probably has some spillover effects. I hope to crunch some numbers on that in the future. Until that, here is to having the efficiency metric: CR% !

Chance Rating One of the most valuable pieces of information in the famous Strata data is the ‘chance rating’. This is a subjective metric, assessed by Stratabet’s well-trained analysts, which assign a rating to each shot. It provides very valuable information by moving forward from conventional shots on/off target stats, as well as binary features like ‘Big Chance’.

We would expect teams that are ahead in the scoreline (matchstate +1,+2, and so on) to have ‘better’ chances. This might be of reasons stated before: trailing teams’ incentive to go all-out-on attack while risking defense. Below, I calculated the number (and %) of shots, by matchstate and chance rating. Stratabet actually classifies shots into 6 categories, but I have grouped them into a reduced form of three: low-medium-high.

High chances, indeed, happen more often when a team’s ahead.

Controlling for two tail-ends of matchstate, a shot is almost twice more likely to be a ‘High’ chance shot (dark green on the graph). When a team’s trailing by one goal, for example, 10% of shots taken by that team is ‘High’ chance. The same figure for +2 goals (i.e. ahead by two goals or more) is 18%. What a lift!

These kind of analyses, by definition, provide insights about the past. But it is possible to estimate teams’ future performances with historic information. How many ‘high’ chances did a team create last year per game? Against which team? Who was on the field on that day? Could that be a proxy for match next weekend? Questions and answers…

Chance rating, it also turned out, is a major factor in determining a shot’s outcome. Let’s leave that for a next post, where I (finally) dive into my Expected Goals model.

Until next time!

This article was written with the aid of StrataData, which is property of Stratagem Technologies. StrataData powers the StrataBet Sports Trading Platform, in addition to StrataBet Premium Recommendations.