Do teams always lose to win?: Performance incentives and the player
draft in the Australian Football League
Jeff Borland, Mark Chicu and Robert Macdonald
(University of Melbourne)*
Abstract
This study examines whether the player draft used in the Australian Football League
(AFL) since 1986 has caused clubs to tank; that is, to seek to lose matches in order to
obtain improved draft choices. A comparison of clubs’ performances in regular
season matches played before and after introduction of the draft provides no evidence
that clubs have engaged in tanking. The main potential explanations for the absence
of tanking in the AFL are the relatively low benefits to clubs from tanking, and
limited opportunities for them to engage in this behaviour.
* This research has been supported by Australian Research Council Discovery grant
0209564. We are grateful for helpful comments from seminar participants at
University of Adelaide, Monash University, Deakin University and Australian
National University, and in particular for comments from Andrew Leigh, Christian
Dustmann and Justin Trogdon. Corresponding author: Professor Jeff Borland,
Department of Economics, University of Melbourne, Melbourne 3010; email:
[email protected]
1
1. Introduction
The 16 club Australian Football League (AFL) is the premier Australian football
competition. Since 1986, a reverse-order draft system has been the primary
mechanism for assigning new players to AFL clubs. A regular feature of recent
media coverage of the AFL competition is the contention that clubs are ‘tanking’; that
is, deliberately losing matches in order to obtain higher selections in the draft. 1
Considerable pressure has therefore been placed on the competition’s organisers to
abolish or significantly reform the player draft system. 2
That mechanisms and policies intended to achieve particular objectives can have
perverse effects is by now well understood. In the case of player drafts in
professional sporting competitions, the way in which perverse effects might arise is
relatively straightforward. Organisers of professional sporting competitions often
seek to enhance competitive balance between clubs by using a reverse-order player
draft to assign new or uncontracted players to clubs. The reverse-order feature of the
draft means that clubs with the poorest win/loss records in a season get to choose the
most highly rated players to add to their playing lists for the next season. Hence,
especially for clubs that have been eliminated from contention for winning the league
championship, there may be an incentive to deliberately lose matches in order to
obtain higher selections in the player draft.
A recent study has examined perverse incentive effects associated with the player
draft mechanism in the US National Basketball Association (NBA). Taylor and
Trogdon (2002) find that the draft for assigning new players to clubs in the NBA
competition appears to have given an incentive to lower-ranked clubs to lose matches
(pp.38-39):
“…in the 1983-84 season we find that eliminated [from playoffs] teams were
approximately 2.5 times more likely to lose than noneliminated teams…This was
expected because the NBA rewarded those teams with lower league rankings with
better draft choices. However, in the 1984-85 season, the NBA restructured its draft
rules to give any nonplayoff team an equal chance at obtaining the most prized draft
2
selections. We find no statistically significant difference between the behaviour of
playoff and non-playoff teams in 1984-85.” 3
In this paper we undertake a similar analysis of the effect on clubs’ performances of
the introduction of a player draft to the AFL. 4 Two main aspects of the AFL player
draft are investigated – first, the National Draft that assigns clubs choices of new
players in reverse order of their league ranking in the preceding season; and second,
the Special Assistance system that has at times provided clubs that win less than a
specified number of games in a season with extra priority draft choices. To examine
both aspects of the player draft a ‘difference-in-difference’ approach is used. This
approach involves comparing the change in performance between the pre-draft and
draft eras of clubs that could have had an incentive to lose matches in draft era, with
the change in performance of clubs that would have had no difference in their
incentive to lose matches between pre-draft and draft eras. In particular we argue that
the Special Assistance system, whereby in some years a club was able to obtain an
extra top-rated draft choice for maintaining a low winning percentage, appears to
provide a strong test for existence of perverse incentive effects.
Our main finding is that there is no evidence that performance of clubs in the AFL has
been systematically affected by the player draft system. We suggest that the main
potential explanations for why the AFL player draft does not appear to have caused
perverse incentive effects are the limited scope for clubs to actually engage in tanking
behaviour, and the relatively small marginal effect of an extra or higher-rated drafted
player on future club performance. The same factors also can explain the difference
in findings in this study from Taylor and Trogdon’s (2002) study of the NBA.
Analysis of the player draft mechanism in the AFL therefore provides an important
case study of how the effect of a mechanism will depend on the environment in which
it is applied and the details of the mechanism. This is consistent with a central
message from the emerging economic design literature that ‘details matter’. 5
In the next section background details on the player draft system in the AFL are
presented. Section 3 develops the main hypotheses to be tested. Section 4 describes
the empirical method, and results from econometric analysis are presented in section
3
5. Section 6 provides some extra empirical analysis. An interpretation of the main
findings is provided in section 7, and concluding remarks are in section 8.
2. About the AFL player draft system
The Victorian Football League (VFL) (renamed the Australian Football League
(AFL) in 1990) introduced a National Draft as a player allocation mechanism in 1986.
Introduction of the player draft was a response to a combination of problems that had
arisen in the Australian football competition in the early 1980s including financial
instability of the VFL and its clubs, escalating and unsustainable player salaries and
transfer fees, declining match attendances, and several challenges to the legality of the
VFL’s labour market controls. 6
The first National Draft occurred in late 1986 with players chosen then being eligible
to play for clubs in the 1987 season. In this draft clubs were each allowed to choose
five players from the Victorian country area and from other states; but players from
the Victorian metropolitan area were still allocated using the geographic zoning
system. Draft choices were made in reverse order of clubs’ finishing positions in the
1986 season. A club’s assigned choice in the National Draft could be traded for
players currently contracted to another club. Drafted players were bound to a club for
three years (Dabscheck, 1989, p.71; Booth, 1996, p.26).
The basic features of the original National Draft – for example, draft choices being
made in reverse order of finishing position in the preceding season, and clubs being
restricted to trading choices in the draft for current players at other clubs – have
remained the same through to the present. Many details of the player draft system
however have undergone reform since its introduction.
First, the significance of the National Draft for clubs has grown over time.
Assignment of new players to clubs has come to occur almost exclusively through the
National Draft, and players chosen in the National Draft have become an increasingly
important determinant of a club’s stock of playing talent. In 1991 the system of
allocating players in metropolitan Victoria through geographic zoning was abolished,
4
and these players were henceforth assigned to clubs through the National Draft. As
well, the entry of clubs from Western Australia and South Australia into the AFL
competition in the 1980s and 1990s meant that local competitions in those states
ceased to provide a viable professional career opportunity for footballers, which has
effectively expanded the pool of talent from which AFL clubs can choose in the
National Draft. Reduction in the maximum size of club player lists, and an increase
in the number of new players that a club could choose in the National Draft to eight in
1988, have also together meant that the National Draft has become an increasingly
important determinant of a club’s stock of playing talent.
Second, the AFL has supplemented the National Draft by allowing at various times
clubs to make ‘priority’ selections (see Booth, 1996, pp.27-28; AFL, 1997, 2003).
Generally these selections are made as the first choices in the National Draft prior to
the regular ‘reverse-order’ draft selections. There have been two main rationales for
granting priority choices. One is to enable new entrants to the AFL competition to
have player lists that make them competitive – for example, Fremantle was granted
priority selections prior to its entry to the competition in 1995. The other ‘Special
Assistance’ rationale has been to assist clubs that have performed poorly.
There were three main phases of provision of Special Assistance for poor
performance prior to 2005. From 1986 to 1993 such assistance was provided on an
ad-hoc basis; for example, priority choices were given to the three clubs at the bottom
of the league table in 1993. From 1994 to 1996 clubs were eligible for a priority
choice on the basis of the weighted average of their winning percentages in the
previous four seasons. From 1997 to 2005 a club that won five matches or less in the
preceding season would receive one priority selection of a player in the National
Draft. The AFL Player Rules (AFL, 2003, p.43) specified that:
“6.2.1 Any club which has 20 premiership points or less [wins five matches or less, or
equivalent] at the end of the Home and Away matches after taking into account any
sanctions imposed by the [AFL] Commission under Rule 17 and subject to any
determination by the Commission in relation to the order of selection, is entitled to
one priority draft selection at the National Draft Selection Meeting.” 7
5
For example, a club that finished at the bottom of the league table and won only 2
matches (and was the only club to win 5 matches or less) would be assigned both the
1st and 2nd choices in the National Draft, whereas if it had finished bottom but won
five matches or more then it would only receive the 1st choice.
Third, some specific details of the National Draft such as age of eligibility have been
regularly revised, and additional draft mechanisms have been progressively
introduced: a pre-season reverse-order draft for out of contract AFL/VFL players or
players cut from a club’s playing list since the National Draft in the previous year
(from 1989 onwards); a mid-season draft for out of contract AFL/VFL players (from
1990 to 1993); and a rookie draft for players aged 18 to 23 years to constitute an
‘emergency’ playing list (from 1997 onwards) (see Booth, 1996, p.27).
Fourth, it is also important to note the existence of other types of labour market
regulations. A salary cap on player payments for each club was introduced for the
1985 season and still exists, being $6.74M. per club in 2005. The size of a club’s
playing list, which was set at a maximum level of 50 players in 1983, has
subsequently been adjusted downwards to the current maximum of 40 players
(Macdonald and Booth, 2007).
3. Theory
For the analysis of incentive effects of the AFL draft we will emphasise two main
aspects of the player draft system: the reverse-order National Draft, and the extra
priority choices that clubs with poor performance have been able to receive through
the Special Assistance system. The central hypothesis to be tested is whether these
aspects of the player draft system provided an incentive for clubs to lose matches in
order to obtain higher draft choices.
In what circumstances would the draft system potentially cause an incentive for a club
to lose matches? One situation where this could occur would be where clubs have
lexicographic-type preferences towards winning the league championship – such as
having an objective function that puts no weight on performance in a season once the
club no longer has a chance of winning the premiership in that season. In this case, a
6
club eliminated from contention for the championship in the current season, would be
expected in that season to adopt the strategy that maximizes the quality of players it
will have available in future seasons. 8 Introduction of the National Draft and the
Special Assistance system mean that a club can potentially improve its quality of
players in future seasons by having the highest draft choices possible in the current
season; that is, achieving the lowest possible finishing position on the league table or
by not having a winning percentage above a specified level. As Taylor and Trogdon
(2002, p.27) then argue: ‘…as the reward for losing increases, teams should exert less
winning effort and more losing effort, and performance should decline.’ Exerting
more ‘losing effort’ might for example involve choosing a team or adopting a training
regime that does not maximize a club’s chance of winning a match. 9
The first main hypothesis to be tested is with regard to the introduction of the reverseorder National Draft. We examine whether introduction of the draft caused a decrease
in the probability of winning matches for clubs eliminated from contention from
finals. The incentive to lose extra matches would be that by doing so a club can
obtain a higher choice in the draft. For example, a club that, by losing extra matches
can change its finishing position in the regular season from 2nd last to bottom, would
move from having the 2nd choice in the National Draft to having the 1st choice. A
supplementary hypothesis is that any effect of the National Draft on the probability of
winning matches for teams eliminated from contention from finals will have increased
over time, since it appears that the importance of the National Draft has grown due to
its greater scope and smaller club playing lists.
The second main hypothesis that we test involves the Special Assistance priority
choice system. There have been three phases of this system, and we argue that
incentive effects to lose matches would have been strongest in the third phase. This is
because it was in this phase that the effect of a club’s performance in the current
season on whether a priority choice would be received in the next season was greatest.
For example, consider the case of a club that has won less than five matches in the
current season. In the first Special Assistance phase (1986-93), priority choices were
provided to clubs on an ad-hoc basis. Hence in this phase a club’s number of games
won could not have been seen as directly related to the likelihood of receiving a
7
priority choice. In the second phase (1994-96), a team might have won five matches
or less in the current season, but if its performance had been better in the previous
seasons, most likely it would not have been eligible for a priority choice. So again
there was no necessary gain from keeping its number of wins at five or less. In the
third phase (1997-2005) a club in this position would definitely be eligible for a
priority choice if it did not win more than five matches. Having such a priority choice
would give a club an extra choice in the National Draft before the main reverse-order
component of the draft. Hence in this phase there is potentially a strong incentive for
a team to win five matches or less in the current season. To test the incentive effects
of the Special Assistance priority choice system therefore we examine whether –
during the third phase of the system between 1997 and 2005 – there was a decrease in
the probability of winning matches for clubs eligible for Special Assistance compared
to clubs that had been in similar positions in earlier seasons.
The value of improved draft choices in the AFL National Draft will be, for example,
the value from having the 1st choice rather than 2nd choice; whereas, as we have noted,
under the AFL Special Assistance system the value is from having an extra top draft
choice compared to no such choice. Hence, other things equal, we expect that the
value of obtaining a Special Assistance draft choice is greater than achieving a higher
choice in the reverse-order draft.
4. Empirical model and data
Estimation is undertaken using a probit model on match-level data. Separate models
are used for testing each of the two main hypotheses. The dependent variable in each
model is a dummy variable equal to one if club i wins its match in round j in season t,
win ijt . 10
To examine the hypothesis that introduction of the reverse-order National Draft
caused clubs eliminated from contention for finals to win less matches in the draft era,
we estimate model (1):
8
n
win ijt = f(∑ D j ⋅ elim ijt ,
j=1
n
∑D
j
n
∑ D j ⋅ elimijt ⋅ post86t , oelimijt ,
j=1
n
∑ D ⋅ elim
j
ijt
⋅ oelimijt ,
j=1
⋅ elimijt ⋅ oelim ijt ⋅ post86 t , winpcijt , owinpcijt , hagrd ijt , ahgrd ijt , hhgrd ijt ,
j=1
hastateijt , ahstateijt , hhstateijt )
(1)
Where i = club; j = round (1, 2,…, (n-1), n); and t = season (1968, 1969,…, 2005). In
this model elimijt is a dummy variable that equals one where club i has been
eliminated from participating in the finals series of matches prior to round j in season
t, oelimijt is a dummy variable for where club i is playing a match against an opposing
club that has been eliminated from participating in the finals prior to round j in season
t, post86 t = 1 for t = 1986,…, 2005, and zero otherwise; and
D j = 1 for j = (n-6),..., (n-1), n, and zero otherwise .
Model (1) tests whether clubs eliminated from participating in the finals were less
likely to win matches (or more likely to lose matches) in the final six rounds of the
regular season in the period after introduction of the draft than in the period prior to
introduction of the draft. To control for the baseline effect of being out of contention
n
for the finals we use a set of dummy variables (∑ D j ⋅ elimijt ) that interact the dummy
j=1
variable for a club having been eliminated from contention for finals (elimijt ) and a set
of dummy variables for each of the last six rounds of the regular season. To test for
whether there is a different effect of being out of contention for finals after
introduction of the National Draft the same dummy variables are interacted with a
dummy variable that equals 1 in any season from 1986
n
onwards (∑ D j ⋅ elimijt ⋅ post86t ) . Significant negative coefficients on these variables
j=1
would be consistent with a negative effect of the reverse-order National Draft on a
club’s likelihood of winning matches. In examining the effect of the National Draft
we also control for matches where a team that had been eliminated from participating
in the finals was playing against another team that had also been eliminated from the
finals, as in such matches we would be considering two teams, both of which may
9
have an incentive to seek to lose the match. We do this by using the sets of dummy
n
n
j=1
j=1
variables (∑ D j ⋅ elimijt ⋅ oelimijt ) and (∑ D j ⋅ elimijt ⋅ oelimijt ⋅ post86 t ) .
In constructing the dummy variable for whether a club has been eliminated from
contention for finals, elimijt , it is assumed that clubs know the number of matches that
it will ultimately be necessary to win to qualify for finals in each season. Hence, for
example, if in a season it will be necessary to win 12 matches to qualify for finals, and
with four rounds remaining, a club has only won seven matches, it would be
considered to have been eliminated from contention for finals. In fact, by the final six
rounds of the season, most clubs that we classify as eliminated from contention for
finals would be unable to qualify for finals regardless of outcomes in other matches
for the remainder of the season. 11
To investigate the effect of the Special Assistance system of priority choices that
operated from 1997 to 2005, two different models are used. Model (2) tests the effect
of a team being eligible for a priority choice:
n
win ijt = f(∑ D j ⋅ papijt ,
j=1
n
∑ D j ⋅ papijt ⋅ opapijt ,
j=1
n
∑D
j
n
∑ D ⋅ pap
j
ijt
⋅ 8692 t ,
j=1
n
n
∑ D ⋅ pap
j
j=1
∑ D j ⋅ papijt ⋅ opapijt ⋅ 8692t ,
j=1
n
∑D
j
ijt
⋅ 9396 t ,
n
∑D
j
⋅ papijt ⋅ 9705t ,
j=1
⋅ pap ijt ⋅ opapijt ⋅ 9396 t ,
j=1
⋅ papijt ⋅ opapijt ⋅ 9705t , opapijt , winpcijt , owinpcijt , hagrd ijt , ahgrd ijt , hhgrd ijt ,
j=1
hastateijt , ahstateijt , hhstateijt )
(2)
In this model papijt is a dummy variable that equals one where club i has won less
than or equal to five matches prior to round j in season t so that it would have retained
eligibility for a priority draft choice in the phase of the Special Assistance system that
operated from 1997 to 2005, opapijt equals one where a club is playing a match
against an opponent that has prior to that round had won five matches or less, 8692t =
10
1 for t = 1986,…, 1992, and zero otherwise, 9396t = 1 for t = 1993,…, 1996, and zero
otherwise, and 9705t = 1 for t = 1997,…, 2005, and zero otherwise.
Model (2) tests whether clubs that could finish the season having won only five
matches or less were more likely to lose matches in the third phase of the Special
Assistance system than in other time periods. To control for the baseline effect of
having won five matches or less in a season we use a set of dummy variables
n
(∑ D j ⋅ papijt ) that interact a dummy variable for club i having won five matches or
j=1
less prior to round j in season t (papijt ) and a set of dummy variables for the last six
rounds. To test for whether there is a different effect of having won five matches or
less under the Special Assistance priority choice system that operated from 1997 to
2005 we use the same dummy variables interacted with a dummy variable that equals
22
1 in any season from 1997 to 2005, (∑ D j ⋅ papijt ⋅ 9705t ) are used. Significant
j=1
negative coefficients on these variables would be consistent with a negative effect of
the Special Assistance priority choice system on a club’s likelihood of winning
matches. Interactions of the baseline dummy variables with dummy variables for the
other Special Assistance eras (1986 to 1992, and 1993 to 1996) are also included.
Since different priority choice systems operated at those times that would have had a
weaker incentive effect than the system from 1997 to 2005 it is predicted that
coefficients on these variables should be smaller in magnitude or insignificant.
Finally, we control for matches where a club that would have retained eligibility for
Special Assistance in the phase from 1997 to 2005 was playing against another club
that also retained this eligibility, as once again in such matches we would be
considering two clubs, both of which may have an incentive to seek to lose the match.
22
We do this by using the set of dummy variables (∑ D j ⋅ papijt ⋅ opapijt ) , as well as the
j=1
same dummy variables interacted with the time intervals corresponding to the
different phases of the Special Assistance system.
Model (3) specialises the analysis of the Special Assistance system to examine effects
on clubs that have ‘borderline’ eligibility for a priority choice. By borderline what is
11
meant is a club that, if it wins its next match, will have won more than five matches,
and hence have lost eligibility for Special Assistance in the phase of that system from
1997 to 2005. A club that had only won four matches or less in a season from 1997 to
2005 could have afforded to win another match and would still have retained
eligibility for Special Assistance; whereas a team that had already won four and a half
(that is, won four matches and drawn one match) or five matches would lose
eligibility if it won another match. Hence it seems possible to argue that the incentive
to lose matches would be greater for a club with ‘borderline’ eligibility for Special
Assistance. To test this we define a new dummy variable, bpapijt , which equals one
where club i would lose eligibility for Special Assistance in the phase from 1997 to
2005 if it wins its match in round j in season t, and obpapijt , which equals one where a
club is playing a match against an opponent that will lose eligibility for Special
Assistance if it wins its match in round j. Model (3) is then:
22
win ijt = f(∑ D j ⋅ bpapijt ,
j=1
n
∑ D j ⋅ bpapijt ⋅ obpapijt ,
j=1
n
∑D
j
n
∑ D ⋅ bpap
j
j=1
ijt
⋅ 8692 t ,
n
∑ D ⋅ bpap
j
ijt
⋅ 9396 t ,
j=1
n
n
j=1
j=1
n
∑ D ⋅ bpap
j
ijt
⋅ 9705t ,
j=1
∑ D j ⋅ papijt ⋅ obpapijt ⋅ 8692t , ∑ D j ⋅ papijt ⋅ obpapijt ⋅ 9396t ,
⋅ papijt ⋅ obpapijt ⋅ 9705t , obpapijt , winpcijt , owinpcijt , hagrd ijt , ahgrd ijt , hhgrd ijt ,
j=1
hastateijt , ahstateijt , hhstateijt )
(3)
Throughout, we test for effects of the draft system on club performance by restricting
attention to the final six rounds of the season. This is because it is only at this stage of
the season, where there just over one-quarter of the regular season remaining, that
clubs are likely to begin to take account of considerations regarding to player draft
choices for the next season. In any case, it will be apparent from our results that this
restriction does not constitute a major constraint on the analysis.
To control for other determinants of a club’s likelihood of winning a match, apart
from the incentive effects of the drafts, we include a variety of explanatory variables.
First, to control for the quality of a club and its opponent in a match, the winning
percentages of that club and its opponent prior to round j in season t,
12
winpcijt , owinpcijt , are included. Second, dummy variables for whether a club is
playing at its home ground, whether a club is playing at its opponent’s home ground,
or whether both clubs are playing at their home ground (hagrdijt , ahgrd ijt , hhgrd ijt )
are included. Previous research has found that ‘home ground advantage’ is an
important determinant of match outcomes in the VFL/AFL (Clarke, 2005). Third,
dummy variables for whether a club is playing in its home state, whether it is playing
in the home state of its opponent, or whether both clubs are playing in their home
state (hastateijt , ahstateijt , hhstateijt ) , are included. Fourth, as an extra control for the
opponent’s quality, a dummy variable for whether the opposing club had been
eliminated from contention for the finals (oelimijt ) is included in the model testing
the effect of the reverse-order National Draft; and dummy variables for whether the
opposing club had won less than 5 matches prior to round j or would lose eligibility
for Special Assistance by winning its match in round j, opapijt and obpapijt
respectively, are included in the models testing for effects of the Special Assistance
system.
Table 1 provides a full list and description of all variables. 12 The data set used for
estimation includes every regular season (home-and-away) match played in VFL/AFL
between 1968 and 2005. The starting date for the sample is chosen to have an
equivalent number of seasons prior to and after introduction of the National Draft.
The primary source of data on individual matches including competing teams, round,
season, score, location, and winning percentages is http://afl.has.it. 13 In 1968, 1969
and 1993 the regular season involved each club contesting 20 matches, and in each
season each club played 22 matches. Each match is associated with two observations
(one for each club) in the data set. The number of teams in the VFL/AFL competition
has increased across time, from 12 at the start of the sample period to 16 at the end of
the sample period. 14 Data on rounds 1 and 2 for each season is excluded to enable
calculation of measures of clubs’ winning percentages prior to each round of matches
within each season. 15 The total number of match-level observations is 10,342.
13
5. Results
a. Effect of reverse-order National Draft
Estimates of model (1), which tests the effect of the reverse-order National Draft, are
reported in Table 2. The first column presents results from a restricted model that
includes as explanatory variables only the dummy variables for clubs having been
eliminated from participating in the finals series of matches. The second column
reports a model that also controls for where a club that has been eliminated from the
finals series plays a match against another club eliminated from the finals series. The
third column includes the full set of explanatory variables. In testing for effects of the
National Draft in the final six rounds of a season, dummy variables for these rounds
have been aggregated into variables for matches played in the 1st or 2nd last rounds, in
the 3rd or 4th last rounds, or in the 5th or 6th last rounds.
The key finding is that there is no evidence of any significant change in the likelihood
of winning matches for clubs that have been eliminated from participating in the
finals series between the periods prior to and after introduction of the reverse-order
National Draft. Clubs eliminated from participation in the finals series are less likely
to win matches in the last six rounds than other clubs; for example, in the full model
the estimated effect on ‘elim12’ implies that a club eliminated from the finals has a
probability of winning a match in the final two rounds that is 17.8 percentage points
lower than a club that had not been eliminated. Hence knowing a club has been
eliminated from participating in the finals series is informative about its quality.
However, none of the interactions between the dummy variable for being eliminated
from the finals series and the dummy variable for the draft period, for example
‘elim1286-05’, is found to be significant. Therefore we can infer that clubs that had
been eliminated from the finals were not more likely to lose matches in the draft era.
To test whether there might have been a change in the likelihood of winning matches
in any sub-period of the draft era we also estimated a version of the model where the
dummy variable for having been eliminated from the finals is interacted with dummy
14
variables for sub-periods of the draft period (1986-90, 1991-95, 1996-2000, and 200105). But for none of these sub-periods was there found to be a significant effect. 16
Further evidence on the effects of the reverse-order feature of the National Draft is
presented in Table 3. Predicted winning percentages for clubs involved in each match
are derived, and a predicted winner of that match is determined as the club with the
highest predicted probability of winning. 17 Matches are then divided between those
that clubs eliminated from participation in the finals series would have been expected
to win and expected to lose, and we compare the outcome of these matches between
the pre-draft (1968-1985) and draft (1986-2005) eras. Tanking could be manifested
either as a club winning a smaller proportion of matches than it was predicted to win,
or losing a larger proportion of matches than it was predicted to lose, in the draft era
than the pre-draft era. Table 3 shows that neither of these outcomes is apparent; in
fact if anything there is some evidence that clubs won more matches that they were
predicted to lose in the draft era, although the effect is not significant.
b. Effect of Special Assistance system
Results from tests for the effect of the Special Assistance system are presented in
Tables 4 and 5. Table 4 shows results from Model (2) which tests for the effect of
eligibility for Special Assistance in the National Draft. Columns (1) and (2) present
results with restricted and full sets of explanatory variables. There is no evidence in
either model of any significant change in the probability of winning matches in the
last six rounds of the season for clubs that had won five matches or less between the
Special Assistance phase, comparing between 1997 to 2005 with other time periods.
The baseline effects for eligibility for Special Assistance, such as ‘pap12’, are
significantly negatively related to the probability of winning a match. There is
however no evidence of a difference in this effect across time. Interactions between
the dummy variables for eligibility for Special Assistance and the dummy variable for
1997 to 2005, such as ‘pap1297-05’, are not significant.
Another way to test the effect of the National Draft is to examine whether the
difference in aggregate scores between winning and losing clubs in matches in which
15
a club eligible for Special Assistance is playing alters between the pre-draft and draft
eras. Clubs that are eligible for Special Assistance are not predicted to win many
matches, and hence it is possible that a greater incentive to lose matches could be
manifested in larger losing margins rather than losing a larger number of matches.
Results from a model which tests this proposition, using the percentage difference in
the aggregate scores of clubs in each match, are shown in columns (3) and (4) in
Table 3. The significant negative coefficients on ‘pap1297-05’ do appear to indicate
that the losing margin is significantly larger in the final two rounds of the season in
matches involving a club eligible for Special Assistance in the phase from 1997 to
2005 than in the pre-Draft era. Before concluding that this is evidence of an effect of
the draft system, it is also necessary however to note that there is a very similar effect
estimated in the final two rounds in the earlier draft phases (‘pap1294-96’ and
‘pap1286-93’) when the effect of the Special Assistance system on incentives would
have been predicted to be smaller. 18 Hence it does not seem possible to conclude that
analysis of the size of losing margins provides evidence of an incentive effect due to
the Special Assistance system.
Results from Model (3), which tests the effect of ‘borderline’ eligibility for Special
Assistance, by comparing the performance of clubs involved in matches where
winning that match would cause them to have won more than five matches, between
the phase from 1997 to 2005 and other time periods, are presented in Table 5.
Columns (1) and (2) present results with restricted and full sets of explanatory
variables. In both columns we control for matches in which a club with borderline
eligibility for Special Assistance played a match against another club with borderline
eligibility. Columns (3) and (4) present the same models but also include dummy
variables to control for matches in which a club with ‘borderline’ eligibility for
Special Assistance played a match against another club that was eligible for Special
Assistance. Once again, there is no evidence of the Special Assistance system having
affected club performance. The baseline effects for a club that is ‘borderline’ are
significant and negative. For example, the coefficient on ‘bpap12’ shows that a club
that was in the position in the final two rounds of having won more than five matches
if it wins its next match, is on average 25.1 percentage points less likely to win that
match than other clubs. However, for none of the versions of the model estimated is
16
the interaction of this effect with the phase from 1997 to 2005, ‘bpap1697-05’,
significant. 19 Hence there is no evidence that the Special Assistance system caused
‘borderline’ clubs to be more likely to lose matches in this phase.
Table 6 reports how the performance of clubs with borderline eligibility for Special
Assistance has compared with predicted performance in the last six rounds of the
season between the pre-Draft and Draft time periods. There is no significant
difference across the time periods in the proportion of matches that clubs with
borderline eligibility lose out of the matches they are predicted to lose. Due to a small
number of observations it is not possible to test for whether there is a statistically
significant difference in the proportions of matches that clubs win out of the matches
they are predicted to win.
c. Other explanatory variables
Other results on the determinants of the likelihood of clubs winning matches seem
plausible and support the robustness of the estimated models. First, the likelihood of
winning is strongly increasing with a club’s winning percentage and decreasing in its
opponent’s winning percentage. Second, clubs are more likely to win matches against
clubs that have been eliminated from finals or are eligible for Special Assistance draft
choices. Third, there is a significant home ground advantage, but no effect where
both clubs are playing at their home ground; and no effect associated with whether a
team is playing inter-state once the home ground effect is controlled for.
d. Robustness checks
We are able to undertake several checks on the robustness of our findings. First, we
examine the predictive power of the model used to estimate the determinants of match
outcomes. We again use estimates from a basic form of the model to predict each
club’s probability of winning a match, and the club with the highest predicted
probability in that match is designated as the predicted winner of the match. This
forecast outcome is compared with the actual outcome. 20 Results from a comparison
of predicted and actual outcomes are shown in Table 7. On average the model
17
correctly predicts the outcome in about 65 per cent of matches. This is very close to
the average success rate of expert forecasters; for example, Amor and Griffiths (2004)
calculate the winning percentages of five expert forecasters in the 1999 and 2000 AFL
seasons to have been between 63.9 per cent and 66.7 per cent. Further support for the
model is that the percentage of correct predictions of match outcomes increases as the
difference between each club’s predicted probability of winning increases. Where the
gap in predicted winning probabilities is less than 10 percentage points then the model
forecasts just over 50 per cent of matches correctly, but when the gap is between 90
and 100 percentage points the model is correct for 75 per cent of matches. Second,
we are able to examine whether our results are likely to have been affected by
structural changes in the AFL competition between the pre-draft and draft eras;
specifically, changes in the number of clubs in the AFL competition, and in the
proportion of AFL clubs making the finals series. The effects of these changes in the
AFL competition on the proportion of matches won by teams eliminated from
participating in the finals series and by teams with borderline eligibility for Special
Assistance are tested. If, for example, a positive relation was found between the
number of AFL clubs and the proportion of games won by clubs eliminated from the
finals series, this might suggest that the increase in the number of clubs between the
pre-draft and draft eras had biased upwards our estimates of the likelihood of clubs
eliminated from the finals series winning matches in the draft era; or putting this the
other way around, biased downwards our estimates of the incidence of tanking in the
draft era. However, we find no significant relation between the performance of clubs
eliminated form the finals series or with borderline eligibility for Special Assistance
and the structural changes in the AFL competition. 21
6. Some extra empirical analysis of tanking
The econometric analysis that has been undertaken thus far suggests there is no
evidence of systematic tanking by AFL clubs that could be attributed to either the
reverse-order National Draft or to the opportunity to obtain Special Assistance. How
would ‘the fan in the street’ respond to our findings? We believe that an Australian
football fan might have two responses. The first would be to cite examples of
particular matches where there might seem to have been strong evidence of tanking
18
by a club. An example (outside our sample period) would be the match between
Carlton and Collingwood that was played in round 18 in the 2007 season. If Carlton
had won that match it would have lost eligibility for Special Assistance. Towards the
end of that match, Carlton, which appeared to be in a position where it might win the
match, took a star forward player off the ground, and during this time the opposing
team, Collingwood, scored goals that allowed it to win by 120 points to 96 points
(Gleeson, 2007). The second response would be to cite descriptive evidence that has
sought to demonstrate the existence of tanking. An example is the article by Stevens
(2007) which presented data showing that between 2000 and 2005 no club finished
the regular season just above the threshold for Special Assistance with six wins,
whereas five clubs finished just below the threshold with five wins.
With regard to the fan in the street’s first response, it is important to acknowledge that
while our findings can be interpreted as showing that there is no evidence of
systematic or significant tanking behaviour by AFL clubs in the draft era, the
econometric method we use is unlikely to have sufficient power to rule out the
possibility that tanking occurred in a small number of matches. As well, our approach
for evaluating whether tanking occurred uses predictions of match outcomes that are
based on clubs’ previous performances within a season, and does not take into
account the actual circumstances that existed in any match. These caveats should not
be taken however to mean that tanking has happened. Rather they define limits on the
extent to which it is possible to conclude from our study that tanking has not occurred.
Our feeling about the second response is that properly constructed descriptive
statistics in fact provide little evidence in support of tanking behaviour by AFL clubs.
Appropriate descriptive statistics for evaluating the Special Assistance system should
compare the whole of the phase of that system from 1997 to 2005 with a pre-draft
period, and need to include both clubs finishing with the equivalent of five and a half
or six wins as ‘just over’ the threshold for eligibility for Special Assistance and four
and a half or five wins as ‘just under’ the threshold. We do this in Figure 1, and find
that, by comparison with 1978-85, there is little difference in the distribution of wins
in the Special Assistance phase from 1997 to 2005. In 1978 to 1985 clubs finished on
four and a half or five wins on nine occasions, whereas in 1997 to 2005 this happened
on seven occasions. Clubs finished on five and a half or six wins in 1978 to 1985 on
19
five occasions, compared to three occasions in 1997 to 2005. Hence our analysis
suggests that the Special Assistance system from 1997 to 2005 did not have any
impact on the relative number of clubs winning either four and a half or five matches,
compared to five and a half or six matches. It is therefore difficult to see evidence of
tanking behaviour in this comparison.
7. Discussion
Why is there no evidence of tanking from either the reverse-order National Draft or
the Special Assistance system in the AFL? One approach to answering this question
is to think in terms of the benefits and costs of tanking. The benefits that a club can
obtain by losing matches will depend on how this will affect the draft choices it
obtains, and on the value of obtaining those higher draft choices. The costs are likely
to consist of any adverse consequences for individual club members (players and/or
coaches) or to club revenue from losing extra matches.
There are a variety of reasons for why the marginal benefit from an improved draft
choice in the AFL may be relatively low. First, for our sample period AFL teams
have consisted of 20-22 players of whom 18 are on the playing field at any time.
Hence, the effect of an extra high-ability player on club performance may be
relatively low, especially compared to other sporting competitions such as the US
NBA where a team consists of a much smaller number of players.
Second, benefits from higher draft choices depend on the capacity to correctly
identify player quality. Other analysis we have undertaken (Borland et al., 2008)
shows that in the AFL, at least in the initial period in which the player draft operated,
clubs had significant difficulty in estimating player quality. For example, for the first
five years of the National Draft, players drafted in the 1st round were selected to play
in only 17 per cent of matches in which they could have been chosen in the seven
years after being drafted (compared to 11 per cent for 2nd and 3rd round draft choices).
This contrasts with the NBA where our analysis shows that – for the seasons analysed
by Taylor and Trogdon (2002) – players drafted at the end of those seasons generally
played 40 to 50 percent of game time in their first seven years after being drafted. 22
20
One reason for this difference may be the existence of collegiate basketball
competitions in the United States. Because of the older age of participants and
participation in college sport over several years, the ability of potential players may be
able to be more accurately identified in the NBA than through junior Australian
football competitions (see for example Spurr, 2000). Hence the benefits from
obtaining a higher draft choice, especially where this was simply a matter of for
example a choice one or two higher-ranked than otherwise, are likely to have been
relatively low in the AFL.
Third, the AFL has mainly drawn players exclusively from within Australia. Drawing
from a relatively small population may also imply a relatively low marginal benefit
from improved draft choices. Assuming that ability can be characterised by a singledimension indicator of ability, this situation could be represented as the AFL drawing
a small sample from a population distribution, whereas a competition such as the US
NBA, which draws players from any country where basketball is played, is effectively
drawing a large sample from the population distribution. This is likely to mean that
the absolute ability of the highest ability players will be higher in the NBA than AFL;
and where the effect of higher player ability on team performance is positive and
increasing, this absolute ability effect suggests that the value of a higher draft choice
will be greater in the NBA than AFL. 23,24
A related possibility is that in the AFL it is a combination of individual characteristics
that determine player ability, and that therefore the distribution of ability will be
relatively concentrated. In the case where any individual characteristic has a
distribution that is skewed to the right, and supposing that the effect of individual
characteristics on player ability in the AFL is multiplicative, then by the central limit
theorem, the distribution of player ability will be more concentrated in the AFL than
in another competition, such as the US NBA, where it might be argued that a singledimension characteristic as ‘height’ is very important as a measure of player ability
(see for example, Berri et al., 2005, and also Roy, 1951). This would provide a
further explanation for why an improved draft choice might have relatively low
marginal benefit in the AFL, again especially compared to a competition such as the
US NBA.
21
The main potential cost to a club from losing extra matches that it seems possible to
identify is that clubs with poor winning records are more likely to dismiss coaching
and playing staff. In Table 8 we report results from a probit model of the relationship
between club performance and coach turnover in the AFL between 1968 and 2005.
Consistent with our hypothesis, a 1 percent increase in winning percentage is found to
lower the probability of coach turnover in that season by 0.75 percentage points in the
AFL. Interestingly, however, this effect is not at all dissimilar to what is found for
NBA. Estimating the same model for the NBA between 1977 and 1995 an increase in
winning percentage is estimated to decrease the probability of coach turnover by 1.05
percentage points. Hence it does not seem likely that high costs of losing extra
matches – at least with respect to effects on coach turnover – can explain the absence
of tanking in the AFL, and why this differs from the US NBA.
Notwithstanding the explanations we have just given for why AFL clubs may not
have engaged in tanking behaviour, it may still be hard to accept that the Special
Assistance system has not induced tanking by clubs, given what seem substantial
benefits to a club from having an extra priority draft choice. There is also however
another explanation for why we do not find evidence of effects of the Special
Assistance system. This is that clubs eligible for Special Assistance do not have
many opportunities to engage in what we would regard as tanking. Few clubs are
eligible for Special Assistance, and clubs which are eligible are performing very
poorly. Hence there are hardly any instances in the pre-draft era where clubs that
would have between 1997 and 2005 been eligible for Special Assistance are predicted
to win matches or where they actually win matches. 25 Therefore there is little
opportunity for clubs eligible for Special Assistance between 1997 and 2005 to
exhibit differences in performance by losing matches they would previously have
won. Another way to put this would be to say that the Special Assistance system in
fact limits the scope for clubs to engage in tanking.
Relatively low benefits from having improved draft choices, and the limited scope for
clubs eligible for Special Assistance to affect match outcomes from what would
otherwise have occurred, therefore appear to be the main potential explanations for
22
why we do not find evidence of systematic tanking behaviour by AFL clubs in the
draft era. The same factors also seem to explain the difference between our findings
for the AFL draft, and the results of Taylor and Trogdon (2000) on the US NBA draft.
Differences in the effects of the AFL and NBA player draft systems on incentives to
lose matches seem consistent primarily with greater benefit from tanking in the US
NBA than AFL. For example, differences in details of the competitions – such as size
of teams and the populations from which clubs’ playing lists are drawn – are
consistent with much higher benefits from having higher draft choices in the reverseorder draft in the NBA than AFL.
8. Conclusion
In this study we have examined whether aspects of the player draft mechanism in
the AFL have been associated with perverse incentive effects. Using data on all
AFL matches played between 1968 and 2005 the study finds that: (a) Clubs
eligible for Special Assistance during the period where any team that won five
matches or less would receive an extra priority choice in the player draft (19972005) were no more likely to lose matches in the last six rounds of the season
than similarly-performed clubs in other seasons; and (b) Clubs eliminated from
the finals in the post-draft era are found to be no more likely to lose matches in
the last six rounds of the season than similarly performed clubs prior to the
introduction of the draft. On the basis of these findings we conclude that there is
no evidence that the AFL player draft has caused an incentive for clubs to lose
matches to receive higher draft choices.
The findings in this study for the AFL draft contrast with recent evidence from
the NBA in the United States, where it was found that the likelihood of teams that
had been eliminated from contention for the playoffs winning matches was
significantly related to the scope to improve a club’s position in the draft for new
players. Our main explanation for the different findings is differences in the
benefits of tanking between the competitions. The nature of the competitions and
details of the draft mechanisms appear to give a greater incentive and opportunity
for NBA clubs to seek to deliberately lose matches than in the AFL.
23
One implication from our findings is to suggest circumstances where sporting leagues
would be likely to need to take greater account of potential perverse incentive effects
in designing draft mechanisms. Such incentive effects seem likely to be greater in
sporting competitions where any individual player who is drafted will constitute a
relatively large fraction of the team; or where clubs are able more precisely to identify
player ability prior to the draft. The study can also be argued to provide an
illustration of the general point that how a mechanism affects behaviour is likely to
depend on its details, and on the environment in which the mechanism is applied.
Even with what might seem a fairly specific mechanism such as a player draft,
applied in what might seem to be the homogenous environment of a professional
sporting competition, it is evident that details matter.
24
Endnotes
1. An impression of this coverage can be obtained from the comment by a leading
current player that: “There seems little incentive for teams in 14th, 15th and 16th spot to
win games” (Judd, 2006). Other recent newspaper articles have titles such as ‘To the
loser the spoils’ (Niall, 2005); ‘No incentive in priority pick system’ (Johnson, 2005);
‘Tanks for nothing, Carlton’ (Butler and O’Donoghue, 2007); and ‘Eliminate the
tanking temptation: Malthouse’ (Broad, 2007).
2. One commentator has for example written that: ‘Forget about using only the
finishing positions after the finals to determine the draft pecking order. It is too
simplistic and open to manipulation’ (Stevens, 2007).
3. Other studies of ‘tanking’ in professional sports are by Balson et al. (2007) and
Wolfers (2006) on US College basketball, and Duggan and Levitt (2002) on sumo
wrestling.
4. See Syzmanski (2003) for a general introduction to design issues associated with
sporting competitions. Other studies have used player draft mechanisms in sporting
competitions to study a variety of aspects of behaviour – such as sunk cost effects
(Staw and Hoang, 1995, and Camerer and Weber, 1999), and biases in decisionmaking (Massey and Thaler, 2005).
5. McMillan (2002, p.14) notes that “For markets…it is the details of design that
determine whether they work well; and equally, it might also be said that it is the
details that determine how any mechanism works (see also Roth, 2002).
6. Prior to the introduction of the player draft system the labour market for VFL
footballers was regulated through a system whereby:
(a) Players living in the state of Victoria were initially allocated to clubs on the basis
of residential location through assignment of exclusive metropolitan and country
geographic zones to each club;
(b) Players living in other states (than Victoria) who had played 100 games in that
state could be recruited by VFL clubs through a reverse-order draft system that
allowed each club to choose two players each year with choices being in reverse order
of finishing position on the league table; and
(c) Players already contracted to a club in the VFL could only transfer to another club
with permission of the ‘owning’ club, which would usually involve payment of a
transfer fee to that club.
For overviews of the evolution of AFL labour market regulations, see Dabscheck
(1975, pp.180-83), (1985, pp.2-4), and (1989, pp.65-66); and Booth (1996, pp.21-25).
7. A club that wins a match is awarded four premiership points, clubs that draw a
match are awarded two premiership points each, and a club that loses a match
receives zero points. Hence a club could attain 20 premiership points in a variety of
ways – for example, winning five matches and losing all other matches, or winning
three matches, drawing two matches, and losing all other matches. To simplify
exposition in the remainder of the paper we refer to the Special Assistance system
from 1997 to 2005 as applying where a club wins five matches or less.
25
8. The championship winner (generally known as the premier) in the AFL is decided
through a finals (playoff) series of matches that involve a specified number of top
teams after a regular season of matches. Hence a club that is eliminated from
contention for competing in the finals series is not able to win the premiership, and so
it is clubs eliminated from competing in the finals that we regard as potentially having
an incentive to engage in tanking behaviour.
9. A difficulty is that some types of behaviour that could be interpreted as tanking
behaviour might also be interpreted as an optimal response to preparing a club
eliminated from playing in the finals for the next season. For example, leaving some
key players out of a team at the end of a season in order to allow them to have surgery
so that they have maximum preparation time for the next season may be an optimal
strategy ‘per se’ and not primarily done with the intention of obtaining higher draft
choices. See for example, Connolly (2006) and Le Grand and Conn (2006).
10. Matches that are drawn are coded as a win for both clubs. Omitting these matches
(which constitute only 0.8 per cent of the total matches) from the analysis has no
effect on the main findings.
11. Our approach may classify some clubs to have been eliminated from finals before
this is actually the case. For example, in the numerical illustration just used, it is
likely that, with four matches remaining, there would be some scenario under which
the number of matches a club would need to win to make the finals would only be 11.
Taylor and Trogdon (2002) use an alternative approach that defines a club to be
eliminated from contention for the playoffs if it is statistically impossible for that club
to qualify for the playoffs; that is, if there are n rounds remaining in the regular season
and a club has n+1 wins less than the club that would currently be the last club to
qualify for the playoffs. This approach has the opposite disadvantage of counting
clubs as still being in contention for finals when this may be a very low probability
event. For example, with 5 rounds remaining a club may be 5 matches behind the
club that would at that stage be the last club to qualify for playoffs – but if the club
that is behind must play 5 clubs above it on the league ladder in the remaining 5
matches, then it is difficult to think the club would put significant weight on the
likelihood of qualifying for finals.
12. Appendix Table 1 provides descriptive statistics for explanatory variables.
13. Checking of data on end of season club order, and on location of matches, was
done using Rodgers and Browne (1996) (for 1968 to 1995), various issues of the
Football Record and AFL Record (for 1996 to 2001), and AFL Match Archives at
http://afl.com.au/default.asp?pg=matcharchive (2002-05).
14. Hence the number of observations in 1986 is 264, and in 2005 is 352. The
number of clubs increased from 12 to 14 in 1987, from 14 to 15 in 1991, and from 15
to 16 in 1995.
15. Prior to round 1 all measures would be indeterminate, and prior to round 2 zero or
100%.
16. Appendix Table 2 reports these results.
26
17. Appendix Table 3 reports the probit model used to predict clubs’ winning
probabilities.
18. An F-test cannot reject the hypothesis that coefficients on the dummy variables for
eligibility for Special Assistance are equal across the time periods 1986-1993, 19941996 and 1997-2005 (at 1% level of significance).
19. Due to a small number of observations it is necessary to aggregate some estimated
effects for models (2) and (3) over the last six rounds of the season, rather than groups
of two rounds.
20. See Appendix Table 3.
21. See Appendix Table 4.
22. See Appendix Table 5.
23. Effects from increasing player ability on team performance could be positive and
increasing where, for example, there are external spill-over effects from a player to
other players on the team (Rosen, 1982). Also relevant is that for a club that has
profit maximization as its objective, there may be ‘superstar’ effects associated with
players of high absolute ability that would make the value of higher ability players
positive and increasing (see for example, Rosen, 1981, and Hausman and Leonard,
1997).
24. It is also important to recognise that a further implication is likely to be that the
relative difference in ability between the highest ability players drafted will be greater
in the AFL than NBA. So that where, for example, the effect of higher player ability
on team performance is roughly linear, this might imply that the value of a higher
draft choice will be greater in the AFL than NBA. To test these predictions we
generated random vectors of N(0,1) observations with an ‘NBA’ vector 1500
observations long, and an ‘AFL vector’ 100 observations long (to correspond to the
difference in size of populations in the US and Australia). Each vector was simulated
10,000 times, and an average of the top 5 order statistics for each vector calculated.
For the AFL these are 2.506, 2.148, 1.941, 1.804, 1.688; and for the NBA are 3.359,
3.076, 2.928, 2.825, 2.745. These findings are consistent with top choices in the NBA
draft having higher absolute ability, but a larger difference between the ability of top
draft choices in the AFL.
25. For example, between 1968 and 1985, clubs that would have had borderline
eligibility for Special Assistance between 1997 and 2005 were only predicted to win 4
matches, and out of the matches they were predicted to lose only won 11 matches.
27
References
Amor, M. and W. Griffiths (2004), ‘Modelling the behaviour and performance of
Australian football tipsters’, mimeo, University of Melbourne.
Australian Football League (1997), ‘Player rules’ mimeo.
Australian Football League (1999), ‘Competitiveness on the field and uncertainty of
outcomes: Equalisation strategies at the AFL’, mimeo.
Australian Football League (2003), ‘Player rules’, mimeo.
Baldson, E., L. Fong and M. Thayer (2007), ‘Corruption in college basketball?
Evidence of tanking in postseason conference tournaments’, Journal of Sports
Economics, 8, 19-38.
Berri, D., S. Brook, B. Frick, A. Fenn and R. Vicente-Mayoral (2005), ‘The short
supply of tall people: Competitive imbalance and the National Basketball
Association’, Journal of Economic Issues, 39, 1029-41.
Booth, R. (1996), ‘History of player recruitment, transfer and payment rules in the
Victorian and Australian football league’, Australian Society of Sports History
Bulletin, 26, 13-33.
Borland, J., R. Macdonald and E. Meisel (2007), ‘Adapting to a new market
environment: The case of the AFL player draft’, mimeo, University of Melbourne.
Broad, B. (2007), ‘Eliminate the tanking temptation: Malthouse’, AFL website,
http://www.afl.com.au/Season2007/News/NewsArticle/tabid/208/Default.aspx?newsI
d=48626.
Butler, S. and C. O’Donoghue (2007), ‘Tanks for nothing, Carlton’, accessed at:
http://www.realfooty.com.au/news/tans-for-nothingcarlton/2007/10/12/1191696179365.html on 15/10/2007.
Camerer, C. and R. Weber (1999), ‘The econometrics and behavioural economics of
escalation of commitment: a re-examination of Staw and Hoang’s NBA data’, Journal
of Economic Behavior and Organization, 39, 59-82.
Clarke, S. (2005), ‘Home advantage in the Australian football league’, Journal of
Sports Sciences, 23, 375-85.
Connolly, R. (2006), ‘Losing clubs prefer to focus on next season, The Age, August 8.
Dabscheck, B. (1975), ‘Sporting equality: Labour market versus product market
control’, Journal of Industrial Relations, 17, 174-90.
Dabscheck, B. (1989), ‘Abolishing transfer fees: The Victorian Football League’s
new employment rules’, Sporting Traditions, 6, 63-87.
28
Gleeson, M. (2007), ‘Fevola cans tanking suggestions’, The Age, August 6, Sport
page 3.
Hausman, J. and G. Leonard (1997), ‘Superstars in the National Basketball
Association: Economic value and policy’, Journal of Labor Economics, 15, 586-624.
Johnson, L. (2005), ‘No incentive in priority pick system’, The Age, July 28.
Judd, C. (2006), ‘Draft picks and nuclear weapons – The hidden link’, The Age,
August 6.
Le Grand, C. and M. Conn (2006), ‘Mothball policy a ‘no-brainer’’, The Australian,
August 8.
Lovett, M. (ed.) (2007), AFL Record Guide to Season 2007 (AFL).
Macdonald, R. and R. Booth (2007), “Around the grounds’: A comparative analysis
of football in Australia’, in B. Stewart (ed.) The Games Are Not the Same: The
Political Economy of Football in Australia (Melbourne University Publishing).
Massey, C. and R. Thaler (2005), ‘The loser’s curse: Overconfidence vs market
efficiency in the NFL draft’, mimeo, Fuqua School of Business, Duke University.
McMillan, J. (2002), Reinventing the Bazaar: A Natural History of Markets (W.W.
Norton).
Niall, J. (2005), ‘To the loser, the spoils’, The Age, August 9.
Rodgers, S. and A. Browne (1996), Every Game Ever Played: VFL/AFL Results
1897-1995 (??).
Rosen, S. (1981), ‘The economics of superstars’, American Economic Review, 71,
845-58.
Rosen, S. (1982), ‘Authority, control and the distribution of earnings’, Bell Journal of
Economics, 13, 311-23.
Roth, A. (2002), ‘The economist as engineer: Game theory, experimentation and
computation as tools for design economics’, Econometrica, 70, 1341-78.
Roy, A. (1951), ‘The distribution of earnings and of individual output’, Economic
Journal, 60, 489-505.
Staw, B. and H. Hoang (1995), ‘Sunk costs in the NBA: Why draft order affects
playing time and survival in professional basketball’, Administrative Science
Quarterly, 40, 474-94.
Stevens, M. (2007), ‘Five won’t survive; Fairer draft order would dispel taking talk’,
The Herald Sun, August 7, page 81.
29
Spurr, S. (2000), ‘The baseball draft: A study of the ability to find talent’, Journal of
Sports Economics, 1, 66-85.
Szymanski, S. (2003), ‘The economic design of sporting contests’, Journal of
Economic Literature, 41, 1137-87.
Taylor, B. and J. Trogdon (2002), ‘Losing to win: Tournament incentives in the
National Basketball Association’, Journal of Labor Economics, 20, 23-41.
Victorian Football League (1985), VFL Football: Establishing the Basis for Future
Success (VFL).
Wolfers, J. (2006), ‘Point shaving: Corruption in NCAA basketball’, American
Economic Review Papers and Proceedings, 96, 279-83.
30
Table 1: Variable definitions
Variable
win ijt
winpcijt
owinpcijt
marginpct
elimijt
oelimijt
papijt
opapijt
bpapijt
obpapijt
draw ijt
haground ijt
ahground ijt
hhground ijt
Definition
Dummy variable for whether club i wins its match in
round j in season t
Club i’s winning percentage in season t prior to round j
Winning percentage of opposing club prior to round j in
match played against club i in season t
(Difference in points between club i and its opponent in
match played in round j in season t)/(Average points of
club i and its opponent in match played in round j in
season t)
Dummy variable for where club i has been eliminated
from contention for finals (playoffs) prior to its match in
round j in season t
Dummy variable for where opponent of club i in match
in round j in season t has been eliminated from
contention for finals (playoffs) prior to that match
Dummy variable for where club i is eligible for a
Special Assistance draft choice prior to its match in
round j in season t. Clubs that have won less than 5
matches prior to the current match are defined as
eligible.
Dummy variable for where club that plays match against
club i in round j in season t is eligible for a Special
Assistance draft choice prior to that match. Clubs that
have won less than 5 matches prior to the current match
are defined as eligible.
Dummy variable for where club i is eligible for a
Special Assistance draft choice prior to its match in
round j in season t, and if it wins this match will be
ineligible for Special Assistance draft choice. Clubs that
have won 4 or 4.5 matches prior to the current match are
defined as ‘borderline’.
Dummy variable for where club that plays match against
club i in round j in season t is eligible for a Special
Assistance draft choice prior to that match, and if it wins
this match will be ineligible for Special Assistance draft
choice. Clubs that have won 4 or 4.5 matches prior to
the current match are defined as ‘borderline’.
Dummy variable for whether match in round j in season
t played by club i was drawn
Dummy variable for where club i is playing its match in
round j in season t at its ‘home’ ground, and for its
opponent the match is at an away ground
Dummy variable for where club i is playing its match in
round j in season t at an ‘away’ ground, and for its
opponent the match is at its ‘home’ ground
Dummy variable for where club i and its opponent are
31
aaground ijt
hastateijt
ahstateijt
hhstateijt
aastateijt
elimxyijt
elimxyjj-kk ijt
elimoxyijt
elimoxyjj-kk ijt
papxyijt
papxyjj-kk ijt
papoxyijt
papoxyjj-kk ijt
bpapxyijt
playing the match in round j in season t at a ‘home’
ground for both clubs
Dummy variable for where club i and its opponent are
playing the match in round j in season t at an ‘away’
ground for both clubs
Dummy variable for where club i is playing its match in
round j in season t in its ‘home’ state, and for its
opponent the match is inter-state
Dummy variable for where club i is playing its match in
round j in season t inter-state, and for its opponent the
match is in its ‘home’ state
Dummy variable for where club i and its opponent are
playing the match in round j in season t in the ‘home’
state for both clubs
Dummy variable for where both club i and its opponent
are playing the match in round j in season t inter-state
Dummy variable for where club i has been eliminated
from contention for finals prior to its match in the last x
to y rounds in season t
Dummy variable for where club i has been eliminated
from contention for finals prior to its match in the last x
to y rounds in seasons between 19jj and 19(20)kk
Dummy variable for where club i, which has been
eliminated from contention for finals prior to this match
in the last x to y rounds in season t, is playing against a
club that has also been eliminated from contention for
the finals
Dummy variable for where club i, which has been
eliminated from contention for finals prior to this mtach
in the last x to y rounds in seasons between 19jj and
19(20)kk, is playing against a club that has also been
eliminated from contention for the finals
Dummy variable for where club i is eligible for a
Special Assistance draft choice prior to its match in the
last x to y rounds in season t
Dummy variable for where club i is eligible for a
Special Assistance draft choice prior to its match in the
last x to y rounds in seasons between 19jj and 19(20)kk
Dummy variable for where club i, which is eligible for a
Special Assistance draft choice prior to this match in the
last x to y rounds in season t, is playing against a club
that is also eligible for Special Assistance
Dummy variable for where club i, which is eligible for a
Special Assistance draft choice prior to its match in the
last x to y rounds in seasons between 19jj and 19(20)kk,
is playing against a club that is also eligible for Special
Assistance
Dummy variable for where club i is eligible for a
Special Assistance draft choice prior to its match in the
32
bpapxyjj-kk ijt
bpapoxyijt
bpapoxyjj-kk ijt
last x to y rounds in season t, and if it wins its match
will be ineligible for Special Assistance draft choice.
Dummy variable for where club i is eligible for a
Special Assistance draft choice prior to its match in the
last x to y rounds in seasons between 19jj and 19(20)kk,
and if it wins its match will be ineligible for Special
Assistance draft choice.
Dummy variable for where club i, which is eligible for a
Special Assistance draft choice prior to this match in the
last x to y rounds in season t, and if it wins its match
will be ineligible for Special Assistance draft choice, is
playing against another club which also has borderline
eligibility for Special Assistance
Dummy variable for where club i, which is eligible for a
Special Assistance draft choice prior to this match in the
last x to y rounds in seasons between 19jj and 19(20)kk,
and if it wins its match will be ineligible for Special
Assistance draft choice is playing against another club
which also has borderline eligibility for Special
Assistance.
33
Table 2: Effect of National Reverse-Order Draft on probability of winning a
match, AFL regular season, 1968-2005
elim12
elim34
elim56
elim1286-05
elim3486-05
elim5686-05
oelim
(1)
-0.272
(0.000)
-0.291
(0.000)
-0.249
(0.000)
0.050
(0.279)
0.061
(0.232)
0.005
(0.920)
0.264
(0.000)
(2)
-0.303
(0.000)
-0.337
(0.000)
-0.247
(0.000)
0.112
(0.111)
0.132
(0.070)
0.008
(0.908)
0.261
(0.000)
0.063
(0.380)
0.112
(0.147)
-0.005
(0.951)
-0.112
(0.231)
-0.133
(0.203)
-0.005
(0.948)
(3)
-0.178
(0.002)
-0.222
(0.000)
-0.115
(0.026)
0.079
(0.279)
0.105
(0.163)
0.007
(0.923)
0.136
(0.000)
0.050
(0.497)
0.100
(0.210)
-0.013
(0.885)
-0.080
(0.409)
-0.108
(0.312)
-0.003
(0.977)
0.476
(0.000)
-0.446
(0.000)
0.086
(0.000)
-0.080
(0.000)
0.005
(0.829)
0.081
(0.150)
-0.071
(0.205)
0.002
(0.961)
0.034
-6918.9
0.035
-6917.0
0.109
-6384.5
elimo12
elimo34
elimo56
elimo1286-05
elimo3486-05
elimo5686-05
winpc
owinpc
haground
ahground
hhground
hastate
ahstate
hhstate
Pseudo R-squared
Log likelihood
34
Observations
10,342
10,342
10,342
Note: All estimates reported are marginal effects from probit models. Marginal
effects are the change in the probability of winning a match of a change in a dummy
variable from zero to one, or for a 1-unit change in a continuous variable. p-values are
in parentheses.
35
Table 3: Data on percentage of games won/lost against predictions by teams
eliminated from finals in last 6 rounds of season, 1968-85 and 1986-2005
1968-1985
7/12 (58%)
Games
won/Games
predicted to win
Games lost/Games 249/306 (81.3%)
predicted to lose
1986-2005
22/44 (50%)
Significant
difference in
proportions (pvalue)
0.608
270/354 (76.2%)
0.110
36
Table 4: Effect of Special Assistance rules on the probability of winning a match
and on winning margin, AFL regular season, 1968-2005
Probability
of winning
match
Winning
margin
(Points
difference)
OLS
Probit –
Marginal
effects
(1)
(2)
-0.002
(0.984)
-0.095
(0.402)
0.005
(0.957)
0.016
(0.872)
-0.014
(0.902)
-0.103
(0.379)
-0.019
(0.841)
-0.011
(0.914)
Constant
pap1297-05
pap3497-05
pap5697-05
pap1694-96
pap1294-96
pap3494-96
pap5694-96
pap1686-93
-0.028
(0.689)
pap3486-93
pap5686-93
pap34
pap56
opap
winpc
owinpc
haground
-0.300
(0.000)
-0.261
(0.000)
-0.263
(0.000)
0.102
(0.000)
(4)
-0.009
(0.843)
-0.244
(0.004)
-0.019
(0.812)
-0.082
(0.163)
-0.327
(0.048)
0.021
(0.868)
-0.113
(0.308)
-0.342
(0.024)
-0.049
(0.676)
-0.105
(0.305)
-0.198
(0.045)
-0.014
(0.865)
0.020
(0.792)
-0.194
(0.000
-0.252
(0.000)
-0.205
(0.000)
0.107
(0.000)
-0.191
(0.036)
-0.053
(0.507)
0.012
(0.860)
-0.038
(0.450)
-0.095
(0.041)
-0.062
(0.152)
0.028
(0.002)
0.469
(0.000)
-0.467
(0.000)
0.075
(0.000)
-0.051
(0.474)
pap1286-93
pap12
(3)
-0.044
(0.000)
-0.232
(0.011)
-0.068
(0.908)
-0.327
(0.393)
-0.178
(0.006)
-0.139
(0.017)
-0.135
(0.013)
0.028
(0.015)
0.492
(0.000)
-0.501
(0.000)
0.086
(0.000)
37
ahground
0.245
(0.063)
0.207
(0.065)
0.208
(0.047)
0.028
(0.847)
-0.017
(0.932)
0.040
(0.790)
-0.079
(0.000)
0.003
(0.876)
0.083
(0.138)
-0.070
(0.211)
0.005
(0.916)
0.147
(0.293)
0.106
(0.369)
0.117
(0.286)
0.052
(0.725)
0.007
(0.971)
0.060
(0.697)
0.021
-7017.6
0.106
-6408.5
hhground
hastate
ahstate
hhstate
papo12
papo34
papo56
papo1686-94
papo1694-96
papo1697-05
Pseudo R-squared
Log likelihood
Adjusted R-squared
Observations
10,342
10,342
0.198
(0.106)
0.141
(0.165)
0.142
(0.134)
0.048
(0.698)
0.078
(0.663)
0.072
(0.557)
-0.076
(0.000)
0.001
(0.936)
0.082
(0.062)
-0.082
(0.062)
-0.001
(0.966)
0.081
(0.465)
0.035
(0.705)
0.045
(0.607)
0.062
(0.586)
0.098
(0.556)
0.082
(0.495)
0.034
10,342
0.185
10342
Note: Estimates reported in columns (1) and (2) are marginal effects from probit
models. Marginal effects are the change in the probability of winning a match of a
change in a dummy variable from zero to one, or for a 1-unit change in a continuous
variable. Estimates reported in columns (3) and (4) are from an OLS model. In
columns (1) and (2) p-values are in parentheses. In columns (3) and (4) standard
errors are in parentheses.
38
Table 5: Effect of Special Assistance rules on the probability of winning a match
– ‘Borderline’ teams, AFL regular season, 1968-2005
Constant
bpap1697-05
bpap1694-96
bpap1686-93
bpap12
bpap34
bpap56
obpap
bpapo1686-96
bpapo1697-05
bpapo16
(1)
(2)
(3)
(4)
-0.105
(0.417)
0.112
(0.439)
0.045
(0.698)
-0.333
(0.000)
-0.174
(0.063)
-0.289
(0.001)
0.013
(0.449)
0.022
(0.959)
0.104
(0.837)
0.154
(0.665)
-0.116
(0.387)
0.081
(0.596)
0.002
(0.981)
-0.251
(0.014)
-0.056
(0.567)
-0.166
(0.078)
0.025
(0.188)
0.071
(0.871)
0.116
(0.821)
0.023
(0.950)
-0.144
(0.326)
0.127
(0.402)
0.063
(0.609)
-0.322
(0.001)
-0.157
(0.131)
-0.271
(0.004)
-0.046
(0.014)
0.022
(0.967)
0.057
(0.924)
-0.099
(0.821)
0.127
(0.000)
0.435
(0.135)
0.223
(0.408)
0.185
(0.518)
-0.141
(0.668)
0.089
(0.811)
-0.162
(0.292)
0.100
(0.524)
0.023
(0.854)
-0.267
(0.015)
-0.050
(0.647)
-0.169
(0.096)
0.006
(0.757)
0.067
(0.903)
-0.000
(0.999)
-0.074
(0.869)
0.037
(0.002)
0.386
(0.234)
0.099
(0.724)
0.122
(0.677)
-0.140
(0.672)
0.162
(0.653)
0.524
(0.000)
-0.499
(0.000)
0.086
(0.000)
-0.079
(0.000)
0.004
(0.869)
0.081
(0.151)
opap
papbo12
papbo34
papbo56
papbo1686-96
papbo1697-05
winpc
owinpc
haground
ahground
hhground
hastate
0.522
(0.000)
-0.535
(0.000)
0.085
(0.000)
-0.079
(0.000)
0.002
(0.906)
0.083
(0.137)
39
ahstate
-0.070
(0.212)
0.006
(0.911)
hhstate
Pseudo R-squared
Log likelihood
Observations
0.003
-7144.5
10,342
0.103
-6424.5
10,342
-0.073
(0.194)
0.003
(0.954)
0.014
-7063.7
10,342
0.104
-6418.0
10,342
Note: Estimates reported in columns (1) and (2) are marginal effects from probit
models. Marginal effects are the change in the probability of winning a match of a
change in a dummy variable from zero to one, or for a 1-unit change in a continuous
variable. Estimates reported in columns (3) and (4) are from an OLS model. In
columns (1) and (2) p-values are in parentheses. In columns (3) and (4) standard
errors are in parentheses.
40
Table 6: Data on percentage of games won/lost against predictions in last 6
rounds by teams eligible for Special Assistance draft choices in matches where
are ‘borderline’ and not playing against another team that is ‘borderline’, 196885 and 1997-2005
Games
won/Games
predicted to
win
Games
lost/Games
predicted to
lose
1968-1985
19861996
1997-2005
3/4 (75%)
4/8
(50%)
1/3
(33.3%)
45/56
(80.3%)
23/31
(74.1%)
24/48
(85.7%)
Significant
difference in
proportions
(p-value)
1968-85
against 19972005
Significant
difference
in
proportions
(p-value)
1986-96
against
1997-2005
0.545
0.272
Table 7: Predictive power of the model for determinants of winning matches
Difference in predicted
probability of winning
Number of
observations
0 to 0.1
Greater than 0.1 to 0.2
Greater than 0.2 to 0.3
Greater than 0.3 to 0.4
Greater than 0.4 to 0.5
Greater than 0.5 to 0.6
Greater than 0.6 to 0.7
Greater than 0.7 to 0.8
Greater than 0.8 to 0.9
Greater than 0.9 to 1
Total
1001
1254
712
994
649
469
288
232
62
20
5671
Proportion
correct
prediction of
winner
0.517
0.558
0.652
0.679
0.718
0.761
0.813
0.772
0.790
0.750
0.643
41
Figure 1: Comparison of distribution of matches won by season,
AFL regular season, 1997-2005 and 1977-1985
Number of club/season observations
25
20
15
1997-2005
1977-1985
10
5
0
21.5/22
20.5/21
19.5/20
18.5/19
17.5/18
16.5/17
15.5/16
14.5/15
13.5/14
12.5/13
11.5/12
10.5/11
9.5/10
8.5/9
7.5/8
6.5/7
5.5/6
4.5/5
3.5/4
2.5/3
1.5/2
0.5/1
0
Number of games won by season
42
Table 8: Determinants of coach turnover - NBA and AFL
Dependent variable: Separation of coach from club in season t
Club winning percentage
in season t
Club misses playoffs in
season t
AFL
-0.755
(0.169)
0.073
(0.064)
NBA
-1.058
(0.263)
-0.024
(0.076)
Sample period
Observations
Log likelihood
Pseudo R-squared
1968-2005
515
-275.84
0.117
1977-1995
454
-286.06
0.063
Note: Coefficient estimates are marginal effects on the probability that a coach
separates from a club in season t estimated from a probit model. Estimates show the
effect of a 1 percentage point change in winning percentage, and a change from 0 to 1
in whether a club misses the playoffs. Standard errors are shown in parentheses.
Data sources: AFL data from Lovett (2007). NBA data from
www.database.basketball.com.
1
Appendix Table 1: Descriptive statistics, 1968 to 2005
VARIABLE
win
winpc
owinpc
marginpct
elim
oelim
pap
opap
draw
haground
ahground
hhground
aaground
hastate
ahstate
hhstate
aastate
elim12
elim34
elim56
elim1286-05
elim3486-05
elim5686-05
pap12
pap34
pap56
pap1297-05
pap3497-05
pap5697-05
pap1294-96
pap3494-96
pap5694-96
pap1286-93
pap3486-93
pap5686-93
bpap12
bpap34
bpap56
STANDARD
MEAN DEVIATION
0.504
0.500
0.499
0.245
0.499
0.245
0.000
0.224
0.146
0.353
0.146
0.353
0.571
0.495
0.571
0.495
0.008
0.092
0.404
0.490
0.404
0.490
0.065
0.246
0.125
0.331
0.156
0.363
0.156
0.363
0.678
0.467
0.008
0.092
0.049
0.215
0.040
0.196
0.031
0.173
0.025
0.158
0.020
0.142
0.015
0.122
0.014
0.117
0.017
0.132
0.021
0.146
0.003
0.058
0.003
0.062
0.004
0.069
0.000
0.027
0.001
0.039
0.002
0.047
0.002
0.053
0.003
0.062
0.004
0.070
0.003
0.062
0.004
0.070
0.005
0.077
MIN
0.000
0.000
0.000
-0.797
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
MAX
1.000
1.000
1.000
0.797
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
RANGE
1.000
1.000
1.000
1.794
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
2
bpap1697-05
bpap1694-96
bpap1686-93
obpap
bpapo16
bpapo1686-96
bpapo1697-05
papbo12
papbo34
papbo56
papbo1686-94
papbo1697-05
0.003
0.001
0.003
0.078
0.000
0.000
0.000
0.000
0.001
0.000
0.001
0.000
0.056
0.040
0.062
0.268
0.024
0.013
0.031
0.017
0.038
0.026
0.034
0.024
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
3
Appendix Table 2: Effect of National Reverse-Order Draft on probability of
winning a match, AFL regular season, Disaggregated time periods, 1968-2005
elim12
elim34
elim56
elim1286-90
elim3486-90
elim5686-90
elim1291-95
elim3491-95
elim5691-95
elim1296-00
elim3496-00
elim5696-00
elim1201-05
elim3401-05
elim5601-05
oelim
elimo12
elimo34
elimo56
elimo1286-05
elimo3486-05
elimo5686-05
winpc
(1)
-0.273
(0.000)
-0.291
(0.000)
-0.249
(0.000)
0.009
(0.888)
0.074
(0.323)
0.034
(0.682)
0.065
(0.363)
0.051
(0.525)
-0.051
(0.605)
0.116
(0.100)
0.059
(0.479)
0.000
(0.995)
0.057
(0.905)
0.020
(0.483)
0.020
(0.822)
0.264
(0.000)
(2)
-0.178
(0.002)
-0.222
(0.000)
-0.115
(0.026)
0.041
(0.658)
0.124
(0.192)
0.016
(0.857)
0.074
(0.425)
0.086
(0.391)
-0.026
(0.814)
0.131
(0.138)
0.097
(0.320)
0.013
(0.915)
0.051
(0.582)
0.112
(0.261))
0.017
(0.867)
0.136
(0.000)
0.050
(0.497)
0.100
(0.210)
-0.013
(0.885)
-0.070
(0.471)
-0.111
(0.305)
-0.002
(0.985)
0.476
4
(0.000)
-0.445
(0.000)
0.085
(0.000)
-0.080
(0.000)
0.004
(0.840)
0.080
(0.152)
-0.072
(0.201)
0.002
(0.969)
owinpc
haground
ahground
hhground
hastate
ahstate
hhstate
Pseudo R-squared
Log likelihood
Observations
0.035
-6917.6
10,342
0.109
-6383.7
10,342
Note: All estimates reported are marginal effects from probit models. Marginal
effects are the change in the probability of winning a match of a change in a dummy
variable from zero to one, or for a 1-unit change in a continuous variable. p-values are
in parentheses.
5
Appendix Table 3: Effect of Special Assistance rules on the likelihood of winning
a match, AFL regular season, 1968-2005
Effect on
probability of
winning match
pap1297-05
pap3497-05
pap5697-05
pap1694-96
pap1686-93
pap12
pap34
pap56
opap
winpc
owinpc
haground
ahground
hhground
hastate
ahstate
hhstate
Pseudo R-squared
Log likelihood
Observations
-0.025
(0.822)
-0.066
(0.523)
-0.003
(0.966)
-0.004
(0.956)
-0.032
(0.601)
-0.151
(0.010)
-0.116
(0.025)
-0.110
(0.022)
0.032
(0.005)
0.493
(0.000)
-0.505
(0.000)
0.086
(0.000)
-0.079
(0.000)
0.004
(0.849)
0.084
(0.136)
-0.070
(0.214)
0.006
(0.906)
0.105
-6411.7
10,342
Note: Estimates reported are marginal effects from probit models. p-values are in
parentheses.
6
Appendix Table 4: Effect of number in clubs in competition and proportion of
clubs included in finals series on clubs eliminated from participating in finals
series and eligible for Special Assistance, Final 6 rounds, 1968-2005
Explanatory
variables
Number of
Proportion Constant Adjusted Observations
Rteams in
of teams
squared
competition making
finals
Dependent Sample
variable
0.002
Proportion All
seasons (0.849)
of games
won in last
6 rounds
by teams
eliminated
from finals
series
Draft
0.041
era
(0.388)
All
Total
games won seasons
in last 6
rounds by
clubs with
borderline
eligibility
for Special
Assistance
Draft
era
-0.357
(0.261)
0.567
(0.000)
-0.002
38
-0.609
(0.324)
0.077
(0.872)
-0.053
20
0.067
(0.085)
-1.359
(0.237)
-0.026
(0.953)
0.038
31
0.276
(0.251)
-3.947
(0.204)
-2.061
(0.402)
-0.014
18
Note: There are 7 seasons in which no matches were played by clubs with borderline
eligibility for Special Assistance in the last 6 rounds of the season. Hence there are
38 observations (1968-1985) for analysis of the determinants of the average winning
percentage of clubs eliminated from finals, but only 31 observations for analysis of
the winning percentage of clubs with borderline eligibility for Special Assistance.
7
Appendix Table 5: NBA First Round Draft Picks: Average number of games
played and average number of minutes played as a proportion of total possible
games and minutes
a.1984
Season after being drafted
Proportion of games
played
Proportion of minutes
played
1
2
3
4
5
6
7
0.827
0.843
0.890
0.841
0.844
0.848
0.827
0.370
0.424
0.485
0.563
0.559
0.553
0.562
1
2
3
4
5
6
7
0.837
0.853
0.953
1.021
0.873
0.980
0.909
0.333
0.446
0.515
0.573
0.520
0.554
0.651
1
2
3
4
5
6
7
0.793
0.759
0.775
0.731
0.777
0.722
0.801
0.331
0.358
0.370
0.399
0.419
0.394
0.455
b. 1985
Season after being drafted
Proportion of games
played
Proportion of minutes
played
c. 1990
Season after being drafted
Proportion of games
played
Proportion of minutes
played
Source: Data from www.database.basketball.com