1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36
The use of team performance indicator characteristics to explain ladder position at the conclusion of the Australian Football League home and away season Carl T. Woodsa* a
Discipline of Sport and Exercise Science, James Cook University, Queensland, Australia *Corresponding Author Carl Woods, Discipline of Sport and Exercise Science, James Cook University, Townsville, Queensland, Australia. Ph: +61 08 4781 6550 Mob: +61 421254329 Email:
[email protected] Abstract Not dissimilar to other elite team sports, the Australian Football League (AFL) implements a ‘ladder’ system to rank the performance of teams relative to their opposition. Position on this ladder is primarily operationalised by the number of wins a team accrues. This study investigated the extent to which an AFL team’s ladder position was associated with their performance indicator characteristics. Thirteen team performance indicators were collated for each AFL team (n = 18), with teams being categorised according to their ladder position (ranked one to 18) at the conclusion of the 2015, 23-round, home and away season (394 observations). A higher ranked position (closer to one) was reflective of a higher performing team. Cumulative link mixed models were fitted to the data, modelling the extent to which ladder position was associated with the team performance indicators. Ladder position was significantly negatively associated with ‘hit-outs’, ‘clearances’, and ‘inside 50’s’, respectively. Comparatively, the remaining team performance indicators were unable to significantly explain ladder position. These results show that teams ranked higher on the AFL ladder (closer to one) possess distinctive performance indicator characteristics. This data may be of use to AFL coaches and performance analysts when developing game-plans and training drill designs. Keywords: performance analysis, team sport, ordinal regression, violin plots
1
37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
The use of team performance indicator characteristics to explain ladder position at the conclusion of the Australian Football League home and away season Abstract Not dissimilar to other elite team sports, the Australian Football League (AFL) implements a ‘ladder’ system to rank the performance of teams relative to their opposition. Position on this ladder is primarily operationalised by the number of wins a team accrues. This study investigated the extent to which an AFL team’s ladder position was associated with their performance indicator characteristics. Thirteen team performance indicators were collated for each AFL team (n = 18), with teams being categorised according to their ladder position (ranked one to 18) at the conclusion of the 2015, 23-round, home and away season (394 observations). A higher ranked position (closer to one) was reflective of a higher performing team. Cumulative link mixed models were fitted to the data, modelling the extent to which ladder position was associated with the team performance indicators. Ladder position was significantly negatively associated with ‘hit-outs’, ‘clearances’, and ‘inside 50’s’, respectively. Comparatively, the remaining team performance indicators were unable to significantly explain ladder position. These results show that teams ranked higher on the AFL ladder (closer to one) possess distinctive performance indicator characteristics. This data may be of use to AFL coaches and performance analysts when developing game-plans and training drill designs. Keywords: performance analysis, team sport, ordinal regression, violin plots
2
64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110
1. Introduction Australian football (AF) is a dynamic team invasion sport that requires players at all stages of development to integrate a range of multidimensional performance qualities (Woods et al., 2016a). The game is played between two teams who field no more than 18 players at a time. A ‘goal’ (six points) is scored when a player kicks the ball through two large goalposts positioned at either end of the field. If the ball misses the two large goalposts, but passes through a smaller goalpost located either side of the large goalposts, a single point is scored. Thus, the winning side is the one who has accumulated the highest number of points at the conclusion of the 120 minute game. Given that the field dimensions range between 130-150 m by 150-190 m (Veale & Pearce, 2009), goals are often the product of a sequence of physical and technical player actions performed at various locations on the field (Sullivan et al., 2014). The premier AF competition is the Australian Football League (AFL), which currently consists of 18 teams competing within a 23-week season. This 23-week season is colloquially termed the ‘home and away season’, and its global purpose is to rank teams on a ladder (one being the highest ranked, 18 being the lowest ranked). At the conclusion of the home and away season, the eight highest ranked teams compete in a four-week, knock-out, final series where they contest for the opportunity to participate within the AFL Grand Final. This team ranking system is primarily operationalised by the number of wins a team accrues throughout the season. Specifically, teams are awarded four points for a win, with their seasonal cumulated score being used to rank them relative to the scores obtained by their opposition. However, the magnitude of a win, reflected as a percentage of the points scored relative to those scored against, contributes to a team’s ladder position when their cumulated points are the same as another team. Thus, the fundamental goal for each AFL team during the home and away season is to win as many games as possible in order to rank as high as possible (i.e., closest to one) on the AFL ladder. Given the importance of accruing a high number of wins within the home and away season, work has looked to explain the relationship between match outcome (win/loss) and team performance indicator characteristics within the AFL. Most recently, Robertson et al. (2016) used both logistic regression and a chi-squared automatic interaction detection (CHAID) analyses to elucidate the team performance indicators most explanatory of match outcome in the AFL. From a suite of team performance indicators, ‘kicks’ and ‘goal conversion’ were shown to be the most explanatory of match outcome. However, the CHAID analysis also resolved ‘contested possessions’, ‘contested marks’, ‘handballs’ and ‘inside 50’s’ as being influential in the explanation of match outcome. Further, Stewart et al. (2007) modelled both team and player performance indicators on the scoring margin in the AFL using multiple regression. Similar to Robertson et al. (2016), Stewart et al. (2007) noted that ‘inside 50’s’ and ‘kicks’ displayed the greatest (positive) association with score margin. Combined, these studies demonstrate that there are distinctive differences between successful (winning) and unsuccessful (losing) AFL teams, which are manifested via their team performance indicator characteristics.
3
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147
Indeed, work in other sports that adopt ranking systems to categorise performance has investigated the relationship between performance indicator characteristics and subsequent rankings. For example, Clarke (2011) ranked tennis players relative to one another using their performance in doubles tournaments. However, to date, work is yet to investigate the relationship between ladder position and a team’s performance indicator characteristics in the AFL. This study aimed to examine the extent with which team performance indicators could be used to explain ladder position at the conclusion of the AFL home and away season. It was hypothesised that AFL teams ranked higher on the ladder (closer to one; indicative of a greater number of wins) would possess distinctive performance indicator characteristics relative to the teams ranked lower on the ladder (closer to 18; indicative of a greater number of losses) given the work of Robertson et al. (2016) and Stewart et al. (2007). The subsequent results of this work are likely to assist with game and training tactics as well as player recruitment practices in the AFL by providing objective data illustrative of collective player actions most associated with high and low ladder positions. 2. Methods 2.1 Data Team performance indicators from the 2015 AFL home and away season were extracted from a commercially accessible source (www.afl.com.au/stats, accessed 2nd April, 2016); Champion Data Pty Ltd (Southbank, Australia), and placed into a custom designed Microsoft Excel spreadsheet (Microsoft, Redmond, USA) for analysis. These performance indicators were chosen owing to their use elsewhere (Robertson et al., 2016; Woods et al., 2016b), enabling the practical translation and relativity of the findings. This dataset consisted of team performance indicators from 23-rounds, where nine games were played per round, with the exception of rounds 11, 12, and 13 where six games were played per round due to each team having one bye during this period. Within the 2015 season, one match was abandoned and was subsequently excluded from the sample. Ethical declaration was granted by the James Cook University Human Research Ethics Committee prior to data extraction. The suite of team performance indicators used are presented in Table 1. Table 1. The team performance indicators and corresponding description as used within this study Performance indicator Description Kicks Disposing of the ball with any part of the leg below the knee including kicks off the ground Handballs Disposing of the ball by striking it with a fist while it rests on the opposing hand Contested possessions Possessions obtained while in congested, and physically pressured situations Uncontested possessions Possessions obtained while a player is under no immediate physical pressure from the opposition Effective disposals A disposal that results in a teammate possessing the ball who was the intended target
4
Clangers Marks
Contested marks Marks inside 50 Hit-outs Clearances Tackles Inside 50 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181
An unforced turnover of ball possession stemming from a disposal When a player catches a kicked ball that has travelled more than 15 metres without another player impeding the ball or it having hit the ground A mark recorded while engaging in a congested, physically pressured situation A mark recorded while a player is in their forward 50 m zone An action of clearing the ball from a ruck contest to a teammate by tapping the ball into space Disposing of the ball from a congested stoppage in play Using physical contact to prevent an opposition in possession of the ball from getting an effective disposal An action of moving the ball into the forward 50 m zone
The data for each team was sorted according to their ladder position at the conclusion of the season; ranked one to 18. Each team contributed 22 observations, with the exception of the two teams who were involved in the abandoned matched, subsequently contributing 21 observations each. 2.2 Statistical Analysis To test the study hypothesis, a set of single-term models were built, with each team performance indicator being coded as a predictor variable and a team’s ladder position being coded as the response variable. Given the categorical nature of the response variable, cumulative link mixed models were built using the ‘ordinal’ package (Christensen, 2015) in the R computing environment (version 3.1.3 R Core Team, 2015). Cumulative link mixed models were chosen as they are a form of ordered regression used when the response variable is categorical, possessing some type of order or sequence. The confidence intervals of the model parameter estimates were calculated using the confint function, with ‘P values’ being estimated using Wald’s method. Prior to this main analysis however, a correlation matrix was constructed to assess the level of collinearity between the predictor variables (team performance indicators). Following this modelling, the data properties of performance indicators identified as being significantly explanatory of ladder position were visualised using violin plots. These plots show the underlying distribution of the data by illustrating the probability density distributions. Thus, violin plots provide a comprehensive visualisation of data with reference to skewness and modality when compared to other forms of data visualisation (Spitzer et al., 2014). These plots were built using the ‘ggplot2’ package in the aforementioned computing environment. 3. Results Collinearity was evident between uncontested possessions, and kicks, handballs, marks and effective disposals (r >0.5), thus the former was removed from further analyses. From the remaining 12 team performance indicators modelled, three were shown to be significantly explanatory of ladder position (Table 2). Specifically, a significant negative association was noted between ‘hit-outs’ and ladder position (β (SE) = -0.034
5
182 183 184 185 186 187 188 189 190 191 192 193
194 195 196 197
(0.005); 95% CI = -0.045 – -0.023), with the count of this indicator declining as the ladder
position increased (moved further away from one) (Figure 1). A significant negative association was noted between ‘clearances’ and ladder position (β (SE) = -0.032 (0.011); 95% CI = -0.054 – -0.011), with the total number of clearances declining as the ladder position increased (Figure 1). Finally, ‘inside 50’s’ was significantly negatively associated with ladder position (β (SE) = -0.020 (0.008); 95% CI = -0.036 – -0.004), with the count of this indicator decreasing as ladder position increased (Figure 1). The remaining team performance indicators were not significantly associated with ladder position (Table 2). Table 2. Parameter estimates of the cumulative link mixed models fitted to ladder position Performance indicator Estimate SE LCI UCI P Kicks -0.007 0.013 -0.034 0.019 0.586 Handballs 0.005 0.016 -0.025 0.036 0.723 Contested possessions -0.003 0.014 -0.031 0.024 0.808 Effective disposals -0.046 0.026 -0.098 0.005 0.080 Clangers 0.015 0.008 -0.001 0.030 0.060 Marks 0.003 0.008 -0.013 0.020 0.661 Contested marks 0.001 0.018 -0.034 0.036 0.945 Marks inside 50 0.007 0.016 -0.025 0.040 0.660 Hit-outs* -0.034 0.005 -0.045 -0.023
