UNC CHARLOTTE ECONOMICS WORKING PAPER SERIES CAREER DURATION IN THE NHL: PUSHING AND PULLING ON EUROPEANS? Crag A. Depken II Johnny Duckng Peter A. Groothus Workng Paper No. 2016-005 THE UNIVERSITY OF NORTH CAROLINA AT CHARLOTTE BELK COLLEGE OF BUSINESS DEPARTMENT OF ECONOMICS 9201 Unversty Cty Blvd Charlotte NC 28223-0001 August 2016 UNC Charlotte economcs workng papers represent research work-n-progress and are crculated for dscusson and comment purposes. They have not been peer revewed and the expressed vews solely represent those of the authors. All rghts reserved to the authors. Unless otherwse ndcated below short sectons of the text not to exceed two paragraphs may be quoted wthout explct permsson provded that full credt s gven to the source.
ABSTRACT Usng a panel of Natonal Hockey League players from 2000 through 2013 we analyze the determnants of career length n the league. In our analyss we nclude both performance varables and natonalty of orgn to determne ther mportance n determnng career length. We fnd that European-born players have shorter careers than North Amercan-born players holdng performance constant and Russan-born players have even shorter careers than other Europeans. We further fnd that Russan-born players have even shorter careers than other players after the 2005 lockout. These shorter careers are consstent wth ext dscrmnaton aganst European and Russan players pushng them out of the league and voluntary ext by European and Russan players due to opportuntes n ther home countres pullng them out of the league. Ironcally voluntary ext by European and Russan players provdes teams wth a fnancal ncentve to dscrmnate aganst European players.
Career Duraton n the NHL: Pushng and Pullng on Europeans? Crag A. Depken II Professor Department of Economcs Unversty of North Carolna Charlotte Charlotte NC Johnny Duckng Assstant Professor Department of Economcs North Carolna A&T State Unversty Greensboro NC Peter A. Groothus Professor Department of Economcs Appalachan State Unversty Boone NC 1
Career Duraton n the NHL: Pushng and Pullng on Europeans? Abstract Usng a panel of Natonal Hockey League players from 2000 through 2013 we analyze the determnants of career length n the league. In our analyss we nclude both performance varables and natonalty of orgn to determne ther mportance n determnng career length. We fnd that European-born players have shorter careers than North Amercan-born players holdng performance constant and Russan-born players have even shorter careers than other Europeans. We further fnd that Russan-born players have even shorter careers than other players after the 2005 lockout. These shorter careers are consstent wth ext dscrmnaton aganst European and Russan players pushng them out of the league and voluntary ext by European and Russan players due to opportuntes n ther home countres pullng them out of the league. Ironcally voluntary ext by European and Russan players provdes teams wth a fnancal ncentve to dscrmnate aganst European players. JEL Classfcatons: Z22 L83 Keywords: Ext dscrmnaton competng leagues professonal sports return mgraton 2
In fact the other sde of the water s the one thng the Wngs can t offer. Ths s about geography upbrngng and prde. Datsyuk sad he wants to play a last season n Russa whle hs sklls are stll mostly ntact and not be broken down performng a vctory lap that sn t warranted. Mtch Albom Detrot Free Press Introducton The mprovement n human captal and the reducton n transportaton costs coupled wth ncreasng returns to qualty has caused many labor markets to become nternatonal. For nstance n the Unted States n the health care professon 27% of surgeons are foregn born whle n the educaton professon 40% of engneerng professors are foregn born. Aslanbegu and Montecnos (1998) estmated that n the 1990s approxmately 30% of US economcs professors were foregn born. The nternatonalzaton of labor markets has led to many questons such as are foregn workers more productve because of self-selecton on the part of mmgrants (Borjas and Bratsberg 1996) or less productve due to language and cultural dfferences than natve born workers (Dustmann and Soest 2002)? Do foregn born workers experence negatve dscrmnaton (Aslund 2014)? Do foregn workers return to ther natve country after some amount of tme or money earned (Galor and Stark 1990 and Dustmann 1997)? Because sport s most clearly a mertocracy whch drves sports teams to fnd the most talented players all sports leagues have nternatonal labor markets. For nstance n 2014 twenty percent of the players n the Natonal Basketball Assocaton were foregn born whle twenty-fve percent of Major League Baseball were foregn born and n the Natonal Hockey League ffty-one percent of the players were born n Canada twenty-four percent born n the Unted States and twenty-fve percent born n Europe. 1 1 From the 2000-2013 perod players also haled from Indonesa (1) Japan (2) Kazakhstan (7) Ngera (1) and Tanzana (1). These players are not ncluded n the emprcal analyss below. 3
Sports leagues provde a fertle ground for research on mmgraton due to the ncreasng degree of nternatonalzaton of sports leagues. For nstance Kahane Longley and Smmons (2013) fnd that NHL teams who employ a hgher proporton of Europeans perform better f the Europeans are from the same country than teams wth fewer Europeans or Europeans from many dfferent countres. Ths productvty ncrease appears to arse because of the communcaton and cultural consstences that a crtcal mass of foregn players from the same country attans. Such mprovements to productvty would seem to extend a player s career on the margn. However prevous studes suggest that there s dscrmnaton aganst certan foregn players whch mght n turn reduce a player s career length. The purpose of ths paper s to nvestgate how natonalty nfluences career length n the Natonal Hockey League. In sports leagues career length mght be nfluenced by natonalty because of nsttutonal factors of the league. For nstance many European soccer leagues lmt the number of foregnborn players on a team. In the Kontnental Hockey League Russan teams are not allowed more than 5 foregn players. In these cases a player mght be pushed out of the league f ther productvty falls below that of the sxth foregn player avalable. Ths would correspond wth a reduced career length for foregn natonals. 2 A dfferent source of reduced career length mght be an ncrease n lucratve opportuntes n the natve country of a foregn worker or a suffcent amount of savngs on the part of a foregn worker so that the worker fnds t more attractve to return to ther natve country rather than reman n the host country. In the case of professonal sports athletes ths would lead to ther natve country pullng them back home. Ths too would correspond wth a reduced career length. 2 Although no U.S. league explctly lmts the number of foregn born players teams mght stll dscrmnate aganst foregn players due to fan player or management preferences. 4
Our emprcal results suggest that European-born and Russan-born players have shorter careers n the NHL all else equal. After the end of the 2004-2005 lockout Russan-born players seem to have even shorter careers compared to other European and North Amercan-born players. Whle these results mght be consstent wth ext dscrmnaton the greatest effects correspond wth ncreasng opportuntes outsde of the NHL most notably the Kontnental Hockey League n Russa whch has many players pad at or near the mean of NHL players. Lterature Revew Ext dscrmnaton has a long hstory n sports economcs. Johnson and Marple (1973) poneered ths branch of dscrmnaton research when they found evdence from 1970-71 NBA data that margnal whte players had longer careers than margnal black players. Hoang and Rascher (1999) more formally developed a model to explore the concept of racally-based retenton barrers n the NBA. They too found evdence that performance beng equal; there was ext dscrmnaton n the NBA. Groothus and Hll (2004) faled to confrm Hoang and Rascher s results usng more recent data addng heght as an added explanatory varable and usng a duraton model that allows for both stock and flow samples. Jobu (1988) found evdence that race decreased career length ceters parbus for black players but not Hspancs n Major League Baseball from 1971-1985. Agan Groothus and Hll (2008) faled to fnd ext dscrmnaton n MLB usng more recent data from 1990-2004 and a model that better accounted for performance decay. Lastly Duckng Groothus and Hll (2013) fnd no ext dscrmnaton n the NFL. Dscrmnaton n the NHL aganst foregn born players In hockey most of the dscrmnaton lterature has focused on dscrmnaton aganst French Canadans. For nstance both Grener and Lavoe (1988) and Jones and Walsh (1988) fnd 5
sgnfcant pay dscrmnaton aganst French-Canadan defensemen usng 1977-78 data. In addton Lavoe (1989) fnds evdence of postonal stackng nvolvng mnorty (French- Canadan) hockey players. Mongeon and Longley (2015) fnd referees n hockey exhbt dscrmnaton wth French Canadan referees showng a bas by callng more penaltes aganst Englsh Canadan players than aganst French Canadan Players ceters parbus. Lastly Chrste and Lavoe (2015) fnd that there s entry dscrmnaton aganst European players and partcularly Russan players. They further suggest a bas aganst hrng players from the KHL. Overall we know of no studes that have examned career length of foregn born players n hockey. To analyze career duraton n the NHL we use a panel descrbng all North Amercan European and Russan nongoales who played for only one team n a gven year from 2000 to 2013 and non-parametrc and semparametrc duraton technques. Non-parametrc Analyss of Career Duraton n the NHL To help understand career duraton n the NHL we calculate yearly hazard functons as: (1) ht = dt / nt where dt s the number of players who end ther career n year t and nt s the number of players at rsk of endng ther career n year t. The hazard rate can be nterpreted as the proporton of players who exted the NHL gven they have survved up to some level of tenure. In Table 1 we report the hazard rate for non-russan European-born NHL players for Russan-born players and for North Amercan-players who were ether born n Canada or the Unted States. We also report the longest career for each group. In Fgure 1 we plot the hazard functons for each group of players based upon the tenure of the player. We fnd that compared to North Amercan-born players the hazard rate for non-russan 6
European-born players s hgher for the frst four years of tenure and the hazard rate for Russanborn players s hgher for the frst seven years of tenure. Not surprsngly we fnd that the North Amercan-born the Russan-born and the non-russan European-born hazard rates ncrease over tme. The ncrease suggests that the wear and tear from playng hockey as player s age ncreases the lkelhood of ext. Interpretng the dfferences n hazard rates as dscrmnaton however s potentally msleadng because there s no control for productvty dfferences across players. Fgure 2 depcts the hazard rates for the three sets of players over tme. As can be seen the Russan players have a notably hgher hazard rate after the end of the 2004-05 lockout. Sem-parametrc Analyss of Career Duraton n the NHL Methodology We estmate sem-parametrc hazard functons followng Berger and Black (1998) and Groothus and Hll (2004). Because the data are reported at the season level we calculate the hazard rate as a dscrete random varable. As wth Groothus and Hll (2004) we model the duratons of a sngle spell and assume a homogeneous envronment so that the length of a partcular spell s uncorrelated wth the calendar tme at whch the spell begns. Ths assumpton lets us treat all the players' tenure as the same regardless of when t occurred n the panel study. For nstance all fourth-year players are consdered to have the same baselne hazard regardless of calendar tme so a fourth-year player n 2000 has the same baselne hazard as a fourth-year player n 2009. To understand how stock data nfluence a lkelhood functon we follow the notaton of Groothus and Hll (2004). Suppose the probablty mass functon (pmf) of duratons s defned as f ( t x ) where t s the duraton of the career x s a vector of performance and personal characterstcs and s a vector of parameters. Denote F(tx) as the cumulatve dstrbuton 7
functon; the probablty that a career lasts at least t years s then 1 - F(tx). Defnng the hazard functon as h(tx) f(tx) / S(tx) and applyng the defnton of condtonal probabltes the pmf can be expressed as t 1 j0 (2) f ( t x ) [1 h( j x )] h( t x ). If we have a sample of n observatons {t1 t2... tn} the lkelhood functon of the sample s n n t 1 ( [1 h( j x )] h( t x ). 1 1 j1 (3) L ) f ( t x ) Often t s not possble to observe all careers untl they end hence careers are often rghtcensored. Let the set A be all observatons where careers are completed durng the sample perod and the set B be all observatons where careers are rght censored. For the set B all we know s that the actual length of the career s greater than t the observed length of the career up through the last year. Because we know that the actual length of the career s longer than we observe then the contrbuton of these observatons to the lkelhood functon s just the survvor functon t1 Stx ( ) [1 hx ( )]. 1 To ntroduce stock samplng let the set C be the careers that were n progress when data collecton began. For these observatons we know that the career for player has lasted for r years before the panel begns so the lkelhood must be adjusted by the condtonal probablty of the career havng length r. Of course some stock-sampled observatons may be rght-hand censored. Let the set D be all observatons that are both rght and left censored. An example of a career that s both rght and left censored would be a player who starts hs career pror to 2000 and ends hs career after 2013. Takng nto account all four sets: A B C and D the lkelhood functon becomes 8
9 (4) D t r j C t r j B t j A t j x j h x t h x j h x j h x t h x j h L 1 1 1 1 1 1 )] ( [1 ) ( )] ( [1 )] ( [1 ) ( )] ( [1 ) ( In equaton (4) the contrbuton of censored stock-sampled observatons to the lkelhood functon s strctly from the last two terms; such observatons smply provde nformaton about the survvor functon between (rt). Thus we as Groothus and Hll (2004) have expressed the lkelhood functon as a functon of the hazard functons. All that remans s to specfy the form of a hazard functon and estmate by means of maxmum lkelhood estmaton. As the hazard functon s the condtonal probablty of extng the NHL gven that the NHL career lasted untl the prevous season the hazard functon must have a range from zero to one. In prncple any mappng wth a range from zero to one wll work. Cox (1972) recommends (5) ) exp( 1 ) ( 1 ) ( x e h h x t h x t h t x t t whch s smply the logt model wth ntercepts that dffer by tme perods. The term ht s a baselne hazard functon whch s common to all observatons. The x term determned by the player s personal and productvty characterstcs shfts the baselne hazard functon but t affects the baselne hazard functon n exactly the same way n each perod. Berger and Black (1998) consder other hazard functons and fnd that ther results are relatvely robust across varous specfcatons of the hazard functon. As the logt model s avalable n many software packages we follow Cox and use the logt model. The ntuton behnd equaton (5) when usng the logt model for the hazard functon s relatvely smple. For each year durng the survey n whch the player s n the NHL the player
ether comes back for another season or ends hs career. If the career ends the dependent varable takes on a value of one; otherwse the dependent varable s zero. The player remans n the panel untl the player exts the NHL or the panel ends. If the panel ends we say the worker s spell s rght censored. Thus a player who begns hs NHL career durng the panel and plays for 6 years wll enter the data set 6 tmes: the value of hs dependent varable wll be zero for the frst 5 years (tenure one through fve) and be equal to one for the sxth year. To llustrate a stock observaton consder another player who enters the panel wth 7 years of NHL job tenure pror to 2000 the frst year of the panel then plays for an addtonal 3 years for a 10-year career. For ths player we gnore hs frst 7 years of tenure because he s left censored. As the equaton of the lkelhood functon wth stock data ndcates the duraton of a NHL career pror to the begnnng of the panel makes no contrbuton to the value of the lkelhood functon. Therefore only years 8 through 10 wll enter the data set wth the dependent varable takng on the value zero for year 8 and 9 and n the 10 th year t takes on a value of one ths player appears n the data set a total of 3 tmes. Note for all players who are rght censored we do not know when ther career ends so ther dependent varables are always coded as zero. Because the players n the panel have varyng degrees of job tenure pror to the begnnng of the panel we dentfy the hazard functon for both long and short careers. The dsadvantage to ths approach s that the vector t of equaton (5) can be very large. In our study t would requre 25 dummy varables. We also run nto problems wth the Cox technque because we have too few players who have long careers. To smplfy the computaton of the lkelhood functon and to be able to keep the long careers we approxmate the t vector wth a 5 th order polynomal of player s tenure whch reduces the number of parameters to be estmated from 25 to 5. Thus the hazard functon becomes 10
(6) ht ( x ) x () t e exp( () t x) 1 ht ( x ) where (t) s a 5 th order polynomal n the worker s tenure. We choose the Taylor seres approxmaton technque over usng tenure dummes due to the small number of observatons for hgh tenures. Ths method provdes a very flexble specfcaton of the baselne hazard but does mpose more restrctons than Cox s model. 3 Estmaton Results In Table 2 we report the means of the varables used n the analyss. We fnd that both non- Russan Europeans and Russans have a hgher ext rate than North Amercans. North Amercans have fewer games played asssts and goals than non-russan Europeans and Russans ndcatng that they are less sklled than ther European counterparts. North Amercans have more penalty mnutes than non-russan Europeans and Russans. Both Russan and non-russan Europeans have a better plus mnus than North Amercans ndcatng that teams are better n terms of scorng dfferental when North Amercans are not on the ce. Both non-russan Europeans and Russans are taller than North Amercans. Russans wegh more than North Amercans and North Amercans wegh more than non-russan Europeans. The average age for all three groups s between 26.5 and 27 years of age. Approxmately 70% of all the observatons occur after the lockout but only 51% of the Russan observatons occur after the lockout. Only 4% of the sample s Russan and 21% s non-russan European. 3 When hgher order polynomals of the sxth and seventh power are ncluded the results do not change suggestng that a ffth order polynomal s flexble enough to capture the nfluence of the baselne hazard. 11
In table 3 we report the percentage of non-russan European and Russan players by year. In 2000 the percentage of non-russan European players n the league was 22.3% and the percentage of Russan players was 6.9%. Over the course of the sample perod the percentage of non-russan European players fell to 15% and the percentage of Russan players fell to 2.2%. Table 3 reveals that there has been a steady declne n the percentage of non-russan European and Russan players over tme. In Table 4 we report the results of estmatng equaton 6 for two specfcatons. The frst controls for the post-lockout perod alone and the second controls for dfferences across non- Russan European and Russan-born players n the post-lockout perod wth North Amercan-born players beng the reference category. In both cases better performance decreases the lkelhood of ext wth the coeffcents on games played goals asssts penalty mnutes and plus mnus 4 all beng negatve and statstcally sgnfcant. In addton we fnd that the coeffcent on age s postve and sgnfcant suggestng that older players are more lkely to ext than younger players. We also fnd that the coeffcent on the year the player played s negatve and statstcally sgnfcant suggestng that recent players are less lkely to ext than past players. We also fnd that heaver players are less lkely to ext and taller players are more lkely to ext the league. As n other studes that fnd dfferental treatment of hockey players based on natonalty we also fnd that country of orgn nfluences whether a player exts the league. Usng North Amercan-born players as the reference category both Russan-born and non-russan Europeanborn hockey players have a hgher probablty of ext n a gven year ceters parbus on 4 The "plus-mnus" statstc s calculated as a ponts dfferental. When an even-strength or shorthanded goal s scored every player on the ce for the scorng team s credted wth a "plus." Every player on the ce for the team scored aganst gets a "mnus." A player's overall plus-mnus s calculated by subtractng the mnuses from the pluses. A hgh plus-mnus s taken to mean the player s a good offensve or defensve player. 12
productvty and age. Our results could be consstent wth customer-based dscrmnaton f fan preference for North Amercan-born players s suffcently hgh or wth co-worker dscrmnaton that mght arse wth possble language or cultural dfferences (see Kahane et al. 2013). However there appears to be lttle emprcal evdence of wde-spread dscrmnaton aganst European (both Russan and non-russan) players n other dmensons of hockey such as salares. Therefore t s not mmedately obvous that the evdence ponts to aspects of the NHL pushng European and Russan players away. However the ncreased lkelhood of Europeans and Russans leavng concdes wth ncreasngly lucratve opportuntes to play hockey n Europe most notably n the KHL n Russa. To get a feel for the magntude of ext lkelhood we convert the coeffcents nto a percentage change by usng 100(exp(β)-1) for each dummy varable. We fnd that a non-russan European-born player has a 125% hgher lkelhood of extng than a North Amercan-born player holdng performance constant whle Russan-born players have a 184% hgher lkelhood of extng ther career than a North Amercan-born player pror to the NHL lockout that clmbs to 423% after the NHL lockout holdng performance constant. Our results suggest that although there has been a large nflux of foregn players from Europe nto the NHL there s somethng pushng and/or pullng that reduces career duraton of players from Europe below what performance would suggest. Nonlnear Decompostons of Career Length Dfferences To further explore the effect of beng foregn born on ext n the NHL we use the Blnder-Oaxaca nonlnear decomposton technque (Snnng Hahn and Bauer 2008). Lke the lnear Blnder (1973) and Oaxaca (1973) method the nonlnear method decomposes the dfference between 13
groups nto dfferences across ndvdual characterstcs and dfferences across coeffcents. In the lnear case the decomposton s: (7) ῩA - ῩB = (XA XB)β* + XA(βA- β*) + XB(βB- β*) where ῩA - ῩB s the total dfference between two groups (XA XB)β* s the dfference due to dfferent ndvdual characterstcs XA(βA- β*) s the advantage of beng n group A XB(βB- β*) s the dsadvantage n beng n group B and β* s a weghted average of the coeffcent vectors βa and βb. In the smple Blnder and Oaxaca (1973) decomposton method β* s ether set to βa or to βb. Two alternatve methods to determne β* are the Cotton (1988) and the Neumark (1988) methods where Cotton uses a weghted average technque and Neumark (1988) uses a pooled model to derve β* (Snnng Hahn and Bauer 2008). The non-lnear technque follows the same pattern decomposng the logt equatons nto the percentage determned by characterstcs n our case performance and the percentage determned by dfferences n coeffcents. In our case we estmate separate logt models for North Amercan-born players for non- Russan European-born players and for Russan-born players. We then perform two sets of decompostons one for North Amercan-born players compared to non-russan European-born players and one for North Amercan-born players compared to Russan-born players. The results for the separate logt models are reported n Table 5. We report the decomposton results n Tables 6A and 6B. In Table 6A we report the results of the non-lnear decompostons between North Amercan-born and non-russan-born Europeans usng the smple Blnder and Oaxaca technque the Cotton technque and the Neumark technque. We fnd that the raw dfference s.0217. We convert the coeffcents nto a percentage change of the raw dfference by usng 100(exp(β)-1) and fnd that non-russan European players have only two percent hgher lkelhood of ext than North 14
Amercan-born players on average when not controllng for productvty dfferences. Usng the decomposton technque however we fnd that dfferences n productvty however lower the lkelhood of ext. The effect ranges from -193 to -237 percent dependng upon what decomposton technque s performed. These results suggest that f only productvty determned career length Europeans should have a sx percent lower probablty of ext and thus longer careers than North Amercans. But when focusng on dfferences n coeffcents we fnd that ths dfference counts from 297% to 337% of the dfferental showng that on average non-russan-born Europeans are more lkely to ext than North Amercan-born players. These results suggest that European-born players have an eght percent hgher lkelhood of ext when controllng for performance. In Table 6B we report the results of the non-lnear decompostons between North Amercan-born and Russan-born players usng the smple Blnder and Oaxaca technque the Cotton technque and the Neumark technque. We fnd that the raw dfference s.066. We convert the coeffcents nto a percentage change of the raw dfference by usng 100(exp(β)-1) and fnd that Russan players have a seven percent hgher lkelhood of ext than North Amercan-born players on average. Usng the decomposton we fnd that dfferences n productvty however lowers the lkelhood of ext. The effect ranges from -169% to -72% percent dependng upon what decomposton technque s performed. These results suggest that f only productvty determned career length Russans should have a fve percent to twelve percent lower probablty of ext and longer careers than North Amercans. But when focusng on dfferences n coeffcents we fnd that ths dfference counts from a 29% to 246% of the dfferental showng that on average Russanborn players are more lkely to ext than North Amercan-born players. 15
Our decomposton results further suggest that although there has been a large nflux of foregn players from Europe somethng s pushng and/or pullng European players from the NHL and makng ther careers shorter than ther productvty would suggest. Ether ext dscrmnaton or voluntary ext due to other factors determnes why European players have a hgher lkelhood of ext. One potental explanaton s that Europeans return to ther natve countres to end ther careers n professonal hockey leagues there (Albom 2016). The NHL Lockout s Influence on NHL Ext The 2004-2005 NHL season was cancelled because of a lockout aganst the players. The lost season provded an opportunty for many players especally European-born players to seek employment n other hockey leagues around the world. After the end of the lockout brought about by a collectve barganng agreement whch dramatcally altered the market prce for hockey talents by ntroducng an ndvdual and team salary cap reducng the average salares to year- 2000 levels and ntroducng greater mnmum wages (see Depken and Lureman 2014 for further analyss of how the market for hockey talents changed after the lockout). These changes n the market for NHL talents mght have provded European-born players a greater ncentve to leave the NHL early to play n European leagues that mght be more compettve n salares. Ths appears to have been the case wth the KHL whch started payng hgher salares startng n 2008 and thereby recrutng heavly from NHL players. To test for changes n ext patterns after the collectve barganng agreement we nteract the dummy varables descrbng country of brth wth a dummy varable that takes a value of one after the lockout ended and zero otherwse. The results ncludng these nteractons are reported n Column 2 of Table 4. To better nterpret the magntude of the lkelhood of ext n both the pre- 16
and post-lockout perod we convert the coeffcents nto a percentage by usng 100(exp(β)-1) for each country-of-orgn dummy varable and post-lockout nteracton. We fnd that a non-russan European-born player had a 125% hgher lkelhood of extng n a gven year than a North Amercan-born players pror to the lockout holdng performance constant. After the lockout ths lkelhood ncreased an addtonal 15 percentage ponts although the post-lockout effect s statstcally nsgnfcant. Russan-born players have a 184% hgher lkelhood of extng ther career than North Amercan-born players pror to the lockout and ths lkelhood ncreases by an addtonal 239 percentage ponts after the lockout for a total 423% hgher lkelhood of ext than North Amercan-born players. Concluson We fnd that European-born hockey players have shorter careers than ther performance statstcs would suggest. In partcular Russan-born hockey players have the hghest lkelhood of ext durng the perod 2000-2013 and ths lkelhood ncreased dramatcally after the end of the 2004-2005 lockout n the NHL whch also corresponded to ncreased salares n the Russan Kontnental Hockey League. Our statstcal evdence suggests that there s somethng pushng and/or pullng European players n the NHL nto extng earler than North Amercan-born players. Whle customer-based coworker-based or employer-based ext dscrmnaton mght be pushng these players to leave the league more lucratve opportuntes to play hockey n ther natve countres mght be pullng players to leave the NHL earler than otherwse predcted. There s anecdotal evdence that many players choose to retre from the NHL early to wnd down ther careers n leagues n ther natve countres wth shorter seasons and fewer njures although wth lower salares. Future research would do well to focus on these optons to foregn-born NHL players. 17
Although t s dffcult to determne from our analyss whether the ncrease n the ext of European players s determned by ext dscrmnaton by the team or voluntary ext by the player t s lkely that a porton of the hgher probablty of ext by European players s due to ext dscrmnaton. In the case of European players beng pulled back to ther natve countres (Albom 2016) ext dscrmnaton s a plausble response by teams. Ths s because European players that ext the league voluntarly to return to ther home countres have the ablty to leave ther team wth salary cap and roster ssues that mpose addtonal costs on the team. These addtonal costs provde teams wth an ncentve to release European players from the team sooner than they would release a North Amercan-born player wthout an ncentve to leave the team and return to hs natve country. Even f there would be no dscrmnatory behavor towards European players n the absence of voluntary exts the presence of voluntary exts by European players provde teams wth an ncentve to dscrmnate aganst European players. 18
References Albom Mtch (2016 Aprl 11) Pavel Datsyuk says he wll leave the Red Wngs after playoffs Detrot Free Press. Retreved from http://www.freep.com/story/sports/ columnsts/mtch-albom/2016/04/10/albom-wngs-datsyuk-column/82835594 Alvarez Jame Davd Forrest Ismael Sanz Juan de Dos Tena (2011) Impact of mportng foregn talent on performance levels of local co-workers Labour Economcs 18 (3): 287 296. Aslanbegu Nahd and Veronca Montecnos (1998) "Foregn Students n U.S. Doctoral Programs" Journal of Economc Perspectves 12(3): 171-182. Åslund Olaf Lena Hensvk and Oskar Nordström Skans (2014) Seekng Smlarty: How Immgrants and Natves Manage n the Labor Market Journal of Labor Economcs 32(3): 405-441. Berger Mark C. and Dan A. Black (1998) "The Duraton of Medcad Spells: An Analyss Usng Flow and Stock Samples" The Revew of Economcs and Statstcs 80(4): 667-674. Blnder Alan S. (1973) Wage Dscrmnaton: Reduced Form and Structural Estmates Journal of Human Resources 8(4): 436 455. Borjas George and Brent Bratsberg (1996). Who Leaves? The Outmgraton of the Foregn Born The Revew of Economcs and Statstcs 78(1): 165-176. Chrste Tom and Marc Lavoe (2015). Entry Dscrmnaton n the NHL: Evoluton and the KHL Effect Eastern Economcs Journal 41(2): 214-229. Cotton Jeremah (1988) On the Decomposton of Wage Dfferentals Revew of Economcs and Statstcs70(2): 236 243. Depken Crag A. and Jeff Lureman (2014) "Wage Dsparty Team Performance and the 2005 NHL Collectve Barganng Agreement" mmeo Unversty of North Carolna at Charlotte. Duckng Johnny Peter A. Groothus and James Rchard Hll (2014) Compensaton Dscrmnaton n the NFL: An Analyss of Career Earnngs Appled Economc Letters 21(10): 679-682. Dustmann Chrstan (1997) Return Mgraton Uncertanty and Precautonary Savngs Journal of Development Economcs 52(2): 295-316. Dustmann Chrstan and Arthur Van Soest (2002) Language and the Earnngs of Immgrants Industral and Labor Relatons Revew 55 (3):473-492 19
Galor Oded and Oded Stark (1990) Mgrants Savngs the Probablty of Return Mgraton and Mgrants Performance Internatonal Economc Revew 31(2): 463-467. Groothus Peter A. and James Rchard Hll (2004) Ext Dscrmnaton n the NBA: A Duraton Analyss of Career Length Usng Flow and Stock Samples Economc Inqury 42(2): 341-49. Groothus Peter A. and James Rchard Hll (2008) Ext Dscrmnaton n Major League Baseball: 1990-2004 Southern Economc Journal (75)2: 574-590. Hoang Ha and Dan Rascher (1999) The NBA Ext Dscrmnaton and Career Earnngs Industral Relatons 38(1): 69-91. Jobu Robert M (1988) Racal Inequalty n a Publc Arena: The Case of Professonal Baseball Socal Forces 67(2): 524-34. Johnson Norrs R. and Davd P. Marple (1973) Racal Dscrmnaton n Professonal Basketball: An Emprcal Test Socologcal Focus (6)4: 6-18. Jones J.C. H. and Wllam D. Walsh (1988) Salary Determnaton n the Natonal Hockey League: The Effects of Sklls Franchse Characterstcs and Dscrmnaton Industral and Labor Relatons Revew (41)4: 592 604. Kahane Leo Nel Longley and Robert Smmons (2013) The Effects of Coworker Heterogenety on Frm-Level Output: Assessng the Impacts of Cultural and Language Dversty n the Natonal Hockey League Revew of Economcs and Statstcs 95(1): 302-314. Kahn Lawrence M. (1991) Dscrmnaton n Professonal Sports: A Survey of the Lterature Industral and Labor Relatons Revew (44)3: 395-418. LaVoe Marc (1989) Stackng Performance Dfferentals and Salary Dscrmnaton n Professonal Ice Hockey: A Survey of the Evdence Socology of Sport Journal (6)1: 17-35. Lavoe Marc Gles Grener and Serge Coulumbe (1987) Dscrmnaton and Performance Dfferentals n the Natuonal Hockey League Canadan Publc Polcy/Analyse de Poltques. (13)4; 407-22. Mongeon Kevn and Nel Longley (2015) Testng for Ethncty Dscrmnaton among NHL Referees: A Duraton Model Approach Eastern Economcs Journal 41(1) 86-101. Neumark Davd (1988) Employers Dscrmnatory Behavor and the Estmaton of Wage Dscrmnaton Journal of Human Resources 23(3): 279 295. 20
Oaxaca Ronald (1973) Male Female Wage Dfferentals n Urban Labor Markets Internatonal Economc Revew 14(3): 693 709. Snnng Mathas Markus Hahn and Thomas K. Bauer T. K. (2008) The Blnder Oaxaca Decomposton for Nonlnear Regresson Models The Stata Journal 8(4): 480 492 21
Table 1: Hazard Rates Tenure Full Sample North Amercan European (Non-Russan) Russan 1.163 n=1447.150 n=1053.205 n=335.169 n=59 2.105 n=1095.071 n=787.192 n=255.207 n=53 3.114 n=1092.104 n=823.132 n=227.214 n=42 4.135 n=1035.109 n=783.201 n=208.272 n=44 5.144 n=877.144 n=672.122 n=171.264 n=34 6.131 n=800.142 n=605.088 n=170.160 n=25 7.141 n=733.132 n=545.152 n=157.225 n=31 8.138 n=643.133 n=478.158 n=139.115 n=26 9.128 n=561.122 n=399.148 n=128.117 n=34 10.143 n=467.141 n=333.134 n=97.189 n=37 11.152 n=458.147 n=326.127 n=102.300 n=30 12.176 n=379.196 n=275.108 n=83.190 n=21 13.201 n=332.218 n=247.157 n=70.133 n=15 14.184 n=271.186 n=193.125 n=64.428 n=14 15.264 n=227.229 n=170.367 n=49.375 n=8 16.206 n=179.210 n=138.222 n=36.000 n=5 17.405 n=148.400 n=110.438 n=32.333 n=6 18.263 n=95.315 n=76.066 n=15.000 n=4 19.448 n=78.442 n=61.461 n=13.500 n=4 20.285 n=42.285 n=35.333 n=6.000 n=1 Longest Career 27 years 27 years 24 years 21 years 22
Table 2: Descrptve Statstcs of the Sample Varable Full Sample North Amercan European (Non-Russan) Russan Ext NHL 0.15 (0.36) 0.14 (0.35) 0.17 (0.37) 0.21 (0.40) Games played 47.44 (28.42) 45.92 (28.67) 51.78 (27.37) 51.72 (26.67) Asssts 12.51 (13.15) 11.26 (12.41) 15.74 (14.25) 17.54 (15.59) Goals 7.32 (8.96) 6.69 (8.45) 8.81 (9.61) 10.60 (11.92) Penalty Mnutes 35.18 (35.61) 36.98 (38.30) 29.92 (25.74) 30.75 (26.22) Plus Mnus -0.25 (9.34) -0.52 (8.94) 0.67 (10.36) -0.26 (10.21) Weght (Pounds) 203.20 (15.69) 203.29 (15.62) 202.46 (15.66) 205.41 (16.74) Heght (Inches) 73.14 (2.05) 73.10 (2.06) 73.25 (2.00) 73.21 (2.18) Age (Years) 26.87 (4.56) 26.90 (4.51) 26.84 (4.63) 26.52 (5.01) Post Lockout (1=Yes).68 0.69 0.65 0.51 European-born (1=Yes).21 - - - Russan-born (1=Yes).04 - - - Post Lockout (1=Yes).14 - - - * European-born Post Lockout (1=Yes).02 - - - * Russan-born Sample Sze 11029 8170 2365 494 (Standard devaton n parentheses) 23
Table 3: Percentage of Non-Russan European and Russan Players by Year Year Proporton European (Non-Russan) Proporton Russan 2000 22.3 6.9 2001 23.8 6.4 2002 24.2 6.1 2003 23.4 6.1 2004 Lockout Lockout 2005 22 4.5 2006 22.9 3.9 2007 22.7 3.4 2008 21.4 3.2 2009 20.8 3.3 2010 19.6 3 2011 19.3 2.7 2012 17.6 2.3 2013 15 2.2 24
Table 4: Determnants of NHL Ext: Logt Model Model (1) Model (2) Games Played -0.021*** -0.021*** (0.003) (0.003) Asssts -0.094*** -0.094*** (0.011) (0.011) Goals -0.045*** -0.044*** (0.012) (0.013) Penalty Mnutes -0.010*** -0.010*** (0.002) (0.002) Plus Mnus -0.028*** -0.028*** (0.005) (0.005) Weght -0.011*** -0.011*** (0.003) (0.003) Heght 0.037* 0.037* (0.021) (0.021) Age 0.233*** 0.236*** (0.015) (0.015) Year -0.036** -0.035** (0.014) (0.014) Post Lockout 0.133 0.059 (0.129) (0.135) European 0.754*** 0.652*** (0.078) (0.126) Russan 1.361*** 1.044*** (0.149) (0.212) European x Post Lockout 0.153 (0.159) Russan x Post Lockout 0.611* (0.318) Constant 63.980** 62.914** (28.510) (28.575) Observatons 11029 11029 Robust standard errors n parentheses. *** p<0.01 ** p<0.05 * p<0.1 25
Table 5: Determnants of NHL Ext: Separate Logt Models for Decompostons (1) (2) (3) North Amercan European (Non-Russan) Russan Games Played -0.030*** -0.005-0.006 (0.003) (0.005) (0.008) Asssts -0.074*** -0.122*** -0.129*** (0.014) (0.019) (0.032) Goals -0.064*** -0.040* 0.025 (0.017) (0.020) (0.025) Penalty Mnutes -0.008*** -0.020*** -0.010 (0.002) (0.004) (0.009) Plus Mnus Goals -0.038*** -0.012-0.005 (0.007) (0.009) (0.020) Weght -0.009*** -0.015*** -0.012 (0.003) (0.006) (0.012) Heght 0.038 0.036-0.045 (0.025) (0.045) (0.092) Age 0.235*** 0.253*** 0.220*** (0.018) (0.030) (0.057) Year -0.010-0.091*** -0.050 (0.017) (0.030) (0.068) Post CBA -0.114 0.657** 0.889 (0.155) (0.260) (0.543) Constant 13.477 175.402*** 98.383 (33.867) (59.632) (137.385) Observatons 8170 2365 494 Robust standard errors n parentheses. *** p<0.01 ** p<0.05 * p<0.1 26
Omega =1 Characterstcs Coeffcents Omega =0 Characterstcs Coeffcents Cotton Technque Productvty Advantage Dsadvantage Neumark Technque Productvty Advantage Dsadvantage Table 6A: Nonlnear Decompostons of Ext Logts: European-born Coeffcent -.0514***.0731*** -.0428***.0645*** -.0498***.0181***.0536*** Standard Error 1.0062.0089.0034.0074.0051.0021.0067 Percentage -237% 337% -197% 297% -229% 83% 247% -.0419***.0494***.0143***.0033.0055.0015-193% 227% 66% Raw Dfference.0217.0079 100% Number of Observatons n European-born Group 2365 Number of Observatons n North Amercan-born Group 8170 1 Standard Errors calculate by 50 Bootstrap Replcatons usng STATA Omega =1 Characterstcs Coeffcents Omega =0 Characterstcs Coeffcents Cotton Technque Productvty Advantage Dsadvantage Neumark Technque Productvty Advantage Table 6B: Nonlnear Decompostons of Ext Logts: Russan-born Coeffcent -.1116***.1777*** -.0488***.1148*** -.1073***.0103***.1630*** -.0475***.1070***.0064*** Standard Error 1.0276.0342.0069.0176.0257.0017.0317.0065.0161.0001 Pct of Raw Dfference -169% 29% -74% 174% -163% 16% 246% -72% 162% 10% Dsadvantage Raw Dfference.0660.0172 100% Number of Observatons n Russan-born Group 494 Number of Observatons n North Amercan-born Group 8170 1 Standard Errors calculated by 50 Bootstrap Replcatons 27
Fgure 1: Hazard Rates for Career Endng by Years of Tenure.05.1.15.2.25.3 0 5 10 Tenure NA EU RU 28
Fgure 2: Hazard Rates for Career Endng by Year (2000-2013).1.15.2.25.3 2000 2005 2010 2015 Year NA EU RU 29