You are on page 1of 10

Managerial Efficiency and Survivability in Professional Team Sports Author(s): Gerald W.

Scully Reviewed work(s): Source: Managerial and Decision Economics, Vol. 15, No. 5, Special Issue: The Economics of Sports Enterprises (Sep. - Oct., 1994), pp. 403-411 Published by: John Wiley & Sons Stable URL: http://www.jstor.org/stable/2487990 . Accessed: 25/05/2012 16:52
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

John Wiley & Sons is collaborating with JSTOR to digitize, preserve and extend access to Managerial and Decision Economics.

http://www.jstor.org

MANAGERIAL

AND DECISION ECONOMICS,

VOL.

15, 403-411 (1994)

Managerial

Efficiency

and

Survivability in

Professional

Team

Sports

Gerald W. Scully
Schoolof Management, University Texas Dallas,TX,USA The at of Studentsof organizational sociologytend to believethat managerialefficiency less to has do with individualtalent than with the environment which firms operate.Economists in and fans know that this is not true in sports. Measures of managerialefficiencyare constructedfor baseball,basketballand footballcoaches. Survivalanalysis is utilizedto in measurecoachingtenureprobabilities these sportsand coachingtenureis shownto be relatedto managerialefficiency.

INTRODUCTION

sociologists and economists think Organizational in differently aboutthe role of management firms. The sociologistsbelieve that individualmanagerial effort matterslittle to the successof organizations; rather, institutional constraints predechoice. termineand narrowthe rangeof individual To the extent that individualsseem to matter frequently it is associated with an institutional attribute such as monopoly. Economists believe that individualsare paramountto organizational success. Even in the case of the classical, sole firm operating in extremelycomproprietorship petitive markets,the residualstrictlydepends on the abilityof owner-managersto controlshirking within a team production setting (Alchian and Demsetz, 1972;cf. Coase, 1937).Also, managerial competence and malfeasanceis a main issue in the marketfor corporatecontrol(Manne, 1965). A lack of empiricalevidence has made it difficult to sort out the validityof these competing hypotheses.Private firms have no obligation to reveal internaldata, other than that requiredby such as law. Measuresof corporateperformance, stock prices, may vary for reasons other than managerial performance. Public organizations, such as the military or bureaucracies,produce ambiguousoutputs.Also, many orgameasurably nizations are rather complex entities, making it difficultto link managerialqualityand firm perbaseball than in basketball (t = 2.51), but not in formancedirectly. CCC0143-6570/94/050403-09 ? 1994by John Wiley & Sons, Ltd.

Sportsoffers a good vehicle for testing some of the issues arising from this controversy.Field managersand coaches are crucial to the performance of clubs. The constraintset facing them is constant across clubs and is time invariant.The firmsare identicalin manyrespects:they produce identical outputs, use the same units of input skills, compete under the same rules, employthe same productionfunction,share a commontechnology,and so on. They differ in marketsize and have different owners and managersat any moment of time. The objectiveof the manageris to transforma set of relative offensive and defensive playing skills into club victories. That this is done with is greatvariability illustratedby the high turnover of managers and head coaches throughout the historyof sports.In baseballand basketballfor all coaches the averagetenure has been about three years. Fewer than 10% survive for a decade or more. Tenure is somewhat longer in football, about 4.2 years. In Fig. 1 the distributionsof tenure for all coaches with two or more years of tenure, the sample employed in the statistical analysis that follows, is presented. The means, standarddeviations,skewnesand kurtosis(in that order) for tenure in each of the sports is as follows:baseball(6.79, 6.15, 2.63, and 14.02),basketball (5.44, 4.58, 1.63, and 5.02), and football (6.39, 5.63, 2.44, and 10.44). Tests of differences in the means reveal that tenure is longer in

404

GERALDW. SCULLY

Baseball

Basketball

.5-

Football
IL

All Sports

c-I

0D 20 40 60

Years

20

40

60

Figure 1. Histograms by sport.

football (t = 0.68); tenure is longer in football than in basketball (t = 1.60). The skewness of tenure is highest in baseball. The kurtosis of tenure is most pronouncedin basketball. In this paper we develop a simple model of managerial effort or efficiency, construct meaefficiencyand link these measures of managerial sures to tenure using survivalanalysis.Managerial efficiencyis found to be a good predictorof managerialsurvival. THE OWNER-MANAGERPROBLEM IN SPORTS

ager or head coach is fewer games won than would be possiblewith the playingtalent at hand and managed with maximumeffort and competence. The link between the win recordand profit is well known(Scully,1989). Consider a simple production process, as in team sports, in which output (wins), W, is proparameter,If, duced with a randomproductivity meathat correspondsto some multidimensional sure of team playingtalent, and coachingor manager effort, e. Increased player productivity, i1, and/or increasedcoachingeffort, E, yield higher

In levels of the manager'sexpectedperformance. the standard construction of the problem, the principal(owner) and the agent (head coach or For economists, the problem of separation of manager), both risk neutral, have the same ownershipfrom control is that it leads to a possi- knowledge about the random productivity ble misalignment goals of principalsand agents parameter,but the principalcannot observe the of in organizations. a worldwith informationand realizationof if or the level of agent effort. In In monitoring costs, managerial shirking, malfea- the deterministic case W= E+f.The ideal outcome sance and incompetencemay exist. The effect of for the owneris that for all realizationsof if, the these agent deficienciesis a reduced residual to coach expends the efficient level of effort, e*(f). the principal.In the context of sports,this is the The line segment e*(f) may be thought of as the or owner(s)-manager owner(s)-headcoach prob- efficiencyfrontier or the productionfrontier for parameter if. If the lem. The effect of a reduced effort by the man- all levels of the productivity

MANAGERiAL EFFICIENCY AND SURVIVABILITY PROFESSIONAL IN TEAM SPORTS

405

coach shirksor is incompetentand expendseffort how efficient his or her manager is at any moof e1, the principalloses a residualper unit equal to ment of time. The probability durationin the = 01. In fact, the ratio e/e* or 0, which is time interval(t, t + At) is conditionalon duration elel* in the unit interval, is a measure of manager at time t. For any specificationof a series of effort or efficiency. events (continuationor survivalversus failure or distribution, in From the owner'sperspective,profits are max- termination) termsof a probability equivalentspecification imized by extractingthe most wins from the ros- there is a mathematically ter of playingtalent fielded.In the main,the costs in terms of a hazardfunction or a survivorfuncof fielding playing talent i1 are incurredat the tion. If the probabilityof durationis time indebeginningof the season. The revenuesassociated pendent and exponential,the hazard rate, A, is with that playing talent are determinedby how the instantaneousprobability the event occurof efficientlythe managertransformsplayingtalent ring at time t. The probabilitydistributionof into club victories, 0ti. An incrementin manage- durationat time t for the exponentialdistribution rial efficiency by increasing wins increase club iS profit. (1) F(t; A) 1-e-At The simple productionprocess may be stochasrather than deterministic.In this case, W is The corresponding density function (instantatic, replaced by its density function and associated neous probability) is: cumulativedistributionfunctions:f(WIE,q) and F(WI,E,Ti).Higher levels of coaching effort de(2) Aef(t; A) dF/dt crease the probabilitythat smaller levels of output will be realized.That is, F,(W IE, 4i) < 0. In The correspondinghazard, H, and survival, S, the case of stochasticproduction,the ratio e/e* functionsare: only partly measures managerialeffort or efficiency. Part of the lost output may be due to (3) A)] A H(t; A) =f(t; A)/[l1-F(t; randomfactorsbeyondthe agent'scontrol. 1 (4) S(t; A) 1-F(t; A) e-Akt This characterizationof the principal-agent problemis static - one owner dealingwith one The exponentialdistributionmay be an inadeagent in a vacuum. In a dynamicsetting of the quate representation,when the sample contains problem,ownerslearn about the level of manage- both long and short durations.In this case, the tenure Weibull distributionmay be a better representarial effort or efficiency.Averageownership in team sports is about eleven years, so most tion. If the probabilityof durationis time indeowners will employ about three or four head pendent and Weibull, the hazard rate is coaches. Sports is a most data-riddenbusiness. Ap(At)P-1. The probability distribution of duraInformationis widespreadand cheaply obtained. tion at time t for the Weibulldistribution is Clubscompetewith one anotherin leagues, more or less playingthe same set of competitorson a (l a) F(t; A) 1 1-exp(-Akt)p regularlyscheduledbasis. Each owner can monitor the effort or efficiencyof his coach compared The correspondingdensity function (instantais: of neous probability) to others.Thus, over time, throughobservation the performanceof his/her club and other clubs Fd and throughhiringand firing,ownerswill learnby f(t; A) dF t Ap(At)P exp(-AktP) (2a) experience and distinguish good field managehazard, H, and survival,S, ment from bad. As a result, the level of coaching The corresponding functionsare: qualitywill tend to rise over the owner'stenure. Inferred from this argument is that competent (3a) H(t; A) = kp( kt)pl managerswill tend to survivelonger and incompetent or shirking coaches will be fired more (4a) S(t; A) exp(-AktP) frequently. The length of managerial tenure shouldbe related to the effort level or efficiency Largely,the question of the shape of the distribution is an empirical matter. This empirical of the manager. More formally,each club owner has to decide issue will be discussedbelow.

406

GERALDW. SCULLY

THE EMPIRICAL MEASUREMENT OF MANAGERIALEFFICIENCY By definition, a game is won when the team outscoresits opponent.Scoring(runs in baseball, points in basketball and football) is determined by a vector of players'offensive skills relative to opponent players' defensive skills: s =f(Xi), where X is the ith offensive skill relative to its defensive counterpart.Opponent scoring is determined by a vector of players'defensive skills relativeto opponentplayers'offensiveskills:os = g(Yi),where Y is the ith defensiveskill relativeto its offensivecounterpart. In baseball, offensive skills mainly are hitting. But once a player is on base the object is to advance the man: steal, hit and run, pinch hit, etc.). Bases advanced yield runs scored.Defensive skills mainly are pitching and fielding.Considerable scouting resources are employed in determiningthe opponenthitters'weaknesses(types or positionof pitches and portionof field hits are likely to occur).In basketball,offensive skills are mainly measured by shooting percentages (field goal and free throw), offensive rebounds,assists and turnovers.Defensive skills are mainly measuredby blockedshots, steals, defensiverebounds and fouls. But coaching inputs seem higher in basketball (and in football) than in baseball. Teams appear to develop certain playing strategies (e.g. fast break or a more deliberatepace of play, set offensive plays and defensive strategies, etc.) that are not equallyviable for all opponents. in Particularly, the last quarterof a close game head coaches make near-continuousplaying adjustments and play-calls. In football, offensive skills are mainlymeasuredby offensive blocking, quarterback pass completionrates and yards per pass, receiver completions and yards per pass, runningback yards per carry, and kicker threepoint field goals and conversions.Defensive skills are measured by defensive blocking, tackles, quarterback sacks, interceptions,fumble recoveries, etc. As with basketball, football teams develop certain styles of play, both offensivelyand defensively.The ability of such teams to adjust their offense and defense at halftimepartlymeasures head coachingability. Each team plays 162 games in baseball, 82 in basketball, and 16 in football, during a regular season. The total regularseason scoring(S) and opponent scoring(OS) are, by definition:S = Es

= Ef(Xi) and OS = Eos = Eg(Yi). Then the team

win per cent for the regularseason, W, is functionally related to scoring relative to opponent
scoring. W= F(S/OS)
= F[Ef(Xi)/Eg(Yi)] (5)

In the form for statisticalestimationthe model for the win per cent is:1 InW= a +,8 ln(S/OS) + e (6)

This formulationhas certain attractivestatistical properties. The win percentage is bound between 0 and 1 and, since one team's victoryis has another'sloss, definitionally a constantmean (0.5). With the large number of contests over a season and throughtime, the varianceis constant across time subsamples. Definitionally, across teams, as with the win per cent, runs or points scored duringa regularseason must equal opponent runs scored.S/OS has a constantmean of 1 (a mean of zero, if the specificationis S - OS). While scoringhas not been constantin professional sports due to rule change and innovation (movementof the pitcher'smound to 60'6"from of 50', the introduction the livelyball, the forward pass and the field goal in football, the jump shot and 24-secondclock in basketball,etc.), whatever change increasesscoringalso increasesopponent of scoring.With the specification the independent variablein relativeratherthan absoluteform the varianceis constant.The expectationthen is that with conthe error term is normallydistributed, stantvarianceacrosstime.2With this formulation it is possible to measuremanagerialperformance throughtime and acrosssports. The objectiveof a manageror head coach is to win as many games as possible with the relative offensiveand defensiveplayingskills at hand.The scoring,minimizgoal is achievedby maximizing that relaing opponent scoringand transforming tive scoringproductioninto wins. S is maximized when all player offensive skills relative to the opponents' defensive skills are equated
= at the margin: i.e. Smax Ejj (dS/dXi)Xi + (dS/dX1)Xj, i = 1, n; j = 1, m; and, i oj. OS is

minimized when all playerdefensiveskillsrelative to the opponents'offensive skills are equated at the margin: OSmin = Eij (dOS/dYi) Yi + (dOS/dYj)Y1. These dimensionsof coachingqual-

MANAGERIAL AND SURVIVABILITY PROFESSIONAL IN TEAM SPORTS EFFICIENCY

407

ity are allocativein nature.Given a set of playing The approach,therefore,considersall deviations skills, scoringis maximizedand opponent scoring from the efficient, frontier function as arising is minimized by allocating those offensive and from technical inefficiency. A criticism is that defensive skills in such a way that not one extra only part of the error term may be deterministic; run or point can be producedor an opponent's part may be truly stochastic.The error term may one extrarun or point stoppedby a redistribution be of the form E = u + v, where u is a one-sided the term representing degree of manof playing assignmentsat any moment of time. disturbance Most of the estimatingof managerial coaching agerial technical inefficiencyand v is a symmetor qualityhas been in this spirit.A problemwith this ric, normallydistributed randominfluence.In the approachis that the empiricalanalysismay suffer case of the win productionfunction,the random from omitted variablesbias. To the extent that componentwould be luck; some games are won some importantindependentdimension of play- and some games lost, not for reasons of relative ing skill has been omitted (e.g. hustle, aggressive- playingskill or coachingabilitybut due to referee ness, anticipation,clutch performance,etc.), as- error or other fates beyond the control of the cribingthe differencebetween potential and ac- contestants. Two stochastic frontier functions tual wins to the coachingfunction may be inap- were estimated;one with a normallydistributed propriate.The number of dimensionsof playing error term, v, and the other with the error term skill measuredin baseball is the greatest and in assumed to be Gamma distributed.Neither of football the least. Moreover,the assumptionof a these specifications were superior to the delinear homogeneousproductionfunction is rea- terministicfrontierestimates.Therefore,only the sonable in baseball,less reasonablein basketball, later utilizedhere. and probablynot valid in football. Means and standard deviations of efficiency For our purposes,we will assumethat managers were calculatedfor all of the managerswith two and coaches allocatetheir offensiveand defensive or more seasons managingor coaching.It is well playingskillsin such a way as to maximizescoring knownthat there is a manageriallearningcurve; and minimizeopponent scoring. We will take it that is, efficiencyrises at a decreasingrate over that the observedS and OS are Smax and OSmin. career length (see Porter and Scully, 1982). To The dimension of coaching quality that is mea- capturethe trendin efficiencythe standarddeviasured here is the actual win per cent relative to tion of efficiencyis includedas a covariatein the the potentialwin per cent, given S and OS. That survivalanalysis. is, the measureof coachingqualityis 0 < W/W* < 1, where W*=W+ Emax. The potentialwin per cent is the predictedwin SURVIVALANALYSIS per cent plus the largest observedpositive residual from the equation estimated in (6). These Differences in the mean tenure and in the disregressionsare reported elsewhere (Scully, 1992 tributionsby sport have been noted. These difp. 60). The largestpositiveerroris associatedwith ferences, while not dramatic,suggest differences that head coach that achieved the largest actual in the hazard (survival)rates by sport. Figure 2 win per cent comparedto the predictedwin per presents the estimated Kaplan-Meier survival cent with the observedratio of scoring to oppo- curves with Greenwood 95% confidence bands tenure in each of the three sports. nent scoring.The efficiencyof this best-practice for managerial curves(without head coach or manageris unity. All other head Also in the figureare the survival coacheswill lie in the unit interval. the confidencebands) for all three sports. Since are The choice of the estimationprocedurefor the these survival probabilities impossibleto read production(win) function in Eqn (6) depends on accuratelyoff of the graphs, Table 1 presents made regarding errorterm, E. them for the tenure interval2-15 years. the the assumption the Three specificationswere employed:(1) the deNaturally, first questionis, do these survival in terministic frontier function; (2) the stochastic probabilities fact differby sport?The approprifrontierfunction;and (3) the maximum-likelihood ate test is the log rank test (Savage, 1956). This test comparesthe actual and the predictedfailGammafrontierfunction. frontierfunctionis estimated ures for each groupand uses the x2 statistic.The The deterministic the by minimizing sum of the absolute residuals. result for the comparisonacrosssportsis X2(2)=

408
{a) Baseball

GERALDW. SCULLY
(b) Basketball

~~~~~~~~~~~~~~~~~~~~()I

&(}
C.5

.5

0 Football

20

Yesrs 60ears20

40

go
&.5

Years

40

60

~~~~~~~~~~~~~~~~~~~0
:

20

Years

40

20

Years

40

60

Figure2. Kaplan-Meiersurvival with Greenwood confidencelimits.

Table 1. SurvivalProbabilitiesof Managersby Sport


Tenure Baseball Basketball Football

2 3 4 5 6 7 8 9 10 11 12 13 14 15

0.790 0.637 0.521 0.427 0.341 0.285 0.243 0.213 0.180 0.161 0.142 0.124 0.105 0.086

0.655 0.486 0.380 0.338 0.303 0.239 0.190 0.148 0.134 0.113 0.106 0.077 0.063 0.056

0.776 0.647 0.506 0.423 0.321 0.250 0.218 0.173 0.141 0.115 0.103 0.090 0.083 0.077

other distributions be employed,they involve can difficulties,in part because the qualitativeshape of the hazard (monotonicity,log concavity,etc.) and in part because they yield complex hazard functionsthat do not have the constanthazardas a special case. One test that distinguishedbetween the exponential and the Weibull is to estimate
ln(-ln(S(t))) = -ln(A) +p ln(t), where S(t) is

the survival function (the hazard function is


Ap(At)P- . Thus, the parameters can be esti-

mated by linear regression.If the underlyingdistribution is exponential, p = 1. The alternative, p # 1, implies a Weibulldistribution if p > 1, and the hazardis increasingand otherwisedecreasing. The results by sport were as follows: baseball
(p = 1.148, t = 129.7), basketball (p = 0.954, t = 112.1), and football (p = 1.190, t = 103.5). All the

5.0, prob. = 0.082. On the basis of the log rank test, the survivalcurves are statisticallydifferent from each other. The next question of importanceis the form the underlying distribution; whetherit follows i.e. the exponential(a one-parameter or distribution) the Weibull(a two-parameter distribution). While

coefficientswere statistically differentfrom unity, accepting that the underlying distribution is Weibull. The problem with the test is that the standarderrorsare seriallycorrelated,so that the results can be taken only as approximate.An alternative test is to estimate the Weibull parameters directly. The estimated p = 1/o-

MANAGERIAL EFFICIENCY AND SURVIVABILITY PROFESSIONAL IN TEAM SPORTS

409

comes from the Weibull regressionsin Table 2. All the p's in the Weibull regressionsare significantly greater than unity. We accept that the distribution Weibull. underlying is There are severalparameterizations the maxof imum-likelihood Weibull regressionmodels. The parameterestimatesmay be in log expectedtime, log relative hazardor hazardratios. The log expected time parameterization chosen, since we is want to focus directlyon expected survivaltime relativeto managerialefficiency.Estimationis of the form:

is successful.All the x2s are significantat well above the 99% level. The efficiency measure is alwayspositive(higherefficiencyis associatedwith longer (log) expected survivaltime and is highly significant(the t-values are asymptotict-values) in all sports.The standarddeviationof efficiency also is positively associated with (log) expected survivaltime, but is not statisticallysignificantin football and only weakly significantin basketball (p = 0.27). In the Weibull regression that combines all the sports,all the variables,includingthe intercept and slope dummies, are highly significant. A(t) = (tWl1f) exp( ,0 + EpiXi + e) (7) The expectedsurvivaltimes related to managerial efficiency(the standarddeviationof efficiency where cr is the shape parameter(or p = 1/cr), is set at the mean) are graphed in Fig. 3. The estimated from the data, and E has an extreme efficiencymeasure is at the mean and -2 o- and value distributionscaled by o-. The method of +2o-. The means and standarddeviationsof the via managerial efficiency measure by sport were: estimation is maximum-likelihood baseball (0.7737, o-= 0.0337), basketball(0.7171, Newton-Raphson. The Weibull regressionsappear in Table 2. In a= 0.0947) and football (0.5771, o-= 0.0734). The general, the estimationof the hypothesizedrela- lower mean efficiencyof head coaches in football tionshipbetweenmanagerial efficiencyand tenure is a naturalconsequenceof a higher noise-to-sig-

Table 2. Weibull Maximum Likelihood Regressions of Survival


Variable Baseball Basketball Football All sports

Constant Efficiency crEfficiency Basketball Football Eff* Basket


Eff* Foot
of Eff* Basket of Eff* Foot

-5.624 (4.77) 9.195 (6.22) 10.353 (4.23)

- 1.759 (3.35) 4.697 (7.03) 1.390 (1.12)

0.142 (0.39) 3.063 (6.08) 0.085 (0.08)

-5.696 (4.99) 9.291 (6.49) 10.578 (4.45) 3.989 (3.14) 5.838


(4.87)

-4.688 (2.93)
-6.232

(4.10)
-9.214

(3.40)
-10.489

Sigma or(Sigma) Log Likelihood


X2

0.716 0.030 -313.5 34.2 267

0.650 0.038 -155.1 33.3 142

0.688 0.037 -174.8 26.3 156

(4.02) 0.693 0.020 - 644.2 100.5 565

t-valuesin parentheses. Note: Asymptotic

410

GERALDW. SCULLY

Expected Tenure 12

Baseball

8 _ 4 Basketball -2 Sigma Mean +2 Sigma Efficiency


~~~~~~Football

Figure3. Expectedsurvival times relatedto managerial efficiency.

nal ratio. The theoretical standard deviation of largest win per cent from a given set of player the win per cent of a 0.500 club (or a league inputs. containing clubs of equal playing strengths) is 0.5/ fg, where g is the numberof games during NOTES the season. The theoreticalstandarddeviationin footballwith 16 season games is 3.2 times that of 1. Data on the win per cent, points or runs scored baseballwith 162 season games.

CONCLUSION In the moderntheory of the firm the functionof the manageris to maximizethe principals'residual claim. In professionalteam sportsthis means the maximizing club'swin per cent with the playclubs mayget ing talent at hand.Bottom-finishing there because they lack talented players or because they are managed poorly. Generally, the two can be distinguished. Here the decision to retain or terminate the coach was modelled and estimated. Managerial tenure was shown to be linked to managerial efficiencyor the abilityof the coach to extractthe

during the season and opponent points or runs scored were collected from standardsports record sources.The data for baseballare from 1876to 1989 All and are containedin TheBaseballEncyclopedia. teams and leagues were included, except for the Union Association teams in the 1884 season. Too manyteams folded too early in the league to make their inclusion valid. The data for basketball are Pro from The SportsEncyclopedia: Basketball,3rd edition, and cover all leagues and teams from the 1937-8 season the 1989-90 season. The data for football are from 1933 to 1989 and for all leagues and teams. Prior to the 1960 season the data are from the Official 1985 National Football League Recordand Fact Book; for the period 1960-89, the Pro Encyclopedia: Football, data are from TheSports 8th edition. The head coach list prior to 1960 was obtainedfrom the Pro FootballGuide,1990 edition. 2. The residualsfrom the regressionsestimatedon the basis of Eqn (6) were examined and no pattern related to trend was found. Severaltests were per-

MANAGERLAL EFFICIENCY AND SURVIVABILITY PROFESSIONAL IN TEAM SPORTS

411

formed on the residuals(e.g. the squaredresiduals P.K. Porter and G.W. Scully (1982). Measuringmanagainsttrend)with no significant results. agerial efficiency: the case of baseball. Southem
Economic Joumal, 48, 642-50.

L.R. Savage(1956).Contributions the theoryof rank to REFERENCES


order statistics. Annals of Mathematical Statistics, 27,

590-615. A.A. Alchian and H. Demsetz (1972). Production,in- G.W. Scully (1989). The Business of Major League Baseball, Chicago:The Universityof ChicagoPress. Ameriformationcosts, and economic organization. G.W. Scully (1992). Coaching quality, turnover, and can EconomicReview,62, 777-95. longevityin professionalteam sports.In Advancesin R.H. Coase(1937).The Natureof the Firm.Economica the Economics of Sport (edited by G.W. Scully), N.S., 4, 386-405. Greenwich,CT:JAI Press,53-65. H.G. Manne (1965). Mergersand the marketfor corporate control. Joumal of Political Economy, 73, 110-20.

You might also like