Cardiff School of Sport DISSERTATION ASSESSMENT PROFORMA: Empirical 1

Student name: Ba rnabas Blake Student ID: 20019089

Programme: SPE

Dissertation title: Evaluating the validity of a basketball performance metric in correlation with winning and losing games Supervisor: Darrell Cobner

Comments Section Title and Abstract (5%) Title to include: A concise indication of the research question/problem. Abstract to include: A concise summary of the empirical study undertaken. Introduction and literature review (25%) To include: outline of context (theoretical/conceptual/applied) for the question; analysis of findings of previous related research including gaps in the literature and relevant contributions; logical flow to, and clear presentation of the research problem/ question; an indication of any research expectations, (i.e., hypotheses if applicable). Methods and Research Design (15%) To include: details of the research design and justification for the methods applied; participant details; comprehensive replicable protocol. Results and Analysis (15%) 2 To include: description and justification of data treatment/ data analysis procedures; appropriate presentation of analysed data within text and in tables or figures; description of critical findings. Discussion and Conclusions (30%) 2 To include: collation of information and ideas and evaluation of those ideas relative to the extant literature/concept/theory and research question/problem; adoption of a personal position on the study by linking and combining different elements of the data reported; discussion of the real-life impact of your research findings for coaches and/or practitioners (i.e. practical implications); discussion of the limitations and a critical reflection of the approach/process adopted; and indication of potential improvements and future developments building on the study; and a conclusion which summarises the relationship between the research question and the major findings. Presentation (10%) To include: academic writing style; depth, scope and accuracy of referencing in the text and final reference list; clarity in organisation, formatting and visual presentation

1 This form should be used for both quantitative and qualitative dissertations. The descriptors associated with both quantitative and qualitative dissertations should be referred to by both students and markers. 2 There is scope within qualitative dissertations for the RESULTS and DISCUSSION sections to be presented as a combined section followed by an appropriate CONCLUSION. The mark distribution and criteria across these two sections should be aggregated in those circumstances.

CARDIFF METROPOLITAN UNIVERSITY Prifysgol Fetropolitan Caerdydd

CARDIFF SCHOOL OF SPORT

DEGREE OF BACHELOR OF SCIENCE (HONOURS)

SPORT AND PHYSICAL EDUCATION

2014-5

TITLE: Evaluating the validity of a basketball performance metric in correlation with winning and losing games

DISCIPLINE: Dissertation submitted under the Performance Analysis area

NAME: Barnabas Blake

STUDENT NUMBER: 20019089

EVALUATING THE VALIDITY OF A BASKETBALL PERFORMANCE METRIC IN CORRELATION WITH WINNING AND LOSING GAMES

Cardiff Metropolitan University Prifysgol Fetropolitan Caerdydd

Certificate of student By submitting this document, I certify that the whole of this work is the result of my individual effort, that all quotations from books and journals have been acknowledged, and that the word count given below is a true and accurate record of the words contained (omitting contents pages, acknowledgements, indices, tables, figures, plates, reference list and appendices). I further certify that the work was either deemed to not need ethical approval or was entirely within the ethical approval granted under the code entered below.

Ethical approval code: 14/5/28U Word count: 10,823 Name: Barnabas Blake Date: 19th March 2015

Certificate of Dissertation Supervisor responsible I am satisfied that this work is the result of the student’s own effort and was either deemed to not need ethical approval (as indicated by 'exempt' above) or was entirely within the ethical approval granted under the code entered above. I have received dissertation verification information from this student

Name: Date:

Notes: The University owns the right to reprint all or part of this document.

TABLE OF CONTENTS

ACKNOWLEDGEMENTS i

ABSTRACT ii

CHAPTER ONE: INTRODUCTION 1

1.1. Performance Analysis 2 1.2. Game-related Statistics 3

CHAPTER TWO: REVIEW OF LITERATURE 4

2.1. Performance Analysis in Basketball 5 2.2. Performance Indicators 5 2.3. Advanced Statistics 7 2.4. Game Pace 7 2.5. Performance Metrics 9 2.6. The Defensive Process 12

CHAPTER THREE: METHODS 14

3.1. Sample 15 3.2. Variables 19 3.2.1. Game-related Statistics 19 3.2.2. Advanced Statistics 20 3.2.3. Preventions 21 3.3. Data Collection 21 3.3.1. Hand Notation 21 3.3.2. Reliability 22 3.4. Data Analysis 23

CHAPTER FOUR: RESULTS 24

4.1. Descriptive Statistics 25 4.2. Tests of Significance 28 4.3. Correlations 33

CHAPTER FIVE: DISCUSSION 38

5.1. Discussion of Results 39 5.1.1. All Games 39 5.1.2. Winners and Losers 40 5.1.3. Close and Unbalanced 42 5.1.4. Correlations 42 5.2. Implications for Practice 44 5.2.1. Defending Shots 44 5.2.2. Defending Passes 45 5.2.3. Motivation 46 5.3. Reliability of the Data Source 46 5.4. Game Pace 47

CHAPTER SIX: CONCLUSION 48

6.1. Concluding Thoughts 49 6.2. Future Research 50

REFERENCES 52

APPENDICES 62

LIST OF TABLES

Table 1a. Pool of close games (score differential: x < 5 points) 15

Table 1b. Pool of unbalanced games (score differential: 20 points < 16 x < 40 points)

Table 2. Game categorisation thresholds 17

Table 3. Excluded outlying games (score differential: x > 40 18 points)

Table 4. Game-related statistics recorded by the FIBA Live Stats 19 to produce the Olympic box scores

Table 5. Common advanced statistics generated from box score 20 data

Table 6. Kappa scores and ratings 22

Table 7. Descriptive statistics and percentage differences of box 26 score performance indicators

Table 8. Descriptive statistics of defensive advanced statistics 27 and Preventions metric outputs

Table 9. Statistical significance of the differences between 28 Winners (W) vs Losers (L)

Table 10. Statistical significance of the differences between Close 29 Winners (CW) and Unbalanced Winners (UW)

Table 11. Statistical significance of the differences between Close 30 Losers (CL) and Unbalanced Losers (UL)

Table 12. Statistical significance of the differences between Close 31 Winners (CW) and Close Losers (CL)

Table 13. Statistical significance of the differences between 32 Unbalanced Winners (UW) and Unbalanced Losers (UL)

Table 14. Pearson’s correlation (Winners) 33

Table 15. Pearson’s correlation (Winners) 34

Table 16. Pearson’s correlation (Losers) 35

Table 17. Pearson’s correlation (Close) 36

Table 18. Pearson’s correlation (Unbalanced) 37

LIST OF FIGURES

Figure 1. Descriptive statistics of box score performance 25 indicators

Figure 2. Descriptive statistics of defensive advanced statistics 27 and Preventions metric outputs

ACKNOWLEDGEMENTS

Thanks to the International Basketball Federation (FIBA) and the International Olympic

Committee (IOC) for their archived statistics for the 2012 Olympic Basketball Tournament for Women. Also, many thanks to the Cardiff School of Sport’s Centre for Performance

Analysis for the footage that formed the sample for this study.

Thanks to Alfredo Rodriguez of WammyRadio.com who provided the inspiration for this project with the invention of his Preventions metric. Finally, thanks to Darrell Cobner who provided clear guidance and support throughout the completion of this dissertation project.

i

ABSTRACT

The aim of this dissertation project was to evaluate the validity of the novel Preventions metric, developed by Rodriguez (2013b), in correlation with performance indicators previously defined as discriminants of winning and losing games. The sample consisted of a selection of games from the 2012 Olympic Basketball Tournament for Women (n=18), categorised into two pools, close (n=9; x < 5 points) and unbalanced (n=9; 20 points < x <

40 points), based on final score-line differential. Box score data was scraped from the official FIBA Olympic archives to produce game-related statistics and advanced statistics; while Preventions metric data was recorded using hand notation. Descriptive statistics were generated to distinguish the differences in absolute values between the sample groups; Mann-Whitney U Tests were then performed to identify the significance of these differences. Winners and losers were differentiated by points prevented (p < 0.05), shots challenged (p < 0.05) and botched passes (p < 0.05). Pearson’s correlation data was produced to determine the strength of relationship between the new Preventions metric and academically proven advanced statistics. The variables correlating with the advanced statistics were points prevented (p < 0.01), shots challenged (p < 0.05) and botched passes (p < 0.05). The findings clearly identified points prevented, shots challenged and botched passes as significant variables to be considered as performance indictors due to their discriminant powers and correlation with peer-reviewed advanced statistics. This suggests that coaches should emphasise their defensive strategies regarding closing-out on shooters and disrupting passing lanes through on-ball pressure and off-ball denial defense.

ii

CHAPTER ONE

INTRODUCTION

1

1. INTRODUCTION

1.1. Performance Analysis

Performance analysis is a discipline that is vastly varied in its methods of data collection and analysis, however, the focus is generally the provision of information about sporting performance (Maslovat and Franks, 2008; O’Donoghue, 2015). In order to enhance sports performance, it is critical that accurate and objective feedback is provided due to the inadequacies of the human capacity to recall events (Maslovat and Franks, 2008). The inadequacy of relying on coach recall is a topic that has been highlighted (Franks and Miller, 1986; 1991; Laird and Waters, 2008) and has led to the development of quantifying performance through notation (Hodges and Franks, 2008). Consequently, performance analysis allows the provision of augmented extrinsic feedback, which Franks et al. (1983) included in an early model of the coaching process. Therefore, to fill this need, analytical systems have been developed to best meet the demands of the coaching industry, providing detailed information regarding sports performance (Hodges and Franks, 2008).

Performance analysis research in basketball has been applied to provide in-depth information that can be used to model performance (Malarranha, 2013), to scout opposition (Rose, 2004) and to analyse styles of play (Tavares and Gomes, 2003). Modelling performance has become a focus of performance analysis in basketball, both in an academic context and in practice. While the aim of this study was to explore the process of developing and testing a procedural-based performance metric, it is more commonplace to find research only using the game-related statistics found in a box score. The study of game-related statistics is vast and such research has aimed to discover the discriminants between winning and losing on a purely outcome-based approach (Sampaio and Janeira, 2003; Ibáñez et al., 2009; García et al., 2013), each concluding differing findings.

However, Oliver (2004) suggested that there was ‘Four Factors’ that stood above the rest to statistically model basketball performance (Oliver, 2004, p.63); shooting percentage, rebound rate, turnover rate and free throw rate (Basketball Reference, 2015c). While the Four Factors are based on the absolute box score variables, they are advanced statistics that derived from these figures by taking into account the game pace and opposition effects to best define performance (Oliver, 2004; Kubatko et al., 2007, Swalgin, 2008). The

2

work of Oliver (2004) and Kubatko et al. (2007) has led to game pace being regarded an important issue in basketball analytics and most current advanced statistics take team possessions and opposition statistics into account when providing performance data.

1.2. Game-related Statistics

While game-related statistics and advanced statistics have been found to discriminate between winning and losing, the results have been varied. This suggests that performance outcome cannot be distinctly tied to differences in microcosmic outcome-based variables. Therefore, research into the procedural elements of gameplay has become a trend in basketball analytics, with the aim being to discover any and all variables that may have an influence (Remmert, 2003; Tavares and Gomes, 2003). Similarly, the aim of this study was to develop and test a performance metric system that evaluates the holistic nature of defensive performance in basketball, rather than simply notating the outcomes of possessions, i.e. steals and blocks. While this approach is newer in basketball research, others have aimed to quantify teams’ defensive processes and strategies. Such metrics include Gomez et al.’s (2006) research into the defensive systems applied by teams in the Spanish Play-offs, while Stanković (2013) recently looked at the notion of collective defense; meaning the way in which players move as a unit on defense.

A new method of analysing defensive play that is yet to be discussed academically is the Preventions metric, created by Rodriguez (2013c) to quantify the defensive efforts of the Los Angeles Clippers. While the metric was created for use in his journalistic endeavours, it has potential to become a valid technique of quantifying the process of team defense. Preventions breaks down defensive play to an even greater extent than Gomez et al. (2006), precisely evaluating the actions of each possession in terms of how a team defends shot attempts and the manner in which they create turnovers. These factors have been mentioned separately and sparsely in literature (Csataljay et al. 2013; Fylaktakidou et al., 2011; Hawkins and Choi, 2009), but Preventions combines these variables in a metric that universally quantifies defensive performance. The Preventions metric is not yet peer-reviewed or academically recognised, so this research project aims to test the reliability and validity of this method of notation by assessing its correlation with previously peer-reviewed models of performance.

3

CHAPTER TWO

REVIEW OF LITERATURE

4

2. REVIEW OF LITERATURE

2.1. Performance Analysis in Basketball

Basketball is a game that has historically been very dependent on statistics to provide an analytical element to both the coaching and broadcasting sides of the sport. Therefore, the natural development of the sport has included the notation of game-related statistics to aid statistical analyses for many years (Rose, 2004). The National Basketball Association (NBA) has recorded game statistics since its inception in 1946, at that point named the Basketball Association of America (BAA), but the number of variables notated were greatly restricted (Basketball Reference, 2015a). The box score of the 1940’s contained the variables relating only to shooting and assists, expanding to include overall rebounding in the 1950’s. The 1970’s saw the NBA greatly expanded their statistical notation; adding steals, blocks, offensive/defensive rebounding and turnovers. The present day box score was completed in 1979 when the NBA adopted the three-point line, which therefore meant separately recording two-point and three-point field goals (Basketball Reference, 2015e).

Olympic basketball has seen a similar timeline of gradual expansion of its statistical notation (FIBA, 2009a; 2009b), from the first tournament in 1936 merely recording the game score. When the Olympics returned in 1948, only the very basic game-related variables were recorded; free throws, player fouls and individual points scored. This is notably different to the NBA, as field goals were not recorded until 1964. The 1970’s and 1980’s saw a rise in the global popularity of the game, so the IOC followed the lead of the NBA in recording rebounds, assists, turnovers, steals and blocks. Similarly, the three-point line was adopted for the 1984 Olympic tournament, meaning the separate notation of three-point field goals.

2.2. Performance Indicators

Quantitative analysis of basketball begins with the notation of game-related statistics, which provide a log of the outcome of each ball possession (Sampaio and Janeira, 2003). This data can then be used by coaches and analysts to derive the discriminants between success and failure in performance, allowing them to set individual and team goals. The statistics can also be used in literature and in practice as performance indicators, i.e. a

5

variable or combination of variables that defines an area of performance (Hughes and Bartlett, 2002). Performance indicators must be valid measures of performance that can be objectively quantified, else they cannot be used to produce generalizable findings (O’Donoghue, 2010; O’Donoghue, 2015). Once found and validated, performance indicators can be used to build hierarchical models that prioritise the factors that contribute to effective performance (Hay and Reid, 1988). In an invasion games context such as basketball, this sort of modelling can be used to prioritise the different skill-based components of the game, e.g. shooting, passing, and defense (Hughes and Bartlett, 2004). Therefore, it is important that analysts have the tools to represent these areas of performance with valid indicators that accurately represent the tactical or technical outputs of performance.

Performance indicators in basketball include the game-related box score statistics (Swalgin, 2008) that have been widely used to discriminate between winning and losing in academic literature throughout the 2000’s, building on earlier research from the 1990’s. Akers et al. (1991) studied NCAA basketball and found that two-point field-goal percentage, turnovers and free-throw percentage were variables that distinguished winning teams. This early work was reviewed by Trninic et al. (2002) and Sampaio and Janeira, 2003) whose samples were European professional basketball; Euroleague Final Fours and Portuguese LPB, respectively. Trninic et al. (2002) judged defensive rebounds, field goal percentage and free throw percentage as having the most discriminant power, while Sampaio and Janeira’s (2003) findings included that made free throws, offensive rebounds and committed fouls were the most crucial factors throughout the varying contexts. Research in this manner continued throughout the 2000’s, with Ibanez et al. (2008) defining assists, steals and blocks as the discriminant factors in the Spanish LEB1; Csataljay et al. (2009) concluding that three-point attempts, field goal percentage, free throws made, free throw percentage and defensive rebounding in EuroBasket 2007 while Pojskic et al. (2009) found that field goal percentage, defensive rebounds and bench points were decisive at the 2008 Beijing Olympics. The differing findings from these similar studies suggest that the game-related statistics do not fully represent the intricacies and nuances of basketball performance. While much the analytical community continues to search for factors that are valid and significant among all samples, other researchers have turned to advanced statistics and performance metric to better quantify the actions of a basketball game. This suggests that knowledge of results is not as vital as knowledge of

6

performance, affected less by game context and opposition effects (Hodges and Franks, 2008).

7

2.3. Advanced Statistics

While the analysis of raw match statistics alone has generated valuable descriptors of successful or unsuccessful performance in basketball, the complexity of the sport must also be considered. Dealing with a highly complex invasion game (Sampaio et al., 2013), academic literature has moved into the area of assessing basketball in regard to its complex context. Examples include research into opposition effects (Sampaio et al., 2010; Moreno et al., 2013), effects of previous quarter scores (Sampaio et al., 2010), regular season / play-off games (García et al., 2013) and the effects of back-to-back games (Ibanez et al., 2009). These studies utilise the basic game-related statistics found in a box score (Swalgin, 2008), but apply them in systems of analysis that assess the bigger picture of sports performance (Hughes and Franks, 2008).

Game-related statistics can also be combined to produce advanced statistics that exist as KPI’s to reflect not only one area of performance, e.g. free throw percentage, but to model performance as a whole (Swalgin, 2008), e.g. offensive efficiency rating (Malarranha et al., 2013). Examples of advanced statistics for individuals include player efficiency rating, usage rate and win shares (Basketball Reference, 2015b), while teams can be evaluated using offensive / defensive efficiency ratings, rebound rate and turnover ratio (ESPN, 2015). Malarranha et al. (2013) led the way in modelling performance using advanced statistics, identifying effective field goal percentage, offensive rebound percentage and efficiency ratings as influential, while including quality of opposition as a distinguishing variable in their analysis.

2.4. Game Pace

The ability to rate a team performance and compare game-related statistics with other games relies on the idea of adjusting data to normalise it by determining the pace of the game. Oliver (2004) first introduced the notation that a box score could be used to define the pace a certain game was played at by using the game-related statistics to track the number of possessions. The formulae presented by Oliver (2004) were designed to be used in practice by basketball analysts, but they were not applied in academia until Kubatko et al. (2007) produced a paper that summarised the potential that this line of research had in an academic context. Their study stated that quantifying game pace and possessions is central when starting to analyse the discriminant powers of game statistics.

8

The research paper described the process of testing various formulae that were in use in the field for quantifying team possessions, with the aim of generating a combined formula to best determine the possession statistics for an NBA game. Kubatko et al.’s (2007) valid formula for possessions then allowed more accurate statistics regarding offensive and defensive efficiency ratings to be produced, i.e. the points scored / allowed per 100 possessions. This research has since been applied in academia by Malarranha et al. (2013) who used offensive and defensive ratings as discriminant variables to model successful performance and by Sampaio et al. (2010) who assessed the dominance of the USA men’s basketball team at the 2008 Olympics with respect to the pace they played at. Recent studies can now be critiqued with reference to whether or not they normalise game-related statistics by game pace or if they use the raw match data. Daskalovski et al. (2014) studied the league variation of game-related statistics between the NBA and leagues around Europe, without normalising statistics by number of possessions. Therefore, Daskalovski et al.’s (2014) findings are skewed towards the NBA because of the differences in game pace and the fact that they play forty-eight minute games in contrast to the FIBA rules applied in Europe of forty minute games (Reimer, 2005).

The prevalence of statistical notation at an elite performance level has increased to the point where the minutest of details are recorded and analysed by academics and journalists. The NBA, via Synergy Sports (Synergy Sports Technology, 2013), records the type of possession that occurs within the flow of a game, e.g. pick-and-roll, while recording the players involved and the outcome. This data can be used to assess an individual player’s tendencies, while also evaluating strengths and weaknesses. For example, Synergy Sports data can be used in scouting to provide information such as the opposition’s point guard shoots most often when they drive right, but when they drive left they are more likely to pass. A recent addition to the statistical insights provided by the NBA is SportVU player tracking (NBA Stats, 2014c) using cameras that track the in-game co-ordinates of all players on the court. This data is highly effective to study performance in both spatial aspects, e.g. opposition field goal percentage in the key when a specific defender is also in the key (NBA Stats, 2014a), and temporal aspects, e.g. the average speed a player runs on a fast break (NBA Stats, 2014b).

9

2.5. Performance Metrics

While research studies referring to a traditional basketball box score and the derived advanced statistics are more common (Malarranha et al., 2013; Swalgin, 2008), recently more academic literature is becoming experimental in producing performance metrics that quantify the areas of performance not recorded in game-related statistics. Evangelos and Nikolaos (2005) and Ribas et al. (2011) created metrics to record and assess the variety in rebounding situations, an imperative part of the game to understand as rebounding percentage is a highly discriminating factor in Malarranha et al.’s (2013) dynamic model of performance. In Evangelos and Nikolaos’ (2005) study, they divided the court into five sectors and then assessed the rebound zone of shots from each sector. This information crucially provides coaches the ability to prioritise areas of the court that must be controlled in order to win the rebound battle. Ribas et al. (2011) looked at the numerical advantages in rebound situations, therefore quantifying the aggressiveness of a team in terms of their offensive rebounding presence. Predictably, their results showed that the team that sent more players to rebound came up with a significantly greater amount of rebounds. However, it can be noted that the offensive team need only send one more player to rebound and they have a statistical significance, so sending too many players is inefficient because the benefits do not outweigh the fact that they would not be able to protect against the fast break.

Further studies have assessed other areas of performance such as a team’s tactical process and have aimed to quantify abstract concepts such as teamwork. Tavares and Gomes (2003) evaluated the offensive process of elite junior international teams, discovering that set offenses dominated the style of play (74.6% of the time). Their study also found that the mean duration of a fast break was four to six seconds, while a set offense took thirteen to eighteen seconds. They also were able to break set offenses down into 1v1 play (i.e. isolation / post play), 2v2 play (give-and-go / pick-and-roll) and 3v3 play (pass and screen away / pass-cut-replace); examples stated are provided by Gels (2015). This found that the most common set play structure was based around 1v1 play, which may be due to the sample of junior players who at their stage of development may rely more on individual talent than an understanding of team concepts. While this study helps the reader understand the different offensive structures used in basketball, it makes no reference to which structure is the most effective in generating efficient scoring opportunities – which could be considered more important than the spread of gameplay styles. 10

Remmert (2003) also studied tactical behaviours in basketball, with a process-orientated model that described the details of offensive structures. The model divided gameplay into offensive interaction units (OUI’s), with OUI’s finishing with a change of ball possession (CBP) or a subsequent offensive phase (i.e. after an offensive rebound). Each OUI is described in minute detail, with branches of options occurring based on the offensive and defensive tactical decisions made. This allows for the data to accurately show which offensive tactics worked most efficiently against certain defensive methods. Remmert (2003) made conclusions that were much more insightful than those of Tavares and Gomes (2003), because of the scale of the data collected. This included the ideas that overlapping OUI’s reduced the defenders’ ability to help against 1-on-1 finishing actions, a variety of screening actions (2-, 3- and 4-person structures) increases scoring success ratio and that defenders should work on switching 2-person screening plays, while sliding through 3- or 4-person plays. These findings are extremely applicable to coaching basketball tactics, which fulfils the major role of performance analysis research; the provision of information.

An important element of statistical research in basketball is the efficiency of possessions, which eludes to the pace that teams’ play at and their effectiveness in relation to the pace (Oliver, 2004; Kubatko et al., 2007). The rate at which a team commits turnovers is one of the four key discriminant variables modelled in Malarranha et al.’s (2013) study, which means therefore that the reasons for and effects of turnovers should be quantified in research. Fylaktakidou et al. (2011) did just this and found that 19.1% of turnovers led to a dead-ball (e.g. out of bounds, etc.) which are the least dangerous of turnovers because they cannot lead to a fast break scoring opportunity. Passing errors (40.2%), ball-handling errors (23.9%) and travelling (23.6%) are the most common types of turnover committed; the majority of which happen during set play rather than fast breaks. Passing turnovers were found to be a significant factor in whether teams could break down a zone defense, with winners only turning the ball over 16.5% of the time in this scenario, while losers turned it over 22.2% of the time.

The final offensive performance metric to be discussed is Bazanov et al.’s (2005) research into offensive teamwork intensity, a process that analyses that the teamwork aspect of technical and tactical structures. Bazanov et al.’s (2005) intensity index comprises of dribbles, passes, on-ball screens, off-ball screens, shots and duration of offensive possession. The study found that more successful possessions were significantly shorter in duration and included a higher intensity index of offensive elements. Their process 11

allows coaches to model offensive activity, in terms of how active and intensity their players are on offense. Concluding, Bazanov et al. (2005) noted that coaches could regulate offensive teamwork by setting time parameters in training sessions to be able to implement all offensive elements in a short space of time. Bazanov et al. (2006) followed up this research with a further study that assessed in detail the factors that affected the teamwork intensity index. This included the separation of different offensive structures, e.g. fast breaks to discuss the data most applicable to that offensive style. Bazanov et al. (2006) found that fast breaks were most successful when they included one dribble and/or one pass with a duration of less than five seconds. The intensity index also described effective set offenses as including three or four off-ball screens, in an average of ten seconds.

The key to the performance metrics that have been created by Ribas et al. (2011), Remmert (2003) and Bazanov et al. (2005; 2006) are the process-goal nature of their analysis. While the traditional game-related statistics recorded in a box score are solely focused on the outcome of match events (Swalgin, 2008), it is increasingly important to balance that information with metric data from process-orientated measurements. The coaching implications of process-goal metric data are far more applicable than outcome- based data because the process is the controllable element of performance, while the outcome is based on the interaction of external constraints and extrinsic influences (Newell, 1986). While knowledge of results (KR) is an important tool for motivating players to outperform their peers, it is an obvious source of feedback that may not be completely beneficial (Maslovat and Franks, 2008). Instructional augmented feedback that stems from the knowledge of performance (KP), i.e. from process-oriented metric data, can be a valuable tool in supporting both intrinsic motivation to improve one’s performance and skill acquisition to identify and correct performance defects (Hodges and Franks, 2008). It is also important to consider the role that process-oriented performance metrics can play in providing quantifiable means for both individual and team goal setting procedures, which in turn also influences athlete motivation.

While motivation has been reviewed in a broader sense in relation to sports performance analysis, business studies literature has studied the specific concept of using process- based metric data in goal setting and motivation strategies. Doerr and Gue (2012) based their research upon order fulfilment within a distribution centre, with performance metrics assessing the efficiency of the production line. A key finding of their study was that goal acceptance was crucial in determining motivation levels among the workforce. This 12

conclusion can be applied to sports coaching in the sense that a player must understand and accept their role on the team in order to be motivated to complete the tasks specific to their position and/or status. If goal acceptance in a team is high, then the process and structure around which the team is organised is likely to run smoother, just as a production line would.

2.6. The Defensive Process

This dissertation project will be focussed on a defensive metric known as Preventions. Past literature focussing on defensive metrics includes Gomez et al.’s (2006) study that classified the defensive systems used and the effect of different systems on offensive gameplay. Gomez et al. (2006) found that winning teams were able to generate better offensive opportunities against a variety of defensive systems, while the defensive system that was able to disrupt the offensive system the most was zone pressure by not allowing the offensive team to complete lots of passes and by limiting their time on offense. Stanković (2013) used a performance metric to quantify the effects of different collective defense strategies, while also assessing the contributions of different positions to their team’s defensive efficiency. The study, unsurprisingly, found that the winning team had a greater defensive efficiency in all cases but one game where the team with the better defense lost because they lacked offensive threats. The research also led to the conclusion that perimeter players (i.e. guards) had more of an impact on their team’s overall defensive effectiveness in comparison to interior players (forwards and centres) who were less mobile and therefore not as important in terms of their defensive contribution.

Álvarez et al. (2009) and Ferreira et al. (2014) developed detailed notation systems to identify performance indicators in basketball that are specific to team defensive actions, such as switches, help defense and ball pressure. Álvarez et al.’s (2009) study found that the most effective strategy was half-court zone defense and that help defense is much more common that completely switching defensive assignments. They also found that only 38.9% of shot attempts were taken with a high level of defensive pressure. On the other hand, Ferreira et al. (2014) focussed more on the components that built up team strategy, concluding that moderate ball pressure and restricting quick transition phases, i.e. fast breaks, limited the opposition’s scoring rate. Taking a different approach to the defensive process, Csataljay et al. (2013) focussed a study into the effects of defensive pressure on 13

shooting performance. Their research assessed the defensive pressure that players were under when attempting shots at the basket, with action variables including distance from the basket and level of pressure. They found that winning teams were able to create more open shooting opportunities, leading to a higher shooting percentage; while they were also shot a better percentage under maximal levels of pressure in comparison to the losing teams. This data collected in Álvarez et al. (2009) and Csataljay et al.’s (2013) research projects bear some resemblance to the Preventions metric that will form the basis of the research put forward in this dissertation project.

The Preventions metric was developed by journalist Alfredo Rodriguez (2013c), to be used in his media ventures in basketball analytics; in part to quantify the defensive efforts of the Los Angeles Clippers throughout the 2013-14 season (Rodriguez, 2013b). The Preventions metric has therefore not yet been proven as valid through academic research, thus providing the rationale for this project. Preventions is a performance metric of defensive effectiveness that combines the possession efficiency element that has been researched by Fylaktakidou et al. (2011) with the shooting pressure component recently studied by Csataljay et al. (2013). While the metric is similar in concept to the systems developed in these articles, the manner in which Preventions data is organised is slightly different, leading to an overall metric of defensive performance.

The metric is split into three main groups, with two additional groups to include minor variables. The main groups are Shots, Possessions and Passes; while the minor groups are Turnover Prevention and Miscellaneous variables. The data collection of these variable groups will be further discussed in the methods, including a breakdown of the individual action variables involved. The metric statistics produced by combining these variables are Points Prevented, Possessions Lost, Shots Contested, Botched Passes, Possessions Protected and Team Defense (Rodriguez, 2013a). This dissertation project aims to prove that the Preventions metric is a discriminant of winning performance, while also comparing its validity with the advanced statistics that are already considered the norm in academic basketball analytics; such as defensive rating and defensive efficiency (Malarranha et al., 2013).

14

CHAPTER THREE

METHODS

15

3. METHODS

3.1. Sample

A selection of 18 matches from the 2012 Olympic Women’s Basketball Tournament were evenly split, analysed in two separate pools: close (n=9) and unbalanced (n=9). The games and score-lines are listed in Table 1.

Table 1a. Pool of close games (score differential: x < 5 points)

(Adapted from FIBA, 2009)

Code Team A Score Team B Score Diff.

C1 Russia 58 Canada 53 5

C2 Turkey 61 Czech Republic 57 4

C3 France 74 Australia 70 4

C4 France 64 Canada 60 4

C5 Australia 70 Russia 66 4

C6 France 80 GB 77 3

C7 Turkey 70 Croatia 65 5

C8 Russia 66 Turkey 63 3

C9 France 71 Czech Republic 68 3

16

Table 1b. Pool of unbalanced games (score differential: 20 points < x < 40 points)

(Adapted from FIBA, 2009)

Code Team A Score Team B Score Diff.

U1 Turkey 72 Angola 50 22

U2 USA 81 Croatia 56 25

U3 China 83 Croatia 58 25

U4 China 76 Angola 52 24

U5 USA 89 Turkey 58 31

U6 Turkey 82 China 55 27

U7 USA 88 Czech Republic 61 27

U8 Czech Republic 82 Angola 47 35

U9 USA 86 France 50 36

The sample was not randomly selected, as it rarely is in performance analysis, rather it was selected on the grounds of differentiating between characteristics in order to allow a balanced comparison of the two game states (O’Donoghue, 2010). The idea to separate the games into close and unbalanced pools is consistent throughout the literature, but with varying thresholds used to define categories. Table 2 displays the sample thresholds from recent performance analysis literature surrounding performance indicators in basketball.

17

Table 2. Game categorisation thresholds

Research Paper Categories

Close Gomez et al., 2006 1-6 points

Close Balanced Unbalanced Sampaio and Janeira, 2003 1-8 points 9-17 points 18+ points

Close Balanced Unbalanced Csataljay et al., 2009 1-9 points 10-22 23+ points

Balanced Unbalanced Very Unbalanced García et al., 2013 1-12 points 13-28 points 29+ points

Balanced Unbalanced Blowout García et al., 2014 1-12 points 13-28 points 29+ points

While it is evident that previous research tends to categorise close/balanced games as having score differential in single digits, the threshold for this project was even more restrictive, set at 5 points or fewer; the rationale being to include only the closest of games in this pool. Similar to the unbalanced / very unbalanced / blowout games described in Table 2, this study defined unbalanced games as having a differential of between 20 and 40 points, meaning that three outlying blowout games were removed from the sample (Table 3).

18

Table 3. Excluded outlying games (score differential: x > 40 points)

(Adapted from FIBA, 2009)

Team A Score Team B Score Diff.

USA 90 Angola 38 52

USA 114 China 66 48

USA 91 Canada 48 43

19

3.2. Variables

3.2.1. Game-related statistics

The game-related statistics listed in Table 4 are collected as the norm in basketball box scores and this data was scraped from the official Olympic box scores, stored on FIBA Olympic archive web page (FIBA, 2009). Only team total data was collected from the archived box scores, since individual player performance was irrelevant for the scope of this project. Microsoft Excel was used to store and format the data to ensure that all variables were usable, i.e. field goals made and field goals attempted had to be separated as isolated variables, rather than notated together as they were in the box score.

Table 4. Game-related statistics recorded by the FIBA Live Stats to produce the Olympic box scores (Adapted from FIBA, 2005; FIBA Europe, 2012)

Game-related statistics

Field Goal Attempt Field Goal Made Free Throw Attempt

Free Throw Made Rebound Offensive Rebound

Defensive Rebound Turnover Assist

Steal Block Personal Foul

20

3.2.2. Advanced statistics

Once the game-related statistics had been acquired from the box score archives, Microsoft Excel was used to generate advanced statistics for each performance, i.e. the common metric statistics that are derived from box score variables, listed in Table 5.

Table 5. Common advanced statistics generated from box score data

(Adapted from Basketball Reference, 2015c; 2015d; ESPN, 2015; Hoopdata, 2009).

Assisted Field Goal True Shooting Percentage Assist Rate Percentage

Assist Ratio Turnover Rate Turnover Ratio

Offensive Rebound Turnover Percentage Offensive Rebound Rate Percentage

Defensive Rebound Defensive Rebound Rate Rebound Rate Percentage

NBA Efficiency Win Score Alternate Win Score

Effective Field Goal Offensive Rating Defensive Rating Percentage

Rating Differential Offensive Efficiency Defensive Efficiency

Efficiency Differential Free Throw Rate Team Possessions

Game Pace

21

3.2.3. Preventions

The final stage of data collection was the notation of the performance metric that is the focus of this study, Rodriguez’ (2013c) Preventions metric. The information gathered from Rodriguez’ previous research (2013a; 2013b) included guidance on how to define the Prevention metric’s action variables and the formulae to calculate the Preventions statistics that are used to represent team performance. A list of action variables and their definitions (Rodriguez, 2013b) are provided in Appendix A, while the formulae used to derive the Preventions metric statistics can be found in Appendix B.

3.3. Data Collection

3.3.1. Hand Notation

The collection method used to generate the data for the Preventions metric was a simple tally hand notation system. While computerised analysis and sequential data collection would have provided more detail into the relationship between Preventions and the temporal aspect of basketball, the tally system simply and easily provided the data needed to generate Prevention team totals for use in data analysis. A hand notation proforma was created to allow the match events to be tallied with respect to each team (Appendix C). A pilot study was undertaken on a game outside of the selected sample to assess the ease of use of the proforma. It was found that the hand notation system did not include the ability to differentiate between 2-point and 3-point shot attempts in the shots challenged category. After repeated pilot studies, final issues were established that certain variables needed to be highlighted in order to further develop ease of use. While the Preventions action variables are mostly active during a defensive possession, five variables occur only in offense mode. These variables, offensive rebound, tip save, loose-ball save, screens and hockey assist, were highlighted with asterisks in order to better differentiate between the offensive and defensive variables (Appendix D).

It was also noted that at this point, issues were found relating to the definitions provided by Rodriguez for the action variables (Appendix A). The definitions were found to be too loosely worded in their description of match events, with grey areas that could lead to definitional error-based differences in data collection because of observer subjectivity and

22

bias (O’Donoghue, 2015). Therefore, before reliability testing, the definitions were updated to be more operational with more descriptive elements, adding specific examples of match events (Appendix E).

3.3.2. Reliability

Before the final data collection phase took place, a reliability study followed the pilot study. The same sample game was used, however in order to provide data that could be sequentially analysed on an instance-by-instance basis, the data collection method was adjusted for the sake of reliability testing. A simple word processed game-log, similar to those provided by Rodriguez (2013d), was generated in Microsoft Notepad alongside the hand notation tally system. This allowed instances from the two reliability records to be compared side-by-side to check for intra-observer reliability and related errors.

Reliability of each data collection group (shots, possessions, passes, turnovers and miscellaneous were assessed separately, producing the reliability matrices to compare the two game-logs. This reliability data was entered into a Microsoft Excel spreadsheet used to calculate a Kappa score for reliability; the scores and ratings generated can be found in Table 6.

Table 6. Kappa scores and ratings

Variable Group Kappa score Kappa rating

Shots κ = 0.65 Strong

Possessions κ = 0.70 Strong

Passes κ = 0.61 Strong

Turnovers κ = 0.53 Moderate

Miscellaneous κ = -0.06 Very Poor

23

Table 6 displays strong kappa ratings for data collected in the shots, possessions and passes groups. However, the reliability testing produced a moderate score (κ = 0.53) for the turnover prevention group and a very poor score (κ = -0.06) for the miscellaneous group. While the moderate kappa score is acceptable, the very poor score relating to help defense and delays is deemed unacceptable so the operational definitions were updated further with more detail in order to reduce this error during the data collection process (Appendix F).

3.4. Data Analysis

Once data had been notated from all 18 games within the sample, the information was stored and formatted in Microsoft Excel, then imported into IBM SPSS Statistics (SPSS v20.0.0.1; IBM, 2015). The vast amount of data collected was reduced down to the key variables using SPSS’ Variables Sets feature. The variables highlighted using this feature were Winner/Loser, Close/Unbalanced, Defensive Rating (Def.Rtg), Defensive Efficiency (Def.Eff), Points Prevented (PtsP), Possessions Lost (PL), Shots Challenged (SC), Botched Passes (BP) and Possessions Protected (PossP).

To analyse the data, the sample was split into Winners and Losers, then broken down further into Close Winners, Close Losers, Unbalanced Winners and Unbalanced Losers. Descriptive statistics (mean and standard deviation) were generated for the four two- layered sample groupings. The relationships between these sample groups was then analysed using a Mann-Whitney U Test, to assess the significance of the differences in mean rank and median of the various data sets (O’Donoghue, 2010). The second element that was considered during the data analysis process was the correlation of the Preventions metric statistics (PtsP, PL, SC, BP and PossP) with the defense-based advanced statistics (Def.Rtg and Def.Eff). The correlations were analysed using a Pearson Correlation (O’Donoghue, 2010).

24

CHAPTER FOUR

RESULTS

25

4. RESULTS

4.1. Descriptive Statistics

The results of this study will be presented in accordance with the aims of the project which were to measure the discriminant power of the Preventions metric and to test the relationships of the metric with other, previously academically researched, advanced statistics. A selection of box score statistics that have been previously identified as significant (Csataljay et al., 2009, Ibáñez et al., 2009, Pojskić et al., 2009; García et al., 2013; García et al., 2014; Csataljay et al., 2012; Sampaio and Janeira, 2003; Ibáñez et al., 2008; Malarranha et al., 2013) are presented in Figure 1, to provide an insight into the absolute differences in these statistics.

Close Winners Close Losers Unbalanced Winners Unbalanced Losers

100.0

90.0

80.0

70.0

60.0

50.0

40.0

30.0

20.0

10.0

0.0 FG% 2PT% 3PT% FT% ORB DRB TRB AST TOV PTS Variables

Figure 1. Descriptive statistics of box score performance indicators

Figure 1 highlights, as expected, that the winning teams invariably out-performed the losing teams in all categories, with the differences in close games being much more marginal than in the unbalanced games. In close games, the greatest percentage changes were the winners having a 9.1% improvement in three-point percentage, a 9.0% increase in defensive rebounds and a 7.4% in total rebounds; while in unbalanced games these 26

variables were the winners showing a 48.1 increase in assists, a 45.9% decrease in turnovers and a 27.3% improvement in three-point percentage (not including the already evident 34.1% increase in points scored used to classify the games) (Table 7).

Table 7. Descriptive statistics and percentage differences of box score performance indicators

Close Unbalanced Unbalanced Close Loser Winner Winner Loser

Variable Mean ±S.D. Mean ±S.D. %Diff Mean ±S.D. Mean ±S.D. %Diff

FG% 40.5 5.6 38.1 4.2 +5.9 45.4 3.1 33.5 6.1 +26.2

2PT% 43.6 8.5 41.7 8.4 +4.4 49.1 5.1 37.2 8.2 +24.2

3PT% 29.6 15.3 26.9 12.5 +9.1 31.1 10.4 22.6 8.2 +27.3

FT% 74.8 9.1 75.1 9.6 -0.4 78.2 12.5 67.4 15.6 +13.8

ORB 13.6 3.0 12.9 4.3 +5.1 15.6 5.8 12.2 5.3 +21.8

DRB 26.8 4.1 24.4 4.4 +9.0 31.4 4.2 23.9 4.0 +23.9

TRB 40.3 3.9 37.3 5.0 +7.4 47 8.5 36.1 5.1 +23.2

AST 14.1 4.3 14.1 3.4 0.0 21.2 3.2 11.0 3.8 +48.1

TOV 14.0 3.1 14.6 3.4 -4.3 13.3 4.3 19.4 2.7 -45.9

PTS 68.2 6.8 64.3 7.2 5.7 82.1 5.5 54.1 4.6 +34.1

The variables that will be focussed on in order to answer the research question are the advanced statistics that quantify overall defensive performance; Defensive Rating and Defensive Efficiency, in comparison with the outputs of the Preventions metric – Points Prevented, Possessions Lost, Shots Challenged and Botched Passes. Descriptive statistics for these six chosen variables are presented in Table 8 and are displayed

27

graphically in Figure 2 in order to best illustrate the spread of data across the four sample groups.

Table 8. Descriptive statistics of defensive advanced statistics and Preventions metric outputs

Unbalanced Unbalanced Close Winner Close Loser Winner Loser

Variable Mean ±S.D. Mean ±S.D. Mean ±S.D. Mean ±S.D.

D.Rtg 88.2 7.0 92.7 7.2 69.9 7.2 105.7 10.3

Def.Eff 91.1 8.7 97.5 8.4 71.4 3.4 108.4 5.8

PtsP 47.7 10.3 43.4 8.8 52.9 9.9 38.7 12.1

PL 20.6 5.7 20.3 5.1 24.2 7.5 18.9 5.4

SC 21.4 5.2 19.4 4.0 23.4 4.1 18.1 5.6

BP 6.4 3.9 4.9 2.4 10.2 2.7 6.1 1.6

28

Close Winner Close Loser Unbalanced Winner Unbalanced Loser

140

120

100

80

60

40

20

0 D.Rtg Def.Eff PtsP PL SC BP Variable

Figure 2. Descriptive statistics of defensive advanced statistics and Preventions metric outputs

4.2. Tests of Significance

After establishing the differences in the descriptive statistics of the defensive performance indicators (Appendix G), said differences were tested for statistical significance using a Mann-Whitney U Test. Presented in the following tables is a summary of the results of the statistical tests of the relationships, with the significant differences highlighted. The full results of the Mann-Whitney U Tests can be found in Appendix H. Table 9 presents the results of the Mann-Whitney U Test, highlighting the variables with significance differences between the Winners and Losers samples, with D.Rtg and Def.Eff significant at the p < 0.001 level, while PtsP, SC and BP were significant at the p < 0.05 level.

Table 9. Statistical significance of the differences between Winners (W) vs Losers (L)

Medians Mean Ranks Significance Variable W L W L U z P

Defensive 78.3 98.4 11.2 25.8 30.0 -4.2 .000**** Rating

29

Defensive 77.2 106.4 11.0 26.0 27.0 -4.3 .000**** Efficiency

Points 49.5 43.0 22.8 14.2 85.0 -2.4 .015* Prevented

Possessions 21.5 19.0 20.7 16.3 123.0 -1.2 .216 Lost

Shots 21.5 19.5 22.1 14.9 97.0 -2.1 .039* Challenged

Botched 8.0 5.0 22.7 14.3 87.0 -2.4 .017* Passes

Possessions 8.5 6.5 21.9 15.1 101.0 -1.9 .052 Protected Table 10 presents the results of the Mann-Whitney U Test, highlighting the variables with significance differences between the Close Winners and Unbalanced Winners samples. D.Rtg and Def.Eff significant at the p < 0.001 level, while BP were significant at the p < 0.05 level.

Table 10. Statistical significance of the differences between Close Winners (CW) and Unbalanced Winners (UW)

Medians Mean Ranks Significance Variable CW UW CW UW U z p

Defensive 87.8 73.7 14.0 5.0 0.0 -3.6 .000**** Rating

Defensive 89.8 72.0 14.0 5.0 0.0 -3.6 .000**** Efficiency

Points 46.0 53.0 7.7 11.3 24.0 -1.5 .145 Prevented

30

Possessions 21.0 23.0 8.3 10.7 30.0 -0.9 .353 Lost

Shots 21.0 24.0 7.7 11.3 24.5 -1.4 .155 Challenged

Botched 6.0 11.0 6.8 12.2 16.0 -2.2 .029* Passes

Possessions 8.0 9.0 8.2 10.8 29.0 -1.0 .306 Protected

31

Table 11 presents the results of the Mann-Whitney U Test, highlighting the variables with significance differences between the Close Losers and Unbalanced Losers samples. Only D.Rtg and Def.Eff were found to be significant, at the p < 0.05 level.

Table 11. Statistical significance of the differences between Close Losers (CL) and Unbalanced Losers (UL)

Medians Mean Ranks Significance Variable CL UL CL UL U z p

Defensive 94.9 107.0 6.4 12.6 13.0 -2.4 .015* Rating

Defensive 97.0 107.7 6.4 12.6 13.0 -2.4 .015* Efficiency

Points 46.0 38.0 11.1 8.0 26.5 -1.2 .216 Prevented

Possessions 19.0 19.0 10.1 88.9 35.0 -0.5 .626 Lost

Shots 20.0 18.0 10.7 8.3 30.0 -0.9 .350 Challenged

Botched 5.0 6.0 8.1 10.9 28.0 -1.1 .259 Passes

Possessions 7.0 6.0 9.8 9.2 37.5 -0.3 .788 Protected

32

Table 12 displays the results of the Mann-Whitney U Tests comparing the Close Winners and Close Losers sample groups, with no variables were found to have significant differences.

Table 12. Statistical significance of the differences between Close Winners (CW) and Close Losers (CL)

Medians Mean Ranks Significance Variable CW CL CW CL U z p

Defensive 87.8 94.9 7.8 11.2 25.0 -1.4 .171 Rating

Defensive 89.8 97.0 7.7 11.3 24.0 -1.5 .145 Efficiency

Points 46.0 46.0 10.1 8.9 35.0 -0.5 .625 Prevented

Possessions 21.0 19.0 9.4 9.6 40.0 -0.0 .965 Lost

Shots 21.0 20.0 9.9 9.1 36.5 -0.4 .721 Challenged

Botched 6.0 5.0 10.4 8.6 32.0 -0.8 .449 Passes

Possessions 8.0 7.0 10.8 8.2 28.5 -1.1 .284 Protected

33

Table 13 presents the results of the Mann-Whitney U Test, highlighting the variables with significance differences between the Unbalanced Winners and Unbalanced Losers samples, with D.Rtg and Def.Eff significant at the p < 0.001 level, BP significant at the p <

0.005 level, while PtsP and SC were significant at the p < 0.05 level.

Table 13. Statistical significance of the differences between Unbalanced Winners (UW) and Unbalanced Losers (UL)

Medians Mean Ranks Significance Variable UW UL UW UL U Z p

Defensive 73.7 107.0 5.0 14.0 0.0 -3.6 .000**** Rating

Defensive 72.0 107.7 5.0 14.0 0.0 -3.6 .000**** Efficiency

Points 53.0 38.0 12.4 6.6 14.0 -2.3 .019* Prevented

Possessions 23.0 19.0 11.4 7.6 23.4 -1.5 .132 Lost

Shots 24.0 18.0 12.0 7.0 18.0 -2.0 .046* Challenged

Botched 11.0 6.0 13.1 5.9 8.0 -2.9 .004*** Passes

Possessions 9.0 6.0 11.7 7.3 21.0 -1.7 .083 Protected

34

4.3. Correlations

Table 14 displays the results of the Pearson Correlation test, for the entire sample of games. The significant correlations at the p < 0.001 level were D.Rtg-PtsP (r = +0.917, p < 0.001), D.Rtg-PtsP (r = -0.569, p < 0.001), PtsP-SC (r = +0.983, p < 0.001); while the correlations significant at the p < 0.005 level were Def.Eff-PtsP (r = -0.477, p = 0.003), D.Rtg-SC (r = -0.520, p = 0.001) and Def.Eff-BP (r = -0.547, p = 0.001). The less significant relationships (p < 0.05 level) were D.Rtg-BP (r = -0.412, p = 0.013) and Def.Eff- SC (r = -0.414, p = 0.012).

Table 14. Pearson’s correlation (entire sample) Correlations Entire Sample D.Rtg Def.Eff PtsP PL SC BP PossP Pearson Correlation 1 .917**** -.569**** -.183 -.520*** -.412* -.180 D.Rtg Sig. (2-tailed) .000 .000 .284 .001 .013 .293

N 36 36 36 36 36 36 36 Pearson Correlation .917**** 1 -.477*** -.240 -.414* -.547*** -.230 Def.Eff Sig. (2-tailed) .000 .003 .159 .012 .001 .177

N 36 36 36 36 36 36 36 Pearson Correlation -.569**** -.477*** 1 .050 .983**** .139 .012 PtsP Sig. (2-tailed) .000 .003 .772 .000 .418 .945

N 36 36 36 36 36 36 36 Pearson Correlation -.183 -.240 .050 1 .023 .137 -.008 PL Sig. (2-tailed) .284 .159 .772 .892 .426 .964

N 36 36 36 36 36 36 36 Pearson Correlation -.520*** -.414* .983**** .023 1 .070 .008 SC Sig. (2-tailed) .001 .012 .000 .892 .687 .962

N 36 36 36 36 36 36 36 Pearson Correlation -.412* -.547*** .139 .137 .070 1 .419* BP Sig. (2-tailed) .013 .001 .418 .426 .687 .011

N 36 36 36 36 36 36 36 Pearson Correlation -.180 -.230 .012 -.008 .008 .419* 1 PossP Sig. (2-tailed) .293 .177 .945 .964 .962 .011

N 36 36 36 36 36 36 36 ****. Correlation is significant at the 0.001 level (2-tailed). ***. Correlation is significant at the 0.005 level (2-tailed). **. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).

35

Table 15 displays the results of the Pearson Correlation test, for just the data from the winning teams. The significant correlations at the p < 0.001 level were D.Rtg-Def.Eff (r = +0.870, p < 0.001) and PtsP-SC (r = +0.978, p < 0.001). A less significant relationship was also found in the Def.Eff-BP correlation (r = -0.523, p = 0.026).

Table 15. Pearson’s correlation (Winners) Correlations Winners D.Rtg Def.Eff PtsP PL SC BP PossP Pearson Correlation 1 .870** -.259 -.204 -.219 -.264 .147 D.Rtg Sig. (2-tailed) .000 .300 .417 .382 .290 .562 N 18 18 18 18 18 18 18 Pearson Correlation .870** 1 -.188 -.281 -.110 -.523* .033 Def.Eff Sig. (2-tailed) .000 .456 .259 .664 .026 .897 N 18 18 18 18 18 18 18 Pearson Correlation -.259 -.188 1 .043 .978** -.006 -.183 PtsP Sig. (2-tailed) .300 .456 .865 .000 .982 .468 N 18 18 18 18 18 18 18 Pearson Correlation -.204 -.281 .043 1 .005 .108 -.065 PL Sig. (2-tailed) .417 .259 .865 .983 .670 .798 N 18 18 18 18 18 18 18 Pearson Correlation -.219 -.110 .978** .005 1 -.098 -.129 SC Sig. (2-tailed) .382 .664 .000 .983 .698 .609 N 18 18 18 18 18 18 18 Pearson Correlation -.264 -.523* -.006 .108 -.098 1 .291 BP Sig. (2-tailed) .290 .026 .982 .670 .698 .242 N 18 18 18 18 18 18 18 Pearson Correlation .147 .033 -.183 -.065 -.129 .291 1 PossP Sig. (2-tailed) .562 .897 .468 .798 .609 .242 N 18 18 18 18 18 18 18 ****. Correlation is significant at the 0.001 level (2-tailed). ***. Correlation is significant at the 0.005 level (2-tailed). **. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).

36

Table 16 displays the correlation data for the losing teams, presenting D.Rtg-Def.Eff as significant at the p < 0.001 level (r = +0.813, p < 0.001). PtsP-SC was also found to be significant at this level (r = 0.983, p < 0.001). At the p < 0.01 confidence level, D.Rtg-PtsP were found to be correlated (r = 0.607, p = 0.008). Finally, D.Rtg-SC was found to be significant at the p < 0.05 level (r = -0.579, p = 0.012).

Table 16. Pearson’s correlation (Losers) Correlations Losers D.Rtg Def.Eff PtsP PL SC BP PossP Pearson Correlation 1 .813**** -.607** .196 -.579* -.041 -.084 D.Rtg Sig. (2-tailed) .000 .008 .435 .012 .873 .739 N 18 18 18 18 18 18 18 Pearson Correlation .813**** 1 -.408 .199 -.387 -.039 -.073 Def.Eff Sig. (2-tailed) .000 .093 .429 .112 .877 .775 N 18 18 18 18 18 18 18 Pearson Correlation -.607** -.408 1 -.176 .983**** -.129 -.085 PtsP Sig. (2-tailed) .008 .093 .486 .000 .609 .738 N 18 18 18 18 18 18 18 Pearson Correlation .196 .199 -.176 1 -.167 -.112 -.117 PL Sig. (2-tailed) .435 .429 .486 .508 .658 .643 N 18 18 18 18 18 18 18 Pearson Correlation -.579* -.387 .983**** -.167 1 -.135 -.110 SC Sig. (2-tailed) .012 .112 .000 .508 .592 .663 N 18 18 18 18 18 18 18 Pearson Correlation -.041 -.039 -.129 -.112 -.135 1 .447 BP Sig. (2-tailed) .873 .877 .609 .658 .592 .063 N 18 18 18 18 18 18 18 Pearson Correlation -.084 -.073 -.085 -.117 -.110 .447 1 PossP Sig. (2-tailed) .739 .775 .738 .643 .663 .063 N 18 18 18 18 18 18 18 ****. Correlation is significant at the 0.001 level (2-tailed). ***. Correlation is significant at the 0.005 level (2-tailed). **. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).

37

Table 17 presents the correlation results for the data relating to just the close games, finding only two significant relationships. Significant at the p < 0.001 level, D.Rtg-Def.Eff (r = +0.759, p < 0.001) and PtsP-SC (r = +0.985, p < 0.001) were the only significant correlations.

Table 17. Pearson’s correlation (Close) Correlations Close D.Rtg Def.Eff PtsP PL SC BP PossP Pearson Correlation 1 .759**** -.294 .234 -.255 -.208 .382 D.Rtg Sig. (2-tailed) .000 .237 .349 .308 .407 .118 N 18 18 18 18 18 18 18 Pearson Correlation .759**** 1 -.134 .167 -.096 -.457 .188 Def.Eff Sig. (2-tailed) .000 .595 .508 .706 .057 .454 N 18 18 18 18 18 18 18 Pearson Correlation -.294 -.134 1 -.365 .985**** -.098 .066 PtsP Sig. (2-tailed) .237 .595 .137 .000 .699 .795 N 18 18 18 18 18 18 18 Pearson Correlation .234 .167 -.365 1 -.358 -.305 .014 PL Sig. (2-tailed) .349 .508 .137 .145 .219 .955 N 18 18 18 18 18 18 18 Pearson Correlation -.255 -.096 .985**** -.358 1 -.177 .053 SC Sig. (2-tailed) .308 .706 .000 .145 .482 .834 N 18 18 18 18 18 18 18 Pearson Correlation -.208 -.457 -.098 -.305 -.177 1 .274 BP Sig. (2-tailed) .407 .057 .699 .219 .482 .271 N 18 18 18 18 18 18 18 Pearson Correlation .382 .188 .066 .014 .053 .274 1 PossP Sig. (2-tailed) .118 .454 .795 .955 .834 .271 N 18 18 18 18 18 18 18 ****. Correlation is significant at the 0.001 level (2-tailed). ***. Correlation is significant at the 0.005 level (2-tailed). **. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).

38

Presented in Table 18 are the correlation results for the unbalanced games. The correlations significant at the p < 0.001 level were D.Rtg-Def.Eff (r = +0.947, p <0.001), PtsP-SC (r = +0.985, p < 0.001). At the p < 0.005 confidence level, D.Rtg-PtsP (r -0.675, p = 0.002) and Def.Eff-BP (r = -.0643, p = 0.004) were found to correlate. Moving onto the p < 0.01 significance level, Def.Eff-PtsP (r = +-0.611, p = 0.007) was found to be significant. Finally, at the p < 0.05 confidence level, D.Rtg-BP (r = -0.561, p = 0.015), Def.Eff-SC (r = - .0557, p = 0.016) and BP-PossP (r = +0.557, p = 0.016) were significant.

Table 18. Pearson’s correlation (Unbalanced) Correlations

Unbalanced D.Rtg Def.Eff PtsP PL SC BP PossP Pearson Correlation 1 .947**** -.675*** -.299 -.644** -.561* -.364 D.Rtg Sig. (2-tailed) .000 .002 .227 .004 .015 .138 N 18 18 18 18 18 18 18

Pearson Correlation .947**** 1 -.611** -.375 -.557* -.643** -.396 Def.Eff Sig. (2-tailed) .000 .007 .126 .016 .004 .104 N 18 18 18 18 18 18 18 Pearson Correlation -.675*** -.611** 1 .280 .985** .345 -.024 PtsP Sig. (2-tailed) .002 .007 .261 .000 .161 .924 N 18 18 18 18 18 18 18 Pearson Correlation -.299 -.375 .280 1 .262 .454 -.037 PL Sig. (2-tailed) .227 .126 .261 .294 .059 .884 N 18 18 18 18 18 18 18 Pearson Correlation -.644** -.557* .985** .262 1 .278 -.030 SC Sig. (2-tailed) .004 .016 .000 .294 .263 .906 N 18 18 18 18 18 18 18 Pearson Correlation -.561* -.643** .345 .454 .278 1 .557* BP Sig. (2-tailed) .015 .004 .161 .059 .263 .016 N 18 18 18 18 18 18 18 Pearson Correlation -.364 -.396 -.024 -.037 -.030 .557* 1 PossP Sig. (2-tailed) .138 .104 .924 .884 .906 .016 N 18 18 18 18 18 18 18 ****. Correlation is significant at the 0.001 level (2-tailed). ***. Correlation is significant at the 0.005 level (2-tailed). **. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).

39

CHAPTER FIVE

DISCUSSION

40

5. DISCUSSION

5.1. Discussion of Results

The aim of this study was to assess whether the unproven Preventions metric (Rodriguez, 2013c) is a discriminant of winning performance, through assessing and comparing the performance of winning and losing teams at the 2012 Olympic Games. Using this data, the aim was also to compare the validity of the Preventions metric with advanced statistics that are derived from box score data (Swalgin, 2008). Throughout data analysis, the results were calculated to relate to either the entire sample of games, to separate the winners and losers and to separate the close and unbalanced samples. Statistical testing was performance on seven variables, two of which represent the box score game-related statistics (defensive rating and defensive efficiency) while five of the variables are outputs of the preventions metric (points prevented, possessions lost, shots challenged, botched passes and possessions protected). This restriction of variables ensured the data analysis process to be highly conceptually focussed upon the defensive process.

5.1.1. All Games

The descriptive statistics that compared the entire sample of games (n = 36), in four two- layered samples, showed that the unbalanced winners out-performed the three other samples in the 10 box score variables selected to be presented (Figure M1). These variables, previously found to be significant, were field goal percentage (Csataljay et al., 2009, Ibáñez et al., 2009, Pojskić et al., 2009), two-point field goal percentage (Csataljay et al., 2009, García et al., 2013, García et al., 2014, Ibáñez et al., 2009, Pojskić et al., 2009), three-point field goal percentage (Csataljay et al., 2009, García et al., 2013; García et al., 2014, Ibáñez et al., 2009, Pojskić et al., 2009), free-throw percentage (Csataljay et al., 2009; Csataljay et al., 2012; Ibáñez et al., 2009, Sampaio and Janeira, 2003), offensive rebounding (Csataljay et al., 2012, Sampaio and Janeira, 2003), defensive rebounding (Csataljay et al., 2009; Csataljay et al., 2012, García et al., 2013, García et al., 2014, Ibáñez et al., 2009, Pojskić et al., 2009), total rebounding (Csataljay et al., 2012, Ibáñez et al., 2009), assists (García et al., 2013, García et al., 2014, Ibáñez et al., 2008, Ibáñez et al., 2009, Pojskić et al., 2009), turnovers (Malarranha et al., 2013), points (Malarranha et al., 2013). 41

For all ten variables the best performance was by the unbalanced winners, with the close winners displaying the second best performance in nine of the ten variables, where the exception was the close winners and close losers recording an equal amount of assists. Similarly when comparing the defensive advanced statistics and the Preventions metric statistics across the four sample grouping, it was found that the unbalanced winners invariably out-performed the other samples in all six categories (defensive rating, defensive efficiency, points prevented, possessions lost, shots challenged and botched passes). Defensive rating and defensive efficiency were selected because of their previous appearances in literature where they were found to be significant in modelling winning performance (Malarranha et al. 2013, Sporiš et al., 2006).

5.1.2. Winners and Losers

To delve deeper into the results of this study, the winning and losing samples were separated to provide a statistical insight into the discriminants between successful and unsuccessful performance. The results of the Mann-Whitney U Test of significance (O’Donoghue, 2010) for the global winners and losers identified defensive rating and defensive efficiency as highly significant, while points prevented, shots challenged and botched passes were also significant. This data identifies defensive rating and defensive efficiency as highly related to winning performance, an outcome which is to be expected since they are advanced statistics that purely represent the oppositions’ scoring rate, the only difference between the two measures being that they quantify a team’s possessions differently (Basketball Reference, 2015d; ESPN, 2015). Therefore, all this data serves to prove is that reducing your opponents’ scoring rate will enable you to win more games, a relatively obvious statement.

The statistical significance of the Preventions metric statistics provides a little more insight into the process of winning performance, by identifying the shooting and passing elements of the game as more important than protecting or gaining extra possessions (possessions protected and possessions lost). Similar to defensive efficiency and defensive rating, points prevented and shots challenged are highly related measures, as points prevented is derived from the value of shots a team successfully challenges. However, this does not diminish the fact that forcing a team to miss is significant in producing a winning performance. The significance of the botched passes variable ascertains that being

42

disruptive on defense and not allowing a team to pass the ball freely is also central to giving a team a better chance to win the game.

Dividing the data up further allowed the comparison of close winners and close losers, which led to the conclusion that the teams could not be meaningfully separated by any of the seven performance indicators, due to no variables being statistically significant between the two samples. This indicates that in close games it is extremely difficult to determine, using these variables, what it is that leads to winning performance. Conversely, when looking at the unbalanced winners and unbalanced losers, the significance of the performance indicators is much greater. Defensive rating and defensive efficiency are extremely significant; as expected due to the variables being based around points conceded, which by the nature of the sample grouping criteria means that the unbalanced winners achieved a substantial score differential. However, these are not the only significant variables, significant differences also found in points prevented, shots challenged and botched passes. These results display similar findings as when the winners and losers were compared globally, disregarding the game state, showing the influence of the unbalanced games on the findings.

The unbalanced games seem to prove the impact of the Preventions metric statistics as discriminant between winning and losing, so it is important to delve deeper into the absolute figures. The highly significant difference in botched passes by the unbalanced winners represents an absolute difference of five passes disrupted, not a vast margin, but noteworthy if you consider that those five possessions interrupted equates to an 83% increase on the losing teams performance. Again, it is sensible to link the significance of points prevented and shots challenged since the variables are derived from one another, but that does not discount the bearing that the differences had on the game. Looking at the median figures again, the winners challenged six more shots per game than the losers, equating to an extra 15 points prevented per game, which is a large portion of the 20-40 point score differential that was the sampling criteria.

43

5.1.3. Close and Unbalanced

The winners and losers were also compared with regards to the differences between the teams that lost in close games and the teams that lost in unbalanced contests. This testing only found the advanced statistics, defensive rating and defensive efficiency, to be significant discriminants. This alludes to the fact that due to the sampling criteria, the unbalanced losers conceded a much greater amount of points than the close losers, which then affects their scoring rates. None of the Preventions metric statistics were found to be significant, which suggests that the defensive process by the losing teams was similarly ineffective in all areas of Preventions. On the other hand, when comparing the winners of close games with the winners of unbalanced games, one of the Preventions metric statistics was found to be significant. Similarly defensive rating and defensive efficiency again were found to be highly significant, but with these samples botched passes was also established as a variable with a substantial difference between the two samples.

5.1.4. Correlations

In order to answer the second half of the research aims for this study, i.e. whether the Preventions metric statistics correlate with the advanced statistics derived from box score variables, the final stage of data analysis was to use a Pearson Correlation to assess the relationships between the performance indicators (O’Donoghue, 2010). As has already been mentioned, the defensive rating and defensive efficiency variables are highly linked as they measure the same phenomenon within performance, a team’s scoring rate per 100 possessions, but use a different method to calculate a team’s possessions. Points prevented and shots challenged have also been identified as variables that are derived from one another, therefore the correlations found between these variables (D.Rtg-Def.Eff and PtsP-SC) will be omitted from this discussion.

When correlating the data for the entire sample, seven significant correlations were found. Both defensive rating and defensive efficiency were found to correlate with points prevented, shots challenged and botched passes; while a significant correlation found within the Preventions metric statistics was botched passes correlating, albeit weakly, with possessions protected. If defensive rating and defensive efficiency are considered valid measures of defensive effectiveness, then the significant correlations found between the

44

advanced statistics and the Preventions outputs suggests that the metric statistics are also a valid measure of whether a team plays effective defense.

When the data set was restricted to just the winners’ performances, the only significant correlation was between defensive efficiency and botched passes, a result that strengthens the case for botched passes to be accepted as a key performance indicator. When looking at the losers’ data separately, defensive rating was found to correlate with points prevented and shots challenged, again suggesting that the ability to influence the oppositions shooting performance affects overall defensive performance.

Splitting the data into close and unbalanced games showed that no significant correlations could be found within the close games, while many variables correlated in the unbalanced sample. Defensive rating and defensive efficiency correlated at various significance levels with points prevented, shots challenges and botched passes. Within the Preventions metric itself, botched passes correlated with possessions protected. This was the same correlations found when assessing the entire sample, which suggests that, as with the Mann-Whitney U Tests, the unbalanced data sets had a substantial impact on the results when looking at the data globally.

The data analysis as a whole seems to have proven that Preventions metric statistics that discriminate between winning and losing performance are points prevented, shots challenged and botched passes; however these differences were not significant enough when considered in the context of close games. Identifying these variables as having discriminative power is also supported by the correlation data that established points prevented, shots challenged and botched passes as correlating with the box score-derived advanced statistics. The conclusion can therefore be drawn that points prevented, shots challenged and botched passes can be labelled as performance indicators that have discriminant power in winning performance and that correlate with previously identified performance indicators.

45

5.2. Implications for Practice

As stated in the review of literature, performance analysis exists as a source of information to underpin and inform coaching decisions and interventions (Maslovat and Franks, 2008). Therefore, for this study to be considered a success, the results must be applicable and usable for basketball coaches to use in practice. The findings of this study are applicable in practice to inform and guide coaches on which components of defensive strategy to emphasise, while also providing a valid tool to motivate athletes by quantifying their procedural-based efforts.

5.2.1. Defending Shots

The significance of shots challenged and points prevented variables is consistent with the findings of Álvarez et al. (2009) and Csataljay et al. (2013), who also found defensive pressure to be a key factor in affecting opposition shooting performance. The Preventions metric has therefore emphasised this component of defensive strategy, so coaches should highlight that, on the perimeter, defenders need to effectively close-out on shooters and apply pressure as they perform their shooting motion (ASEP, 2007; Csataljay et al., 2013). This relies on the coaches having taught a defined team defensive strategy in terms of how off-ball defenders move as the ball is passed around the court (ASEP, 2007). Wissel (2012) explains the weak-side help principle that is important in limiting driving lanes and passes to cutters. While help defense is important, this leaves players open to receive passes that can lead to open shots. The Preventions metric stresses that players should be able to close out on these players receiving the ball in order to prevent open shot attempts, in turn preventing scored baskets (ASEP, 2007; Csataljay et al., 2013). The closing out technique is something that is widely discussed among coaching literature as a crucial skill to master in order to be able to play effective individual and team defense (ASEP, 2012). Coaches can use the findings from the Preventions metric to further underpin and evidence the fact that closing out on shooters and forcing misses is a significant factor in winning games (Csataljay et al., 2013).

Similarly, it is vital that when playing defense in the key, defensive players must contest inside shot attempts with their hands up while creating legal contact which affects the focus of the shooter (ASEP, 2007). Throughout the data collection process, it was noted that a two-point shot contest close to the basket was often followed by an offensive

46

rebound for the player who shot the ball. Coaches can use this knowledge to focus their players on quickly transitioning from contesting a shot to boxing out their player in order to secure the rebound (Krause et al., 2008; Wissel, 2012).

5.2.2. Defending Passes

Data analysis also underlined disrupting passes and forcing turnovers as a significant match event, congruent with the findings of previous research into turnovers (Fylaktakidou et al. 2011; Hawkins and Choi, 2009). The ability to defend opposition passing phases had a vast impact on both their defensive and offensive effectiveness, meaning that coaches should place a lot of emphasis on deny passing lanes and playing tight on-ball defense to disrupt the opposition’s offensive flow (ASEP, 2007). The factors that will affect a teams’ ability to forces turnovers by intercept passes are strong-side denial defense and active hands when defending on-ball (Wissel, 2012). Denial defense is when an off-ball defender’s assignment is close enough to the ball to receive a pass, so they must deny that pass being made (ASEP, 2012; Wissel, 2012). This is achieved by taking a closed stance to their player, with their lead foot and hand in the passing lane, restricting the window for the pass to be made (ASEP, 2012; Wissel, 2012; Wootten and Wootten, 2013). Additionally, effective on-ball defense heavily influences the ability of a team to pass the ball fluidly because players defending the ball with their hands up places intense pressure on the ball-carrier as they attempt to find an open team mate (ASEP, 2007; 2012).

The identification of shooting and passing as offensive actions that must be limited is not ground-breaking, however, it is useful evidence to back up the coaching adage that defense wins championships (Oliver, 1997). Defensive effectiveness can also have a vast impact on the style of offense a team can play because if they are adept at forcing turnovers then they will have more fast break opportunities (Tavares and Gomes, 2003). This in turn leads to easier shot attempts because the defense are not able to get back in transition and set up as a team to defend the possession (Wootten, 2003; Tsamourtzis et al., 2005; Krause et al., 2008).

47

5.2.3. Motivation

The Preventions metric can also be used as a motivational tool to quantify the effects of players who commit to the defensive process, through providing extrinsic feedback that leads to knowledge of performance (Hodges and Franks, 2008). This feedback applied in an athlete-centred approach can become a powerful source of motivation, used to promote learning, skill development and persistence (Hendry and Hodges, 2013). Accelerating the skill acquisition and development phases of defensive performance in basketball through process-oriented feedback from the gathered data is an element of this study that has great potential to motivate (Wulf and Shea, 2004). However, this provision of information must be used sparingly, finely balanced by coaches, as too much extrinsic feedback can lead to guidance dependency (Magill and Anderson, 2012).

5.3. Reliability of the Data Source

The reliability of the data collected is a crucial element of this study that requires discussion. As illustrated in the methods section of this study, issues were found regarding the data collection system, specifically the reliability of the notation of certain variables. The shots, possessions and passes group all performed well when tested for reliability, producing strong kappa values, whereas the turnovers group only produced a moderate score (Table 6). The miscellaneous group, consisting of help defense and delays, was found to be very unreliable, producing a negative kappa score. This was noted throughout data analysis so, while the operational definitions were updated to try to improve reliability (Appendix 6), the metric statistics associated with this group were removed from the study. Research conducted by Huciñski and Tymañski (2008) found that even expert observers, when using hand notation, were unreliable when assessing defensive effectiveness. Therefore, it should be expected that evaluating team defense, at a complex procedural level, would be a difficult task for an observer. The operational definitions provided by Rodriguez (2013b) (Appendix 1) were found to be too vague in their description of match events, but these have been updated in a more recent publication (Rodriguez, 2015a) that was released after data collection occurs. The new more rigorous and exhaustive definitions for all action variables would likely have an effect on reliability, if the study was to be repeated (Rodriguez, 2015b).

48

49

5.4. Game Pace

Finally, a concept that was discussed in the review of literature that ought to be mentioned was the notion of normalising data by game pace. It should be noted at this point that this study did not normalise the Preventions data using game pace, which may have had an impact on the results due to the different number of possessions in each game because of the varied styles of basketball that were evident throughout the sample. As Kubatko et al. (2007) and Csataljay et al. (2011) highlight, it is important that data from different games is made relative to the entire sample because the absolute data is affected by the individual characteristics of each independent sample of gameplay. For example, rebounding in absolute terms, e.g. total rebounds per game, is not as descriptive as rebounding rates that consider both the number of rebounds available and the number of rebounds the opposition record (Csataljay et al., 2011). In further studies in this area, it is suggested that the Preventions data should be adjusted by game pace, in order to better assess its validity in comparison to advanced statistics that have been normalised by possessions data, e.g. defensive efficiency rating (Oliver, 2004; Kubatko et al., 2007). This would allow the researcher to make more accurate claims that are generalisable to games played at any pace.

50

CHAPTER SIX

CONCLUSION

51

6. CONCLUSION

6.1. Concluding Thoughts

The sample of 18 games from the London 2012 Olympic Basketball Tournament for Women produced shots challenged, points prevented and botched passes as significant variables that discriminate between winning and losing performances. When categorised by score-line differential, it was found that in close games none of the Preventions metric statistics were significant, while in these unbalanced games the three variables that were globally significant were also factors in discriminating winning performance. The performances of the winners and losers in the close games were too similar for significant differences to be found, suggesting that winning these tight matches comes down to the narrowest of margins. Whereas, in the unbalanced games, it was straight-forward to identify the factors that led to the greater margin of victory.

Identifying the variables found to lead to winning performance means that coaches can better prioritise the skills that are necessary for their team to be successful. Since pressuring shooters is an important facet of defensive strategy (Csataljay et al., 2013), coaches should highlight closing-out and contesting shots as vital to win games (ASEP, 2012). Similarly, with the Preventions data in mind, emphasising pressure on ball-handlers and denying passing lanes is crucial for a team to force turnovers and intercept passes (ASEP, 2007; Fylaktakidou et al., 2011; Hawkins and Choi, 2009), leading to fast break scoring opportunities (Tavares and Gomes, 2003).

While the Preventions metric in its entirety is still to be proven academically as a valid performance indicator, this study has underlined the potential that this type of procedural metric data has to underpin coaching decision making (Hughes and Bartlett, 2002; Maslovat and Franks, 2008). By notating the process of match events, the coaches and players can better understand the efficacy of their actions rather than the results themselves, which are influenced by external uncontrollable constraints (Newell, 1986). This strand of performance analysis is crucial in informing and supporting the process-goal nature of coaching in basketball, due to the complexity of the game itself (Sampaio, Ibáñez and Lorenzo, 2013).

52

6.2. Future Research

While this research has begun to explore the Preventions metric in an academic setting, there is still a vast amount that can be done in future research to further look at the procedural elements of defense in basketball. Firstly, the Preventions metric has three statistically significant variables (points prevented, shots challenged and botched passes), which suggests that the Preventions metric could be a basis for future research because this study has aligned the metric as a discriminant of winning. The other two variables (possessions lost and possessions protected) cannot yet be considered as valid performance indicators because they were not found to be significant in this project (Hughes and Bartlett, 2002). However, that does not mean that in future studies in a different context that these variables would not be significant, so future research could review the application of this metric in other settings. The sample could be adjusted to apply Preventions to the men’s game, junior basketball or domestic competitions rather than international.

Changing the Preventions system should also be considered to better describe the defensive process and its outcomes. More detailed studies include the work by Ferreira et al. (2014) who assessed defensive strategy in three main areas – ball pressure, defensive phase and type of defense. The thorough data collection method meant that the most influential factors were identified, particularly those that affected the game result during the critical periods. A significant variable found in this study was the number of shots challenged, so, in a future study, this notation could be expanded to record the success or failure of the shot attempt in relation to the level of pressure the defender put on the shooter. This would allow the data to accurately show the effects of contesting shots and indicate the level to which a defender must contest a shot for the attempt to likely be unsuccessful.

The instance-by-instance simplicity of the data collection system meant that the sequential nature of the match event was lost, therefore an option for further research in this area would be to preserve the sequential data so that individual actions could be reviewed. This would lead to a better understanding of the sequences of possessions that were either considered successful or unsuccessful for the defense. The author noted in the discussion section that contested shots close to the basket were often recovered by the offense because of the defenders inability to quickly transition to boxing-out (Krause et al., 2008). However, because of the simplicity of the system, there is no data to support this claim so

53

it remains objective conjecture, the exact opposite of what is needed from performance analysis (Hodges and Franks, 2008). Computer-based notation and feedback would improve the reliability of the system and provide an important learning tool for coaches (Liebermann and Franks, 2008). Since the usefulness of this metric is determined by the extent to which coaches trust the data and find it applicable in practice, it would be interesting understand the views of coaches on this facet of the game. Consequently, further research could include interviews and focus groups with coaches to discuss whether the Preventions metric is applicable in their eyes and whether this tool for performance analysis would affect their coaching cycle (Franks et al., 1983).

54

REFERENCES

55

REFERENCES

Akers, M., Wolff, S, and Buttross, T. (1991). An empirical examination of the factors affecting the success of NCAA Division I College Basketball teams. Journal of Business and Economic Studies, 1(2), 57-70.

Álvarez, A., Ortega, E., Gómez, M. Á. and Salado, J. (2009). Study of the defensive performance indicators in peak performance basketball. Revista de Psicología del Deporte, 18(3), 379-384.

American Sport Education Program (ASEP) (2007). Coaching basketball: technical and tactical skills. Leeds: Human Kinetics.

American Sport Education Program (ASEP) (2012). Coaching youth basketball (5th Ed.). Leeds: Human Kinetics.

Basketball Reference. (2015a). 1946-47 BAA Season Summary [online]. Available at: http://www.basketball-reference.com/leagues/BAA_1947.html#team::none [Accessed on: 04/03/14].

Basketball Reference. (2015b). 2013-14 NBA Stats: Advanced [online]. Available at: http://www.basketball-reference.com/leagues/NBA_2014_advanced.html [Accessed on: 04/03/14].

Basketball Reference. (2015c). Four Factors [online]. Available at: http://www.basketball- reference.com/about/factors.html [Accessed on: 04/03/14].

Basketball Reference. (2015d). Glossary [online]. Available at: http://www.basketball- reference.com/about/glossary.html [Accessed on: 04/03/14].

Basketball Reference. (2015e). NBA & ABA League Index [online]. Available at: http://www.basketball-reference.com/leagues/ [Accessed on: 04/03/14].

Bazanov, B., Haljand, R., and Võhandu, P. (2005). Offensive teamwork intensity as a factor influencing the result in basketball. International Journal of Performance Analysis in Sport, 5(2), 9-16.

Bazanov, B., Võhandu, P., and Haljand, R. (2006). Factors influencing the teamwork intensity in basketball. International Journal of Performance Analysis in Sport, 6(2), 88-96.

56

Csataljay, G., Hughes, M., James, N. and Dancs, H. (2011). Pace as an influencing factor in basketball, In Research Methods and Performance Analysis edited by M. Hughes, H. Dancs, K. Nagyváradi, T. Polgár, N. James, G. Sporiš and G. Vuckovic, pp.181-190, Hungary: University of West Hungary.

Csataljay, G., James, N., Hughes, M., and Dancs, H. (2012). Performances differences between winning and losing basketball teams during close, balanced and unbalanced quartiers, Journal of human sport and exercise, 7(2), 356-364.

Csataljay, G., James, N., Hughes, M., and Dancs, H. (2013). Effects of defensive pressure on basketball shooting performance. International Journal of Performance Analysis in Sport, 13(3), 594-601.

Csataljay, G., O'Donoghue, P., Hughes, M., and Dancs, H. (2009). Performance indicators that distinguish winning and losing teams in basketball. International Journal of Performance Analysis in Sport, 9(1), 60-66.

Dancs, H., Hughes, M. and O’Donoghue, P. (2006). Performance Analysis of Sport 7: 23rd–26th Aug 2006. Cardiff: CPA Press, UWIC.

Daskalovski, B., Naumovski, M., and Kocic, M. (2014). Multivariate and univariate differences in offense-defense variables among the best basketball leagues. Research in Physical Education, Sport and Health, 3(1), 103-107.

Doerr, K., and Gue, K. (2013). A Performance Metric and Goal‐Setting Procedure for Deadline‐Oriented Processes. Production and Operations Management, 22(3), 726-738.

ESPN. (2015). Hollinger’s NBA Team Stats [online]. Available at: http://espn.go.com/nba/hollinger/teamstats [Accessed on: 04/03/14].

Evangelos, T., and Nikolaos, A. (2004). Registration of rebound possession zones in basketball. International Journal of Performance Analysis in Sport, 4(1), 34-39.

Ferreira, A., Volossovitch, A., Gomes, F. and Didier, S. (2014). Defensive strategy and critical match episodes in basketball game: analysing the teams’ success, In Performance Analysis of Sport IX edited by D. Peters and P. O’Donoghue, pp.129-135. London: Routledge.

57

FIBA Europe. (2012). Official Basketball Statisticians’ Manual 2012 [online]. Available at: http://www.fibaeurope.com/default.asp?coid={74D20120-82E3-480C-857A- 08ED59BE3539}andsubimage=anddo=download [Accessed on: 03/04/2014].

FIBA. (2005). FIBA Assist Magazine [online]. Available at: http://www.fiba.com/asp_includes/download.asp?file_id=554 [Accessed on: 06/03/15].

FIBA. (2009a). FIBA Archive: Olympic Games: Tournament for Men [online]. Available at: http://archive.fiba.com/pages/eng/fa/p/rpp//fromseason/1930/toseason/2015/q/Olympic%2 0Games%3A%20Tournament%20for%20Men/cid/WOLYM/_//events.html [Accessed on: 04/03/14].

FIBA. (2009b). FIBA Archive: Olympic Games: Tournament for Women [online]. Available at: http://archive.fiba.com/pages/eng/fa/p/rpp//fromseason/1930/toseason/2015/q/Olympic%2 0Games%3A%20Tournament%20for%20Women/cid/WOLYW/_//events.html [Accessed on: 04/03/14].

Franks, I. (2004). The need for feedback, In Notational Analysis of Sport: Systems for better coaching and performance in sport (2nd Ed.) edited by M. Hughes and I. M. Franks, pp.8-16. London: Routledge.

Franks, I. and Miller, G. (1986). Eyewitness testimony in sport. Journal of Sport Behaviour, 9, 39-45.

Franks, I. and Miller, G. (1991). Training coaches to observe and remember. Journal of Sports Sciences, 9, 285-297.

Franks, I., Goodman, D. and Miller, G. (1983). Analysis of performance: qualitative or quantitative, Sports, March.

Fylaktakidou, A., Tsamourtzis, E., and Zaggelidis, G. (2011). The Turnovers Analysis to the Women's National League Basketball Games. Sport Science Review, 20(3-4), 69-83.

García, J., Ibáñez, S. J., Gómez, M. A., and Sampaio, J. (2014). Basketball Game-related statistics discriminating ACB league teams according to game location, game outcome and final score differences. International Journal of Performance Analysis in Sport, 14(2), 443-452.

58

García, J., Ibáñez, S., Martinez De Santos, R., Leite, N., and Sampaio, J. (2013). Identifying basketball performance indicators in regular season and playoff games. Journal of human kinetics, 36(1), 161-168.

Gels, J. (2015). Basketball Plays – 3-on-3 Plays [online]. Available at: http://www.coachesclipboard.net/3on3Plays.html [Accessed on: 05/03/15].

Gomez, A., Evangelos, T., and Alberto, L. (2006). Defensive systems in basketball ball possessions. International Journal of Performance Analysis in Sport, 6(1), 98-107.

Hawkins, M. and Choi, H. (2009). Analysis of different types of turnovers between winning and losing performances in men’s NCAA basketball, In Current trends in performance analysis: World Congress of Performance Analysis of Sport VIII edited by A. Hokelmann, K. White and P. O’Donoghue, pp.105-113. Germany: Shaker Verlag.

Hay, J., and Reid, J. (1988). Anatomy, Mechanics and Human Motion. Englewood Cliffs, NJ: Prentice Hall.

Hendry, D. and Hodges, N. (2013). Getting on the right track: athlete-centred practice for expert performance in sport, In Routledge Handbook of Sports Performance Analysis edited by T. McGarry, P. O’Donoghue and J. Sampaio, pp.5-20. London: Routledge.

Hodges, N., and Franks, I. (2008). The provision of information, In The essentials of performance analysis: an introduction (2nd Ed.) edited by M. Hughes and I. Franks, pp.21- 39. London: Routledge.

Hokelmann, A., White, K. and O’Donoghue, P. (2009). Current trends in performance analysis: World Congress of Performance Analysis of Sport VIII. Germany: Shaker Verlag.

Hoopdata (2009). NBA Advanced Player Stats. Available at: http://hoopdata.com/advancedstats.aspx [Accessed on: 04/03/14].

Huciñski, T. and Tymañski, R. (2008). The observation sheet method in the assessment of the effectiveness of a sports struggle. Research yearbook, 14(2), 65-68.

Hughes, M. and Bartlett, R. (2008). What is performance analysis?, In The essentials of performance analysis: an introduction (2nd Ed.) edited by M. Hughes and I. Franks, pp.8- 20. London: Routledge.

59

Hughes, M. and Franks, I. (2008). Sports Analysis, In Notational Analysis of Sport: Systems for better coaching and performance in sport (2nd Ed.) edited by M. Hughes and I. M. Franks, pp.107-117. London: Routledge.

Hughes, M. and Franks, I. (eds.). (2004). Notational Analysis of Sport: Systems for better coaching and performance in sport (2nd Ed.). London: Routledge.

Hughes, M. and Franks, I. (eds.). (2008). The essentials of performance analysis: an introduction (2nd Ed.). London: Routledge.

Hughes, M., and Bartlett, R. (2002). The use of performance indicators in performance analysis. Journal of sports sciences, 20(10), 739-754.

Hughes, M., and Bartlett, R. (2004). The use of performance indicators in performance analysis, In Notational Analysis of Sport: Systems for better coaching and performance in sport (2nd Ed.) edited by M. Hughes and I. M. Franks, pp.168-188. London: Routledge.

Hughes, M., Dancs, H., Nagyváradi, K., Polgár, T., James, N., Sporiš, G. and Vuckovic, G. (2011). Research Methods and Performance Analysis. Hungary: University of West Hungary.

Ibáñez, S., García, J., Feu, S., Lorenzo, A., and Sampaio, J. (2009). Effects of consecutive basketball games on the game-related statistics that discriminate winner and losing teams. Journal of sports science and medicine, 8(3), 458.

Ibáñez, S., Sampaio, J., Feu, S., Lorenzo, A., Gómez, M., and Ortega, E. (2008). Basketball game-related statistics that discriminate between teams’ season-long success. European Journal of Sport Science, 8(6), 369-372.

IBM. (2015). IBM – Products – SPSS Statistics [online]. Available at: http://www- 03.ibm.com/software/products/en/spss-statistics [Accessed on 08/03/15].

Jenkins, R., Morgan, L. and O’Donoghue, P. (2007). A case study into the effectiveness of computerized match analysis and motivational videos within the coaching of a league netball team, International Journal of Performance Analysis of Sport, 7(2), 59-80.

Krause, J., Meyer, D. and Meyer, J. (2008). Basketball Skills and Drills (3rd ed.). Champaign, IL: Human Kinetics.

60

Kubatko, J., Oliver, D., Pelton, K., and Rosenbaum, D. (2007). A starting point for analyzing basketball statistics. Journal of Quantitative Analysis in Sports, 3(3), 1-22.

Laird, P. and Waters, L. (2008). Eye-witness recollection of sports coaches. International Journal of Performance Analysis of Sport, 8(1), 76-84.

Liebermann, D and Franks, I. (2008). Video feedback and information technologies, In The essentials of performance analysis: an introduction (2nd Ed.) edited by M. Hughes and I. Franks, pp.40-50. London: Routledge.

Magill, R. and Anderson, R. (2012). The roles and uses of augmented feedback in motor skill acquisition, In Skill acquisition in sport: Research, theory and practice (2nd Ed.) edited by A. Williams and N. Hodges, pp.3-21. London: Routledge.

Malarranha, J., Figueira, B., Leite, N., and Sampaio, J. (2013). Dynamic Modeling of Performance in Basketball. International Journal of Performance Analysis in Sport, 13(2), 377-387.

Maslovat, D. and Franks, I. (2008). The need for feedback, In The essentials of performance analysis: an introduction (2nd Ed.) edited by M. Hughes and I. Franks, pp.1-7. London: Routledge.

McGarry, T., O’Donoghue, P. and Sampaio, J. (eds.) (2013). Routledge Handbook of Sports Performance Analysis. London: Routledge.

Moreno, E., Gomez, M., Lago, C., and Sampaio, J. (2013). Effects of starting quarter score, game location, and quality of opposition in quarter score in elite women’s basketball. Kinesiology, 45(1), 48-54.

NBA Stats. (2014a). Player Tracking – Defensive Impact [online]. Available at: http://stats.nba.com/tracking/#!/player/defense/ [Accessed on: 04/03/14].

NBA Stats. (2014b). Player Tracking – Speed and Distance [online]. Available at: http://stats.nba.com/tracking/#!/player/speed/ [Accessed on: 04/03/14].

NBA Stats. (2014c). Player Tracking – What is player tracking? [online]. Available at: http://stats.nba.com/tracking/#!/player/ [Accessed on: 04/03/14].

Newell, K. (1986). Constraints on the development of coordination. Motor development in children: Aspects of coordination and control, 34, 341-360. 61

O’Donoghue, P. (2010). Research Methods for Sports Performance Analysis. Oxon, UK: Routledge.

O’Donoghue, P. (2015). An introduction to performance analysis of sport. London: Routledge.

Oliver, D. (2004). Basketball on paper: rules and tools for performance analysis. Potomac Books, Inc.

Olivier, D. (1997). Journal of Basketball Studies - They say defense wins championships [online]. Available at: http://www.rawbw.com/~deano/ [Accessed on: 12/03/15].

Peters, D. and O’Donoghue, P. (2014). Performance Analysis of Sport IX. London: Routledge.

Pojskić, H., Šeparović, V., and Užičanin, E. (2009). Differences between successful and unsuccessful basketball teams on the final Olympic tournament. Acta Kinesiologica, 3(2), 110-114.

Reimer, A. (2005). FIBA vs North American rules comparison. FIBA Assist Magazine, 14, 40-44.

Remmert, H. (2003). Analysis of group-tactical offensive behavior in elite basketball on the basis of a process orientated model. European Journal of Sport Science, 3(3), 1-12.

Ribas, R. L., Navarro, R., Tavares, F., and Gómez, M. Á. (2011). Analysis of number of players involved in rebound situations in Euroleague basketball games. Open Sports Sciences Journal, 4, 10-13.

Rodriguez, A. (2013a). 2013-14 Clippers Preventions Data [online]. Available at: http://www.wammyradio.com/personal/clippers/lacpre.xlsx [Accessed on: 20/03/14].

Rodriguez, A. (2013b). LAC Preventions: Readme [online]. Available at: https://docs.google.com/document/d/1pjuxsoKPF8FiuvpkAGVyNepW_iJYHjpYzKa5BluTo NY/edit [Accessed on: 20/03/14].

Rodriguez, A. (2013c). Preventions: The Math of Defense [online]. Available at: http://www.wammyradio.wordpress.com/2013/09/28/preventions-the-math-of- defense/#more-46 [Accessed on: 20/03/14].

62

Rodriguez, A. (2013d). Prevention Thoughts: LAC 115, SAS 92. Available at: https://wammygiveaway.wordpress.com/2013/12/18/lacgame26/#more-346 [Accessed on: 07/03/14].

Rodriguez, A. (2015a). Preventions - The Math of Defense [online]. Available at: http://preventions.wammyradio.com/ [Accessed on: 10/03/15].

Rodriguez, A. (2015b). Preventions - The Math of Defense: Glossary [online]. Available at: http://preventions.wammyradio.com/glossary [Accessed on: 10/03/15].

Rose, L. (2004). The basketball handbook. Leeds: Human Kinetics.

Sampaio, J., and Janeira, M. (2003). Statistical analyses of basketball team performance: understanding teams' wins and losses according to a different index of ball possessions. International Journal of Performance Analysis in Sport, 3(1), 40-49.

Sampaio, J., Ibáñez, S., and Lorenzo, L. (2013). Basketball, In Routledge Handbook of Sports Performance Analysis edited by T. McGarry, P. O’Donoghue and J. Sampaio, pp.357-367. London: Routledge.

Sampaio, J., Lago, C., Casais, L., and Leite, N. (2010). Effects of starting score-line, game location, and quality of opposition in basketball quarter score. European Journal of Sport Science, 10(6), 391-396.

Sporiš, G., Šango, J., Vučetić, V. and Mašina, T. (2006). The latent structure of standard game efficiency indicators in basketball. International Journal of Performance Analysis in Sport, 6(1), 120-129.

Stanković, A. (2013). Analysis of collective defense in finals of 2012 Radivoj Korac Cup. Fizička kultura, 67(1), 68-77.

Swalgin, K. (2008). Building a Better Basketball ‘Box’ Score: How Division I Coaches Define and Value Field Goal Shooting and Taking-Charges [online]. Available at: http://grfx.cstv.com/photos/schools/nabc/genrel/auto_pdf/building-better-box-score.pdf [Accessed on: 14/03/15].

Synergy Sports Technology. (2013). Basketball Products – NBA Stats Software [online]. Available at: http://corp.synergysportstech.com/basketball-products-nba [Accessed on: 04/03/14].

63

Tavares, F., and Gomes, N. (2003). The offensive process in basketball–a study in high performance junior teams. International Journal of Performance Analysis in Sport, 3(1), 34- 39.

Trninić, S., Dizdar, D., and Lukšić, E. (2002). Differences between winning and defeated top quality basketball teams in final tournaments of European club championship. Collegium Antropologicum, 26(2), 521-531.

Tsamourtzis, E., Karypidis, A. and Athanasiou, N. (2005). Analysis of fast breaks in basketball. International Journal of Performance Analysis in Sport, 5(2), 17-22.

Williams, A. and Hodges, N. (2004). Skill acquisition in sport: Research, theory and practice. London: Routledge.

Williams, A. and Hodges, N. (2012). Skill acquisition in sport: Research, theory and practice (2nd Ed.). London: Routledge.

Wissel, H. (2012). Basketball: steps to success (3rd Ed.). Leeds: Human Kinetics.

Wootten, W. and Wootten, J. (2013). Coaching basketball successfully (3rd Ed.). Leeds: Human Kinetics.

Wulf, G. and Shea, C. (2004). Understanding the role of augmented feedback: The good, the bad, and the ugly, In Skill acquisition in sport: Research, theory and practice edited by A. Williams and N. Hodges, pp.121-144. London: Routledge.

64

APPENDICES

65

APPENDIX A

PREVENTIONS METRIC VARIABLES

Table A1. Preventions metric action variables and definitions

(Adapted from Rodriguez, 2013b)

Variable Definition

Tallies the number of shot attempts the defense successfully stopped. Shots can be denied by directly challenging the shot Shots Group man-to-man, blocking the shot, stealing the ball during a shot (SHOT) attempt, taking a charge, or forcing a turnover while a shot is being attempted. Successfully prevented shots are converted to points based on the value of the shot.

Block (BLK) Challenged by a block. All blocks are final.

Challenged by a steal. Usually occurs near the basket, but points Steal (STL) lost on a steal are very rare.

Offensive Foul Challenged by taking a charge. Scorer must be attempting a (OFF) basket.

Turnover Challenged by forcing a turnover. Defender(s) must be within (TOV) range of the scorer.

Counts the number of possessions that were lost or forfeited to the defensive team. Possessions can be prevented defensively by stealing the ball, drawing an offensive foul, or forcing a Possessions turnover via basketball error or a time-based violation. They can Group (POSS) also be prevented offensively by collecting an offensive rebound, tipping the ball to an offensive teammate, or saving the ball from going out-of-bounds or following a jump ball.

A - 1

Lost on a steal. Steals can be discredited if defender loses the Steal (STL) ball within three seconds of the steal.

Offensive Foul Forfeited on an offensive foul. Defender must be within range of (OFF) the primary ball handler.

Turnover Forfeited on a turnover. Defender(s) must be within range of the (TOV) primary ball handler.

Shot Clock Forfeited on time-based violations. Defender(s) must be within (CLK) range of the primary ball handler.

Offensive Preserved by an offensive rebound. Known as extra Rebound possessions. Occurs only in offense mode. (ORB)

Preserved by a tip back. Known as extra possessions. Occurs Tip Save (TIP) only in offense mode.

Loose-ball Preserved by saving the ball following a jump ball, or from an out- Save (SAV) of-bounds call following an errant execution.

Counts the number of assists the defense successfully Passes Group interrupted or botched by drawing a steal or forcing the offense to (PASS) commit a bad pass.

Steal (STL) Intercepted by a steal.

Turnover Thrown away by a turnover. Known as bad pass. Defender(s) (TOV) must be within range of the primary ball handler.

Turnovers Actions in offense mode that prevent a team from turning the ball Prevented over. (TOVS)

Number of picks that were set. Occurs only in offense mode. Screen (SCR) Picks prevent turnovers. The basket must be scored on the same possession that the pick was set to receive credit.

A - 2

Hockey Assist Number of hockey assists (pass to an assist) that were executed. (HAS) Occurs only in offense mode. Hockey assists prevent turnovers.

Miscellaneous Miscellaneous variables not part of the shots, possessions, Group (MISC) passes or turnovers groups that also contribute to Preventions.

Number of times a defender helped another defender to Help (HLP) complete a defensive assignment.

Number of times a defender knocked the ball out of bounds or Delay (DLY) drew a jump ball.

A - 3

APPENDIX B

PREVENTIONS METRIC FORMULAE

Table B1. Preventions metric statistics formulae

Variable Formula

Shots Group Value Total:

Points Prevented (PtsP) (2 * (2CON + 2BLK + 2STL + 2OFF + 2TOV)) + (3 * (3CON + 3BLK + 3STL + 3OFF + 3TOV))

Shots Group Total:

Shots Challenged (SC) (2CON + 2BLK + 2STL + 2OFF + 2TOV + 3CON + 3BLK + 3STL + 3OFF + 3TOV)

Passes Group Total: Botched Passes (BP) (STL + TOV)

Turnovers Group Total: Possessions Protected (PossP) (SCR + HAS)

Help Defense: Team Defense (TD) (HLP)

3PT Shots Group Total:

Three-pointers Prevented (3PP) (3CON + 3BLK + 3STL + 3OFF + 3TOV)

B - 1

Delays: Disruption Value (DV) (DLY)

B - 2

APPENDIX C

HAND NOTATION PROFORMA (V1)

Figure C1. Hand Notation Proforma (version 1)

C - 1

APPENDIX D

HAND NOTATION PROFORMA (V2)

Figure D1. Hand Notation Proforma (version 2)

D - 1

APPENDIX E

UPDATED PREVENTIONS METRIC DEFINITIONS

Table E1. Updates made to preventions metric action variable definitions

(Adapted from Rodriguez, 2013b)

Variable Definition

Challenged by man-to-man defense. Defender(s) must be within range of the scorer.

Challenged by man-to-man or closely guarded match-up within Shot Contest zone defense. Defenders must contest shot with at least one arm (CON) up towards the shooter and within a proximity of 3 feet.

Shot contests within the restricted area must include some legal physical contact (verticality principle).

Forfeited on a turnover. Defender(s) must be within range of the Possession primary ball handler. Turnover (TOV) Specific match event examples – travelling, 5-second violation, 3- second violation, etc.

Preserved by an offensive rebound. Known as extra Offensive possessions. Occurs only in offense mode. Rebound (ORB) Specific match event examples – award ORB for tip-in attempt and for team rebound when loose-ball is tapped out of bounds.

E - 1

Number of picks that were set. Occurs only in offense mode. Picks prevent turnovers. The basket must be scored on the same

Screen (SCR) possession that the pick was set to receive credit.

Specific match event examples – award for both on-ball and off- ball screening actions.

Number of hockey assists (pass to an assist) that were executed.

Hockey Assist Occurs only in offense mode. Hockey assists prevent turnovers. (HAS) Specific match event examples – award if an assist is created through quick ball movement.

Number of times a defender helped another defender to complete a defensive assignment. Help (HLP) Award when defenders have to leave their match-up to defend the ball on a shot attempt.

Number of times a defender knocked the ball out of bounds or drew a jump ball. Delay (DLY) Specific match event examples – deflecting a pass out-of- bounds.

E - 2

APPENDIX F

FUTHER UPDATES TO PREVENTIONS METRIC

Table F1. Further updates made to preventions metric action variable definitions.

(Adapted from Rodriguez, 2013b)

Variable Definition

Number of picks that were set. Occurs only in offense mode. Picks prevent turnovers. The basket must be scored on the same possession that the pick was set to receive credit.

Screen (SCR) Specific match event examples – award for on-ball screen action where ball-handler or screener score, award for off-ball screen

action where cutter or screener score, do not award for screen action causing defensive rotations that allow scores off extended ball movement.

Number of hockey assists (pass to an assist) that were executed. Occurs only in offense mode. Hockey assists prevent turnovers. Hockey Assist (HAS) Specific match event examples – only award if the second-to-last pass directly created the chance for the assist to occur via quick ball movement.

F - 1

Number of times a defender helped another defender to complete a defensive assignment.

In man-to-man, award when help defense rotates from another match-up to contest a missed shot. In zone defense, award when Help (HLP) the zone collapses and more than one player defend a shot, forcing a miss.

Specific match event examples – only award if the help defender’s contest is the action that causes the miss, or contributes to the missed shot.

Number of times a defender knocked the ball out of bounds or drew a jump ball.

Delay (DLY) Specific match event examples – deflecting a pass out-of- bounds, forcing a jump-ball when on defense, disrupting a possession with a foot violation.

F - 2

APPENDIX G

DESCRIPTIVE STATISTICS

Table G1. Medians of sample groupings

Median

Winner/Loser D.Rtg Def.Eff PtsP PL SC BP PossP Winner 78.3 77.2 49.5 21.5 21.5 8.0 8.5 Loser 98.4 106.4 43.0 19.0 19.5 5.0 6.5 Total 89.6 95.0 46.0 20.0 21.0 6.0 7.5

Median

Close/Unbalanced D.Rtg Def.Eff PtsP PL SC BP PossP Close Winner 87.8 89.8 46.0 21.0 21.0 6.0 8.0 Loser 94.9 97.0 46.0 19.0 20.0 5.0 7.0 Total 89.6 95.0 46.0 20.5 20.5 5.0 7.0 Unbalanced Winner 73.7 72.0 53.0 23.0 24.0 11.0 9.0 Loser 107.0 107.7 38.0 19.0 18.0 6.0 6.0 Total 83.2 86.4 46.5 19.5 21.5 7.5 8.0

G - 1

APPENDIX H

INFERENTIAL STATISTICS

Table H1. Mann-Whitney U Test Results

Table H1a. Relationship of Winners vs Losers

Ranks Winners vs Losers N Mean Rank Sum of Ranks D.Rtg Winner 18 11.17 201.00 Loser 18 25.83 465.00 Total 36 Def.Eff Winner 18 11.00 198.00 Loser 18 26.00 468.00 Total 36 PtsP Winner 18 22.78 410.00 Loser 18 14.22 256.00 Total 36 PL Winner 18 20.67 372.00 Loser 18 16.33 294.00 Total 36 SC Winner 18 22.11 398.00 Loser 18 14.89 268.00 Total 36 BP Winner 18 22.67 408.00 Loser 18 14.33 258.00 Total 36 PossP Winner 18 21.89 394.00 Loser 18 15.11 272.00 Total 36 Test Statisticsa

Winners vs Losers D.Rtg Def.Eff PtsP PL SC BP PossP Mann-Whitney U 30.000 27.000 85.000 123.000 97.000 87.000 101.000 Wilcoxon W 201.000 198.000 256.000 294.000 268.000 258.000 272.000 Z -4.176 -4.271 -2.439 -1.237 -2.065 -2.390 -1.941 Asymp. Sig. (2-tailed) .000**** .000**** .015* .216 .039* .017* .052

Exact Sig. [2*(1-tailed Sig.)] .000b .000b .014b .226b .040b .017b .055b a. Grouping Variable: Winner/Loser b. Not corrected for ties. H - 1

Table H1b. Relationship of Close Winners vs Unbalanced Winners Ranks

Close Winners vs Unbalanced Winners N Mean Rank Sum of Ranks Winners D.Rtg Close 9 14.00 126.00

Unbalanced 9 5.00 45.00

Total 18

Def.Eff Close 9 14.00 126.00

Unbalanced 9 5.00 45.00

Total 18

PtsP Close 9 7.67 69.00

Unbalanced 9 11.33 102.00

Total 18

PL Close 9 8.33 75.00

Unbalanced 9 10.67 96.00

Total 18

SC Close 9 7.72 69.50

Unbalanced 9 11.28 101.50

Total 18

BP Close 9 6.78 61.00

Unbalanced 9 12.22 110.00

Total 18

PossP Close 9 8.22 74.00

Unbalanced 9 10.78 97.00

Total 18

Test Statisticsa

Close Winners vs Unbalanced Winners D.Rtg Def.Eff PtsP PL SC BP PossP Winner Mann-Whitney U 0.000 0.000 24.000 30.000 24.500 16.000 29.000

Wilcoxon W 45.000 45.000 69.000 75.000 69.500 61.000 74.000

Z -3.576 -3.576 -1.458 -.929 -1.423 -2.177 -1.025

Asymp. Sig. (2-tailed) .000**** .000**** .145 .353 .155 .029* .306

Exact Sig. [2*(1-tailed Sig.)] .000b .000b .161b .387b .161b .031b .340b

H - 2

Table H1c. Relationship of Close Losers vs Unbalanced Losers Ranks

Close Losers vs Unbalanced Losers N Mean Rank Sum of Ranks Loser D.Rtg Close 9 6.44 58.00 Unbalanced 9 12.56 113.00 Total 18 Def.Eff Close 9 6.44 58.00 Unbalanced 9 12.56 113.00 Total 18 PtsP Close 9 11.06 99.50 Unbalanced 9 7.94 71.50 Total 18 PL Close 9 10.11 91.00 Unbalanced 9 8.89 80.00 Total 18 SC Close 9 10.67 96.00 Unbalanced 9 8.33 75.00 Total 18 BP Close 9 8.11 73.00 Unbalanced 9 10.89 98.00 Total 18 PossP Close 9 9.83 88.50 Unbalanced 9 9.17 82.50 Total 18

Test Statisticsa

Close Losers vs Unbalanced Losers D.Rtg Def.Eff PtsP PL SC BP PossP Loser Mann-Whitney U 13.000 13.000 26.500 35.000 30.000 28.000 37.500

Wilcoxon W 58.000 58.000 71.500 80.000 75.000 73.000 82.500

Z -2.428 -2.428 -1.238 -.487 -.935 -1.129 -.268

Asymp. Sig. (2-tailed) .015* .015* .216 .626 .350 .259 .788

Exact Sig. [2*(1-tailed Sig.)] .014b .014b .222b .666b .387b .297b .796b a. Grouping Variable: Close/Unbalanced b. Not corrected for ties.

H - 3

Table H1d. Relationship of Close Winners vs Close Losers Ranks

Close Winners vs Close Losers N Mean Rank Sum of Ranks Close D.Rtg Winners 9 7.78 70.00 Losers 9 11.22 101.00 Total 18 Def.Eff Winners 9 7.67 69.00 Losers 9 11.33 102.00 Total 18 PtsP Winners 9 10.11 91.00 Losers 9 8.89 80.00 Total 18 PL Winners 9 9.44 85.00 Losers 9 9.56 86.00 Total 18 SC Winners 9 9.94 89.50 Losers 9 9.06 81.50 Total 18 BP Winners 9 10.44 94.00 Losers 9 8.56 77.00 Total 18 PossP Winners 9 10.83 97.50 Losers 9 8.17 73.50 Total 18

Test Statisticsa

Close Winners vs Close Losers D.Rtg Def.Eff PtsP PL SC BP PossP Close Mann-Whitney U 25.000 24.000 35.000 40.000 36.500 32.000 28.500

Wilcoxon W 70.000 69.000 80.000 85.000 81.500 77.000 73.500

Z -1.369 -1.457 -.488 -.044 -.357 -.758 -1.071

Asymp. Sig. (2-tailed) .171 .145 .625 .965 .721 .449 .284

Exact Sig. [2*(1-tailed Sig.)] .190b .161b .666b 1.000b .730b .489b .297b a. Grouping Variable: Winner/Loser b. Not corrected for ties.

H - 4

Table H1e. Relationship of Unbalanced Winners vs Unbalanced Losers Ranks

Unbalanced Winners vs Unbalanced Losers N Mean Rank Sum of Ranks Unbalanced D.Rtg Winners 9 5.00 45.00

Losers 9 14.00 126.00

Total 18

Def.Eff Winners 9 5.00 45.00

Losers 9 14.00 126.00

Total 18

PtsP Winners 9 12.44 112.00

Losers 9 6.56 59.00

Total 18

PL Winners 9 11.39 102.50

Losers 9 7.61 68.50

Total 18

SC Winners 9 12.00 108.00

Losers 9 7.00 63.00

Total 18

BP Winners 9 13.11 118.00

Losers 9 5.89 53.00

Total 18

PossP Winners 9 11.67 105.00

Losers 9 7.33 66.00

Total 18

Test Statisticsa

Unbalanced Winners vs Unbalanced Losers D.Rtg Def.Eff PtsP PL SC BP PossP Unbalanced Mann-Whitney U 0.000 0.000 14.000 23.500 18.000 8.000 21.000

Wilcoxon W 45.000 45.000 59.000 68.500 63.000 53.000 66.000

Z -3.576 -3.576 -2.344 -1.508 -1.994 -2.892 -1.731

Asymp. Sig. (2-tailed) .000**** .000**** .019* .132 .046* .004*** .083

Exact Sig. [2*(1-tailed Sig.)] .000b .000b .019b .136b .050b .003b .094b

H - 5