The Impact of Shots, Shots Against and Total Shots Ratio in College Soccer
Without question the publication of Michael Lewis’s seminal book Moneyball, which examined the data analytic approach of the Oakland A’s baseball team, advanced interest in the use of statistical analysis in professional sports
While slower to recognize the value of analytics, many professional soccer clubs now embrace the “big data” movement (Anderson and Sally 2014). Yet so far, below the pro level, where the vast majority of players and coaches perform, analytics is largely absent. We hope to demonstrate the value of analytics by utilizing typical box score metrics available at all levels of college soccer.
Because there are many more shots than goals, and good fortune is often the antecedent of goals (Anderson and Sally 2014), shots are a more stable and reliable measure to evaluate team performance (Grayson 2012a). To examine shots in the college game, we collected 2014 data from the NCAA web site that provides season box score team totals for the three men’s soccer divisions (NCAA 2014). Data was gathered for 205 teams in Division 1, 204 in Division 2* and 406 in Division 3.
*Division 2 had three teams listed - Pureto Rico Mayaguez, Pureto Rico Rio Piedras, and Sioux Falls that did not provide data.
Table 1. Mean Team Shots and Standard Deviation for NCAA Men’s Soccer 2014 Season
Division Shots per game Std. dev
1 12.4 2.1
2 13.4 2.6
3 13.6 3.3
Table 1 shows that average team shots per game were remarkably similar across divisions. Division 1 teams took 12.4 shots, Division 2, 13.4 and Division 3, 13.6 shots per game. The standard deviation, however, which is a measure of how widely teams are spread across the distribution of shots per match, is notably higher in Division 3. For example, St. Scholastica outpaced other Division 3 teams at 26 shots per match while Iowa Wesleyan was last at a curious .83. Simon Fraser topped Division 2 at 23 shots per game and Dominican of California recorded the fewest shots at 7.1. Wofford enjoyed the highest average in Division 1 at 18.2 while Howard produced a low of 6.4 shots per game.
Shots and Shots Against
Manufacturing a high number of shots is perhaps desirable but shots should be considered in relation to the opponents’ offensive efforts. Table 2 shows the relationship between shots per game and shots against. We use correlations between shots and shots against to establish the direction and strength of association. Correlations range from -1 to +1. The closer the correlation to +1 or -1, the stronger the association between shots and shots against. If the correlation is close to 0, there is little to no relationship between shots and shots against. A negative sign suggests that greater numbers of shots per game are associated with fewer shots against. That is, the more a team shoots, the less shots for the opposition.The -.44 correlation for Division 1 suggest a moderately negative relationship. However, in Division 2 and 3, the correlations are notably stronger yet still negative.
Table 2. Correlations between Team Shots and Team Shots Against- 2014 Season
Division Correlations between Shots and Shots Against
1 - . 44
2 - . 63
3 - . 62
By comparing shots to shots against, we draw two conclusions. First, across three Divisions, high shot teams concede on average significantly fewer shots. Second, the stronger correlations between shots and shots against in Division 2 and 3 suggest a greater number of extreme cases. That is, significant disparities between average shots and shots against are more prevalent in Division 2 and 3 – thus greater shot imbalance among teams. The substantially weaker correlation exhibited in Division 1 implies relatively greater shot balance, where shots and shots against are more equally distributed across teams.
The Consequences of Shots and Shots Against
Shots naturally represent offensive chances; they bring to mind aggressive movement forward and frequent access to the final third. Shots against, by contrast, denote defense and how well or poorly it performs. An obvious question concerns the impact of shots and shots against.
Do these measures influence the bottom line – wins?
Our approach to this question is first to examine the teams that produced the most shots and those that conceded the fewest and compare their win percentages.
The comparisons in Table 3 show that Creighton allowed the fewest shot attempts while producing the highest win percentage in Division 1 at 79.5%. Wofford was best at generating shots but ranked 47th in win percentage at 63.9%. Similarly, in Division 2, Southern New Hampshire held opponents to fewest shots ranking 4th in win percentage at 90%. Simon Fraser led in shots per game yet ranked 42nd in win percentage at 66.7%. Division 3 does not follow the pattern. Methodist held teams to fewest shots but ranked 24th in win percentage while Saint Scholastica was first in shots but 4th in win percentage.
In general, Table 3 suggests that limiting the offensive chances of competitors produces more wins than generating the most shots.
Table 3. Most Shots, Fewest Conceded and Win Percentage – NCAA Men’s Soccer 2014
Division Most Shots Win % Fewest Shots Against Win %
1 Wofford 63.9 Creighton 79.5
2 Simon Fraser 66.7 Southern NH 90.0
3 St. Scholastica 87.0 Methodist 79.4
We next used a more sophisticated statistical method called regression analysis to assess the effects of shots and shots against on win percentage. This technique allows us to predict the win percentage of teams based on offensive and defensive shot measures, and determine which measure is the strongest predictor. Regression analysis yields estimates that translate one additional shot per game, and one fewer shot against, into an increment of win percentage. In this way we are able to determine how much each measure impacts win percentage while controlling for the impact of the other measure.
Results show that in Division 1, manufacturing one more shot per game was worth an additional 2.8 percent for a team’s season winning percentage. Conceding one fewer shot per game increases winning percent by 3.3. The difference, .5, represents the larger impact of the defensive measure – conceding fewer shots. For example, if a team was able to reduce the number of shots against per game by two, win percentage for season would increase 6.6 percent – for a 20 game season this is equivalent to one more win.
Additional analyses showed a nearly identical effect in Division 2. One more shot per game in Division 2 increases winning percent for the season by 2.8 and one less shot conceded raises win percentage by 3.4 percent.
For Division 3, shots and shots against contributed about equally to win percentage. One more shot per game increases season win percent by 2.1, and one less against enhances winning by 2.4 percent.
Our analyses, then, points to the importance of tactical balance between offense and defense, but slightly favors defense: It is better to concede one less shot than to produce one more.
Total Shots Ratio
It appears that successful teams shoot a lot and hold their opponents to fewer shots as well. Analysts applied this logic in developing a measure called Total Shots Ratio (TSR), which is a difference ratio of shots and shots against (Goodman 2013; Grayson 2012b). The measure is expressed as the ratio of shots a team takes versus the number of total shots.
The equation, Total Shots For/ (Total Shots For + Total Shots Against) is a simple way of quantifying how frequently a team shoots compared to their opponents. A ratio of .5 means the teams are matching shots, whereas a value over or under .5 indicates one team outshooting the other.
Total Shot Ratio and Win Percentage
Figure 1 (below), which depicts the relationship between TSR and winning percentage in D3, shows that larger TSR values are in fact associated with higher winning percentages. The correlation is very strong and positive at .83, nearly the same as in D2 .80 and a bit weaker in D1 .72. The upper right quadrant of Figure 1 includes winning programs that also maintain high TSR values. It is in this quadrant that the vast majority of NCAA D3 tournament teams are located – NCAA tourney teams colored in red. The percentage of teams that qualified for the tournament possessing TSR values of .5 or higher, and winning 50 percent plus, are in fact striking across all Divisions – D3 92%, D2 95% and D1 75%. The teams in the upper left quadrant of Figure 1 are likely successful counter attack teams, losing the shot contest but nevertheless winning the game.
We draw several conclusions from the analyses.
First, team shots per game were remarkably similar in the three college divisions. In terms of shot production, a college soccer fan would experience comparable games across divisions. However, the spread between the best and worst shot producing teams was largest in Division 3.
Second, the data showed that shots and shots conceded were significant predictors of win percentage. Shots conceded were a slightly better predictor, especially in Division 1 and 2. Conceding 1 less shot per game increased win percentage to a greater extent than producing one more shot. We now have estimates of the expected increase in win percentage given a specific coaching adjustment. If a D1 coach, for example, adjusts defensive tactics for the coming season, and his/her team concedes three fewer shots per game as a result, while maintaining typical shot production totals, win percentage would be expected to increase by nearly 10 percent – equivalent to 2 games for a 20 game season. Conceding three less shots a game is no easy task, but the payoff is considerable.
Third, TSR, which is the share of total shots one team produces,is a powerful predictor of season win percentage. TSR outperformed the singular measures that comprise it (shots and shots against), and among other box score metrics that precede goals – corners, saves, fouls - it is the most potent predictor of win percentage.The idea is straightforward: Good teams shoot a lot, and they keep their opponents from shooting. Given these results, it may be tempting to increase TSR by encouraging more shooting, or perhaps recruiting players that take every shot opportunity. Shot volume matters, but TSR speaks to more than quantity. Rather, we believe TSR is a reflection of a team’s control of play. Shooting more than opponents generally implies control, and meaningful attacking possession. TSR may then be an indicator of possession.
We cannot test this assertion directly. Unfortunately NCAA box scores do not include a possession statistic. However, we can examine whether possession is related to TSR at the pro level. For the 2014 MLS season we collected data from WhoScored.com and calculated the correlation between team possession percentage and TSR. A significant positive correlation of .62 was discovered. MLS teams that were excellent possession sides were also able to outshoot their opponents – exemplars were Sporting KC and LA Galaxy. In the Spanish top flight La Liga, the correlation was even higher at .81. Barcelona and Real Madrid maintained both possession and TSR levels higher than 60 percent. Finally, in EPL’s 2013-14 season, the association between possession and TSR was .82. Manchester City ranked first in TSR and possession. This evidence strongly supports our contention that TSR evokes control of play and indeed possession. Though the correlation between TSR and possession is not a perfect one, especially in MLS, in the absence of a possession statistic college coaches may turn to TSR as a useful proxy.
Finally, private firms such as Opta and Prozone are using technological advances to collect and maintain large inventories of advanced player and team data for professional clubs. These data likely strengthen the role of statistical analyses for on and off the field decisions.
By comparison, NCAA box scores are rudimentary and do not include such valuable metrics as shot location, possession, clearances, tackles, interceptions, to name a few. While we analyzed team box score totals for the entire season, individual match data offers greater variation of key metrics. However, collecting and coding well over 15,000 box scores represents a significant challenge. It is precisely these types of barriers, - the resources required and expertise needed – that deter some programs from pursuing analytics (Hanlon 2015). Here we offered several metrics that can be quickly calculated from box scores and used to evaluate divisions, teams, and tactics.
About the authors:
Louis R. Joslyn played soccer at Simpson College where he was a Capital One CoSIDA Division 3 Men’s Soccer Academic All-American in 2014 and 2015. He is currentlypursing a PhD in Bioinformatics at the University of Michigan.
Nicholas J. Joslyn is currently a Junior Captain on Simpson College men’s soccer team and is pursing a Bachelor’s degree in Physics and Mathematics.
Mark R. Joslynis a Professor in the Political Science Department of the University of Kansas and was the Baldwin Kansas High School Boys Soccer Coach from 2010 through 2014.
Anderson, Chris and David Sally. 2014. The Numbers Game: Why Everything You Know about Soccer is Wrong. Penguin Books.
Goodman, Mike. 2013. “What is Total Shots Ratio? And How Can It Improve Your Understanding of Soccer?” Grantland.com Accessed here: http://grantland.com/the-triangle/what-is-total-shots-ratio-and-how-can-it-improve-your-understanding-of-soccer/
Grayson James. 2012a. “How Quickly Do Advanced Metrics Regress to their Final Values.” James Blog. Accessed at: https://jameswgrayson.wordpress.com/2012/11/27/how-quickly-do-the-advanced-stats-regress-to-their-final-values/
Grayson, James. 2012b. “Another post about TSR” James Blog. Accessed at: https://jameswgrayson.wordpress.com/2012/07/15/another-post-about-tsr/
Hanlon, Michael. 2014. “Current Practices and Perceptions of Notational Analysis among United States Soccer Coaches.” Soccer Journal. Vol. 60, No. 1 p. 56-64.
Lewis, Michael. 2003. MoneyBall: The Art of Winning an Unfair Game.” W.W. Norton & Co. New York.
NCAA 2014. http://www.ncaa.com/sports/soccer-men/d1