Ball possession effectiveness in men’s and women’s elite basketball
according to situational variables in different game periods
MIGUEL-ANGEL GÓMEZ
1
, ALBERTO LORENZO
1
, SERGIO-JOSÉ IBAÑEZ
2
,&
JAIME SAMPAIO
3
1
Faculty of Physical Activity and Sport Sciences, Polytechnic University of Madrid, Madrid, Spain,
2
University of
Extremadura, Faculty of Sport Sciences, Av. Universidad s/n, Caceres, 10171 Spain, and
3
Universidade de Trás-Os-Montes e
Alto Douro, Sport Sciences Department, Quinta de Prados, ap 202, Vila Real, 5000 Portugal
(Accepted 22 March 2013)
Abstract
The aim of the present study was to identify the importance of basketball performance indicators in predicting the
effectiveness of ball possessions in men’s and women’s basketball, when controlling for situational variables and game
periods. The sample consisted of 7234 ball possessions, corresponding to 40 games from the Spanish professional leagues.
The effects of the predictor variables on successful ball possessions according to game period were analysed using binary
logistic regressions. Results from men’s teams show interactions with number of passes and ending player during the first
five minutes, with starting and ending zone, defensive systems, screens used and possession duration during the middle
thirty minutes, and there were interactions with passes used, possession duration and players involved during the last five
minutes. Results from women’s teams show interactions with starting and ending zone, passes used, defensive systems and
ending player during the first five minutes, and with starting and ending zone, and screens used during the middle thirty
minutes. The results show no interaction with situational variables in men’s basketball, while league stage was important
during the middle thirty minutes and last five minutes in women’s basketball, whereas match status was only important
during the last five minutes.
Keywords: situational variables, team sports, performance analysis, binary regression, performance indicators
Introduction
One of the most important tasks for basketball coa-
ches is to prepare practice sessions according to
competition constraints (Hughes & Bartlett, 2002;
Sampaio, Lago, & Drinkwater, 2010). The available
research has helped to identify and describe the per-
formance indicators that allow discrimination of
teams’ performances according to game final out-
come in different situations (Gómez, Lorenzo,
Sampaio, Ibáñez, & Ortega, 2008; Ittenbach &
Esters, 1995; Karipidis, Fotinak is, Taxildaris, &
Fatouros, 2001; Kozar, Vaughn, Whitfield, Lord, &
Dye, 1994; Sampaio, Drinkwater, & Leite, 2010;
Sampaio, Lago, Casais, & Leite, 2010). For exam-
ple, the defensive rebounds and the percentage of
successful field-goals are strongly related to the out-
come of male competitions (Ibáñez, Sampaio,
Sáenz-López, Giménez, & Janeira, 2003; Sampaio
& Janeira, 2003; Trninić, Dizdar, & Lukšić, 2002).
However, the percentage of successful 3-point field-
goals and assists are the best discriminators between
women’s winning and losing teams (Gómez,
Lorenzo, Sampaio, & Ibáñez, 2006; Gómez,
Lorenzo, Ortega, Sampaio, & Ibáñez, 2009).
All these results suggest that effectiveness in col-
lective movement patterns varies according to gen-
der. In fact, Sampaio, Ibáñez, and Feu (2004) found
that men’s team performances were best discrimi-
nated from women’s teams by their higher percen-
tage of blocks, lower percentage of steals and
unsuccessful 2-point field-goals. Accordingly, João,
Leite, Mesquita, and Sampaio (2010) identified gen-
der differences in volleyball game-related statistics,
with men’s performances being associated with
terminal actions (errors of service), but women’s
performances being characterised by continuous
actions (in defence and attack). These differences
were seen as a consequence of anthropometric and
physiological differences between genders and need
to be accounted for when preparing game strategies.
Correspondence: Miguel-Angel Gómez, Faculty of Physical Activity and Sport Sciences. Polytechnic University of Madrid, Madrid, Spain. E-mail:
magor_2@yahoo.es
Journal of Sports Sciences, 2013
Vol. 31, No. 14, 1578–1587, http://dx.doi.org/10.1080/02640414.2013.792942
© 2013 Taylor & Francis
Downloaded by [UPM] at 02:57 21 December 2015
In the last few years, several studies examined the
effects of situational variables such as game location,
match status and quality of the opponent on perfor-
mance indicators (Gómez & Pollard, 2011; Lago,
2009; Lago & Martín, 2007; Marcelino, Mesquita,
& Sampaio, 2011; Sampaio, Lago, & Drinkwater,
2010; Sampaio, Lago, Casais, et al., 2010; Taylor,
Mellalieu, James, & Shearer, 2008; Tucker,
Mellalieu, James, & Taylor, 2005). In basketball,
other situational variables such as the stage of the
league m ay highlight the differences between regular
season and playoff games. In fact, different strategies
may be used when teams are pla ying for accumulat-
ing points in a regular season compared with com-
peting in a playoff series, with one team facing
immediate elimination. Therefore, this specific
game context accounts for differences due to the
importance of the game perceived in different stages
of the league (Gómez et al., 2008; Sampaio & Janeira,
2003). The game period is also a situational variable
of interest in basketball, according to research on
critical moments. Some research identified the end
of a game (last five minutes) as the most critical
moment (Bar-Eli & Tractinsky, 2000; Kozar,
Whitfield, Lord, & Mechikoff, 1993; Mechikoff,
Kozar, Lord, Whitfield, & Brandenburg, 1990;
Navarro, Lorenzo, Gómez, & Sampaio, 2009). On
the other hand, there is also recent research addres-
sing the importance of the starting (first five minutes)
periods of the game (Sampaio, Lago, & Drinkwater,
2010; Sampaio, Lago, Casais, et al., 2010).
Few studies have examined how basketball teams
use their opportunities to score (i.e., their ball pos-
session effectiveness) and none has addressed how
the situational variables may affect scoring strategies
and tactics (Gómez et al., 2010; Remmert, 2003). In
fact, players’ interactions are constantly present in
the games and may influence the differen t tactical
approaches to score or prevent the opponents from
scoring (Remmert, 2003). Available research has
identified the number of passes, the number of par-
ticipants and the possession duration of each ball
possession as relevant variables to analyse the ball
possessions (Gómez, Tsamourtzis, & Lorenzo, 2006;
Ortega, Cárdenas, Sainz de Baranda, & Palao,
2006a, 2006b). However, other authors have sug-
gested the importance of group tactical offensive
and defensive behaviours, such as screens on and
off the bal l, multiple screens and defensive systems
(Gómez et al., 2010; Mexas, Tsiskaris, Kyriakou, &
Garefis, 2005; Mikes, 1987; Remmert, 2003).
Although there is interest in identifying and
describing these performance indicators and their
effects on ball possession effectiveness, available
research is still very scarce when all these dimensions
are addressed simultaneously. Therefore, the aim of
the present study was to identify the importance of
performance indicators in predicting the effective-
ness of ball possessions in men’s and women’s bas-
ketball, when controlling for situational variables
(league stage, game location and match status) and
game periods (first five minutes, middle thirty min-
utes and last five minutes). For each gender, we
hypothesised that playing tacti cs are influenced by
the situational variables in each game period.
Method
Sample and variables
The sample consisted of 7234 ball possessions
(men’s teams = 3523; women’s teams = 3711), cor-
responding to 40 games (10 regular season and 10
playoff games for each gender) from the 2006–2007
Spanish men’s and women’s professional basketball
league, with mean sc ore differences of 7.1 ± 0.8 and
6.3 ± 0.7, respectively. The games were provided by
the Spanish Baske tball Federation after being ran-
domly selected from those available on the public
TV. The games ending with overtime were excluded
from the sample. Ethics approval was obtained both
from the Spanish Basketball Federation and the
Faculty of Physical Activity and Sport Sciences of
the Polytechnic University of Madrid.
The bal l possession effectiveness was transformed
into a dichotomous dependent variable: the success-
ful ball possessions (when the offensive team scored
a 2 or a 3-point field-goal, recovered a ball, secured a
rebound or received a foul, including foul shot), and
the unsuccessful ball possessions (when the offensive
team missed a 2 or 3-point field-goal, received a
block shot, committed a foul, made a turnover or
made any other rule violation).
The independent variables were related to zone,
task and players’ position. The zone was studied by
the possession starting and possession ending areas
of the court. Sixteen different basketball court zones
were established (Hughes & Franks, 2004), namely
zone A, B, C, D, E, F, G and zone O in the defen-
sive half and zone I, J, K, L, M, N, H and zone P in
Figure 1. Basketball court zones used in relation to playing tactics
(adapted from Hughes & Franks, 2004).
Ball possession effectiveness in elite basketball 1579
Downloaded by [UPM] at 02:57 21 December 2015
the offensive half of the playing court (Figure 1). The
players’ positions on court were defined as guards,
forwards or centres (Ackland, Schreiner, & Kerr,
1997; Carter, Ackland, Kerr, & Stappf, 2005;
Trninić, Dizdar, & Dežman, 2000).
The task-related variables included (i) the number
of passes used by each team during the ball posses-
sion (one, two, three, four, or five or more passes
used); ii) the number of players involved in the ball
possession (one, two, three, fou r or five partici-
pants); iii) the defensive systems used by the defen-
sive team (man-to-man, zone, press and combined
defences); iv) the bal l possession duration (0 to 4
seconds, 5 to 10 seconds, 11 to 15 seconds, 16 to 20
seconds, and 21 to 24 seconds); and v) the scree ns
used (no screen used, only screens on used – when
the screener sets a screen to the offensive player that
handles the ball, only screens off used – when the
screener sets a screen to an offensive player without
the ball, and screen on and screen off used – when
both types of screen were used).
In order to control for the effects of the situational
variables, game location (playing at home or away),
league stage (regular season and playoff games) and
match status were introduced in the models as cov-
ariates. Match status was obtained using the accu-
mulative differences between points scored and
allowed in each ball possession and then converted
into a categorical variable using a two-step cluster
analysis (Marcelino et al., 2011; Sampaio,
Drinkwater, et al., 2010; Sampaio, Lago, &
Drinkwater, 2010). Five clusters were identified
and categorised as “high disadvantage” (differences
between -10 and -7 points), “moderate disadvan-
tage” (differences between -6 and -3 points),
“balanced” (differences between -2 and 2 points ),
“moderate advantage” (differences between 3 and 7
points) and “high advantage” (differences between 7
and 13 points). For each gender, the sample was
stratified in order to build separate model s for three
game periods (first five minutes, middle thirty min-
utes, and the last five minutes of the games)
(Sampaio, Lago, Casais, et al., 2010; Sampaio,
Lago, & Drinkwater, 2010).
Procedures
The 40 games were analysed through systematic
observation performed by four expert technicians
trained for this task, all graduated in Sports
Sciences with a minimum of five years’ experience
as basketball coaches. After a 3-week period, to
prevent any learning effect, each team re-analysed
one randomly selected game. Weighted kappa
correlation coefficients were calculated to assess
inter-observer and intra-observer reliability
(O’Donoghue, 2010; Robinson & O’Donoghue,
2007). The obtained results showed very good
kappa values (range = 0.84–0.95) for intra-observer
reliability, while inter-observer reliability showed
good and very good values (range = 0.80–0.91)
according to Altman (1991).
Statistical analysis
Binary logistic regression was used to estimate regres-
sion weights and odds ratios of the relation between
performance indicators and covariates according to
ball possession effectiveness (Bar-Eli, Tenenbaum, &
Geister, 2006; Marcelino et al., 2011). In this non-
linear model of regression, the estimated regression
coefficients represent the estimated change in the log-
odds, corresponding to a unit change in the corre-
sponding explanatory variable conditional on the
other explanatory variables remaining constant
(Landau & Everitt, 2004). The performance indicators
were tested individually and later, the adjusted model
was performed with all variables that previously
showed relation to ball possession effectiveness
(Landau & Everitt, 2004). Odds ratios (OR) and
their 95% confidence intervals (CI) were calculated
and adjusted for ball possession effectiveness. The
successful ball possessions was the level of the depen-
dent variable with the OR baseline value (OR = 1). The
observations were considered as independent sam-
pling units, assuming that behaviour during ball pos-
sessions configure unique interactions between
combinations of players and opponents regulated by
unpredictable task and environment-related functional
information (Duarte, Araújo, Correia, & Davids,
2012). The team ability was disregarded in this analysis
because the focus of the study was not to compare
between teams and there are unlikely to be large differ-
ences in ability between teams competing in elite pro-
fessional leagues (Pollard & Pollard, 2005). Also, the
approach to study ball possessions separating three
game periods would be conflicting with an overall
game measure such as team ability. The statistical
analyses were performed using SPSS for Windows,
version 16.0 (SPSS Inc., Chicago IL), and statistical
significance was set at P <0.05.
Results
The distribution of relative frequencies from the
studied variables across the three game periods for
men’s and women’s basket ball teams are showed in
Table I and Table II, respectively.
In the first stage, when the models of the binary
logistic regression were computed with one vari-
able at each step (Table III), the results showed
that in both genders, during the firs t five minutes,
there were relations between ball possession effec-
tiveness and number of passes used and ending
1580 M.-A. Gómez et al.
Downloaded by [UPM] at 02:57 21 December 2015
player. Men’sandwomen’s teams’ results showed
relations with starting and ending zone and
screens used during t he middle thirty minutes.
Conversely, no relations we re identified for both
genders during t he last five minutes. In addition
for men’s teams, there were rela tions between
effectiveness and screens used during the first
fi ve minu tes (Table III), with defensive syst ems
and possession duration during the middle thirty
minutes, and with passes used , possession du ra-
tion and players involved during the last five min-
utes. For women’s t eams, there were additional
relations between effectiveness and starting and
ending zone and defensive systems during the
first five minu tes, a nd no addi tional relations
were found during the middle thirty minutes and
Table I. Distribution of relative frequencies from the studied variables across the three game periods in men’s basketball.
Performance
indicators
First five
min
Middle
thirty min Last five min
Performance
indicators
First five
min
Middle thirty
min
Last five
min
(n = 464) (%) (n = 2576) (%) (n = 483) (%) (n = 464) (%) (n = 2576) (%) (n = 483) (%)
Efficacy Task (cont.)
Successful 50.0 49.5 56.6 Screens used
Unsuccessful 50.0 50.5 43.5 No screens 27.6 29.7 44.9
Space Screens on 16.6 18.2 23.2
Starting zone Screens off 31.5 29.7 18.4
A 32.1 38.4 35.0 Screens on/off 24.4 22.4 13.5
B 15.1 15.8 17.0 Defensive system
C 2.2 2.5 2.5 Man-to-man 92.0 87.0 72.9
D 2.2 3.0 2.1 Zone 1.9 5.4 8.1
E 11.6 12.2 14.1 Press 5.6 6.6 19.0
F 3.4 2.2 1.7 Combined 0.4 1.0 0.0
G 1.9 3.1 2.9 Possession
duration (s)
H 5.2 2.4 2.1 0–4 21.3 20.7 25.3
I 2.6 2.3 2.3 5–10 23.7 18.8 25.5
J 1.5 2.1 3.1 11–15 28.2 29.3 23.6
K 7.8 6.1 6.6 16–20 21.6 22.5 17.0
L 2.6 2.2 1.0 21–24 5.2 8.7 8.7
M 3.9 1.5 1.7 Players’ position
N 2.8 2.6 5.0 Starting Player
O 3.2 1.7 1.0 Guard 55.6 56.3 55.7
P 1.9 1.9 2.1 Forward 21.1 19.1 18.8
Ending zone Centre 23.3 24.6 25.5
A 1.1 0.9 3.5 Ending Player
B 0.2 0.3 0.8 Guard 21.8 23.5 26.7
C 0.2 0.2 0.2 Forward 34.5 36.8 37.3
D 0.2 0.3 2.1 Centre 43.8 39.8 36.0
E 0.6 0.3 0.6 Players involved
F 1.7 0.2 3.1 1 9.5 8.4 9.5
G 9.3 0.3 11.4 2 20.3 19.6 24.4
H 7.3 1.6 6.0 3 31.9 31.8 33.3
I 49.8 10.5 41.4 4 26.9 30.2 26.9
J 11.0 8.7 5.8 5 11.4 10.1 5.8
K 9.7 50.5 10.4
L 2.2 7.4 4.6 Covariates
M 6.7 9.5 1.0 Game Location
N 1.1 1.4 9.1 Home 52.5 49.7 51.8
O 2.2 0.4 3.5 Away 47.5 50.3 48.2
P 6.2 7.4 0.8 League Stage
Task Regular season 49.5 50.7 50.1
Passes used Playoff 50.5 49.3 49.1
0 9.5 8.3 9.1 Match Status
1 13.8 15.6 19.9 High
disadvantage
0.2 7.9 13.9
2 20.9 20.2 25.5 Mod
disadvantage
15.5 21.9 28.4
3 19.2 17.5 17.8 Balanced 68.4 39.2 18.2
4 15.1 15.7 12.0 Mod advantage 15.4 25.0 30.6
+5 21.6 22.8 15.7 High advantage 0.5 6.0 8.9
Ball possession effectiveness in elite basketball 1581
Downloaded by [UPM] at 02:57 21 December 2015
the last five minutes. The likelihood ratio tests
(LRT) identified that the covariates (league
stage, game location and match status) were not
related with ball possession effectiveness in men’s
teams. In women’s teams, the covariates match
status was related with ball possession effective-
ness during the first five minutes, and league
stage was related with ball possession effective-
ness during the middle thirty minutes and last
five minutes. Finally, match status was r elated
with effectiveness during the last five minutes of
the games in women’steams(TableIII).
In Table III, the adjusted model fitted the three
game period in men ’s(first fi ve minutes: LRT =
80.6, df = 10, P < 0.01; middle thirty minutes:
LRT = 320.5, df = 41, P < 0.0001; last five minutes:
LRT = 120.1, df = 13, P < 0.0001) and women’s
teams (first five minutes: LRT = 164.8, df = 44,
Table II. Distribution of relative frequencies from the studied variables across the three game periods in women’s basketball.
Performance
indicators
First five
min
Middle thirty
min
Last five
min
Performance
indicators
First five
min
Middle thirty
min
Last five
min
(n = 822) (%) (n = 2538) (%) (n = 351) (%) (n = 822) (%) (n = 2538) (%) (n = 351) (%)
Efficacy Task (cont.)
Successful 41.0 41.8 45.0 Defensive system
Unsuccessful 59.0 58.2 55.0 Man-to-man 90.4 84.4 74.1
Space Zone 5.2 11.2 7.7
Starting zone Press 4.0 3.1 17.7
A 34.3 37.1 35.9 Combined 0.4 1.3 0.6
B 13.5 11.5 12.3 Screens used
C 4.0 3.4 5.1 No screens 49.4 52.2 57.0
D 2.9 2.7 4.0 Screens on 8.4 9.7 10.0
E 13.3 14.0 10.8 Screens off 32.1 27.1 23.9
F 3.0 4.1 4.6 Screens on and off 10.1 10.9 9.1
G 2.6 2.6 3.7 Possession
duration (s)
H 2.1 2.1 3.1 0–4 17.5 18.3 24.8
I 2.6 3.3 1.7 5–10 24.5 21.1 20.8
J 1.7 3.0 3.4 11–15 29.9 28.8 25.4
K 9.2 7.9 4.3 16–20 22.6 23.3 20.2
L 2.2 2.2 2.8 21–24 5.5 8.4 8.8
M 3.2 2.1 2.3 Players’ position
N 1.6 1.5 2.3 Starting Player
O 1.9 1.6 1.7 Guard 56.1 48.3 36.8
P 1.9 0.9 1.7 Forward 21.9 24.3 31.3
Ending zone Centre 22.0 27.3 31.9
A 0.7 0.9 3.1 Ending Player
B 1.1 0.4 1.1 Guard 20.2 18.6 21.9
C 0.2 0.3 0.3 Forward 38.1 42.6 49.6
D 0.2 0.3 1.4 Centre 41.7 38.8 28.5
E 0.1 0.5 1.7 Players involved
F 1.6 0.2 0.6 1 15.2 12.8 14.0
G 0.4 0.2 0.6 2 23.1 21.5 25.9
H 1.6 1.6 2.3 3 28.5 28.7 29.3
I 6.3 8.3 8.5 4 22.5 24.8 24.5
J 12.8 10.6 8.8 5 10.7 12.2 6.3
K 47.6 51.7 40.7
L 12.7 11.0 9.7 Covariates
M 8.0 7.4 9.4 Game Location
N 1.9 0.8 3.4 Home 51.0 49.5 48.2
O 0.4 0.8 0.9 Away 49.0 50.5 51.2
P 4.4 3.9 7.4 League Stage
Task Regular season 59.3 46.4 44.8
Passes used Playoff 40.7 53.6 55.2
0 15.1 12.7 14.0 Match Status
1 16.9 17.1 21.9 High disadvantage 5.2 11.0 13.9
2 21.3 21.0 19.9 Moderate
disadvantage
17.8 20.7 28.9
3 22.0 18.4 14.0 Balanced 52.1 37.4 20.4
4 13.4 13.9 17.9 Moderate advantage 19.3 18.9 13.3
+5 11.3 17.0 12.3 High advantage 5.6 12.0 25.5
1582 M.-A. Gómez et al.
Downloaded by [UPM] at 02:57 21 December 2015
