You are on page 1of 14

Journal of the Operational Research Society (2008) 59, 729 --742  2008 Operational Research Society Ltd.

Ltd. All rights reserved. 0160-5682/08 $30.00

www.palgrave-journals.com/jors

Extending the range of player-performance measures


in one-day cricket
AJ Lewis∗
Oxford Brookes University, Oxford, UK
The author’s previous paper on this topic showed that the Duckworth/Lewis methodology has considerable
potential for providing relevant and objective measures of performance in one-day cricket. The present paper
evaluates how these measures can apply in the longer term. This assessment is based upon several series
of matches involving the main players from most of the major cricketing nations. Although data used are
not exhaustive, they are of sufficient quantity to provide strong indications of the long-term viability of the
measures, thus leading to the realistic expectation of inclusion of the methodology into the existing player-
rankings or the development of an independent set of performance indicators that reflect more validly the
value of players’ inputs to matches. Such measures are expected to be of valuable use to team management,
media commentators and to the public.
Journal of the Operational Research Society (2008) 59, 729 – 742. doi:10.1057/palgrave.jors.2602379
Published online 7 March 2007

Keywords: sports statistics; cricket; performance measurement

Introduction Reflecting a growing interest in sporting performance mea-


sures, in recent times other authors have addressed the cre-
This paper is concerned with further evaluating developments
ation of alternative cricketing performance measures. Barr
in measures of player-performance in one-day cricket. In
and Kantor (2004) have devised a criterion for comparing
Lewis (2005) it was shown that the Duckworth/Lewis (D/L)
batsmen, which combines batting average and strike rate.
methodology of target resetting in interrupted limited overs
They used the Cricket World Cup of 2003 in South Africa to
cricket matches could be used to assess performances of
calibrate a model that compares the top batsmen of that tour-
batsmen and bowlers as a contrast to the traditional measures
nament. It uses the two dimensions of probability of dismissal
of batting and bowling averages, and their subsidiaries of
and strike rate to create an efficient frontier, analogous to that
strike and economy rates.
of portfolio theory (Ali, 1975), by which to compare batsmen.
In addition, by using other aspects of the D/L methodology,
By using an analogy with stochastic dominance within port-
Lewis (2005) and Beaudoin and Swartz (2003), have created
folio theory, Damodaran (2006) has evaluated performances
additional measures that objectively assess ‘all-round’ perfor-
of the Indian batsmen of recent times with cautious sugges-
mance, which is not possible using the traditional measures
tions that his results can shed light on how certain players
of performance.
are more suited than others in meeting the particular require-
Lewis showed that the D/L-based measures take into
ments of team situations during its innings.
account the context of the match in which runs are scored,
Swartz et al (2006) have used simulation techniques,
or conceded, and wickets taken, or lost, thereby arguably
which have also involved the performances in recent years of
more objectively comparing batsmen, comparing bowlers,
the leading Indian batsmen, to assist in determining optimal
comparing batsmen with bowlers and in assessing all-round
batting orders.
performance. These measures were shown to help team
Although Lewis (2005) showed how the D/L measures of
management
performance were useful in the short term for single matches
in assessing contributions to the team effort from individual
and for short series, he concluded that analysis of many more
players, not only for a particular match, but also for a series
matches was necessary in order to investigate the reliability
of matches. In addition, they were shown to provide aids to
and validity of these measures in the longer term.
adjudicators in deciding on ‘man-of-the match’ and ‘player
Bryman and Bell (2003) discuss reliability and validity in
of the series’ awards.
relation to quantitative social metrics. Although cricket is a
∗ Correspondence: AJ Lewis, The Business School, Oxford Brookes sport, the style of performance measurement being discussed
University, Wheatley, Oxford, OX33 1HX, UK. in this paper is distinctly different from measures made
E-mail: ajlewis@brookes.ac.uk under more controlled scientific conditions (Hopkins, 2000).
730 Journal of the Operational Research Society Vol. 59, No. 6

Cricket’s performance measurements are therefore regarded Net contribution


as of a ‘social’ nature so that the criteria of Bryman and Bell
The measure called net batting (or bowling) contribution is a
will be used to assess the long-term reliability and validity
summation, on a ball-by-ball basis, of what a batsman scores
of the measures of performance discussed in this paper.
(or a bowler concedes) compared with the runs expected
Bryman and Bell regard reliability as the stability of
from the D/L model (Lewis, 2005). This summation includes
the measure following additional observations of the same
the ball that takes the wicket, since a batsman thereby loses
characteristic. In cricketing terms reliability would then be
run-scoring potential for his team, and the bowler saves his
exhibited if the metrics were stable, that is that they did not
side runs by the taking of the wicket. Because cricket is a
fluctuate too much (Bryman and Bell, 2003) if a batsman
zero-sum game, adjustments to these net runs are necessary to
or bowler were performing consistently to the same stan-
allow for the asymmetric way that some forms of ‘extras’ are
dard. Bryman and Bell define validity as whether the metric
debited against bowler but not credited to the batsman.
that has been designed to measure a certain concept really
The measure for each player is in the form of a positive
measures that concept. In particular, convergent validity can
or negative number of runs, which is usually non-integral,
be gauged (Bryman and Bell, 2003) by comparing a metric
that represents the overall extent to which the player has
to other measures of the same basic concept. In cricketing
performed above, or below, expectations for the match situa-
terms all the measures presented in this Introduction are
tion for the balls received or delivered.
intended to assess the players’ cricketing abilities as batsmen
or bowlers. Convergent validity would be evident in the Resource percentage average
proposed D/L measures if there were statistically significant
agreement with the appropriate standard measures and with The second measure of player-performance proposed in Lewis
other external performance measures. (2005) was the batting (or bowling) resource percentage
In subsequent sections of the paper, firstly the concepts of average. According to the D/L methodology, the delivering
the D/L measures are reviewed. Secondly, the measures are and receiving of a ball and the loss and taking of a wicket
applied to a large number of matches covering a majority of involve consumption of resources by the batsman and contri-
the major countries playing one-day internationals (ODIs) and bution of resources by the bowler. The respective resource
a majority of the leading players in ODIs from those countries. percentage averages are the average runs scored per unit
Thirdly, the results of the analysis obtained on these players of resource consumed by the batsman and the average runs
are discussed and finally, conclusions are drawn on the validity conceded, including wides and no-balls, per unit of resource
and reliability of the measures in the longer term. In drawing contributed by the bowler.
these conclusions, comparisons are made with the popular Beaudoin and Swartz (2003) have produced a measure
LG ICC ratings (http://www.lgiccrankings.com/, accessed 8 called ‘runs per match’ (RM) for both batsmen and bowlers.
November 2006). These are produced for the International These measures are multiples of 100 of the respective batting
Cricket Council (ICC) and sponsored by the LG marketing and bowling resource percentage averages. They are inter-
company. preted as the average number of runs scored or conceded were
Further work is then suggested for additional refinements the player to do all of the batting or the bowling himself.
to the measures. According to the D/L model of one-day cricket (Duckworth
and Lewis, 1998, 2004), the average runs in 50 overs, with all
wickets available, is represented by Z(50,0) and has a value of
235. An RM measure above 235 represents an above average
Review of D/L performance measures batsman with the converse true for a bowler.
The performance measures proposed in Lewis (2005), of net
All-round performance
contributions and resource percentage averages are based on
the D/L methodology (Duckworth and Lewis, 1998, 2004). Lewis (2005) showed that both of these two types of measures
There follows a review of these measures. In this review, and arguably produced better assessment of players’ performances
in the subsequent developments, players are referred to in separately for batting and for bowling in single matches, and
the male gender. This is partly because the data used have in a series of matches. Further, he showed that the net contri-
come solely from the men’s series of ODIs between the rele- butions of bowlers and batsmen, being on a common and addi-
vant countries. Clearly, however, the measures discussed are tive scale, can be compared directly and that performances
equally applicable in women’s cricket but, to smooth the in batting and bowling can be added to help assess all-round
narrative in this paper, cross gender terms, such as ‘he/she’, performance.
will not be employed. In addition, it is of interest to note Although the RM measures of bowlers and batsmen are not
that in women’s cricket, male terminology is still used such directly comparable, Beaudoin and Swartz (2003) proposed
as ‘batsman’ and not ‘batswoman’ for the player batting, that all round performance can be assessed by finding the
and ‘third man’ for one of the traditional fielding positions difference of a player’s batting and bowling RMs. Their
(Engels, 2006). interpretation of this measure is the average performance of
AJ Lewis—Extending the range of player-performance measures 731

the player were he to do all of the batting and the bowling All major ODI countries have featured in at least three of
himself. these series with the exceptions of Pakistan, Zimbabwe and
Bangladesh. Consequently, the principal players from these
Assessing performances in the longer term three countries are not included in subsequent analysis.
Because the sources of the author’s electronic data are located
Batting and bowling averages are the traditional ways of eval-
in the United Kingdom and Australia, there is a prepon-
uating player performance in the long term, though couched
derance of matches involving England and Australia. Such
in qualifying terms because of the known weaknesses in
volume of data aids in establishing the longer-term stability
these measures (Lewis, 2005; Beaudoin and Swartz, 2003).
of the proposed D/L measures. However, a priori, there is
Using their RM measures, Beaudoin and Swartz (2003) under-
no reason to believe that bias towards results from these
took an analysis of the performances of several high-profile
two countries would invalidate the general conclusions to be
players between 1998 and September 2002 in order to achieve
drawn.
more meaningful measures of performance over those players’
Because of the turnover of players during the three-year
careers during that period. Some of these players also appear
period of the data for this research, many players have played
in the analysis that follows. Several useful comparisons will
insufficiently for their overall performances to be regarded as
be made, which will be used also for the assessment of the reli-
representative. In order to reduce bias due to small volumes of
ability and validity of the D/L-based performance measures.
input, players included in subsequent analyses are only those
The analysis that follows examines whether net contri-
who have played a role in more than 12 innings. Moreover, in
butions and resource percentage averages would assist in
order that comparisons are as current as possible, only players
the longer term in providing measures of player-performance
who have retained a ranking in the top 100 of the LG ICC
that are more valid than existing measures of performance,
ratings are included.
and whether the all-round performance measures outlined are
valid indicators of such performances.
In order to gain some indications as to how the measures The LG ICC ratings
would work in the longer term, several ODI series have been
There are two sets of LG ICC player rankings, one for Test
analysed, in exactly the same way as the Victoria Bitter
Matches, and one for ODIs. This paper is concerned only
(VB) Series for 2003, which is discussed in some detail in
with the latter set, which includes ratings and rankings for
Lewis (2005). The series of matches summarized in subse-
batsmen and bowlers and more recently, rankings for all-
quent analyses, in chronological order, are as follows. This list
round performance. The ratings are provided on a scale from
defines several abbreviations subsequently employed in this
0 to 1000. The Internet website cited gives a general overview
paper.
of the principal features of the ratings but does not provide
(i) VB 2002 (VB02): between Australia (Aus), South full details of the calculations due to commercial sensitivity.
Africa (RSA) and New Zealand (NZ) [14 matches] However, it is known that the ‘all-round’ rating is the product
(ii) NatWest Series 2002 (NW02): England (Eng), India of the player’s batting and bowling rating. The administrators
(Ind) and Sri Lanka (SriL) [10] of the ratings produce this only for selected players deemed
(iii) VB 2003 (VB03): Australia, England, Sri Lanka [14] to be all-rounders.
(iv) NatWest Series 2003 (NW03S): England, South Africa, The mechanism of the LG ICC ratings uses only infor-
NZ [10] mation provided by a ‘standard’ summary scorecard as
(v) NatWest Challenge 2003 (NW03C): England, Pakistan typically reproduced in Engels (2006). For batsmen, infor-
(Pak) [3] mation recorded includes runs scored, which bowler took his
(vi) Tour 2003: West Indies (WI), Australia [7] wicket if dismissed, and the batsman’s strike rate. For bowlers,
(vii) TV Sundaram Iyengar & Sons Cup 2003 (TVS03): information recorded includes the names of the batsmen
India, Australia, NZ [10] dismissed, the numbers of overs bowled, wickets taken and
(viii) VB 2004 (VB04): Australia, India, Zimbabwe (Zim) runs conceded including wides and no-balls, and the economy
[14] rate, that is, runs conceded per over bowled.
(ix) Tour 2004: West Indies, England [5] In these ratings, bowlers get extra credit for the quality of
(x) Tour 2004: Sri Lanka, Australia [5] the batsmen dismissed, but no account is taken of the stage
(xi) NatWest Series 2004 (NW04S): England, NZ, of the team’s innings that the wickets are taken. The reverse
West Indies [10] consideration, however, cannot be given to the batsmen for
(xii) NatWest Challenge 2004 (NW04C): England, India [3] scoring runs against the better-ranked bowlers since such data
(xiii) Tour 2005: South Africa, England [7] are not recorded in the scorecard. Consequently, the infor-
(xiv) VB 2005 (VB05): Australia, NZ, Pakistan [14] mation captured by these ratings is likely to be more limited
(xv) Tour 2005: NZ, Australia [5] in scope than that used for analysis in this paper. Further-
(xvi) NatWest Series 2005: England, Australia, Bangladesh more, because of a discounting factor over time in the LG
(Bang) [10] ICC ratings, greater weight is given to players’ performances
732 Journal of the Operational Research Society Vol. 59, No. 6

that are more recent, so that players who are not selected for The Beaudoin and Swartz batting RM measure is not included
play, quickly lose their status in the rankings. as it is merely a multiple by 100 of the batting resource
Whereas the information in the traditional scorecard percentage average. Table 1 also provides the number of
provides the details for calculation of the averages, it is not innings in which each player had batted, together with the
sufficient for the analysis undertaken in this and the previous LG ICC rating applicable to 31st May 2005. This was the
paper (Lewis, 2005); substantial details on the outcome of date of the rating’s publication as near as possible to the last
every ball are required. The author’s data-sources, from match of the final series included in the database, that of the
Australia and the United Kingdom that are acknowledged NatWest Series of 2005 which concluded on 3rd July 2005.
later, provide detail to the level of who bowled to whom, how Several features emerge from Table 1. One is the variation
many runs were scored, or byes and leg-byes, wides and no- in the rankings of the players under the various methods
balls conceded and the mode of dismissal when the wicket of measuring batting performances. Kallis (RSA) has a
is taken. These details are supplied in electronic form so that high batting average of 54.87, yet ranks more lowly under
the processing into the summary information needed for the net contribution and resource percentage average. Quite
proposed measures is relatively straightforward compared probably, this is due to the nature of his play, which is to
with the approach by Beaudoin and Swartz (2003). They build an innings more slowly than his counterparts thereby
transcribed verbal commentaries such as those on Cricinfo consuming more resources and run-scoring potential for his
(http://uk.cricinfo.com/db/ARCHIVE/, accessed 8 November innings, which drops him down the D/L-based rankings.
2006) into a form for digital processing. Beaudoin and Swartz Beaudoin and Swartz (2003) obtained a similar pattern for
admit that the workload involved in this process was very the measures of Kallis. Conversely, Gilchrist (Aus) who has a
onerous. modest 23rd-ranked batting average of less than 40, is ranked
1st and 4th under net contribution and resource percentage
Analysis of performance measures average, respectively and 2nd by the LG ICC ratings. His
role has been to score runs more quickly, particularly during
The author stresses that the analysis that follows is intended
the earlier phase of an innings.
to indicate how the measures of net contribution and of
Pietersen (Eng) tops the standard batting averages partly
resource percentage average would operate over the several
because his figures are based on several high scores in
series of matches. Although some 141 matches have been
relatively few matches and even fewer completed innings.
included to produce these statistics, in the main they are still
However, he is highly ranked also by resource percentage
a minority of matches from most players’ careers and so con-
average and by net contribution. The mechanism of the LG
clusions on ratings and rankings can only be indicative at this
ICC ratings means that there is a necessity for sustained high
stage. Consequently, there are no suggestions that the rank-
performance in order to climb the rankings, which takes more
ings subsequently derived in this paper, will be definitive for
time than he had spent playing ODIs, hence his relatively
players’ career performances. To some extent, however, com-
low LG ICC ranking compared with the other measures.
parison will be made with the current LG ICC international
player ratings and with the results of the study by Beaudoin
and Swartz (2003). As well as being of considerable interest, Batting correlations
such comparisons will be relevant to assess the validity and
Table 2 summarizes the correlations of the players’ batting
reliability of the D/L-based performance measures.
performance measures with each other and with the number of
In order to standardize the analysis, the net contribu-
innings in which the batsmen have contributed. The table also
tions and resource percentage averages will be based on the
summarizes the significance of the sample correlations from
D/L model and tables of September 2002 (http://www.icc-
zero on a two-tailed alternative hypothesis. These correlations
cricket.com/rules/d-l method.pdf, accessed 8 November
are obtained from all the 47 batsmen who have batted in more
2006) even though two of the series analysed (VB02 and
than 12 innings and have an LG ICC rating.
NW02) took place before this date. The former edition of
The correlations between the three performance measures
the tables (see Duckworth and Lewis, 2004; or Beaudoin and
are statistically significant and, using verbal descriptors
Swartz, 2003) would have been applicable to any D/L target
defined by Cohen (1988), they are moderate to large, the
calculations in those two series.
largest being between the D/L-based measures (as might be
expected), and the lowest between the standard batting
Evaluating the batting performance measures
average and net contribution. Beaudoin and Swartz (2003)
Table 1 provides the performances of the eligible players found a correlation of 0.67 between batting average and their
(according to the criteria previously defined) over the one-day RM measure, which is not inconsistent with the equivalent
series mentioned above. The ranking of players in the table is 0.746 for the study of this paper.
in accordance with their traditional batting averages, but their The correlations of the performance measures with the LG
ranks by the two measures suggested by this paper (called the ICC ratings are moderate in all cases with the highest being
D/L-based measures) are included for comparative purposes. for net contribution.
AJ Lewis—Extending the range of player-performance measures 733

Table 1 Summary of batting performance measures for qualifying batsmen


Batting
Batting Batting Batting res% LG LG
Batting Batting avge Batting contrib res% avge ICC ICC
Player Country innings avge rank contrib rank avge rank rating rank *
Pietersen Eng 13 99.57 1 252.3 7 3.473 1 611 23
Lehmann Aus 14 68.12 2 126.5 14 2.836 5 544 31
Kallis RSA 20 54.87 3 40.6 26 2.306 31 708 8
Tendulkar Ind 21 54.68 4 280.3 5 2.996 3 692 10
Sarwan WI 23 50.82 5 156.1 11 2.661 10 672 13
Kaif Ind 13 46.87 6 66.8 20 2.652 11 532 34
Clarke Aus 33 46.66 7 243.9 8 2.649 12 705 9
Flintoff Eng 31 43.96 8 309.8 3 2.998 2 711 5
Sangakkara SriL 17 42.86 9 127.7 13 2.703 7 709 7
Laxman Ind 21 41.66 10 57.8 22 2.389 24 532 35
Strauss Eng 29 41.40 11 120.3 15 2.483 16 488 39
Lara WI 22 41.15 12 158.5 10 2.681 8 646 20
Smith RSA 14 41.00 13 40.0 27 2.368 26 722 3
Hayden Aus 49 40.33 14 294.8 4 2.622 13 671 14
Chanderpaul WI 16 39.73 15 69.3 19 2.444 20 625 21
Martyn Aus 55 39.67 16 4.2 38 2.192 39 654 19
Ponting Aus 61 39.52 17 253.3 6 2.478 18 759 1
Trescothick Eng 53 38.57 18 396.3 2 2.747 6 658 18
Jayawardene SriL 19 38.33 19 72.8 18 2.409 23 555 29
Symonds Aus 47 37.75 20 241.0 9 2.596 14 660 17
Dravid Ind 24 37.04 21 −1.5 40 2.184 40 711 5
Gilchrist Aus 56 36.96 22 508.0 1 2.945 4 724 2
Jayasuriya SriL 18 35.83 23 133.3 12 2.663 9 670 15
Gayle WI 24 35.36 24 10.7 36 2.217 36 717 4
Gibbs RSA 24 33.3 25 34.6 30 2.306 32 674 12
Y. Singh Ind 25 32.63 26 82.9 17 2.477 19 546 30
Fleming NZ 23 32.04 27 38.4 28 2.280 34 670 15
Solanki Eng 22 31.85 28 49.8 23 2.370 25 490 38
McMillan NZ 24 31.71 29 14.1 34 2.211 37 578 25
Cairns NZ 17 31.64 30 58.9 21 2.501 15 588 24
Vaughan Eng 41 31.64 31 0.7 39 2.198 38 569 26
Boucher RSA 20 30.5 32 45.1 24 2.434 21 542 32
Collingwood Eng 37 30.34 33 95.4 16 2.431 22 512 36
Sehwag Ind 22 28.81 34 41.6 25 2.332 28 618 22
Atapattu SriL 19 28.63 35 −24.3 43 2.043 43 675 11
Katich Aus 14 28.54 36 37.9 29 2.480 17 348 46
JonesG Eng 21 28.43 37 21.4 31 2.325 29 419 43
Astle NZ 15 27.46 38 −54.8 47 1.921 46 569 27
Hogg Aus 22 26.25 39 −44.8 45 1.928 45 407 44
Ganguly Ind 21 25.85 40 −37.6 44 2.024 44 557 28
Harris NZ 16 24.58 41 −50.4 46 1.844 47 426 42
Styris NZ 15 24.15 42 −9.7 42 2.063 42 535 33
McCullum NZ 14 24.08 43 −6.6 41 2.120 41 450 41
Powell WI 19 23.93 44 20.0 32 2.315 30 503 37
Pollock RSA 17 21.90 45 5.9 37 2.271 35 472 40
LeeB Aus 23 18.93 46 13.9 35 2.287 33 346 47
Vettori NZ 16 15.61 47 16.2 33 2.359 27 368 45
∗ The LG ICC ranking displayed is for the 47 players of this table having excluded players from the LG ICC lists not qualifying by the criteria of this
paper.

It will be seen that the correlation of the number of innings performance evaluation of the particular aspect of batting that
with the standard batting average, 0.004, is not only insignif- it is meant to reflect.
icant in the statistical sense but it is insignificant also in the On a two-tailed alternative hypothesis, the number of
cricketing sense. Such insignificance provides justification for innings by batsmen, correlates lowly, at 0.169, and statistically
using the standard batting average as a long-term measure of insignificantly, with resource percentage average. This is
734 Journal of the Operational Research Society Vol. 59, No. 6

Table 2 Correlations of 47 players’ batting performances and number of innings


Batting Batting Batting Batting Batting LG ICC
correlations innings avge contrib res% avge rating
Batting innings 1
Batting avge 0.004 (NS) 1
Batting contrib 0.590 ( p < 0.0001) 0.479 ( p < 0.001) 1
Batting res% avge 0.169 (NS) 0.746 ( p < 0.0001) 0.810 ( p < 0.0001) 1
LG ICC rating 0.400 ( p < 0.001) 0.437 ( p < 0.001) 0.499 ( p < 0.001) 0.35 ( p < 0.05) 1

Table 3 Progressive batting performance measures for Ponting (Aus)


Prog. Cum. Batting Prog.
ODI Times Batting batting Net net res% Res% res% Batting
series Runs out avge avge contribution contrib contrib avge avge innings
VB02 254 7 31.75 31.75 15.32 15.32 106.9 2.376 2.376 8
VB03 306 8 43.71 37.33 43.14 58.46 118.9 2.573 2.479 8
WI-Aus03 200 5 40.00 38.00 24.32 82.78 80.6 2.482 2.480 7
TVS03 257 6 42.83 39.12 29.98 112.76 106.1 2.421 2.465 7
VB04 315 9 35.00 38.06 43.55 156.31 125.5 2.509 2.475 9
SL-Aus04 257 4 64.25 40.74 73.44 229.75 83.5 3.078 2.556 4
VB05 184 7 26.29 38.54 −33.82 195.93 100.7 1.827 2.455 8
NZ-Aus-05 266 3 88.67 41.61 78.66 274.59 83.8 3.173 2.529 4
NatWest05 135 6 22.50 39.53 −21.26 253.33 71.2 1.896 2.478 6
Totals 2174 55 39.53 253.33 877.4 2.478 61

suggesting, as is the case with standard batting average, that 300 3.0

with sufficient volumes of input, the resource percentage 250


2.8
Runs - Avge or Cum

average will stabilize in the medium to long term enabling 200


Contribution

Res % Avge
players with unequal volumes of input to be compared 2.6
150
directly. However, resource percentage average will not 2.4
have the deficiency of the standard batting average of being 100
2.2
distorted by the number of ‘not-out’ innings in the players’ 50
careers. 0 2.0
Table 2 shows that the number of innings by a batsman
02

03

04

04

05

05

S
s0

S0

05
VB

VB

VB

us

VB

s-
Au

TV

W
Au
-A

correlates much more highly, at 0.590, with net contribution N


I-

SL

Z-
W

and is statistically significantly different from zero. This is ODI Series over time
suggesting the more that batsmen play the greater the effect
on their net contribution. As there is a clear cause and effect, Figure 1 Progressive batting measures: Ponting (Aus).
this indicates strongly that the net contributions of the better
batsmen will generally increase the more they play, whereas average is fairly stable and above the average of 2.35 (or
for weak batsmen net contributions will generally decrease 235 RM), his net contribution has a strongly upward trend.
the more they play. Consequently, for such players as Ponting, their net contri-
In order to illustrate this feature, Table 3 summarizes butions do not stabilize in order for this measure to be used
the progressive batting performance of Ponting (Aus), who to compare players with different volumes of input.
has appeared in all the tournaments of this study involving This characteristic has implications on how net contribution
Australia. For a previous period of his career, Beaudoin and can be used as a long-term measure of batting performance. It
Swartz (2003) found Ponting’s batting average to be 40.6 suggests that the measure might not be helpful in comparing
with an RM of 251.4, which are very similar to the batting active players who have played in significantly different
average of 39.53 and the equivalent resource percentage numbers of matches. However, the measure’s use over
average of 2.478 for the study in this paper. Figure 1, in players’ careers would still be of value, and maybe more so,
which the horizontal axis is not to a time scale but represents than another commonly used measure of career performance,
the various ODI series above in chronological order, illus- namely the aggregate of the batsmen’s runs. Aggregates of
trates very clearly that while Ponting’s resource percentage runs are monotonic increasing measures that reflect, to a
AJ Lewis—Extending the range of player-performance measures 735

Table 4 Career runs for selected ODI batsmen


Career Career Batting D/L net D/L
Batsman Country runs matches average contribution innings
Tendulkar Ind 14 146 362 44.20 280.3 21
Jayasuriya SriL 10 625 357 31.90 133.3 18
Ganguly Ind 10 123 279 40.65 −37.6 21
Azaruddin Ind 9378 308 36.92 Retired
Lara WI 9359 259 41.41 158.5 22
Dravid Ind 9048 280 40.21 −1.5 24
Ponting Aus 8958 247 42.25 253.3 61
Gilchrist Aus 7999 235 36.35 508.0 56
Kallis RSA 7954 229 44.18 40.6 20
Waugh S Aus 7569 325 32.90 De-selected
Bevan Aus 6918 232 53.58 De-selected
Border Aus 6524 273 30.62 Retired
Hayden Aus 4131 119 40.10 294.8 49
Trescothick Eng 3923 114 36.66 396.3 53
Lehmann Aus 3078 117 38.96 126.5 14
Flintoff Eng 2601 98 35.14 309.8 31
Pietersen Eng 888 25 68.30 252.3 13

substantial extent, the sheer volume of matches played. If a player’s aggregates could decline as well as rise then
Table 4 lists the career totals of runs for the batsmen in assessment of merit would be based more on the quality of
Table 1 whose career aggregates of runs in ODIs are in the their performances and less on their quantity.
top 25, as summarized in Engels (2006), and updated to 1st
March 2006 (http://www.stats.cricinfo.com/guru, accessed Reliability and validity
8 November 2006). Several players are included in the list The discussion of the previous section illustrates the validity
of Table 4 whose ODI careers have terminated due to de- of the D/L measures for assessing batting performance. As
selection or retirement. The net contributions of batsmen required by Bryman and Bell (2003), there are significant
qualifying for this study are provided for comparative correlations of these with the traditional measure of batting
purposes. The correlation of career runs to career matches average, although high correlation is not expected because of
of the 25 players is 0.927, whereas the correlation of D/L the different construct of the D/L measures compared with
net contribution to number of innings, for the smaller set of the construct of batting average. There is also significant,
qualifying players in Table 4, is 0.661 so that this measure yet moderate, correlation of the D/L measures with the LG
is less dependent on volume of input. In addition, since the ICC ratings, with variations expected for the same reasons of
aggregate D/L net contribution can decline as well as rise, construct.
periods of loss of form or declining ability are reflected in Bryman and Bell’s requirement for reliability for the
the reduction of net contribution, it becomes a basis for the resource percentage average is met by its insignificant corre-
comparison of players’ performances over whole careers and lation with number of innings, and supported by the similarity
for comparison of players of different eras. of the results of this study to that of Beaudoin and Swartz
For example, in Table 4 we see that Tendulkar (Ind) (2003). In the long term, therefore, resource percentage
has scored more runs in ODIs than any other player. He average would reliably compare batsmen’s performances.
heads this performance measure by a considerable margin. Net contribution is also a valid measure according to
However, he has played in far more matches than any other Bryman and Bell by similar arguments to the previous
player. Although his batting average would also be mentioned paragraph. However, net contribution does not stabilize, as
in discussion in order to help justify his top ranking, it required by Bryman and Bell, consequently it is unreliable
is argued that a career total net contribution would be a for comparing batsmen’s careers while they are in progress,
better measure by which to compare the longest serving but it has considerable relevance when comparing batsmen’s
players. whole careers.
In addition, Table 4 shows Ponting has accumulated almost
9000 runs from his 247 matches, and has a slightly lower
batting average than that of Tendulkar. For the series covered Evaluating the bowling performance measures
in this study, his net contribution from 61 matches is slightly Table 5 summarizes the three bowling performance measures
below that of Tendulkar but, from Table 3, it has not risen of standard average, net contribution and resource percentage
continuously as does his career aggregate of runs; two of average. These cover the 36 players in this study who have
his last three series have shown negative net contributions. played in more than 12 innings and whose appearances have
736 Journal of the Operational Research Society Vol. 59, No. 6

Table 5 Summary of ODI bowling performance measures


Bowling Bowling Bowling Res% LG LG
Bowling Bowling avge Bowling contrib res% avge ICC ICC
Player Country innings avge rank contrib rank avge rank rating rank *
McGrath Aus 34 19.30 1 388.8 1 1.686 1 865 1
LeeB Aus 48 22.32 2 246.3 2 2.015 3 819 3
Kasprowicz Aus 13 22.50 3 69.8 6 1.985 2 632 12
Ntini RSA 22 22.82 4 90.8 4 2.222 8 757 6
Flintoff Eng 31 24.50 5 55.7 7 2.163 5 727 8
Styris NZ 14 25.53 6 −1.8 15 2.294 12 629 13
Khan Ind 13 25.75 7 −63.5 28 2.509 20 625 16
Anderson Eng 29 25.95 8 24.9 11 2.243 9 652 11
Lehmann Aus 13 26.53 9 −37.9 19 2.592 24 436 30
Agarkar Ind 17 26.96 10 −36.4 18 2.395 17 596 20
Gough Eng 42 28.01 11 79.7 5 2.178 6 732 7
Harmison Eng 27 28.27 12 −3.7 16 2.297 13 618 18
Cairns NZ 18 29.45 13 −14.2 17 2.365 16 624 17
Gayle WI 18 29.50 14 14.7 13 2.412 18 525 24
Hogg Aus 37 29.64 15 18.2 12 2.249 10 627 15
Pollock RSA 23 29.92 16 136.6 3 2.020 4 805 4
Bravo WI 13 30.15 17 32.6 9 2.302 14 452 29
Vaas SriL 18 31.33 18 −38.3 20 2.702 29 831 2
Clarke Aus 33 32.61 19 −89.6 31 2.677 28 281 35
Boje RSA 15 32.73 20 12.8 14 2.354 15 560 22
Gillespie Aus 39 33.24 21 −133.7 34 2.503 19 786 5
Pathan Ind 13 33.68 22 −60.7 26 2.520 21 629 14
Watson Aus 17 34.11 23 −81.4 29 2.660 25 370 33
Kallis RSA 19 34.85 24 −60.5 25 2.712 30 470 28
Vettori NZ 25 35.36 25 26.4 10 2.263 11 722 9
Symonds Aus 45 37.42 26 −164.6 35 2.528 23 515 26
Collingwood Eng 31 38.83 27 −132.2 33 2.672 27 393 31
Nehra Ind 13 40.61 28 −104.4 32 2.819 34 607 19
Ganguly Ind 14 40.66 29 −63.2 27 2.733 32 314 34
Giles Eng 31 43.03 30 41.5 8 2.196 7 675 10
Sehwag Ind 14 45.50 31 −53.1 23 2.671 26 390 32
Vaughan Eng 16 46.00 32 −57.5 24 2.71 31 279 36
Jayasuriya SriL 18 48.30 33 −84.4 30 2.815 33 553 23
Harris NZ 17 49.54 34 −46.3 22 2.525 22 474 27
Dillon WI 15 57.83 35 −40.0 21 2.897 35 594 21
Kumble Ind 16 59.33 36 −168.4 36 2.968 36 519 25
∗ The LG ICC rankings quoted are for the 36 players of this table, suitably adjusted for ODI players not qualifying for the table.

been sufficiently recent that they have acquired and retain Elsewhere in Table 5, it will be seen that there is less
an LG ICC rating. A notable, and regrettable, absentee from variation in the rankings than with the batsmen in Table 1.
the list is Muralitharan (SriL), who has an LG ICC rating of Significant variations, however, are evident such as for Khan
742 and is ranked 3rd in the complete listings of 31st May (Ind), Pollock (RSA), and Giles (Eng). Whereas Khan has a
2005. His omission from Table 5 is because injury prevented 7th-ranked bowling average of 25.75 (and 26.6 by Beaudoin
him from playing sufficient innings in the few matches of the and Swartz, 2003) which relies heavily on wickets taken, the
limited number of series covered by this paper in which Sri runs conceded for the stage of innings bowled has led to a
Lanka took part. negative net contribution. The reverse is the case for Pollock.
Table 5 ranks the players by their standard bowling aver- Although his bowling average of 29.92 is modest for this study
ages. There would likely be no disagreement in the conclu- (formerly 23.0 by Beaudoin and Swartz, 2003), his relative
sion that the best bowler within the study of this paper has economy is reflected in a strong net contribution of +136.6
been McGrath (Aus); there is a clear difference in all his runs during the study and a much better than normal resource
performance measures from the next-best bowler. In addition, percentage average at 2.020 (formerly 1.866 by Beaudoin and
there is consistency in his performances with the Beaudoin Swartz, 2003).
and Swartz study, which found for him a bowling average of Despite a 30th-ranked bowling average (partly due to the
20.1 and an RM of 164.0. In their study, Muralitharan just small number of wickets taken), Giles (Eng) has been an
edged McGrath into 2nd place. effective one-day bowler by bowling relatively economically
AJ Lewis—Extending the range of player-performance measures 737

Table 6 Correlations of players’ bowling performances and number of innings


Bowling Bowling Bowling Bowling Bowling LG ICC
correlations innings avge contrib res% avge rating
Bowling innings 1
Bowling avge −0.272 (NS) 1
Bowling contrib 0.300 (NS) −0.580 ( p < 0.001) 1
Bowling res% avge −0.422 ( p < 0.05) 0.746 ( p < 0.0001) −0.855 ( p < 0.0001) 1
LG ICC rating 0.375 ( p < 0.001) −0.455 ( p < 0.001) 0.595 ( p < 0.001) −0.628 ( p < 0.0001) 1

to produce a net contribution of +41.5 and a slightly better mances in the medium term and also between player’s who
than normal resource percentage average of 2.196. These have played different numbers of times for their countries.
figures have helped raise his ranking to the top ten by all the
D/L-based measures. Reliability and validity
In a similar vein to that of batsmen, the discussion of the
Bowling correlations previous section illustrates the validity of the D/L measures for
assessing bowling performance. There are significant correla-
Table 6 summarizes the correlations between the bowling tions of these with the traditional measure of bowling average
performance measures and the number of innings played that that confirm their validity (Bryman and Bell, 2003). Further-
produced them. more, high correlation is not expected because of the different
Similar to the situation with the batting measures, the construct of the D/L measures compared with the construct
correlations between the three bowling performance measures of bowling average. As further evidence of validity, there are
are all statistically significant and are moderate to large, the also significant yet moderate-to-large correlations of the D/L
largest again being between the D/L-based measures and the measures with the LG ICC ratings, with variations expected
lowest between the standard bowling average and net contri- for the same reasons of construct.
bution. Bowling average and bowling resource percentage In contrast with the batting measures, Bryman and Bell’s
average are very highly correlated (0.746). In order to illus- requirement for reliability for the resource percentage average
trate this feature, in Table 5 it can be seen that there is is not quite met as there is a significant correlation ( p < 0.05)
only one name different in the top 12 players by these two with the number of innings. Nevertheless, the results of this
measures, and there is close agreement in their rankings. study are similar to the study by Beaudoin and Swartz (2003),
Beaudoin and Swartz found an even higher correlation of and, for reasons stated in the previous section, it is reasonable
0.93 but they used only the top 12 bowlers active at the time. to believe that this measure would reliably compare bowlers’
In contrast, however, with the situation relating to the performances in the longer term.
batting measures, the bowling net contribution is lowly corre- By arguments similar to those for the resource percentage
lated (0.300) with the number of innings in which the bowler average, bowlers’ net contribution is also a valid measure
contributes and it is not statistically significantly different (Bryman and Bell, 2003). In assessing its reliability, however,
from zero. It would appear, therefore, that for medium and it is noted that net contribution is not significantly correlated
long-term purposes, net bowling contribution may well be with number of innings. This would suggest that it would be
suited for comparison of players’ performances. This would stable in the longer term, which is counter-intuitive. However,
be applicable within careers and over whole careers. It would possibly due to the small sample size, and that bowlers only
be an additional measure not only to the traditional career get the opportunity to bowl 20% of the overs of the innings
bowling average but also in addition to the aggregate number (Engels, 2006), it seems likely that the pattern and conclu-
of wickets, which is also used to assess bowlers’ career sion for batsmen has not yet been confirmed, in the way
performances (Engels, 2006). that logic suggests it should. Further analysis of more data
The correlation of bowlers’ number of innings with and players could indicate that net contribution is unreliable,
resource percentage average, at −0.422, is significant at the according to Bryman and Bell, for comparing bowlers’ careers
5% level. Contrasted with the situation for the equivalent while in progress. However, again it would have consider-
batting measure, this might suggest that resource percentage able relevance when comparing bowlers’ whole careers as a
average would not stabilize in the medium to long term in contrast with career-total of wickets taken, which is mono-
order to be used as a performance measure. Nevertheless, tonic increasing.
because of the apparent equivalence with the situation for
the batting measure, there is considerable logic, (supported
Evaluating all-round performances
by the fact the correlation with bowling average is highly
significant) that resource percentage average would still be Although the batting net contribution depends upon frequency
a valuable measure by which to compare bowling perfor- of appearances, nevertheless when combined with bowling
738 Journal of the Operational Research Society Vol. 59, No. 6

Table 7 Summary of ‘all-round’ performance measures for qualifying players


Total RM LG
Total contrib Batting Bowling RM Diff ICC
Player Country contrib rank RM RM Diff rank rank
Flintoff Eng 365.4 1 299.8 216.3 83.5 1 1
LeeB Aus 260.2 2 228.7 201.5 27.2 2
Pollock RSA 142.5 4 227.1 202.0 25.1 3 2
Lehmann Aus 88.6 5 283.6 259.2 24.4 4
Cairns NZ 44.7 8 250.1 236.5 13.7 5
Vettori NZ 42.6 9 235.9 226.3 9.6 6
Symonds Aus 76.3 6 259.6 252.8 6.8 7
Clarke Aus 154.3 3 264.9 267.7 −2.7 8
Jayasuriya SriL 48.9 7 266.3 281.5 −15.2 9
Gayle WI 25.4 10 221.7 241.2 −19.6 10 4
Styris NZ −11.5 11 206.3 229.4 −23.1 11
Collingwood Eng −36.8 15 243.1 267.2 −24.1 12
Hogg Aus −26.6 14 192.8 224.9 −32.2 13
Sehwag Ind −11.5 12 233.2 267.1 −33.8 14
Kallis RSA −19.9 13 230.6 271.2 −40.6 15
Vaughan Eng −56.8 16 219.8 271.4 −51.6 16
Harris NZ −96.7 17 184.4 252.5 −68.1 17
Ganguly Ind −100.9 18 202.4 273.3 −70.9 18

net contribution, their sum, identified as total net contribu- Collingwood (Eng) also is thought of as an all-round player,
tion may still have value in order to provide an indicator of a but his bowling performances in these series, summarized in
player’s all-round ability. All-round performance will also be Table 5, have not supported his stronger batting so that his
assessed using the Beaudoin and Swartz (2003) RM difference all-round rankings are rather low at 12 and 15.
and will be compared with the recently introduced LG ICC
all-round ranking. Table 7 provides the all-round information Reliability and validity
on the 18 players in this study who had batted and bowled Subject to the previous caveat in relation to the reliability
in more than 12 innings and whose performances were suffi- of the bowling resource percentage average, the derivative
ciently current that they retained an LG ICC rating for both all-round measure of the RM difference would carry the prop-
batting and bowling. The ranking of these players is provided erties of its two parts and would therefore exhibit reliability
according to their RM difference over the 16 one-day series (Bryman and Bell, 2003). Although batting and bowling net
considered in this paper. contributions do not exhibit reliability, because of the very
Again, several interesting aspects of their rankings are high correlation of the all-round measures of total net contri-
apparent. There is substantial agreement in the rankings with a butions and resource percentage average there is good support
Spearman’s rank correlation coefficient of 0.930. By a consid- for the use of total net contribution as an all-round measure
erable margin, the best all-rounder is Flintoff (Eng) by both of performance.
of the D/L-based measures. Whereas the two D/L-based measures of all-round ability
Since all bowlers get some opportunities to bat, in the are consistent among themselves, there is limited external
modern game it is becoming more and more important for evidence available to ascertain their validity (Bryman and
bowlers to improve their batting skills. This is not only to Bell, 2003). However, the LG ICC all-rounder ratings of Table
score runs but also, towards the end of the innings, to support 7 are clearly consistent and therefore are not unsupportive of
the other batsman who may be a higher-order player capable their validity.
of scoring many runs and guiding the team to victory. Lee
(Aus) is regarded as a specialist bowler but he has acquired
Proposals for D/L performance indices
batting expertise of the nature mentioned in the VB series of
2003 (Lewis, 2005) which, when aggregated with his perfor- Although suspect in the same way as standard batting and
mances in the other series in this study, rates his all-round bowling averages to low volumes of input, the resource
performance as second. Pollock (RSA) is also regarded as an percentage averages, and their RM derivatives, arguably
all-round player. He is a better batsman than Lee but not quite provide a reliable and valid way of comparing batsmen and
as effective as a strike bowler and has been ranked just below comparing bowlers for the medium and longer term and when
Lee by this study. there are differing volumes of input. However, in their current
AJ Lewis—Extending the range of player-performance measures 739

form they are incompatible for the purposes of comparing of batsmen with bowlers impossible. In order to preserve the
batsmen with bowlers. A difficulty also is in their popular nature of high indices representing good performance and
interpretation, although Beaudoin and Swartz have made a high ranking, the proposal for the bowler’s index Ib is the
valiant attempt with the RM measures themselves. But the reciprocal of the formula for the batsman so that Ib is defined
interpretation of the all-round RM difference, as summarized as Ib = Z (50, 0)/Pb .
in Table 7, as being the average net performance of the player As an example, Lehmann from Table 5 would have a
were he to do all of the batting and bowling himself (Beau- bowling index of 235/2.592=90.7. His bowling performances
doin and Swartz, 2003) is perhaps rather a severe stretch would be 90.7% of an ODI average bowler, which is 9.3%
of the imagination for the typical cricket supporter. This below average.
paper now discusses a further derivative of the D/L-based
performance measures in order to create player-performance
All round index
indices that could overcome the problems of interpretation.
An all-round index Ir could be defined as the geometric √ mean
of the batting and bowling indices so that Ir = Ia × Ib .
D/L batting and bowling indices
For
√ Lehmann, his all-round index would have the value of
The LG ICC ratings of teams, introduced in 2002 120.7 × 90.7=104.6 representing an all-round performance
(http://www.lgiccrankings.com/, accessed 8 November 2006), 4.6% above average.
rank the major cricket playing countries using an index based It will be seen that Z (50, 0) is √ eliminated and the index
upon 100 as the ‘average’ both for Test matches and ODIs. As is thereby defined as merely 100 Pa /Pb . The removal of
an example, on 30th June 2005, among the 11 ODI-playing the D/L reference point means that an index of, say, 104.6
countries, Australia topped the ratings with 137, England may have arisen from many combinations of batting resource
were 6th on 106, West Indies 9th on 95 and Zimbabwe’s percentage average Pa and bowling resource percentage
9th-ranked rating was 50. Since such ranking indices are average Pb . For Lehmann these were 2.836 and 2.592,
now familiar to cricket enthusiasts, equivalent performance respectively. Another combination of say 1.845 and 1.686,
measures for players are now proposed. respectively would yield the same all-round index of 104.6
In Lewis (2005), a batsman’s
 resource percentage average but with a much stronger bowling than batting performance.
for a match is defined as si / pi where the si are the runs Far from being a disadvantage to the proposed index, it is
scored on the ith ball remaining. The pi are the resources suggested that the endless permutations in obtaining a given
consumed for the ith ball remaining including the resource value of the index would be appreciated by aficionados of
percentage consumed for the loss of the wicket at the terminal cricket statistics as providing quantitative information to
ball. These are then summed over the balls received during discussions on who is, or was, the best all-round cricketer!
his innings. This is easily extended to the many matches in a Although the geometric mean is a recognized mechanism
player’s career. The resource percentageaverage  Pa for
 the for the averaging of indices, there is arguably a very prac-
batsman’s career to date would be Pa = j i si j / j i pi j tical cricketing interpretation to its square. This interpretation
where si j now represents the runs scored from the ith ball would be that of the compound effect of the player’s perfor-
remaining in the jth innings in his career. There would be a mance in the two disciplines. For Lehmann, his all-round
similar quotient Pb for a bowler where the si j of runs con- index would then be 109.4. Lehmann has a batting index of
ceded, would be replaced by si j +h i j , which includes the num- 120.7 representing performances 20.7% above average, but his
ber of extras, that is, wides and no balls h i j , that are charged bowling index of 90.7 represents bowling performances 9.3%
against the bowler. The pi j are the resource percentages below average. The multiplication of these indices gives 109.4
contributed by the bowling of balls and the taking of wickets. and provides a compounded all-round performance measure
The average value of P(.) is Z (50, 0)/100, namely 2.35 representing 9.4% above that of an ‘average’ ODI cricketer.
in the Standard Edition of the D/L model (Duckworth and Table 8 summarizes the various indices of all the players
Lewis, 2004) that has been used in the analysis and has been in this study who had contributed in more than 12 one-day
applicable for the majority of the series of matches identified innings and held LG ICC ratings that were current for both
above. The proposal is that a player’s batting index Ia is of cricket’s main disciplines. These players’ ratings and rank-
defined as 100Pa /Z (50, 0) and expressed as a percentage; the ings include their D/L, RM and LG ICC ratings for batting
higher the figure the better the evaluation of the batsman. For and bowling, and their rankings as all-rounders. Clearly, for
a batsman such as Lehmann (Aus) from Table 1, his batting ranking purposes, there is no difference between using the
index, Ia , would be 100 × 2.836/2.35 = 120.7. This would be geometric mean or its square. Table 8, however, records the
interpreted as assessing Lehmann as a batsman performing at square of the geometric mean as the D/L all-round index.
20.7% above that of an average ODI batsman. The LG ICC all-round rankings are also based on the square
The problem with the equivalent bowler’s index is that of the geometric mean, or simply, the product of the player’s
good performance would generate a low index, contrary to the LG ICC ratings. With such a mechanism a player with LG
batting index, which would again make direct comparisons ICC batting and bowling ratings each of 500 would rate better
740 Journal of the Operational Research Society Vol. 59, No. 6

Table 8 D/L indices for batting, bowling and all-round performance measures for qualifying players
RM LG ICC LG ICC LG ICC D/L D/L D/L D/L
RM diff batting bowling all-round batting bowling all-round all-round
Player Country diff rank rating rating rank index index index rank
Flintoff Eng 83.5 1 711 727 1 127.6 108.7 138.6 1
LeeB Aus 27.2 2 346 819 97.3 116.6 113.5 2
Pollock RSA 25.1 3 472 805 2 96.6 116.3 112.4 3
Lehmann Aus 24.4 4 544 436 120.7 90.7 109.4 4
Cairns NZ 13.7 5 588 624 106.4 99.4 105.8 5
Vettori NZ 9.6 6 368 722 100.4 103.8 104.3 6
Symonds Aus 6.8 7 660 515 110.4 93.0 102.7 7
Clarke Aus −2.7 8 705 281 112.7 87.8 99.0 8
Jayasuriya SriL −15.2 9 670 553 113.3 83.5 94.6 9
Gayle WI −19.6 10 717 525 4 94.3 97.4 91.9 10
Collingwood Eng −24.1 12 512 393 103.4 88.0 91.0 11
Styris NZ −23.1 11 535 629 87.8 102.5 89.9 12
Sehwag Ind −33.8 14 618 390 99.3 88.0 87.3 13
Hogg Aus −32.2 13 407 627 82.0 104.5 85.7 14
Kallis RSA −40.6 15 708 470 98.1 86.6 85.0 15
Vaughan Eng −51.6 16 569 279 93.5 86.6 81.0 16
Ganguly Ind −70.9 18 557 314 86.1 86.0 74.1 17
Harris NZ −68.1 17 426 474 78.5 93.1 73.0 18

as an all-rounder than say a player with ratings of 900 and criteria above, there is no need for a subjective judgment as
100 (in either permutation). The website for the LG ICC ODI to whether the player is considered an all-rounder. A substan-
player ratings initially posted the actual ratings for just the tial proportion of batsmen do not bowl and they would not,
top five players deemed to be international all-rounders. More therefore, meet the minimum bowling contribution require-
recently, this list has been extended to the top 10 all-rounders ment for inclusion. However, all bowlers will be required to
(http://www.cricketratings.com, accessed 8 November 2006). bat at some stage. Subject to minimum participation require-
The relative rankings of batsmen by the D/L index are, of ment, in due course bowlers will acquire a batting index and
course, identical to that using the batting RM since the former hence an all-round index. It would be their ability in both
is a constant divisor of the latter. The relative bowling rankings disciplines of cricket that would indicate, or otherwise, that a
by the D/L index are slightly different from the RM bowling bowler could be regarded as an all-round player, with subjec-
rankings due to the reciprocal nature of the two measures. The tive judgment thereby removed. For example, in Table 5 Lee
D/L all-round index provides identical rankings to the RM (Aus) is a highly rated bowler by all measures. From Table 1,
difference for the top ten players. Thereafter there are minor he has also made a small but positive net contribution of 13.9
differences in the rankings of no more than one position. It is runs with the bat even though his batting average is below
argued, however, that Flintoff’s clear first-place ranking, with 20 and his D/L batting index is just below 100. In this study,
an index of 138.6, assessing the player’s combined perfor- his combined performances have placed him in second place
mances as 38.6% above average would have more meaning of the all-rounder list of Table 8. Compared with compatriot
than the equivalent RM difference. To rate him as scoring on Lehmann, who is a better batsman but a poorer bowler, he
average 83.5 runs per match more than the opposition were has performed slightly better overall as an all-rounder.
he to do all of the batting and bowling himself, might stretch
abilities of interpretation rather more than the proposed D/L Reliability and validity
all-round index.
Previous sections have discussed the reliability and validity of
There is little information provided on the ICC website for
the batting and bowling resource percentage average. Since
the all-rounder rankings. The rankings provided in Table 8,
these measures are the basis of the construct of the D/L
as applicable on 31st May 2005, are consistent among them-
indices, it is argued that these indices will therefore exhibit the
selves but omit two all-rounders from Pakistan who have not
same characteristics. In addition there is consistency with the
qualified for inclusion in the analysis of this paper due to
limited information from the LG ICC all-rounder rankings.
insufficient contributions with either bat or ball or both. The
LG ICC ratings are also unclear on the criteria by which a
player is deemed an all-rounder. Indeed, at the time of this Applications for the D/L performance measures
study, the site only ranked five players suggesting a dearth of There are two areas where it is anticipated the measures
players purporting to be international all-rounders. Using the produced in this paper could be useful within the cricketing
AJ Lewis—Extending the range of player-performance measures 741

world. The obvious one, as indicated in the paper’s title, is time. Further work might suggest that performances could
in the extension of valid measures of performance of players. be discounted exponentially. If the function were analogous
For cricket enthusiasts there would be more objective ways of to a financial discount factor, recent performances would be
comparing batsmen and comparing bowlers. In addition, some discounted more quickly than earlier performances. This is
of the measures devised enable batsmen and bowlers to be likely to be contrary to the intent of the LG ICC ratings and
compared directly, hitherto impossible with the old measures, there would be no natural cut-off point when a player could be
along with the ability to combine batting and bowling perfor- removed from the ratings as having lost all currency of perfor-
mances into an all-rounder measure. mance. A more realistic function might be one that discounts
The second major area of usage would be as aids to team recent data slowly, but earlier data to zero more quickly. Such
management and coaches. The measures are not intended to functions might include those from the family of exponential
have predictive power but, for individual matches or short smoothing functions.
series, net contribution in particular highlights, clearly, where
batsmen and bowlers achieve positive or negative contribu-
Summary
tions, which can aid in coaching strategies, in team selection
and in deciding the batting order. This paper has looked at whether the measures of player
In addition, the measures produced arguably are fairer in performance called net contribution and resource percentage
evaluating performance and contributions within a team and average for batsmen and bowlers as defined in Lewis (2005)
that self-sacrifice by a batsman towards the end of the innings are valid and reliable for the purposes of evaluating ODI
is not counted significantly against him. Similarly, a bowler cricket players in the longer term and over players’ careers.
credited with sacrificed wickets is not unjustifiably flattered. Through the analysis of many series of matches, with a
three-year time span and involving several of the world’s
regular and leading players, it has been shown that there are
Further study on weightings of contributions
some limitations to the net contribution measure. Whereas this
There is one area of further study that suggests itself in order measure was shown to be very effective in evaluating perfor-
to improve the D/L-based measures that have been discussed. mances in single matches and for a series of matches with
This is in relation to the weighting of performances during a major advantage of being able to assess all-round ability,
players’ careers. the analyses in this paper reveal that the measure depends
The data used in this study date from January 2002 to July on the frequency of playing. This suggests that net contri-
2005. During this time many players have appeared for their bution would be a valid and reliable measure for comparing
respective countries. Some of these have played many times the careers of players overall. Traditionally aggregate runs
but others have made few appearances. Some have been near are used to evaluate batsmen’s careers and total wickets used
the end of their careers and others either have since retired or to evaluate bowlers’ careers. Whereas these measures cannot
have no longer been selected for their national teams. decline, the net contribution would have this characteristic so
Of those players who have met the minimum contribution that continuing playing but to a lower standard would reduce
criteria described, several have made appearances throughout the net contribution and arguably better reflect career-length
the time-span of the study. Their D/L-based performance performance.
measures and their standard batting and bowling measures The resource percentage average was shown to be unstable
include all performances with equal weight so that, subject to and therefore not a reliable measure when applied in small
the limitations already outlined, the measures can be used to numbers of matches (Lewis, 2005), but it becomes a more
compare players’ careers, both ongoing and as a whole. From reliable measure of performance over the longer term. This
the outset, this has been a primary focus of the study and, measure, for batsmen, combines features of quantity of runs
it is argued, has been quite successful in defining potentially scored but suitably weights the value of the loss of the
valid and reliable additional mechanisms for this purpose. batsmen’s wickets for the stage of the innings it was lost
The LG ICC ratings and rankings assess current form and through the resources that were consumed. Similarly, the
contributions. Consequently, they have an inbuilt discounting measure encompasses aspects of the strike-rate measure by
procedure that weights recent performances greater than those better evaluating their merit according to stage of the innings
earlier in players’ careers and depreciates players’ earlier the runs were scored.
contributions more quickly especially when they are not There is, arguably, a practical difficulty in the interpretation
selected, or are not available. The LG ICC website already of resource percentage averages. Beaudoin and Swartz (2003)
cited provides an overview of the process of evaluating solve this by multiplying by 100 and calling the result ‘runs
players but does not provide the detail of their calculations per match’ that they interpret as the average performance of
due to commercial confidentiality. the batsman, or bowler, were he to do all of the batting, or
In order to compare the D/L-based performance measures bowling, himself. As an alternative, this paper proposes a D/L
with the LG ICC ratings more accurately, it would be desir- index, based on the player’s resource percentage average as
able to define a mechanism that discounts performances over a percentage of the ODI average of 2.35 runs per unit of
742 Journal of the Operational Research Society Vol. 59, No. 6

resource. As international country rankings are given in index • The two anonymous referees for very helpful comments to improve the
form, it is suggested that the cricketing public might more paper’s content and its presentation.
easily accept a measure of performance in this form.
An all-round measure of performance index is suggested
References
as the product of a player’s batting and bowling indices,
which has the virtue of reflecting the compounding effect of Ali M (1975). Stochastic dominance and portfolio analysis. J Fin
performance in the two disciplines. With a ‘norm’ of 100, Econ 2: 205–230.
the index would be more easily interpreted than the Beaudoin Barr GDI and Kantor BS (2004). A criterion for comparing and
selecting batsmen in limited overs cricket. J Opl Res Soc 55:
and Swartz RM difference and its ‘hard-to-swallow’ interpre-
1266–1274.
tation that it captures the average performance of the player Beaudoin D and Swartz TB (2003). The best batsmen and bowlers in
were he to do all of the batting and the bowling himself. one-day cricket. South African Statist J 37: 203–222.
Comparisons of rankings have been made with the LG Bryman A and Bell E (2003). Business Research Methods. Oxford
ICC ratings, which show reasonable agreement for the players University Press: Oxford, UK.
Cohen J (1988). Statistical Power Analysis for the Behavioral
qualifying for the analysis summarized in this paper. There
Sciences, 2nd edition. Lawrence Erlbaum: New Jersey.
are major differences in the methodology of these ratings in Damodaran U (2006). Stochastic dominance and analysis of ODI
comparison with the methods proposed in this paper. One batting performance: the Indian cricket team, 1989–2005. In:
difference relates to the limited nature of the information in Hammond J and de Mestre N (eds). Proceedings of 8th Australasian
a ‘standard’ scorecard from which the LG ICC ratings are Conference on Mathematics and Computers in Sport. Australian
Mathematical Society Inc.: Canberra, Australia, pp 73–80.
developed. A second difference is the use of a discounting
Duckworth FC and Lewis AJ (1998). A fair method for resetting the
factor in the LG ICC ratings that attempts to assess current target in interrupted one-day cricket matches. J Opl Res Soc 49:
rather than career levels of performance. Further work on 220–227.
this aspect is suggested if the methodology of the measures Duckworth FC and Lewis AJ (2004). A successful Operational
proposed in this and the previous paper (Lewis, 2005) are to Research intervention in one-day cricket. J Opl Res Soc 55:
749–759.
be incorporated into the LG ICC ratings, or are to be used to
Engels M (ed) (2006) Wisden Cricketers’ Almanack. John Wisden &
create a separate set of D/L performance measures. Co. Ltd.: Hampshire, UK.
Hopkins WG (2000). Measures of reliability in sports medicine and
science. J Sports Med 30: 1–15.
Acknowledgements — The author would like to acknowledge, with thanks, Lewis AJ (2005). Towards fairer measures of player performance in
assistance received in preparing this paper:
one-day cricket. J Opl Res Soc 56: 804–815.
Swartz TB, Gill PS, Beaudoin D and de Silva BM (2006). Optimal
• His colleague Frank Duckworth for valuable comments on the draft batting orders in one-day cricket. Comput Opns Res 33: 1939–
manuscripts
• Champion Data (Australia) and the Press Association (Sports) for the 1950.
provision, in electronic form, of the ball-by-ball details of the matches
absolutely essential for undertaking the analyses summarized in this Received March 2006;
paper. accepted November 2006 after three revisions

You might also like