Collective Turnover at the Group, Unit, And Organizational Levels Evidence (2)

NOTES WHO WINS THE OLYMPIC GAMES: ECONOMIC RESOURCES AND MEDAL TOTALS Andrew B. Bernard and Meghan R. Busse* Abstract—This paper examines determinants of Olympic success at the country level. Does the United States win its fair share of Olympic medals? Why does China win only 6% of the medals even though it has one-fth of the world’s population? We consider the role of population and economic resources in determining medal totals from 1960 to 1996. A
of 5
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  NOTES WHO WINS THE OLYMPIC GAMES: ECONOMIC RESOURCES AND MEDAL TOTALS Andrew B. Bernard and Meghan R. Busse*  Abstract  —This paper examines determinants of Olympic success at thecountry level. Does the United States win its fair share of Olympicmedals? Why does China win only 6% of the medals even though it hasone-fth of the world’s population? We consider the role of population andeconomic resources in determining medal totals from 1960 to 1996. At themargin, population and income per capita have similar effects, suggestingthat both a large population and high per capita GDP are needed togenerate high medal totals. We also provide out-of-sample predictions forthe 2000 Olympics in Sydney. I. Introduction I N this paper, we ask the straightforward question of how manyOlympic medalscountriesshouldbe expectedto win by consideringwhat factors inuence national Olympic success. 1 Most Olympicmedal predictions assess athletic talent sport by sport and predictwinnersin eachevent.We followa differentpathby generalizingfromindividual sports. Although this has the disadvantage of missingnation-specic expertise in a particular event, it has the advantage of averaging the random component inherent in individual competition,enabling us to make more accurate predictions of national medaltotals.Even the most ardent xenophobeswould not suggest that a singlecountryshouldwin all the medalsat a givenOlympic Games.The realquestion is how many medals qualify as a successful performancebya nationalteam. Clearly, populationshould play a role in determiningcountry medal totals. Larger countries have a deeper pool of talentedathletes and thus a greater chance at elding medal winners. Wepresentand test a simple theory of medal successbased on populationbut nd that pure population levels are not sufcient to explainnationaltotals. If they were, China, India, Indonesia,and Bangladesh,with just over 43% of the world’s population, would have won morethan6% oftotalmedalsin1996.To thisend,we extendthe population-based model to include a measure of resourcesper person in the formof GDP per capita. 2 The additionof per capitaGDPdramaticallyimprovesthe abilityof the model to t the data. Whereas China, India, Indonesia, andBangladesh have a huge share of world population, together theyaccount for under 5% of world GDP in 1996, roughly equal to theirmedal share. Real GDP is the best single predictor of a country’sOlympic performance. Population and per capita GDP contributeequallyat the margin, implyingthat two countrieswith identicalGDPbut different populations and per capita GDP will win the samenumber of medals.GDP is not the whole story. Host countries typically win anadditional 1.8% of the medals beyond what would be predicted bytheir GDP alone. The forced mobilization of resources by govern-ments clearly can also play a role in medal totals. On average, theSoviet Union and Eastern Bloc countrieshad medal shares more than3 percentage points higher than predicted by their GDP. 3 The rest of the paper is organized as follows. Section II presentstwo simple models for analyzing country medal totals. Section IIIdescribes the data and the empirical results. Section IV presentsout-of-sampleforecasts for the Sydney 2000 Olympics, and sectionVconcludes. II. A Simple Theory of Population and Olympic Success We start by considering the underlying distribution of athletictalent.If we thinkof countriesas beingarbitrarydivisionsof the worldpopulation, then we should expect to nd medal-caliber athletes inproportion to the country’s share of world population. One caveat isthat not every country participates in the Olympics. The actualrelationship predicted by the talent distribution is that the expectedmedal share accruing to a country should be equal to its share of thetotal population of countries participating in the Olympics:E ~ medalshare it  ! 5 medals it  ¥  j  medals  jt  5  population it  ¥  j  population  jt  5  popshare it  . (1)Equation (1) can be tested empirically in the form of a tobit, withthe results shown in column I of table 1. Although the populationshare is positive and signicant, the estimated coefcient is signi-cantly below 1. 4 There are several reasons related to the structure of the Olympicsthat help explain why the relationship in equation (1) does not hold.First, countriescannotsend athletes in proportionto their populationsfor eachevent, for example,in team competitions,where eachcountryhas at most one entry. Second, in medal counts, team events count asone medal even though a country must provide a number of athletes.Finally,the number of athletesfrom each countryis determinedby theIOC in negotiationwith the country’s Olympic committee.As a result,not all the Olympic caliber athletes from a large country are able toparticipate.  A. Economic Resources and Olympic Medals To augment our specication, we examine the role of economicresources in generating Olympic medals. We choose to frame ouranalysis in terms of a production technology.In the previous section,we assumed that talented athletes were randomly distributed in the Received for publication June 5, 2001. Revision accepted for publicationDecember 4, 2002.* Tuck School of Business at Dartmouth and National Bureau of Economic Research; and Haas School of Business, UC Berkeley, respec-tively.We thank Julio Teran for valuable research assistance. A version of thepaper was distributed before the Sydney Olympics on August 28, 2000and is available at Any opinions ex-pressed are those of the authors and not those of the National Bureau of Economic Research. 1 All references to the Olympics or Games refer to the Summer Games. 2 In a contemporaneous paper, Johnson and Ali (2000) also investigatethe economic and political determinants of participation and medal totalsat the summer games and make out-of-sample predictions. Earlier papersin the area include Ball (1972), Levine (1974), and Grimes, Kelly, andRubin (1974). 3 Shughart and Tollison (1993) argue that the change in the structure ineconomic incentives in the former Soviet countries is responsible for theirlower medal totals in the 1992 Olympics. 4 In all our estimated specications we include year dummies and correctfor heteroskedasticity and autocorrelation in the tobit standard errors. SeeBusse and Bernard (2002) for the econometric methodology. The Review of Economics and Statistics,  February 2004, 86(1): 413–417 ©  2004 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology  world population. However, developing Olympic caliber athletesrequiresconsiderableexpenditureon facilitiesand personnel.Wealth-ier countries are more likely to have individuals, organizations, orgovernmentswilling to make such an investment.Wealthiercountriesare also more likelyto haveathleticsas a part of schoolingand to haveleisure time to devote to sports. As a result we include a measure of real GDP per capita in a model of Olympic medal production.Our productionfunctionfor generatingOlympic caliberathletesfora country  i  in year  t   requires people, money, and some organizationalability: T  it  5  f  ~  N  it  ,  Y  it  ,  A it  ! , (2)where  N  it   is the population,  Y  it   is the GDP, and  A it   is the organiza-tional ability of the country. The share of Olympic medals,  M  * it  , wonby a country is a function of the talent in it:E S  medals it  ¥  j  medals  jt  D  5  M  * it  5 g ~ T  it  ! .There is no theoretical guidance on the precise form of either  f  [ or  g [ . We use a Cobb-Douglas production function in population(  N  it  ) and national income ( Y  it  ) for the production of Olympic talent,and a log functionfor the translationof relativetalentto medal shares: T  it  5  A it   N  it  g Y   it  u ,(3)  M  * it  5 ln T  it  ¥  j  T   jt  .This yields the following specication for medal shares:  M  it  5 H  ln  A it  1 g  ln  N  it  1 u  ln  Y  it  2 ln  O  j T   jt  if   M  * it  $ 0,0 if   M  * it  , 0.(4)Because national income can be expressed as the product of popula-tion and per capita income,we will actuallyestimatea specicationof the form  M  it  5 H C  1 a  ln  N  it  1 b  ln  ~ Y   /   N  ! it  1 d  t  1 n i 1 e it   if   M  * it  $ 0,0 if   M  * it  , 0,(5)where  d  t   is a yeardummy includedto capturechangesin the totalpoolof talent and in the number of countries participating, as well as thechanging number of sports;  n i  is a country random effect; and  e it   is anormally distributed error term. III. Data and Results The data for this projectconsistof two main components:Olympicmedal counts and socioeconomic indicators. We obtained the medaldata from Wallechinsky (1992) and direct correspondence with theInternational Olympic Committee (IOC). We would prefer to have arange of socioeconomic indicators for each country. However, thedifcultyof obtainingsuchmeasuresfor more than150 countriesover30 years precludes us from considering anything but GDP andpopulation. Our primary source for population and GDP data is theWorld Bank. 5 Population gures could be found fairly readily; GDPmeasures were more difcult. For some countries,it was necessarytointerpolate or extrapolate using either reported or imputed growthrates.All GDP guresare convertedto 1995 U.S. dollarsusingcurrentexchange rates.  A. Results In this section,we report results on the relationshipbetween medalshares and population, income per capita, and total GDP. Univariatespecicationswith log populationand log GDP per capita,reportedincolumns II and III of table 1, raise the log likelihood over that withpopulation shares alone.Column IV estimates equation (5), including both log populationand log GDP per capita. The variables are positive and signicant atthe 1% leveland have similarmagnitude.The log likelihoodincreasesmarkedly over the univariate specications. The similarity of thecoefcients on log population and log GDP per capita suggests thatlog GDP is the relevant determinant of country medal shares. How-ever, a likelihoodratio test rejects the equality of these coefcients atthe 5% level.The coefcients estimatedin this sectioncan be looselyinterpretedto mean that if the average country were to double its total GDP, itcould expect its medal share to rise by 1%–1.5% of the total medalsawarded.  B. Additions to the Model The empiricalspecication given in equation (5) shifts all country-specic informationnot includedin GDP and populationinto the errorterm. In this section we explore some of the additional factors thatmight augment or diminish medal shares,includingthe advantagesof hosting, the medal premium enjoyed by the former Soviet Union andits satellites, and the role of large-scale boycotts.Hosts have several potential advantages over other Olympic par-ticipants. First, the cost of attending the Olympics for individualathletesis minimized.In addition,host countriescan tailor facilitiestomeet the needs of their athletes and may gain an edge if home crowdenthusiasm sways judges. Individual athletes may be more motivatedto achieve Olympic fame when the events are conducted in front of friends and family. Finally, host countries are inuential in theaddition of new sports to the Games themselves. 5 We also used United Nations data sources, the  CIA Factbook,  the  Economist   magazine, and the  Taiwan Statistical Planning Book. T ABLE  1.—T OBIT OF  M EDAL  S HARE ON  P OPULATION AND  GDPIndependentVariableMedal ShareI II III IVPopulation share 0.5753(0.1758)Log population 0.0152 0.0163(0.0024) (0.0022)Log GDP per capita 0.0115 0.0142(0.0018) (0.0017)LR test 4.435 (0.03)Year dummies Yes Yes Yes YesLog likelihood 468.72 581.46 476.66 723.93Observations 1254 1254 1254 1254 Note: Heteroskedastic- and autocorrelation-consistentstandard errors are in parentheses(see Busse &Bernard, 2002).The LR test reportsthe test of the equality of the coefcients on log GDP per capita andlog population (and the  p -value). It is distributed  chi -squared with 1 degree of freedom. THE REVIEW OF ECONOMICS AND STATISTICS414  One of the most interesting questions regarding Olympic medaltotals concerns the ability of countries, especially the former SovietUnion and EasternEuropeancountries,to “manufacture”goldmedals.Unconditionalmedal totals cannottell us how successfulthey were atmobilizingresources.We create two dummy variablesto capturetheseeffects.The rst covers countriesdistinctlyinside the Soviet sphereof inuence, and the second includes other nonmarket, typically com-munist,countries. 6 We considerthe additionalmedalsfor these groupsafter controlling for income and population to provide an estimate of the power of central planning in the Olympic race. The resultingspecication is  M  it  5 C  1 a  ln  N  it  1 b  ln  S  Y  N  D it  1  Host  it  1 Soviet  it  1 Planned  it  1 d  t  1 n i 1 e it  . (6)Two Olympics were subject to large-scale boycotts: those in 1980and in 1984. 7 The coefcients on the host dummy and the dummiesfor the centrally planned economies are likely to be particularlysensitive to the inclusion of these Games. Table 2 reports the tobitspecicationfor medal shareswith and withoutthe boycottedyears,incolumns I and II.The results for population and GDP per capita are largely un-changed in that they remain positive and signicant. Now, however,we cannot reject the hypothesisthat the coefcients are equal to eachother. The Soviet countries have medal shares more than 6.1 percent-age points higher than other countries. The other planned economieshave shares that are higher by roughly 1.6 percentage points. Neitherof these effects is sensitive to the exclusion of the boycotted Games.The host effect on medal totals is also positiveand signicant.Thebump in medal share from hosting a nonboycottedOlympics is morethan 2 percentagepoints.During the boycottedGames the hosteffectswere enormous, on the order of 19 percentagepoints, suggestingthatthe United States and Soviet Union were the prime beneciaries fromeach other’s boycotts in terms of medal counts. In column III, wereport results for nonboycott years with country random effects tocontrol for any persistent country-specic ability to produce medals.The results are broadly similar, although we reject the equality of thecoefcients on population and GDP per capita.We nish by considering the adequacy of our sparse specicationforthe purposesof predictionby presentingthe resultsvisually.Figure1 shows the relationship between the predicted medal shares andactual medal shares for 1996 from our augmented tobit excluding theboycotted years (column II of table 2). The model underpredictsmedal shares at both the low and high ends of the range andoverpredicts in the middle. Although the additions of log GDP percapita and several dummies have improved the t substantially, thecurrent model is lacking in overall predictive power. C. Time to Build  Until now we have implicitly modeled the production of Olympicathletes and medals as a within-period ow process with potentiallypersistent country-specic organizationalcapabilities. However, it isquite likely that Olympic athletes are more similar to durable capitalgoods in that they may provide medal potential over several Olym-pics. This would suggest that investments for one Olympics mayincrease the chance of winning medals in subsequent Olympics. Tocapture such effects we add lagged medal shares to our empiricalspecication 8 :  M  it  5 C  1 ~ 1 2 d!  M  it  2 1 1 a  ln  N  it  1 b  ln  S  Y  N  D it  1 d t  1 e it  . (7)Results from this specication are given in column IV of table 2.Becauseof the inclusionof the laggeddependentvariable,we omit theboycotted games from the sample, and also the 1988 games, becausethe 1984 medal shares are distorted by the Soviet-led boycott. Thecoefcients on population and per capita GDP are again signicantand statistically equal to each other. Lagged medal share has acoefcient of 0.73 and is strongly signicant. The estimated hosteffectis 1.8 percentagepoints;the Soviet effectand planned-economyeffect are 3.4 and 1.0 percentage points respectively.The statisticalsignicanceof GDP per capitaindicates,perhapsnotsurprisingly, that economic resources are important in producingOlympic medalists. More surprising is the persistentsimilarity of thecoefcients on log population and log GDP per capita. This suggeststhat it is a country’s total GDP that matters in producing Olympicathletes. This in turn has the implication that two countries with thesame GDP will win approximatelythe same number of medals, evenif one is more populouswith lower per capita income and the other issmaller with higher per capita GDP. Furthermore, this section hasidentied some important country characteristics that boost medaltotals, including Soviet and host effects. Finally, there is strongevidence for durability to a country’s Olympic investments. Pastsuccess is an indicator of current success; including lagged medalshare further improves the t of the model. 6 The  Soviet   dummy includes Bulgaria, Czechoslovakia, Poland, theUSSR, East Germany, Hungary, and Romania from 1960 to 1988; theUnied Team in 1992; and Cuba throughout the period. The  Planned  dummy includes China, Albania, Yugoslavia (through 1988), and NorthKorea. 7 All previous results are robust to the omission of these boycottedOlympics. 8 Bernard and Busse (2000) present a talent investment model thatgenerates equation (7). T ABLE  2.—T OBIT OF  M EDAL  S HARE ON  E XPANDED  E XPLANATORY  S ET IndependentVariableMedal ShareI II III IVLog population 0.0128 0.0127 0.0083 0.0064(0.0016) (0.0017) (0.0011) (0.0005)Log GDP per capita 0.0126 0.0125 0.0098 0.0062(0.0014) (0.0015) (0.0010) (0.0006)Host 0.0605 0.0241 0.0122 0.0179(0.0221) (0.0077) (0.0077) (0.0067)Soviet 0.0666 0.0610 0.0300 0.0338(0.0088) (0.0085) (0.0136) (0.0031)Planned (non-Soviet) 0.0177 0.0161 0.0174 0.0101(0.0027) (0.0031) (0.0041) (0.0051)Lagged medal share 0.7333(0.0340)LR test 0.063 (0.80) 0.059 (0.81) 8.400 (0.00) 0.031 (0.86)Boycott years Yes No No NoYear dummies Yes Yes Yes YesRandom effects No No Yes NoLog likelihood 862.06 738.23 984.04 794.03Observations 1254 1036 1036 885 Note: Heteroskedastic- and autocorrelation-consistentstandard errors are in parentheses(see Busse &Bernard, 2002).The LR test reportsthe test of the equality of the coefcients on log GDP per capita andlog population (and the  p -value). It is distributed  chi -squared with 1 degree of freedom. NOTES 415  IV. Predicting Medals in Sydney To provide a sterner test of our framework, we evaluate theout-of-sampleperformance.We made public predictionsbased on themodel several weeks before the 2000 Sydney Games by estimatingequation (7) on the 1996 cross section. 9 This cross-section specica-tion cannotestimate a host effect, so we employ the coefcient on thehost dummy for the same specication pooled over all nonboycottyears (column IV of table 2). Figure 2 shows actual numbers of medals won in 1996 and the predicted numbers from this specica-tion. The model does quite well in predicting totals for a number of countries, including the United States, Russia, and China. However,France, Italy, and Austria won more medals than predicted, andGermany won fewer than predicted.For the 2000 games, we predicted total medal counts for the 36countries that won at least ve medals in 1996. Table 3 contains twosets of predictionsfor the Sydney Games as well as the actual medaltotals for the 36 countries.Columns 1 and 2 represent the predictionsand standard errors the model would have made if it had been 9 See the  New York Times,  September 9, 2000. Johnson and Ali (2000)also predicted country medal totals for the Sydney Games, using differentdependent and independent variables. F IGURE  1.—P REDICTED AND  A CTUAL  M EDAL  T OTALS FOR  1996  FROM THE  S PECIFICATION IN  T ABLE  2, C OLUMN  II Points on the diagonal line represent perfect predictions. F IGURE  2.—P REDICTED AND  A CTUAL  M EDAL  T OTALS FOR  1996  FROM THE  S PECIFICATION IN  T ABLE  2, C OLUMN  IV Points on the diagonal line represent perfect predictions. THE REVIEW OF ECONOMICS AND STATISTICS416


Jul 22, 2017
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!