Download as pdf
Download as pdf
You are on page 1of 15
Production Efficiency: The Case of Professional Basketball ‘Thomas A. Zak; Cliff J. Huang; John J, Siegfried The Journal of Business, Vol. 52, No. 3 (Jul., 1979), 379-392. Stable URL: butp//links jstor.org/sici?sict=0021-9398% 28 197907%2952%3A3%3C379%3APETCOPS%3E2.0,CO%3B2-%23 The Journal of Business is currently published by The University of Chicago Press. ‘Your use of the ISTOR archive indicates your acceptance of JSTOR’s Terms and Conditions of Use, available at hhup:/www.jstororg/about/terms.hml. JSTOR’s Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at hup:/www jstor.org/jouralsuepress.himl, Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the sereen or printed page of such transmission, STOR is an independent not-for-profit organization dedicated to creating and preserving a digital archive of scholarly journals, For more information regarding JSTOR, please contact jstor-info@ umich edu, upswww jstor.org/ ‘Tue Feb 24 03:08:11 2004 Thomas A. Zak* IMiinois State University Cliff J. Huang and John J. Siegfried* Vanderbilt University Production Efficiency: The Case of Professional Basketball Sports, like politics and religion, is a subject to be avoided if peace and quiet are to be maximized. Asking a fan “whose team is best” is akin to speculation about whose god is best. Yet an important difference exists: Even if gods have won-lost records, the statistics are not readily available. Sport, however, results in violent dis- agreements despite well-documented records. Many disputes center on teams with “great po- tential” somehow wasted or, conversely, those teams with seemingly inferior talent that consis- tently “‘play over their heads” and compile ex. cellent records. Does the best team always win? By estimating a production frontier, this paper considers the possibility that a team with the greatest potential may not prevail as league champion due to inefficient use of its available resources. Estimating the production function also allows us to examine the determinants of a team’s performance. ‘Method of Estimating the Production Frontier and Efficiency Actually, no theoretical distinction can be drawn between a production function and a production + We would like to thank the National Basketball Asso tion, especially Matt Winnick (director of media informatior {for providing data, and an anonymous referee (ot, however, in black-and-white-tiped shir) for useful comments on air draft ofthis paper. Computer time was provided by Vanderbilt University Ulournal of Business, 1979, vol. $2, n0. 3) {©1979 by The University of Chicago (0021-9398/79/503-0004801 25 a ‘This paper estimates production function for professional basketball teams. Techniques well known in the produc- tion-function literature are applied to data collected during the 1976-77 National Bas- ketball Association sea: son to determine the production frontier. Production-frontier estimates yield infor ‘mation about the impact of various inputs used in the production process as well as allowing interteam comparisons of potential output and average team efficiency In addition, the paper considers the role of the ‘home-court advan- tage” and examines the source of this empir- cally observed phenom- 380 Journal of Business frontier. Both define the maximum output attainable for a given combi- nation of inputs. Differences in the economics literature arise because until the late 1950s empirical estimates of production functions were actually estimates of an average production function. Recent econometric research has provided a number of techniques for es- timating the frontier production function. Timmer (1971) has applied linear programming techniques in his estimates, while Afriat (1972) and ‘hmond (1974) rely on a multiplicative error term to arrive at an estimate of the frontier. For a given vector of inputs x, the maximum output, or the frontier- production function, is denoted by F(x). The observed output ¥ differs from the frontier output by a factor u, that is, 7 Fo) w, a If we restrict w to between 0 and I, then u can be used as a measure of production efficiency." At the extremes, if a firm always receives the ‘maximum output from a given set of inputs, then observed output equals frontier output, and the firm is 100% efficient. The term u has the value 1. If, however, a firm receives no output from its inputs (a case as unlikely as the first), then it rates a zero measure of efficiency, that is, w = 0. Obviously the efficiency, u, is a random variable in the sense that it varies from firm to firm or from game to game. Afriat suggests that a beta distribution may be appropriate for the efficiency term u. In the case of a Cobb-Douglas form for the produc- tion frontier F(x), however, he and Richmond assume that v = —In wu has a gamma distribution with parameter \. Richmond goes on to show that the mean of the efficiency term, u, is equal to 2-». Geometrically, when 1, the distribution ofu indicates that mostobservationsare relatively inefficient. ‘Thus, byestimating the gammadistribution parameter, wecanrevealthe distribution of efficiency across firms or across games of a team and evaluate a team’s average efficiency. ‘Taking the logarithm of the production function, equation (1) yields In ¥ = In Fx) + Inu @ = [In F(x) = A) + A = vd 1. More recently, Aigner, Lovell, and Schmidt (1977) decomposed the term into two parts: a random component and a measure of production efficiency. Obviously. this Technique is theoretically superior: measurement error may cause observed output to be larger than the frontier. However, inthe case of profesional haskethall. measurement problems are more unlikely than for other data sources so, or simplicy's sake, we employ the Richmond technique in our empitical work Production Efceney 38 Since the mean and variance of a gamma distribution are both equal to the parameter X, the random variable (\ — v) has mean zero and constant variance 4. Using the ordinary least squares (OLS) method to estimate equation (2), we can therefore estimate 2 (i.e., the variance of the regression), hence the mean efficiency 2~* and the production frontier Fx). A critical aspect of production-function estimation is the require- ‘ment that output and all inputs be measurable. Accurate measurement is one of the principal advantages of using professional sports team data. Statistics are extensive, and, unlike industrial data, few disclo- sure problems are encountered. Furthermore, the unit of observation, a contest, is analogous to a production run in a factory; this enables us to estimate the production function for individual firms (teams) as well as for the industry (league) as a whole.” Empirical estimation of the production frontier for professional bas- ketball teams using the Richmond method enables us to separate team potential (its frontier) from a measure of its technical efficiency (how close it normally comes to the frontier). Combining the two elements associated with each team’s production function—efficiency and potential—allows us to calculate an index of team quality. In the absence of random disturbances (luck), the quality index should result in a ranking identical with the actual records. In addition, the production-function approach serves as an indicator of the relationship between inputs and their contribution to output. Computation of team-specific marginal products can identify those areas in which a team’s best opportunities for improvement lie. Of course this assumes that one input is as easily expanded as the next, an assumption that may conflict with reality. The seriousness of this assumption is mitigated somewhat by the fact that the measured inputs. (rebounds, assists, etc.) are jointly produced by individual ballplayers. Therefore, given a choice between two ballplayers (willing to work for the same Salary), a team can choose the one making the larger con- tribution to output, Empirical Example: Professional Basketball Using a Cobb-Douglas production function and data for individual ‘games during the 1976-77 National Basketball Association (NBA) sea- son, we estimate the equation Y = F(x) + u, where the production frontier is Fa) (fl Xa) emir ® 2. For more on the question whether the individual team or an entire league is analogous tothe frm. see Neale (1964) and Demmert (1973. pp. 15-20), aa Journal of Business and ratio of the final scores, atio of field goal percentages (+), ratio of free throw percentages (+), ratio of offensive rebounds (+), ratio of defensive rebounds (+), tio of assists (++), ratio of personal fouls (—), ratio of steals (+), X, = ratio of turnovers (—), inary variable for location (home = 1) (+), and difference in number of blocked shots (+) (expected signs are shown in parentheses). ‘The specification of the error term u is multiplicative since the ratio u = Y/E(x) measures the degree of efficiency. For the convenience of estimation it is natural to specify the production frontier F(x) to be a multiplicative (Cobb-Douglas) functional form too. In addition to this attribute, the Cobb-Douglas form has other desirable properties. First, the marginal products of each input in the Cobb-Douglas form depend on the levels of other inputs. This is an intuitively attractive character- istic for professional basketball. For example, the marginal product of shooting skills should depend on the number of times the team acquires the ball (via rebounds, steals, turnovers by the opposing team, and opposing-team scores) and has an opportunity to shoot. Second, the coefficients of the Cobb-Douglas form are easy to interpret—namely, as elasticities. Finally, the Cobb-Douglas form is a widely used specification, with which many people are familiar, It possesses an appealing simplicity. Consequently, we chose the Cobb-Douglas form for the estimated production frontier. Measuring output as the ratio of final scores captures the relative closeness of each game. Alternatively, one might use absolute score differences to measure output. The two measures will not yield identi cal results. For example, a game with a final score of 130 t0 120 is a "closer" contest than a game ending 90 to 81 when using ratios, although the latter contest has a smaller point difference. Our principal concern, however, is the use of a “relative” measure. Using total points scored is sensitive to an opponent's playing style (a freewheel- ing, fast-breaking club or a slowdown-pattern one). Besides, if a team averages 120 points per game (the highest in the league) but yields 125 points per game on average, one would not want {0 say that team has the largest output. Infact, the essence of sports competition isto gauge ‘one’s performance relative to others’. The possibility of using a binary variable for winning/losing is rejected because of the Cobb-Douglas Production Efficiency 38 form of the production function (estimated using logarithms). ‘The decision to use a ratio output measure instead of differences is arbi- trary; however, we believe that normalizing the margin of victory by the number of points scored reflects some of those aspects of defense not included in the direct input measures, as well as controlling for teams with different offensive tactics. ‘On the input side the case for using ratios is stronger. Once again, logic argues for a comparative measure of inputs. Does it matter if a team gets 50 rebounds if the opposing team has 75? If both teams capture the same number of rebounds, the use of differences would result (in the Cobb-Douglas specification) in zero output. Con- sequently, we employ ratios on the input side as well, except for those variables which might have an observed value of zero (location and blocked shots in our equations). Relative shooting percentages, both field goal and free throw, should make an important contribution to team output. Everything else equal, the better a team shoots relative to its opponent, the larger its output. Shooting percentages are used because they reflect how effectively a team shoots the ball more accurately than does the total number of shots. Moreover, the quality of a team’s shooting is, for our purposes, as important as the quantity of shots. Rebounding is also expected to have a relatively large impact on the outcome of a game. Outrebounding the opposition should improve a team’s chances of winning. Unlike shooting percentages, where we believe field-goal shooting makes a larger contribution to output than free-throw shooting (more points are scored in a game from the field than from the free-throw line), we cannot infer greater a priori impor- tance for either offensive or defensive rebounds. Assists are included to capture those aspects of ball handling and teamwork not captured in improved shooting percentages. Of course, ‘we anticipate that assists will make a positive contribution to output. ‘The ratio of steals and the difference in blocked shots should reflect aspects of defense that do not appear in reduced shooting percentages. While not expected to add a great deal to output, we expect a positive effect. Inputs that should reduce output are personal fouls (essentially al- lowing the other team to shoot more free throws) and turnovers (relinquishing the ball without a shot). All else equal, as the number of turnovers increases, output will be adversely affected. In one sense turnovers are the converse of assists; they may measure the lack of cohesiveness of a team during a game. They also proxy aspects of an ‘opponent's defense. ‘A binary variable, taking the value of one if team is playing at home and zero if an away game, is also included in the estimated equations to a ournal of Business account for the possibility that location affects the outcome of a game. Examination of estimated coefficients for this variable should indicate the magnitude of any advantage the home team enjoys. In terms of quantifiable measures, the above list exhausts the data source and covers nearly every aspect of the game (shooting. re- bounding, ball handling, and defense) Data were gathered from official NBA box scores for the five Atan- tic Division teams (Boston Celtics, Buffalo Braves, New York Knicks, New York Nets, Philadelphia 76ers) during the 1976-77 season. AF though the sample is based on these five teams, every other team in the NBA js represented approximately 20 times. Each team plays an 82- game schedule against the other 21 teams in the league. Therefore, teams face each other approximately four times during a season.” Some data are lost because the official score sheet was not received, or, im some instances, all of the information was not available for a given game. This accounts for the disparity between the number of observations in the estimates and 82. Since each team faced common foes a similar number of times, the information can be pooled for the Atlantic Division teams to estimate a production function for the entire NBA. For the aggregate NBA estimate the number of observations is further reduced to avoid double counting. For example, a Boston versus Buffalo contest is included only once, instead of as both Boston versus Buffalo and Buffalo versus Boston. This problem is encoun- tered, however, only when two Atlantic Division clubs play one an- other. Empirical Results Regression results are reported in table 1. The logarithmic form of the estimated relationship permits us to interpret most of the coefficients as elasticities of output with respect to the various inputs. Overall the results confirm our hypotheses. At the 5% level all statistically significant coefficients exhibit the predicted signs, and most of the coefficients are statistically significant. As expected, the largest output, elasticities are associated with shooting percentages, especially field- goal percentage. This means that the larger a team’s field-goal-shooting percentage is relative to its opponent's field-goal percentage, the larger its final point total will be relative to the opposition’s. However. a 1% increase in the ratio of shooting percentages results in a less than 1% inctease in the ratio of points scored. In the case of the aggregate NBA 3. Bach team faced 19 other teams four times during the season and two opponents three times (4% 19+ 23 = 82) “4 This isto be expected because an increase in shooting percentage substitutes for alternative ways of scoring (e-8.ofesive rebounding coupled with poorer shooting) Production Effiieney sowed ON, a S048 poyPOHEL -spxouiny, seg S100} uossag spunogar auisu9;2q _spamnogat ansuauo, «sr 2389 or00— zorusa 1 oun eum SorgeUeA, AN. eqioeg feuoRa}ONG 40} SIMUIBET YoHURA-UORDNPOS| ——T ATHVL 386 ournal of Business regression, a 1% increase in field-goal-shooting percentage leads to a 0.61% increase in the ratio of final scores. he elasticity of score ratios with respect to free-throw-shooting. percentage is less than that for field goals. This is to be expected for several reasons: first, free throws made contribute only one point to score rather than the two points awarded for field goals, and, second, missed free throws provide less opportunity for offensive rebounds due to the positioning of defensive players closest to the basket. Also, since free throws are sometimes awarded without a field-goal attempt, they are a less effective method of scoring, Other variables making a substantial contribution to output include offensive and defensive rebounds and, in several cases, steals. As expected, personal fouls and tumovers reduce output. The difference in blocked shots and the ratio of assists proved to be insignificant. The difference in blocked shots probably is insignificant because so few ‘occur in the average game. The average frequency of blocked shots is about 10 per game, which could have only a small impact on the ratio of final scores. The assists variable should probably be interpreted jointly with shooting percentage, since assists frequently lead to easy shots, which are missed less often. Consequently, the absence of an impact of assists separate from shooting percentage is not so surprising and, furthermore, does not indicate the irrelevance of assists. This is a case like the offensive lineman’s blocking assisting the running back in football, where joint production confounds the estimation of separate independent effects. The relatively high coefficient of determination for cross-sectional regression with so many observations confirms our belief that the list of included inputs is comprehensive. Contrary to the hypothesis that the home team has an advantage, the coefficient on the locational variable is consistently insignificant. Does this imply that a home-court advantage does not exist? Not necessar- ily, since the location variable does not include the effect of playing at home on the level of other inputs. Differences between teams can be examined by comparing coefficients for the five teams. Buffalo's and Philadelphia's output is a 200d deal more responsive to field-goal-shooting percentage than is Boston’s or either of the New York teams’. On the other hand, the New York Knicks’ output is relatively insensitive to offensive re- bounds, yet, compared to the other teams, is more responsive to the ratio of defensive rebounds. Since the Buffalo franchise changed coaches in midseason we esti- mated separate equations for both coaches. The final two columns of table | report these results. Of particular interest is the hypothesis that performance improves after a coaching change. In fact, a frequent Justification for replacing a coach is that the incumbent is not getting as Production Efciney 37 much out of the players as is possible. Team potentials and mean efficiencies are reported in table 2. An interesting development oc- curred in the Buffalo case; while team efficiency improved (from .99897 to .99926) with the coaching change, the frontier output declined. Replacing the coach resulted in a decline in expected output for the Buffalo team. In other words, the change actually led to a reduction in team output, contrary to its (presumed) intended purpose. Efficiencies for the other teams are also reported in table 2. Buffalo is the most efficient team; even the pre-coaching-change (BulTalo 1) effi- ciency exceeded that of the next most elficient team. Somewhat sur prising is the relatively high level of efficiency achieved by the Philadelphia franchise. Many sports commentators have suggested that, while the 76ers are loaded with talented players, they do not play well together. Thus, one expects a rather poor showing by the Philadelphia club in terms of efficiency. One explanation of our result is that, while the combination of Philadelphia players may not be com: plementary, given the particular personnel, the coach assembled and directed them in an effective manner. In other words, Philadelphia's output might increase if team members were more compatible, but, given that they are not, they are combined in a fairly efficient manner.* ‘The lowest level of efficiency is .99849. associated with the New York Knicks. Compared to other studies estimating production efficiencies, these ‘numbers are extremely large. Two explanations are plausible. First, in a competitive industry one expects inefficient firms to be driven out of business. While the sale of professional sports entertainment is not competitive,® the actual athletic contests are probably highly competi- tive, since the players and coaches have managerial control over strategic decisions affecting individual contests. Coaches and players that do not perform well are quickly rationed out of the market Second. previous studies used more aggregate data. Estimates made at the industry level, where outputs are not as homogeneous or inputs are measured less accurately than in this study, should exhibit lower esti- mated efficiency levels. ‘The estimated production-frontier output of each team is also re- ported in table 2. The frontier is calculated using the mean values for all inputs and the estimated coefficients. Of course, a team’s actual per- formance is a combination of its potential (the frontier output) and its efficiency. Only in the case of 100% efficiency is performance equal to potential. Multiplying the frontier output by the level of efficiency yields expected output, and teams can be ranked on this basis. Table 2 5. The Philadelphia coach was replaced jus afew games into the next season 6: Indeed, most teams have local monopolies in the output market ournal of Business ' ‘ z Y € yous emoy 1 § z , £ re86 eae sur ise stot S66 Loot 001 92666 866 9866 666 21666 1366 1366" 0100 $5100 ozo” zone L100 Sa100" 66 6 ‘a0 6386 Oster FoR ‘root Sz0"T dino Janos Zomuna 1 orune igdlopenua ero some aN XN ‘Supura pus dng wRoPeLE PaITUTT — TATE Production Egiciency 39 reveals a ranking identical with one based on actual won/lost records for the 1976 season. Given the extraordinarily high levels of team efficiency. this is to be expected. The elasticities in table 1 can easily be transformed to find the marginal productivity of each input. Multiplying the output elasticity by the ratio of the mean of the output to the mean of the input gives the marginal product (MP), (uP), a The result of these calculations is reported in table 3. Although in most cases a higher output elasticity implies a larger marginal product. this need not be the case. For example, Philadelphia has the second most inelastic output coefficient for offensive rebounds (.0829), yet the mar- sxinal product of offensive rebounds for the 76ers exceeds, by a sub- stantial margin, the marginal product of offensive rebounds for all other teams in our study. Apparently the Philadelphia team has a consider able offensive rebounding advantage over its opponents relative to its point margin, The same story applies to comparisons between vari- ables. The Boston team provides an example. Output is more elastic with respect (0 offensive rebounds than with respect to free-throw percentage, yet the marginal product of free-throw percentage is greater for the Celtics than the marginal product associated with offen- sive rebounding. To reconcile our finding against a home-team advantage with the fact that most teams have a better winning percentage at home than on the road. itis necessary to consider two possibilities. First. the production function at home might not be the same as the production function for a basketball team playing on its opponent's court. The other possibility i that, while the production function is the same at home and on the road, the level of some inputs varies systematically with the site of the contest Chow tests (Chow 1960) are performed to investigate the possibility that teams have two distinct production functions—one for home games and one for games played in another city. To test this hypothe- sis. we ran regressions nearly identical with the specification in equa- tion (3) (deleting only the binary locational variable) for three different samples: all games, home games only, and away games only. The hypothesis that home and away production functions are identical is accepted for all five teams at the 5% significance level. ‘The mean values of inputs for home games can be compared to the mean values for away games to determine whether teams employ more inputs at home than in games played on the road. The s-values are presented in table 4 for differences in mean inputs and output at home and on the road. All teams except the New York Nets performed Estimated Marginal Products of Professional Baskethall Inputs TABLE 3 NY NY. Kicks Buffalo 1 Philadelphia Buffalo Nets Boston League Variables Bulla 2 325 i ournal of Business Production Efceney vt ‘TABLE 4 Significance of Differences in Means of Inputs and Ouputs between Home and. Away Gamest NY NY. Variables Boston Buffalo Knicks Nets Philadelphia Points scored” 2a 4217 ao 1am «38251" Fiel-goal hooting percentage” 1.1785 3.1388" 2.268% 8570 2.3670" Pree-throw Shooting percentage 2865. SSD 3112, 13855097 Offensive rebounds" 190 iss). “1a ari “3798 Defensive rebounds? 121s a7si 30393" size 3aseTe Assiats™ 2l6sai* 4.6799" 4.3995" 2.0002" 5.2479" Personal foul 890 14311 —1'k036" 133000048 Steals" vais 4134s saz 152070867 Turnovers! = 16713" 20116" 938177 Blocks" 2670" 1.3628 3.1986" 1653818959" Degrees of freedom 15 7 79 76 7 “Galatea home game valve minus away game vale ‘Mesre atthe log of te ai. ‘Meavared atthe diferenee in shots blocked ‘Significant a the evel significantly better as hosts than as visitors. The primary reason for this is traced to relatively better shooting and rebounding by teams when playing in front of hometown fans. In most cases field-goal percentage, \d offensive and defensive rebounds, have significantly higher values home than for away games, ‘The difference in assists may be misleading. While it may signal a more cooperative attitude by players on the home team, or less effec: tive defense by the visitors, it could well result from padding the statistics of local favorites. Very little discretion is available to scorekeeper for most of the measures, but whether a pass followed by a basket is counted as an assist or not is highly subjective. Teams also block relatively more shots at home than on the road. Unlike assists, there is little reason to believe measurement error is at fault. Perhaps officials are influenced by the crowd and unwittingly shade close calls (between goaltending and a legal block) in favor of the home team. Or pethaps the adrenalin pumps harder at home in anticipation of local advertising endorsement contracts. At any rate, although statistically significant for three of the teams, blocked shots’ contribution to output iso small as to be of no practical importance in determining the source of the home-court advantage. Lack of a statistically significant difference in the ratio of personal fouls between home and away games, except inthe case of the Knicks, provides evidence that officials appear to be unbiased. This, of course, assumes that teams do not play “rougher"” at home, in which case they would actually deserve more fouls. on Journal of Business Conclusion Overall, the empirical results are encouraging. Application of the Richmond technique allows us to estimate the potential output of each team and calculate the efficiency with which inputs are combined, Output is most responsive to field-goal percentage, free-throw percent: age, and rebounding. Other inputs significantly affecting output are turnovers and personal fouls. Careful examination of location leads us to believe the observed home-court advantage in professional basket ball results from superior performance by the home team and not preferential treatment by officials. We also conclude that the determi- nants of performance are the same at home as on the road. The marginal products presented here could be of value to professional basketball teams. The same combination of inputs contributes different amounts to different teams; a team can evaluate players on the basis of their contribution to output and choose those players that increase ‘output the most relative to their salary. Finally, the interaction of team. potential and efficiency used to evaluate performance and rank teams results in a ranking identical with one based on actual wonilost records; given the extremely high estimates of efficiency, this outcome is to be expected and confirms the reliability of our estimating technique. References Ait. 8. N. 1972, EMciency estimation of production functions. Invernariomal Kew omic Review 13 (October): 568-58. Aigner. D.; Lovell, C. A. K.: and Schmidt, P. 1977. Formulation and estimat ‘Statistic production function models. Journal of Eeanometries § uly) 2 Chow, G. L- 1960. Tests of equality between sets of coeticients in two linear reares ‘ions, Econometrica 28 (July) 591-605 Demmert, H. G. 1978. The Beonomics of Professional Sports Teams, Lexington, Mass. Heath Neale. W. C. 1964. The peculiar economics of professional sports. Quarterly Journal of Jn of Economies 78 (Febeuary): 1-18 Richmond, J. 1974, Estimating the efficiency of production. ternational Economic Review 15 (lune) S1S-21 Timmer. C. P. 1971. Using a probabilistic frontier production function to measure Technical effcleney. Journal of Poltcal Economs 79. no. Uuly/Augist): 776-94,

You might also like