Article

Journal of Sports Economics

2015, Vol. 16(7) 760-783
ª The Author(s) 2013
20-Over Versus Reprints and permission:
sagepub.com/journalsPermissions.nav
50-Over Cricket: DOI: 10.1177/1527002513505284
jse.sagepub.com
Is There a Difference?

Colin Cannonier1, Bibhudutta Panda2,

and Sudipta Sarangi3

Abstract
Using 2008-2009 data, we identify winning input combinations for the game of cricket in
two different formats: 50-over one-day internationals and 20-over games from
Twenty20 internationals and the Indian Premier League. We find that attacking batting
and bowling are the best determinants of the winning probability in both formats in an
international setting despite their 30-over difference. Interestingly, attacking batting and
defensive bowling are the optimal input combinations in the league version of the 20-
over format. Finally, we speculate about their implications for the future of cricket,
especially for the popularity of formats, choices of players, and player development.

Keywords
cricket, match results, IPL, ODI, T20I

Introduction
The game of cricket has gone through many transformations to keep up with the
changing times. Modern society, as reflected by attendance at games, has been
creating pressure to adopt shorter, result-oriented versions of the game. The conven-
tional form of five-day Test matches was complimented by the introduction of

1
College of Business Administration, Belmont University, Nashville, TN, USA
2
Division of Social Sciences, University of Minnesota, Morris, MN, USA
3
Department of Economics, Louisiana State University, Baton Rouge, LA, USA

Corresponding Author:
Sudipta Sarangi, Department of Economics, Louisiana State University, Baton Rouge, LA 70803, USA.
Email: [email protected]

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
Cannonier et al. 761

limited-over one-day cricket in the 1960s. More than 40 years later in 2003, an even
shorter 20-over format called Twenty20 (T20) was introduced in order to revive
spectator interest in the game. This version has not only transformed the game by
introducing a fast-paced and TV-friendly format but also created new business inter-
ests in the game by attracting millions of spectators. Though several formats coexist
today, ranging from the traditional five-day Test match to the one-evening T20 both
at international and domestic levels of cricket, there has been an increasing demand
for the T20 format from the time of its emergence.
The objective of our article is to compare the 50-over one-day cricket with the
shorter version T20. We will examine the factors that determine a team’s ability
to win matches in these two different formats, given that there is a difference of
30 overs between them. As Rumford (2011a) states, the T20 is often considered
to be more formulaic than the 50-over version with relatively little scope for strategy
implementation. In fact, he observes that the traditional ‘‘bat versus ball’’ contest has
been reduced to an attacking ‘‘six-hitting’’ batting versus a defensive ‘‘in-swinging-
yorker’’ bowling contest with the introduction of T20 cricket. Clearly, the veracity of
such observations needs to be examined.
To study these issues, we use a production function approach that determines the
outcome of a cricket match as a function of batting, bowling, fielding, and other rel-
evant inputs. Our utilized data set covers 50-over one-day international (ODI)
matches, 20-over international (T20I) matches, and matches from the Indian Premier
League (IPL). We use conditional-logit models to determine the (relative) impor-
tance of batting and bowling inputs for winning matches. This will be used to iden-
tify the optimal combination of winning inputs in both formats.
The rest of the article is structured as follows: The second section briefly reviews
the history, nature, and rules pertaining to the limited-over cricket. The third section
reviews the relevant literature. The fourth section presents the empirical framework.
The fifth section describes the data sources and construction of variables. The sixth
section discusses the empirical results, and the seventh section concludes and pro-
vides implications for the future of cricket.

Brief History of Limited-Over Cricket

The limited-over one-day cricket game was first introduced in the domestic cricket-
ing circles of England in the 1960s.1 This variant was specifically introduced to
revive the popular appeal of the game in England by providing result-orientated and
fast-paced cricket in contrast to conventional first-class cricket which was prone to
being inconclusive. The format became popular very quickly and was adopted across
the cricketing nations. The internationalization of this format began with the first
ever ODI game played between England and Australia on January 5, 1971, at the
Melbourne Cricket Ground in Australia, which was won by Australia. This format
was further popularized with the introduction of the World Cup cricket in 1975.

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
762 Journal of Sports Economics 16(7)

While the last three decades have witnessed a significant increase in popularity
and growth in limited-over cricket, this period coincided with a fall in popularity
of the game at the domestic level in the cricketing nations. This was particularly true
for county cricket in England. In an attempt to revive the mass appeal of the game,
the England and Wales Cricket Board introduced an even shorter 20-over version
of the game in its domestic circuits in 2003. This format was introduced to attract
the traditional spectators who were constrained with time, as well as non-traditional
spectators like young and women. The format became an immediate success, largely
owing to its non-stop action over a three-hour period in compared to the whole-day
commitment associated with the longer fifty-over version. Other test-playing nations
quickly followed suit.2 The formal introduction of this format at the international
level occurred on February 17, 2005 when Australia played and defeated the host
New Zealand in a match at Eden Park, Auckland. With the growing demand for this
format, the International Cricket Council (ICC) organized the first T20 International
World Cup in 2007, which added further momentum to the popularity of this format.
Following India’s victory in the inaugural T20 International World Cup, the
Board of Control for Cricket in India decided to establish a T20 cricket league in
India known as the Indian Premier League (IPL) along the lines of soccer’s English
Premier League. The aim was to establish a global league where teams were allowed
to draw from a pool of international players while still maintaining a local identity by
tying the team to a major city and associating an iconic Indian player to it (Gupta,
2011).3 Each team was allowed to choose players from a pool of international and
regional players through auctions with a cap on the amount a team could spend
on player acquisition.4 IPL rules permit a team to include a maximum of 4 overseas
players in its playing 11, while the remaining players must be drawn from India.
Teams in the IPL are owned by wealthy Indian businessmen and Bollywood stars
which adds further local flavor and glamor to the event. On an important note, the
financial payoff from a few weeks of commitment makes this league very lucrative
for international players.5 The inaugural competition took place over 2 months
(April–May) in 2008 between eight teams competing on a ‘‘home’’ and ‘‘away’’
basis.6 The commercial viability of IPL through advertising, media campaigns, sat-
ellite and TV rights, and ticket sales in the presence of the South Asian market con-
sisting of 1.5 billion people has made it the most successful and cash-rich league in
the world.7
Earlier in the introduction, we have already noted that this article uses data from
50-over ODI matches and T20 games consisting of T20I matches and matches from
the IPL. The objective of using international games is fairly straightforward. Players
in the international format have years of international and domestic cricketing
experience and are trained by long-term strategists and coaches. To obtain insights
about how the optimal input combination changes once we change the format, we
need to focus on games involving such players. Our choice of a club-level league
was guided by the fact that IPL has emerged as the most successful T20 league, pool-
ing players from across the globe and is completely different in its setup in contrast

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
Cannonier et al. 763

to international cricket. Given the huge financial incentives in IPL, it would be inter-
esting to see if the optimal input combination changes once we move from interna-
tional cricket to league cricket using this significantly successful league.

Differences in Rules
Given the limited-over nature of both ODI and T20, the rules are similar in many
respects. However, some rules have been modified in T20 to create fast-paced and
TV-friendly cricket within a three-hour time frame.8 Obviously, as opposed to 50
overs in ODI, in T20 each side gets 20 overs to bat with the highest scoring team
deemed the winner. Similarly, each bowler is capped to bowl a maximum of 4 overs
in T20 as opposed to 10 overs in ODI. The rules have been modified in T20 to pro-
vide additional advantage to the batsmen. For example, a bowler who oversteps the
crease in ODI is considered to have bowled a ‘‘no-ball’’, resulting in an additional
run being awarded to the batting team. However, in the case of a ‘‘no-ball’’ in
T20, two additional runs are awarded to the batting team with the next ball bowled
being considered as a ‘‘free-hit’’ for batsman who can only be dismissed via run-out.
Another significant difference in the rules relates to time-wasting in T20. Owing to
the three-hour time constraint, umpires are authorized to penalize a bowling team if
it is found guilty of wasting time by awarding five additional runs to the batting
team. To make the game more result oriented, in case of a tie in T20, the winner
of the game is decided through a one-over per side ‘‘super-over’’ where the team
scoring the most runs in the ‘‘super-over’’ is declared the winner.

Literature Review

We now provide a brief overview of the relevant literature starting with the produc-
tion function approach in cricket. Schofield (1988) in his seminal article estimates
the outcome of two different formats in English county cricket: the three-day County
Championship and the limited-over (40-over per innings) John Player League for the
1981-1983 seasons. The author finds attacking batting measured as runs scored per
over (RPO) has greater importance in both formats, while the strength of bowling
inputs differs across formats. Attacking bowling measured as overs bowled per
wicket taken and wickets taken per game has greater importance in the three-day for-
mat, while defensive bowling measured as runs conceded per over emerges as a key
factor for winning limited-over games. Bairam, Howells, and Turner (1990a) extend
this approach to include the longer format of first-class cricket matches played in
Australia and New Zealand. Using a Box-Cox general transformation function, they
find attacking batting and attacking bowling as the best input combination to max-
imize the probability of winning in New Zealand cricket. By contrast, winning prob-
abilities are maximized by using a mix of attacking batting and defensive bowling in
Australia. In a subsequent article, Bairam, Howells, and Turner (1990b) compare the

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
764 Journal of Sports Economics 16(7)

optimal input combinations across the three-day and one-day matches played in the
domestic cricketing circles in New Zealand. Their findings suggest that while the
optimal mix is attacking batting and attacking bowling for the three-day matches,
the optimal mix is attacking batting and defensive bowling in the limited-over
one-day cricket. This echoes the conclusion made by Schofield (1988). Brooks, Faff,
and Sokulsky (2002) apply the production function approach to the longer format
Test cricket. Methodologically, their article differs from the previous articles, as it
uses an ordered response model to accommodate the presence of a qualitative depen-
dent variable in contrast to the ordinary least squares (OLS) used in other articles.
They find average batting and average bowling intents implying a mix of both
attacking and defensive intents to be the optimal mix for most test-playing nations,
with the model correctly predicting 71% of the match outcomes.9
Besides batting and bowling inputs, researchers have also studied the impact of
other important factors such as toss and home-team bias on match results in cricket.
de Silva and Swartz (1997) provide evidence of significant home team advantage in
ODI, but fail to find similar evidence for the effect of toss advantage. Allsopp and
Clarke (2004) provide similar conclusions for ODI and five-day Test matches.10 On
the other hand, Morley and Thomas (2005) find significant positive effects on match
outcomes in English one-day cricket conditional on the home team winning the toss,
but such effects disappear after controlling for other variables such as team quality
and match importance. Bhaskar (2009) shows significant advantage for a team that
wins the toss and opts to bat first in a day–night ODI match. On the other hand, he
finds significant disadvantage for a team which opts for a similar strategy in a day
match. Dawson, Morley, Paton, and Thomas (2009) further corroborate this finding
and show that the toss outcome combined with the decision to bat first increases the
probability of winning by 31% in day–night ODI matches.
To our knowledge, there are only a limited number of studies examining T20
matches. Lemmer (2011) using data on 2009 IPL matches asserts that traditional bat-
ting average is not an appropriate measure to compare batsmen’s performance, given
the different batting conditions. Saikia, Bhattacharjee, and Bhattacharjee (2012) uti-
lize data on IPL and T20 World Cup matches for 2009 and 2010 to evaluate crick-
eters’ performances and conclude that the poor performance of players in T20 World
Cup is uncorrelated with their IPL performance.

Empirical Framework

The contribution of batting, bowling, and fielding inputs to a team’s ability to win
games within the production function approach has been widely explored in the
sports economics literature.11 This approach expresses a team’s winning ability as
a function of the inputs shown below:
Win ¼ f ðBatting; Bowling; FieldingÞ: ð1Þ

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
Cannonier et al. 765

Observe that for the game of cricket each of these inputs can be measured in differ-
ent ways. In this article, we have created two different categories for the first two
inputs of Equation 1: attacking and defensive. These two categories can be thought
of as capturing the intent of the teams. Fielding is ascribed only to one category since
intent is not really a choice in the context of fielding.
Each input measure depends on a set of observable and unobservable factors.
Observable factors include toss outcomes, home-team bias, and weather conditions.
For example, although the toss outcome is merely decided by flipping a coin with a
50% probability of winning for each team and does not directly affect any input mea-
sures, it confers a comparative advantage to the team winning the toss in choosing its
input measures given the pitch, outfield, and weather conditions. Similarly, playing
at home can provide a boost to a team’s input measures, given the familiarity with
pitch conditions and home crowd support. Examples of unobservable factors include
player ability and form, captaincy skills, coaching skills, and team management
skills. While ability and form is player-specific, coaching, captaincy, and manage-
ment skills tend to be team-specific and may be thought as the team’s think tank that
devises its strategies for a given match.
The fielding variable has been ignored by most of the studies mentioned above.
However, we believe that it plays a significant role and there are several reasons to
consider its inclusion. While bowling measures may or may not affect fielding mea-
sures, it is definitely true that measures of bowling performance depend to a great
extent on fielding performance. Further, in the shorter format of the game, run rate
is key to winning a match, making fielding a crucial input. Since one of our goals is
to determine whether or not a good fielding side increases its chances of winning, we
include this variable in our empirical model.12
The literature using the production function approach for the game is divided over
the choice of the empirical methodology for estimating the model discussed above.
The difference in opinion revolves around the construction of the dependent variable
that in part is dictated by the nature of the data. Schofield (1988) and Bairam et al.
(1990a, 1990b) focus on first-class cricket where teams play each other on a seasonal
basis, and present the output and input measures relative to seasonal averages. Since
the output variable (expressed as a percentage of matches won or points scored) is a
continuous one, they estimate the model using OLS. On the other hand, Brooks et al.
(2002), who use data from international Test cricket where teams do not play each
other on a seasonal basis, estimate an ordered response model in which the depen-
dent variable is a categorical variable taking values based on win, loss, or draw. In
our data set, though the IPL tournament is played on a seasonal basis, the ODI and
T20I competitions are scheduled in advance by the ICC’s Future Tours Program.
Hence, we opt for a logit model using a binary dependent variable based on the win
or loss outcome of the match.
The latent variable model can be expressed as
yj ¼ xj b þ ej ; ð2Þ

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
766 Journal of Sports Economics 16(7)

2
where ej has a standard logistic distribution independent of xj , with ej  ð0; p3 Þ, and
y is the unobserved latent variable indicating a team’s probability of winning. The
vector of explanatory variables comprising batting, bowling, and fielding inputs
along with variables related to toss and home team advantage is represented by xj
while b is the matrix of parameters to be estimated. Given that y is unobserved,
what we really observe is whether a team has won (¼1) or not (¼0) where:
y ¼ 1 if y > 0 and y ¼ 0 if y  0:
So, given the explanatory variables, the logit model can be represented as

  e xj b

P yj¼1 jxj ¼ P yj > 0jxj ¼ x b
¼ L xj b ; ð3Þ
1þe j

where L(.) is the cumulative distribution function for a standard logistic

distribution.
Further, it should be noted that each match generates a pair of observations in the
data set: one for the winning team and the other for the losing team. This involves a
one-to-one matching between the two choices: win and loss. It stands to reason that
within each pair, the outcome of a match is linearly dependent and hence the error
terms are correlated. So, we estimate a ‘‘Conditional (Fixed-Effect) Logit’’ model
with ‘‘match’’ as the grouping variable that accounts for this correlation due to fixed
effects (different intercept across matches) in the model.
Before proceeding to the empirical section, we focus on two issues that warrant
further discussion. First, the article identifies the input measures that have the great-
est impact on team success. This not only identifies the critical inputs but also has
implications for both player selection and team composition. For example, the
importance of attacking batting over the defensive one suggests selecting an attack-
ing batsman over a defensive batsman and/or also requires adopting an attacking bat-
ting style. Second, it is possible that the selection of batting or bowling measures or
intent is also conditional on the choices made by the rival team, but at the time of
making its own choice a team may not be fully aware of the other team’s choice.
Hence, in keeping with the literature, our focus will be to identify the optimal com-
bination of input measures independent of the opposing team’s choices.

Data

We collect individual match-specific information from the ESPN-owned website

CRICINFO (http://www.espncricinfo.com) for all IPL, T20I, and ODI matches in
2008 and 2009. Our data set comprises 276 ODI, 77 T20I, and 118 IPL games that
took place during these 2 years. For every match, we obtain data on batting, bowling,
fielding, and other related inputs for both teams. Using this, we create different bat-
ting and bowling variables to measure different types of intent for each team. The

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
Cannonier et al. 767

dependent variable is binary and equals 1 when a team wins a match and 0 otherwise.
Each match generates two observations, and the observations are stacked by one
team’s input variables followed by the same variables for the other team.
We divide the batting and bowling inputs into two categories based on attacking
and defensive intents. The batting variables are runs scored through boundaries
(fours and sixes) as a percentage of total runs scored (BOUND) and nonboundary
runs (excluding extras) as a percentage of total runs scored (NBR). BOUND may
be thought of as an ‘‘attacking’’ batting measure whereby teams attempt to accumu-
late runs at a faster pace in maximizing the number of runs scored. Quite naturally,
this method involves higher risks as batsmen aim to acquire more boundaries. The
downside of this batting measure is that the team might lose wickets quickly and end
up with a lower total score if this risk does not pay off. On the contrary, NBR can be
used to depict a ‘‘defensive’’ batting measure where the batting team is conservative
in its approach and takes less chances by accumulating runs through singles, dou-
bles, and triples by skillfully finding open spaces in the field and hence, limiting the
number of wickets lost while batting. Traditionally, the literature exploring produc-
tion function in cricket used RPO as a measure of an ‘‘attacking’’ batting intent.13
While we have included this variable in our empirical analysis for a comparison with
the earlier findings in the literature, we emphasize here that this variable conveys
little information regarding the intent of the batting input. This is because the anal-
ysis involves only limited-over cricket and a team with higher RPO (hence, higher
total score) wins the match by definition if both teams use the allocated overs
entirely.14
The set of bowling variables comprise the following: opposition wickets taken
per over while bowling expressed as a percentage of overs (OWPO) and the number
of runs scored through boundaries (fours and sixes) by the opposition as a percentage
of total runs scored (OBOUND). OWPO reflects an attacking bowling intent where a
team aims to bowl out the opposition team quickly.15 On the contrary, OBOUND
expresses a defensive bowling intent. Intuitively, the aim of this defensive bowling
measure is to minimize the total runs scored by opposition regardless whether the
opposition gets out or not. This can be effectively achieved by restricting the number
of boundaries (fours and sixes) hit by the opposition. So, a low OBOUND implies a
more successful implementation of defensive bowling by restricting the opposition’s
run accumulation through boundaries, hence limiting them to a lower total.16 We
also include a comparable traditional defensive bowling measure: runs scored by
opposition per over bowled (ORPO) in our empirical analysis. Following our discus-
sion on RPO in the previous paragraph, it can be argued that ORPO, which is the
opposition’s RPO, does not provide much insight in defining the bowling intent
given the limited-over format of the matches.
The fielding variable (Fielding) is defined as the wickets due to fielding (catches,
stumpings, run-outs) as a proportion of total wickets taken.17 Other independent
variables include dummy variables reflecting whether a team played at home
(Home), whether a team won the toss (Toss) and whether a team batted first (Bat1).

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
768 Journal of Sports Economics 16(7)

Table 1. Summary Statistics for One-Day International (ODI) and Twenty20 (T20) Matches
for 2008-2009.

ODI T20
Variable Variable Definition M (SD) M (SD)

Win Equals 1 if team wins 0.500 0.500

(0.500) (0.501)
BOUND Runs scored in boundaries as a percentage of 42.074 51.450***
(OBOUND) total runs scored (by opposition) (9.852) (11.523)
NBR Nonboundary runs (excluding extras) as a 51.379 42.100***
percentage of total runs scored (9.524) (10.865)
RPO (ORPO) Runs scored per overs faced (by opposition) 5.039 7.743***
(1.158) (1.632)
OWPO Wickets taken per over while bowling 17.075 33.276***
(as a percentage) (7.047) (14.139)
Fielding Sum of catches, stumpings, and run-outs as a 0.694 0.750***
proportion of total wickets taken (0.208) (0.200)
Home Equals 1 if team plays at home 0.402 0.249***
(0.491) (0.433)
Toss Equals 1 if team wins the toss 0.490 0.497
(0.500) (0.501)
Bat1 Equals 1 if team bats first 0.500 0.500
(0.500) (0.501)
Observations 514 374

Note. T20 consists of all T20 International (T20I) matches and Indian Premier League (IPL) matches during
this period. The summary statistics are reported after dropping the abandoned and tied matches. Statis-
tical levels of significance, based on the difference between means of the two groups, are as follows:
**p <.05, ***p <.01.

Tables 1 and 2 report the summary statistics for ODI and T20 matches, and for
IPL and T20I matches, respectively. The reported summary statistics exclude the
information on 17 ODI, 4 T20I, and 4 IPL matches that were abandoned or tied.
As is evident from both tables, there are no significant differences in the outcomes
between ODI and T20 matches or between IPL and T20I matches. With respect to
independent variables, the batting inputs are significantly different between ODI and
T20 as well as between IPL and T20I. In Table 1, the average value for BOUND, a
measure of attacking batting, is understandably higher for the shorter format T20I
and IPL matches as there is a higher premium on accelerated scoring in form of hit-
ting boundaries. This can be further confirmed by the higher average for RPO in T20
where RPO is a traditional measure of attacking batting and indicates a higher rate of
run accumulation. On the contrary, a higher average NBR in ODI confirms the
importance of a defensive batting approach where runs are accumulated by taking
fewer risks by placing the ball in the gaps where there are no fielders. This is par-
ticularly important since the batting team needs to bat through a longer innings in
order to maximize its runs scored. Meanwhile, OWPO, an attacking bowling

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
Cannonier et al. 769

Table 2. Summary Statistics for Indian Premier League (IPL) and 20-Over International (T20I)
Matches for 2008-2009.

IPL T20I
Variable Variable Definition M (SD) M (SD)

Win Equals 1 if team wins 0.500 0.500

(0.501) (0.502)
BOUND Runs scored in boundaries as a percentage of 54.508 46.675***
(OBOUND) total runs scored (by opposition) (9.963) (12.186)
NBR Nonboundary runs (excluding extras) as a 39.223 46.594***
percentage of total runs scored (9.514) (11.342)
RPO (ORPO) Runs scored per overs faced (by opposition) 7.794 7.436***
(1.505) (1.774)
OWPO Wickets taken per over while bowling 31.669 35.785***
(as a percentage) (14.027) (13.993)
Fielding Sum of catches, stumpings, and run-outs as a 0.772 0.717***
proportion of total wickets taken (0.199) (0.198)
Home Equals 1 if team plays at home 0.241 0.260
(0.429) (0.440)
Toss Equals 1 if team wins the toss 0.500 0.493
(0.501) (0.502)
Bat1 Equals 1 if team bats first 0.500 0.500
(0.501) (0.502)
Observations 228 146

Note. The summary statistics are reported after dropping the abandoned and tied matches. Statistical lev-
els of significance, based on the difference between means of the two groups, are as follows: **p <.05,
***p <.01.

measure is significantly higher for T20 in comparison to ODI. Given that the bowl-
ing team has an entire 50 overs at its disposal to bowl out the opposition in ODI as
opposed to the 20 overs in the T20 format, the OWPO measured in percentage terms
is higher for T20. Since OBOUND (ORPO) is the opposition team’s BOUND (RPO),
the mean values are very similar to those shown under batting inputs and are not
reported separately to avoid repetition. The values in Table 2 illustrate that the
attacking batting input BOUND in IPL matches is statistically significantly larger
than that for T20I. This is also reflected through a higher average of run rate (RPO)
for IPL. To the extent that league-level games such as the IPL allow for the possi-
bility of exceptional pooling of talent from across the world, this result is not surpris-
ing. Moreover, the higher financial incentives may also be driving performance. The
average value of the attacking bowling measure OWPO is higher for T20I where
most bowlers are specialized in this form of the game and often possess years of
international cricketing experience.
The average value for the fielding variable is significantly higher for T20 in com-
parison to ODI. This is possibly due to the fact that the faster rate of run accumula-
tion creates more chances for the fielders in T20. The same argument can be

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
770 Journal of Sports Economics 16(7)

extended to support a higher mean value of the fielding variable in IPL than in T20I.
With respect to other variables, there is no significant difference between ODI and
T20 matches or between IPL and T20I matches. The one exception where the differ-
ence is statistically significant between ODI and T20 matches is the Home variable.
This is because the documented mean value of 0.249 under T20 is smaller than usual
as a result of 2009 IPL matches (which constitute a portion of the T20 matches)
played at neutral venues in South Africa. Having found that the descriptive statistics
for batting and bowling inputs are different in the two different formats of the game,
we now move on to answer the main question of this article: What inputs are respon-
sible for winning matches in the two different formats?

Results

Our objective is to find the combinations of batting and bowling inputs that increase
the log odds of winning in different formats of the game. Therefore, our results are
neither specific to a team nor in response to a specific team’s choices—they identify
different winning combinations of batting and bowling inputs that a team can adopt
based on the format of the game. The estimated conditional logit coefficients for dif-
ferent combinations of batting and bowling inputs, as well as fielding and other fac-
tors are presented in Tables 3–5 for ODI, T20I, and IPL matches, respectively.18 We
include year fixed effects to separate out any qualitative changes in a team between
2008 and 2009. We discuss and compare the importance of each input across the dif-
ferent formats below.

Batting Inputs. Table 3 reports the estimated coefficients for ODI matches. It is evi-
dent from columns (1) through (3) that the coefficient of BOUND enters at the 1%
significance level in the regressions and has the expected positive sign. This implies
that a team increases its likelihood of winning a match if it adopts an attacking bat-
ting intent, hence maximizing its total score by accumulating runs at faster pace
through boundaries. However, as earlier discussed, there is a risk involved with this
intent and a team might end up losing wickets quickly with a lower total score if this
intent fails to succeed on the field. On the other hand, columns (4) through (6) reveal
that our other variable of interest NBR enters at the 1% significance level in the
regressions and has a negative impact on the likelihood of winning. By adopting
defensive batting, while a team can limit its fall of wickets to bat through the entire
50 overs, it may end up with a relatively lower total score, hence reducing its
chances of winning the match. Our results can be corroborated from the fact that
in recent times, international teams have adopted a more attacking intent in the
50-overs format, hence attributing an increasingly important role to attacking bat-
ting.19 Columns (7) through (9) report the results for the traditional attacking batting
measure RPO. It can be seen that the variable enters at the 1% significance level and
has the strongest impact on the log odds of winning. As earlier argued, this is not

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
 Table 3. Conditional Logit Results from One-Day International (ODI) Matches.

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Variables Win Win Win Win Win Win Win Win Win

BOUND 0.069*** 0.049*** 0.068***

(0.023) (0.009) (0.015)
NBR 0.057*** 0.041*** 0.046***
(0.022) (0.010) (0.014)
RPO 1.236*** 1.288*** 3.956***
(0.261) (0.159) (1.058)
OWPO 0.421*** 0.418*** 0.414***
(0.043) (0.041) (0.046)
OBOUND 0.042*** 0.045*** 0.064***
(0.009) (0.010) (0.015)
ORPO 1.253*** 1.186*** 3.930***
(0.152) (0.147) (1.053)
Fielding 0.371 0.683 0.209 0.337 0.713 0.116 0.151 1.327** 0.098
(0.879) (0.447) (0.517) (0.862) (0.441) (0.505) (1.008) (0.540) (0.904)
Home 0.051 0.102 0.114 0.015 0.128 0.136 0.087 0.069 0.137
(0.263) (0.168) (0.171) (0.256) (0.166) (0.171) (0.282) (0.176) (0.265)
Toss 0.313 0.033 0.344 0.339 0.038 0.355 0.452 0.067 0.206
(0.429) (0.202) (0.288) (0.425) (0.204) (0.275) (0.547) (0.317) (0.318)
Bat1 0.205 0.132 0.252 0.180 0.106 0.281 0.043 0.037 0.174
(0.402) (0.197) (0.310) (0.401) (0.200) (0.314) (0.454) (0.272) (0.297)
Toss  Bat1 0.358 0.076 0.438 0.489 0.138 0.506 0.424 0.385 0.418

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
(0.650) (0.106) (0.435) (0.640) (0.120) (0.427) (0.848) (0.444) (0.393)
N 514 514 514 514 514 514 514 514 514
Pseudo R2 .627 .103 .331 .618 .086 .299 .704 .339 .753
Count R2 .960 .658 .852 .953 .664 .832 .973 .886 .993
Log likelihood 82.52 198.7 148.1 84.66 202.3 155.1 65.57 146.2 54.78
LR test 277.7 45.38 146.6 273.4 38.18 132.5 311.6 150.2 333.2

771
Note. The sample excludes the abandoned and tied matches. The numbers in the parentheses represent the robust standard errors. *p <.1, **p <.05, ***p <.01.
 Table 4. Conditional Logit Results from 20-Over International (T20I) Matches.

772
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Variables Win Win Win Win Win Win Win Win Win

BOUND 0.070** 0.087*** 0.132***

(0.031) (0.022) (0.037)
NBR 0.064* 0.073*** 0.102***
(0.035) (0.021) (0.030)
RPO 0.626* 1.020*** 2.118**
(0.321) (0.276) (1.009)
OWPO 0.108*** 0.112*** 0.110***
(0.021) (0.022) (0.024)
OBOUND 0.078*** 0.074*** 0.120***
(0.022) (0.019) (0.037)
ORPO 1.048*** 0.897*** 2.113**
(0.281) (0.209) (1.027)
Fielding 1.754 0.130 0.659 1.554 0.058 0.221 1.395 0.085 0.168
(1.085) (0.748) (0.845) (1.013) (0.721) (0.731) (1.094) (0.894) (0.971)
Home 0.132 0.615 0.626 0.067 0.551 0.624 0.106 0.650 0.702
(0.679) (0.570) (0.643) (0.697) (0.550) (0.626) (0.668) (0.598) (0.507)
Toss 0.264 0.247 0.160 0.287 0.240 0.135 0.122 0.565 0.371
(0.785) (0.531) (0.634) (0.848) (0.489) (0.569) (0.821) (0.764) (0.715)
Bat1 0.723 0.574 0.862 0.546 0.412 0.681 0.588 0.275 0.442
(1.090) (0.497) (0.719) (1.083) (0.482) (0.683) (1.043) (0.639) (0.713)
Toss  Bat1 1.755 0.912 1.180 1.565 0.662 0.984 1.529 0.192 0.700

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
(1.463) (0.714) (1.194) (1.443) (0.698) (1.144) (1.440) (1.131) (1.230)
N 146 146 146 146 146 146 146 146 146
Pseudo R2 .478 .309 .471 .460 .255 .384 .502 .447 .694
Count R2 .936 .745 .872 .915 .723 .830 .894 .872 .957
Log likelihood 31.89 42.22 32.34 33.00 45.54 37.69 30.47 33.80 18.69
LR test 58.51 37.84 57.60 56.29 31.21 46.90 61.35 54.69 84.90

Note. The sample excludes the abandoned and tied matches. The numbers in the parentheses represent the robust standard errors. *p <.1, **p <.05, ***p <.01.
 Table 5. Conditional Logit Results from Indian Premier League (IPL) Matches.

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Variables Win Win Win Win Win Win Win Win Win

BOUND 0.087*** 0.155*** 0.153***

(0.020) (0.031) (0.026)
NBR 0.088*** 0.165*** 0.149***
(0.026) (0.044) (0.031)
RPO 0.827*** 1.107*** 2.253**
(0.116) (0.246) (0.961)
OWPO 0.112*** 0.111*** 0.130***
(0.017) (0.016) (0.017)
OBOUND 0.151*** 0.144*** 0.152***
(0.030) (0.028) (0.031)
ORPO 1.134*** 1.018*** 2.328**
(0.223) (0.192) (0.988)
Fielding 0.858 0.756 0.159 0.783 0.633 0.162 1.206 1.323 0.765
(0.875) (0.997) (0.902) (0.859) (1.000) (0.889) (1.165) (1.207) (0.857)
Home 0.690 1.115*** 0.721 0.540 0.733** 0.394 0.417 0.602 0.304
(0.484) (0.426) (0.449) (0.494) (0.359) (0.473) (0.620) (0.481) (0.527)
Toss 0.440 0.159 0.292 0.462 0.244 0.388 0.589 0.349 0.121
(0.556) (0.374) (0.380) (0.517) (0.409) (0.391) (0.620) (0.415) (0.407)
Bat1 0.257 0.287 0.548 0.182 0.092 0.393 0.321 0.411 0.757*
(0.488) (0.452) (0.644) (0.468) (0.417) (0.618) (0.524) (0.344) (0.435)
Toss  Bat1 0.274 0.034 0.013 0.225 0.085 0.075 0.166 0.222 0.316

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
(0.561) (0.629) (0.694) (0.576) (0.656) (0.719) (0.558) (0.576) (0.324)
N 225 225 225 225 225 225 225 225 225
Pseudo R2 .336 .278 .330 .326 .266 .302 .397 .321 .629
Count R2 .875 .875 .875 .875 .875 .875 1 1 .966
Log likelihood 91.06 99.03 91.83 92.44 100.7 95.66 82.74 93.12 36.79
LR test 92.12 76.17 90.59 89.37 72.93 82.93 108.8 87.99 124.9

773
Note. The sample excludes the abandoned and tied matches. The numbers in the parentheses represent the robust standard errors. *p <.1, **p <.05, ***p <.01.
774 Journal of Sports Economics 16(7)

surprising, given that a team with higher run rate (RPO) wins the match by amassing
higher total score if it bats through the entire 50 overs. This result is comparable to
those of Schofield (1988) who finds a greater importance for attacking batting (RPO)
in the limited-over John Player League in English county cricket.
Tables 4 and 5 report the estimated coefficients for T20I and IPL matches, respec-
tively. As before, our variables of interest BOUND and NBR enter significantly with
the expected signs in both sets of regressions.20 The attacking batting measure
BOUND clearly emerges as the best input choice in affecting the log odds of winning
in this shortened 20-over format. In fact, the relative strength of BOUND increases,
and the negative impact of NBR gets even stronger as we move from ODI to T20I
and from T20I to IPL. Since there is relatively lower burden to limit the fall of wick-
ets in the shorter duration of T20 cricket, there is significant premium on risk taking
by adopting an attacking batting intent, thereby providing a larger target for the
opposition. The premium is even higher in IPL given the financial incentives. The
regression results involving the traditional batting measure (RPO) are reported in
columns (7) through (9) in Tables 4 and 5. It can be seen that RPO enters signifi-
cantly with positive signs in the regressions drawing support for observation made
by Schofield (1988).

Bowling Inputs. While a team needs both attacking and defensive bowlers in its bowl-
ing portfolio, it is important to examine the relative importance of defensive and
attacking bowling intents in different formats. In Table 3, the measure of attacking
bowling OWPO enters at the 1% significance level in the regressions with the
expected positive sign. This is an expected result since a team will increase its like-
lihood of winning by quickly dismissing the opposition whose total score will be
minimal, regardless of whether the opposition bats first or second. While attacking
bowling is important, given the length of the ODI innings, one may argue that
whether or not the opposition gets bowled out, it is more important to restrict the
opposition to a lower total score by restricting its run accumulation. This can be
effectively done by restricting the boundaries scored by the opposition by adopting
a defensive bowling intent. This is evident from Table 3 where the defensive bowl-
ing measure enters significantly in the regressions at the 1% significance level with
the expected negative signs. Recall that a lower value of our defensive bowling mea-
sure OBOUND implies a more successful implementation of defensive bowling,
hence increases the likelihood of winning. However, when compared in terms of
absolute values, the attacking bowling OWPO emerges as the best bowling measure
in improving the log odds of winning.21
Similar to ODI, the attacking bowling intent OWPO in Table 4 emerges as a clear
winner for T20I regardless of the chosen batting measure. However, it can be seen
that while the relative strength of OWPO falls, the relative strength of OBOUND
improves compared to the results of ODI. This is expected as T20 is an even shorter
format where a team can still win if it restricts the opposition to a lower total without
even dismissing it. On the contrary, in Table 5, defensive bowling OBOUND clearly

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
Cannonier et al. 775

outperforms the attacking bowling OWPO in IPL which implies that it is more
important to limit the opposition to a relatively lower score total to win the match
in IPL than to bowl them out.22

Optimal Input Combination. We use the relative strength of the input measures and
selected test statistics to identify the optimal combination. Looking at the strength
of the batting and bowling inputs in columns (1) through (6) in Table 3, attacking
batting and attacking bowling intents clearly emerge as the optimal batting and
bowling combination for a team in ODI. A count R2 value of .96 suggests that this
combination explains 96% of the match outcomes. Similar conclusions can be drawn
from the reported pseudo-R2 which suggests that 62.7% of the match outcomes can
be explained by this combination. Similar to ODI, attacking batting and attacking
bowling also emerge as the optimal input combination in T20I. The reported count
R2 and pseudo-R2 (column 1, Table 4) clearly suggest that this combination has the
greatest explanatory power in determining match outcomes. In contrast to ODI and
T20I, the optimal combination for IPL in Table 5 is attacking batting and defensive
bowling. It can also be seen from count R2 (column 2, Table 5) that the optimal input
combination explains 87.5% of the match outcomes.23 One of the striking features of
these results is that while the optimal input combination is identical for ODI and
T20I, it is different for the IPL. One possible explanation is that both ODI and
T20I involve national teams where players have years of international and domestic
cricketing experience and teams are developed through long-run training from pro-
fessional coaches and tips from think tanks and managers. With such skilled batsmen
and bowlers at its disposal, a team is expected to choose attacking batting and bowl-
ing measures where the respective objectives are to maximize run accumulation and
to bowl out the opposition quickly. On the contrary, teams are a mix of domestic and
a maximum of four international players in the IPL. Hence, it could be argued that
the IPL teams lack the skill set and specialization that are present in a national team.
Further, given that there is a premium on run accumulation, an IPL team might focus
to strengthen its batting line up by recruiting international batsman and all-rounders,
while relying more on local talent for bowling. Given this, one possibility might be
that IPL teams lack the skill set and resources to pursue attacking bowling and in
turn, focus on limiting the opposition to a relatively lower total to maximize the win-
ning probability.

Fielding. Our article also contributes to this literature by including the fielding input
as an additional explanatory variable in the model. Though Schofield (1988) consid-
ers fielding an important input in determining the success of a team, he does not test
it explicitly due to lack of data. We construct a proxy for fielding as dismissals due to
catches, run-outs, and stumpings as a proportion of total wickets taken. We discuss
our results on the fielding variable only in conjunction with the optimal inputs. The
fielding variable does not have any significant impact on the log odds of winning in
ODI (column 1, Table 3), T20I (column 1, Table 4), and IPL (column 2, Table 5). In

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
776 Journal of Sports Economics 16(7)

a limited-over format, a good fielding side not only plays an important role by cre-
ating better chances for the bowlers in dismissing the opposition, but it also plays a
more important role in limiting the opposition’s run accumulation. However, our
fielding variable is only based on the wickets resulting from fielding. Hence, it is
likely that in a way it undermines the dual role played by fielding and therefore does
not possess any significant explanatory power in the regressions. The construction of
a fielding variable that takes its dual role into account requires a ball-by-ball analysis
of each match which is beyond the scope of this article relying on aggregate match-
level information.

Other Inputs. Additionally, we test the importance of toss and the presence of home
team advantage for our optimal input combinations. Our empirical exercise does not
provide any support for the importance of toss. Similarly, we do not find any evi-
dence that winning the toss and batting first improves winning probabilities.24
Finally, unlike the findings reported in de Silva and Swartz (1997), we do not find
any evidence for home team advantage in ODI and T20I but find significant evi-
dence for home team advantage in the IPL.

Predicting Outcomes: An Illustration

In this section, we provide both in-sample and out-of-sample predictions for our
optimal input combination. Note that due to the nature of the estimation, as
explained below, our predictions can at best be viewed as indicative and no more
than an illustration. Table 6 reports the predictions for the optimal input combination
using the conditional (fixed-effect) logit model. For comparison purposes, we also
report the predictions using the logit estimates. The first two columns in the table
refer to the in-sample predictions using the 2008-2009 data and the last two columns
refer to the out-of-sample predictions using 2010 data.25
We begin with two specific observations on our choice of models. Recall, in our
sample, each match generates a pair of outcomes: One team wins while the other
does not. It stands to reason that within each pair, the outcomes of a match are lin-
early dependent and hence the error terms are correlated. Therefore, the model cho-
sen must account for correlation due to fixed effects (different intercept across
matches) as well as correlation with other covariates in the model. The conditional
logit is designed to address these issues and is thus preferred to the standard logit
model. Under these conditions, the conditional logit produces more accurate and
efficient estimates. Although some variant of the conventional logit model may
be used (e.g., adding match dummies as covariates in the model), these require
extremely large data sets.
Second, the results in Table 6 were derived from postestimation procedures.
However, in both models, the postestimation procedures make the restrictive
assumption that the estimated model has no fixed effects (i.e., the intercept is the
same across all observations). This assumption is inherent in the logit model which,

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
Cannonier et al. 777

Table 6. Prediction Based on Optimal Input Combination.

In-Sample Out-of-Sample

Clogit Logit Clogit Logit

1 2 3 4

ODI 50.00 80.12 50.00 60.36

T20I 50.00 77.40 50.00 75.00
IPL 72.37 72.81 72.88 72.88
Note. IPL ¼ Indian Premier League; ODI ¼ One-day international; T20I ¼ 20-over international.
The reported numbers refer to the percentage of observations correctly predicted for each match
outcome.

as already pointed out, will provide less efficient estimates. In terms of prediction,
we find that in many instances the logit model outperforms the conditional logit
model. This may partly be due to the fact that although the logit model does not
include match-fixed effects, it does incorporate team-fixed effects through the team
dummies.26 In spite of the fact that predictions are not as precise as we would like
them to be, we feel that this exercise is instructive.
Table 6 shows that while the conditional logit model successfully predicts 50% of
the match outcomes for ODI and T20I, the logit model predicts more than 80% of the
match outcomes for ODI and more than 77% of the outcomes for T20I for 2008-2009.
This can be attributed to the fact that the logit controls for team-fixed effects using
team dummies, whereas the conditional logit does not, resulting in lower prediction
of outcomes. The out-of-sample predictions for 2010 using conditional logit model
have remained steady at 50% for both ODI and T20I. However, the predictions for
2010 using logit are relatively lower at 60% for ODI and are fairly stable at 75% for
T20I in comparison to their in-sample counterparts. One possible explanation for the
low out-of-sample logit predictions for ODI may be related to greater variation in team
quality arising from a number of factors related to coaching, team composition, and
the increased participation of other teams over time. Contrast these results with those
of the IPL where both models correctly predict almost 72% of the in-sample outcomes
on average. The average out-of-sample predictions are also similar around 73%.
Recall that in IPL the teams are constructed through auctions where the primary goal
is to win, as opposed to national teams competing in ODI and T20I where other goals
such as rebuilding the team with younger players (thus compromising quality) may be
important. For the first 3 years of the IPL, the teams retained their core players, so that
team composition and quality remained relatively stable over the period.

Conclusion

In this article, we use the production function approach to determine the outcome of
a cricket game as a function of batting, bowling, fielding, and other variables. Using

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
778 Journal of Sports Economics 16(7)

data from ODI and T20 games (comprising of IPL and T20I), the article determines
the optimal combination of inputs for winning games. Our empirical results indicate
that a team’s best winning input is a combination of attacking batting and attacking
bowling in an international (ODI and T20I) setting. Thus, while a batting team
should be committed to scoring as many runs by hitting boundaries by adopting
an attacking batting and combine this with bowling that aims to dismiss the opposi-
tion quickly. Given that the international players have years of (international and
domestic) cricketing experience and national teams are developed over the long run
through professional coaching and team strategists, a team is expected to choose
attacking inputs to maximize its winning probabilities. Interestingly, attacking bat-
ting and defensive bowling turn out to be the optimal input combination in the league
version (IPL) of the 20-over format. Given the strong financial incentives and rules
favoring batsmen in T20, it is important to adopt an attacking batting strategy to
increase the scoring rate in a game. Quite possibly, since IPL teams pool interna-
tional and domestic players for a limited period only, they lack the skill set and spe-
cialization that is present in a national team; therefore it might be necessary to
restrict the opposition to a relatively lower score to maximize winning probabilities.
Another possibility is that given its focus on batting or all-rounders, IPL teams may
not have the ability to pursue an attacking bowling strategy.
Although bowling choices differ between international and league settings, the
skill sets are fairly similar across formats. Consequently, this has implications for
the choices that the players will make in the future and also for the future of the game
as well. Given the popularity of T20 cricket and the significant monetary benefits
associated with leagues such as IPL, players will possibly prefer the shorter format.
In a recent survey of 45 players conducted by the Federation of International Crick-
eters’ Association, 40% of the players indicated a preference to play in the IPL over
their country, while 32% of the respondents reported they could retire prematurely in
order to keep playing unconditionally in these lucrative leagues (Times of India, June
03, 2011). Rumford (2011b) studies this phenomenon in more detail and expresses
concern over the emergence of what he calls ‘‘freelance players’’ and ‘‘portfolio
players.’’ Rumford (2011b) identifies successful Test match players like Shane
Bond, Adam Gilchrist, Shaun Tait, and Brett Lee as ‘‘freelance players’’ who he
argues retired prematurely from conventional cricket in order to keep playing in the
lucrative T20 leagues. He is particularly concerned about the younger generation
‘‘portfolio’’ players who prefer to maintain a portfolio of T20 contracts from differ-
ent countries in order to obtain significant financial benefits rather than to play con-
ventional cricket for the national team.
Rumford (2011b) also argues that this has consequences for player development.
The rise of T20 leagues along with the rise of the ‘‘portfolio players’’ poses a serious
threat to player development since these leagues have little interest in developing
cricketing skills as players are pooled over for few weeks in return for large sums
of money. He argues strongly in favor of a ‘‘nested’’ version of cricket where players
grow by playing different formats in domestic cricket across the year and gradually

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
Cannonier et al. 779

rise from the local level to the national level. On the other hand, it is also likely that
in the future, cricketing nations will have specialized teams for the different formats.
In fact, it is already the case that different national squads are chosen for Test cricket
and ODI. Hence, we believe that due to the considerably more aggressive nature of
T20, this will lead to very different squads for Test cricket and T20 with somewhat
dissimilar squads for the two shorter formats.
Rumford (2011a) argues that the success of T20 poses a threat to the future of
the game itself. The growing popularity of T20 is creating a pressure on national
boards to accommodate more T20 matches in both their domestic and international
calendars. This threat is more serious to the 50-over cricket, given the rise of
too much of ‘‘context less’’ ODI cricket in recent years. Further, the emergence
of IPL as a lucrative model has attracted many national boards to set up domestic
T20 leagues which will further crowd out other formats.27 On the spectator side,
it is sometimes claimed that T20 has made a dent in the viewership and popularity
of the 50-over format of the game (see, for instance, Raghunath, 2009). In his 2011
Bradman Oration lecture at Canberra, the well-known Indian batsman Rahul Dra-
vid referred to the empty stands in a recently concluded India-England one-day
series and suggested a reduction in the number of ODI matches. He went on to say
that the game itself was at a crossroad and the future needs to be thought out care-
fully (Dravid, 2011). Quite likely though both ODI and Test cricket will continue
to exist for the purists who will prefer formats where players can demonstrate finer
talents, but there might be fewer games available to watch. Moreover, since the
majority of the ICC’s revenues currently derive from ODI, their decline in impor-
tance will not be immediate, though it is hard to speculate about how fast this
might occur.

Acknowledgments
We thank the editor and two anonymous referees for their careful reading and suggestions. We
also thank Areendam Chanda, Rajnish Kumar, David Paton, Abhinav Sacheti, and Arabinda
Sarangi for their insightful comments on the article.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship,
and/or publication of this article.

Funding
The authors received no financial support for the research, authorship, and/or publication of
this article.

Notes
1. See Kitchin (2008) for a detailed discussion on the evolution of the limited-overs cricket.
This section is based on his discussion to a large extent.

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
780 Journal of Sports Economics 16(7)

2. T20 cricket was introduced in domestic circuits in Pakistan and South Africa in 2004,
Australia in 2005, and in West Indies in 2006. See Kitchin (2008) and Saikia et al.
(2012) for more details.
3. Gupta (2011) coins the event as a ‘‘glocal’’ event that has both global and local elements in it.
4. See Karnik (2010) for a review of the IPL auction.
5. Andrew Flintoff and Kevin Pietersen from England each attracted bids worth US$1.55
million in 2009 IPL auction (Rumford, 2011a).
6. The eight teams were Bangalore Royal Challengers, Chennai Super Kings, Delhi Dare-
devils, Hyderabad Deccan Chargers, Jaipur Rajasthan Royals, Kolkota Knight Riders,
Mohali Kings XI Punjab, and Mumbai Indians.
7. See Gupta (2011) for more on IPL as a successful business model.
8. See Axford and Huggins (2011) for further details on T20 cricketing rules. Also see http://
www.icc-cricket.com/rules_and_regulations.php, and http://www.indiatwenty20.com/
twenty20-rules.htm
9. While the production function approach relies on aggregate match-level information to
estimate the optimal inputs (aggregate) to win a match, the economics of cricket is rich
with literature that identifies the optimal inputs at different stages of a match. Clarke
(1988) uses a dynamic programming formulation that identifies the optimal scoring rate
at any stage of batting in one-day cricket. Preston and Thomas (2000) extends the formu-
lation made by Clarke and finds that optimal batting strategies differ over the course when
first innings is compared with second innings in the limited-over one-day cricket. On the
other hand, Swartz, Gill, Beaudoin, and deSilva (2006) simulate a model to identify opti-
mal batting orders in one-day cricket. Their identified batting order for Indian national
team for 2003 World Cup final performs better than the actual adopted batting order.
10. Allsopp and Clarke (2004) test the effect of toss only for Test matches and find the effect
to be insignificant. Additionally, the authors find Australia and South Africa as the best
teams based on their ratings on batting and bowling performances for ODI and Test
matches. In an earlier article, Clarke and Allsopp (2001) fit a model using margin of vic-
tory calculated from the Duckworth and Lewis method to rank teams in 1999 ODI World
Cup. They also fail to find any significant effect for toss advantage in their model.
11. See for instance, Schofield (1988), Bairam et al. (1990a, 1990b), and Brooks et al. (2002).
12. Bairam et al. (1990a) suggest possible collinearity as a reason for the exclusion of field-
ing. In our data, we (simultaneously) verify that there is no evidence of multicollinearity
with the inclusion of both bowling and fielding inputs.
13. See Schofield (1988), Bairam et al. (1990a, 1990b), and Brooks et al. (2002).
14. We thank an anonymous referee for providing valuable suggestions in defining the input
measures.
15. Previously, the literature on production functions in cricket used overs per wickets taken
(i.e., the inverse of OWPO) as a measure of attacking bowling. We use OWPO, as it pro-
vides a more direct measure of attacking bowling. We would like to thank an anonymous
referee for bringing this to our attention.
16. Alternatively, one can use nonboundary runs (excluding extras) scored by the opposition
as a percentage of total runs scored (ONBR) as a measure of defensive bowling where a

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
Cannonier et al. 781

high ONBR implies a better implementation of defensive bowling. However, our conclu-
sions are unaltered by this alternative bowling measure.
17. Alternatively, we also introduced the fielding variable as wickets due to fielding (catches,
stumpings, run-outs) as a proportion of nonfielding wickets (due to bowled and leg before
wickets) in our regressions. Due to the construction of the variable, we lost approximately
16% observations in ODI, 20% observations in T20I, and 27% observations in IPL in the
process since the denominator (nonfielding wickets) was zero in these cases. These
results which provide no additional insights are not reported, and are available upon
request.
18. The reported results are based on the sample that excludes the observations on abandoned
and tied matches. However, including those observations do not alter our conclusions and
these results are available on request.
19. Among the 391 highest innings totals listed in the ESPN-CRICINFO database, approxi-
mately 50% of these occurred post-July 2005, after the introduction of field restrictions in
terms of ‘‘Powerplays’’ and approximately 30% occurred between 2008 and 2011 (ESPN-
CRICINFO Statsguru). This suggests the adoption of an increasingly attacking batting
intent in the limited-over format in recent times.
20. In conjunction with an attacking bowling input OWPO, NBR enters significantly only at
the 10% level for T20I in Table 4.
21. Our regressions with ONBR as an alternate measure of defensive bowling also support our
conclusions. A higher value of ONBR can be argued as a successful implementation of
defensive bowling. Additionally, this finding is in sharp contrast to the conclusions made
by Schofield (1988) who finds relatively greater importance of defensive bowling in the
limited-over John Player Leagues from English county cricket. However, it must be noted
that Schofield used ORPO as a measure of defensive bowling. Table 3 also reports the
additional results with ORPO as an alternative measure of defensive bowling. The mea-
sure ORPO outperforms the rest of the bowling measures in the regressions. Due to the
nature of the problem discussed earlier, this variable provides little to no insights about
bowling intent.
22. As conjectured earlier, it can be seen from columns (7) through (9) of Table 5 that the
traditional measure of defensive bowling (ORPO) has the highest explanatory power.
23. However, the pseudo R2 for this combination is slightly lower at 27.8% in comparison to
the other input combinations.
24. We also ran additional regressions similar to Dawson et al. (2009) by dropping the
dummies for toss and batting first and introducing the interaction of the dummies for
toss and bat first and toss and bowl first. Those results do not provide any additional
insights.
25. The data set for 2010 is also constructed using the individual match-specific information
from the ESPN-owned CRICINFO website.
26. On a cautionary note, we wish to emphasize that despite its better predictive power, the
logit model will provide inaccurate estimates.
27. To name a few, the Big Bash League of Australia, Bangladesh Premier League of
Bangladesh, Sri Lanka Premier League of Sri Lanka.

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
782 Journal of Sports Economics 16(7)

References
Allsopp, P. E., & Clarke, S. R. (2004). Rating teams and analysing outcomes in one-day and
test cricket. Journal of the Royal Statistical Society: Series A, 167, 657–667.
Axford, B., & Huggins, R. (2011). Cricket for people who don’t like cricket?: Twenty20 as
expression of the cultural and media zeitgeist. Sport in Society: Cultures, Commerce,
Media and Politics, 14, 1326–1339.
Bairam, E. I., Howells, J. M., & Turner, G. M. (1990a). Production functions in cricket: The
Australian and New Zealand experience. Applied Economics, 22, 871–879.
Bairam, E. I., Howells, J. M., & Turner, G. M. (1990b). Production functions and the strategy
implications for cricket in New Zealand. Sporting Traditions: The Australian Society for
Sports History, 6, 202–217.
Bhaskar, V. (2009). Rational adversaries? Evidence from randomized trial in one day cricket.
The Economic Journal, 119, 1–23.
Brooks, R. D., Faff, R. W., & Sokulsky, D. (2002). An ordered response model of test cricket
performance. Applied Economics, 34, 2353–2365.
Clarke, S. R. (1988). Dynamic programming in one-day cricket-optimal scoring rates. Journal
of the Operational Research Society, 39, 331–337.
Clarke, S. R., & Allsopp, P. (2001). Fair measures of performance: The world cup of cricket.
Journal of the Operational Research Society, 52, 471–479.
Dawson, P., Morley, B., Paton, D., & Thomas, D. (2009). To bat or not to bat: An examination
of match outcomes in day-night limited overs cricket. Journal of the Operational Research
Society, 60, 1786–1793.
de Silva, B. M., & Swartz, T. B. (1997). Winning the coin toss and the home team advantage
in one-day international cricket matches. The New Zealand Statistician, 32, 16–22.
Dravid, R. (2011). Dravid’s Bradman oration. Cricketnext. Retrieved from http://cricketnext.
in.com/live/news/watchread-dravids-bradman-oration/62281-13.htm
Gupta, A. (2011). The IPL and the Indian domination of global cricket. Sport in Society: Cul-
tures, Commerce, Media and Politics, 14, 1316–1325.
‘‘IPL over country for 40% players.’’ (2011, June 3). The Times of India. Retrieved June 30,
2011, from http://timesofindia.indiatimes.com//sports/cricket/top-stories/IPL-over-coun-
try-for-40-players/articleshow/8694729.cms
Karnik, A. (2010). Valuing cricketers using hedonic price models. Journal of Sports Econom-
ics, 11, 456–469.
Kitchin, P. (2008). The development of limited overs cricket: London’s loss of power. London
Journal of Tourism, Sport and Creative Industries, 2, 70–75.
Lemmer, H. H. (2011). The single match approach to strike rate adjustments in batting per-
formance measures in cricket. Journal of Sport Science and Medicine, 10, 630–634.
Morley, B., & Thomas, D. (2005). An investigation of home advantage and other factors
affecting outcomes in English one-day cricket matches. Journal of Sports Sciences, 23,
261–268.
Preston, I., & Thomas, J. (2000). Batting strategy in limited overs cricket. The Statistician, 49,
95–106.

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016
Cannonier et al. 783

Raghunath, A. (2009, October 7). Twenty killed the ODI star. Forbes India. Retrieved June
30, 2011, from http://business.in.com//article/what-if/twenty-killed-the-odi-star/4402/1
Rumford, C. (2011a). Guest editor’s introduction: Twenty20 and the future of cricket. Sport in
Society: Cultures, Commerce, Media and Politics, 14, 1311–1315.
Rumford, C. (2011b). Twenty20, global disembedding and the rise of the ‘portfolio player’.
Sport in Society: Cultures, Commerce, Media and Politics, 14, 1358–1368.
Saikia, H., Bhattacharjee, D., & Bhattacharjee, A. (2012). Is IPL responsible for cricketer’s
performance in twenty20 world cup? International Journal of Sports Science and Engi-
neering, 6, 96–110.
Schofield, J. A. (1988). Production functions in sports industry: An empirical analysis of pro-
fessional cricket. Applied Economics, 20, 177–193.
Swartz, T. B., Gill, P. S., Beaudoin, D., & deSilva, B. M. (2006). Optimal batting orders in
one-day cricket. Computers & Operations Research, 33, 1939–1950.

Author Biographies
Colin Cannonier is a former first-class cricketer and central banker who started his career as
a country economist in the Eastern Caribbean. He is currently an assistant professor of eco-
nomics at Belmont University. His research interests include health economics, economic
demography, applied economics, public policy, labor economics and sports economics.
Bibhudutta Panda is an Assistant Professor of economics at the University of Minnesota-
Morris. His primary research interests include economic growth and development, and
macroeconomics.
Sudipta Sarangi is the Gulf Coast Coca-Cola Distinguished Professor of Business Adminis-
tration in the Department of Economics at Louisiana State University. His primary research
interests lie in the area of networks, development economics and experimental economics.
This is the first time he has ventured into sports economics.

Downloaded from jse.sagepub.com at Middle East Technical Univ on February 16, 2016