NBER WORKING PAPER SERIES PROFESSIONALS DO NOT PLAY MINIMAX: EVIDENCE FROM MAJOR LEAGUE BASEBALL AND THE NATIONAL FOOTBALL LEAGUE
|
|
- Darcy Harris
- 6 years ago
- Views:
Transcription
1 NBER WORKING PAPER SERIES PROFESSIONALS DO NOT PLAY MINIMAX: EVIDENCE FROM MAJOR LEAGUE BASEBALL AND THE NATIONAL FOOTBALL LEAGUE Kenneth Kovash Steven D. Levitt Working Paper NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambridge, MA September 2009 We are grateful to John List, Ben Baumer, Andrew Stein, Jeff Ma, Mark Kamal, Lucas Ruprecht, and Paraag Marathe for helpful comments and discussions. Daniel Hirschberg provided outstanding research assistance. Thanks to Baseball Info Solutions and Citizen Sports Network for providing data. Correspondence should be addressed to: Professor Steven Levitt, Department of Economics, University of Chicago, 1126 E. 59th Street, Chicago, IL 60637, or The views expressed herein are those of the author(s) and do not necessarily reflect the views of the National Bureau of Economic Research. NBER working papers are circulated for discussion and comment purposes. They have not been peerreviewed or been subject to the review by the NBER Board of Directors that accompanies official NBER publications by Kenneth Kovash and Steven D. Levitt. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including notice, is given to the source.
2 Professionals Do Not Play Minimax: Evidence from Major League Baseball and the National Football League Kenneth Kovash and Steven D. Levitt NBER Working Paper No September 2009 JEL No. D01,D82 ABSTRACT Game theory makes strong predictions about how individuals should behave in two player, zero sum games. When players follow a mixed strategy, equilibrium payoffs should be equalized across actions, and choices should be serially uncorrelated. Laboratory experiments have generated large and systematic deviations from the minimax predictions. Data gleaned from real-world settings have been more consistent with minimax, but these latter studies have often been based on small samples with low power to reject. In this paper, we explore minimax play in two high stakes, real world settings that are data rich: choice of pitch type in Major League Baseball and whether to run or pass in the National Football League. We observe more than three million pitches in baseball and 125,000 play choices for football. We find systematic deviations from minimax play in both data sets. Pitchers appear to throw too many fastballs; football teams pass less than they should. In both sports, there is negative serial correlation in play calling. Back of the envelope calculations suggest that correcting these decision making errors could be worth as many as two additional victories a year to a Major League Baseball franchise, and more than a half win per season for a professional football team. Kenneth Kovash Mozilla 5807 S Woodlawn Ave Chicago, IL kkovash@gmail.com Steven D. Levitt Department of Economics University of Chicago 1126 East 59th Street Chicago, IL and NBER slevitt@midway.uchicago.edu
3 Aspiriteddebatehasarisenregardingthequestionoftheextenttowhichactions intwo-playerzerosumgamesconformtothepredictionsofgametheory.von Neumann sminimaxtheorymakesthreebasicpredictionsaboutbehaviorinsuchgames. First,sincetheplayermustbeindifferentbetweenactionsinordertomix,theexpected payoffsacrossallactionsthatarepartofthemixingequilibriummustbeequalized. Second,theexpectedpayoffforallactionsthatareplayedwithpositiveprobabilitymust begreaterthantheexpectedpayoffforallactionsthatarenotplayedwithpositive probability. 2 Third,thechoiceofactionsispredictedtobeseriallyindependent,sinceif thepatternofplayispredictable,itcanbeexploitedbyanoptimizingopponent. Laboratorytestsofminimaxhave,almostwithoutexception,shownsubstantial deviationsfromthesetheoreticalpredictions(lieberman,1960,1962;brayer1964, Messick1967,Fox1972,BrownandRosenthal1990,Rosenthaletal.2003).One remarkableexceptiontothispatternispalacios-huertaandvolij(2008),inwhichsoccer playerswhoarebroughtintothelaboratoryplayminimaxinbotha2x2gameand O Neill s(1987)4x4game.levitt,list,andreiley(2008),however,areunableto replicatethesefindings,eitherusingprofessionalsoccerplayersorworldclasspoker players,callingthepalacios-huertaandvolijresultsintoquestion. Instarkcontrasttothelab,theexistingliteratureonminimaxplayinthefieldhas generallyprovidedsupporttothepredictionsofgametheory.whetherthetaskisthe directionofaserveintennis(walkerandwooders2001,hsu,huang,andtang2007), 2 Inmostpriorempiricalanalyses,thispredictionhaseitherbeenirrelevantofuntestable.Ifthereareonly twoactionsavailable(e.g.walkerandwooders2001,hsu,huang,andtang2007),thenbydefinitionany mixingstrategymustincludebothactions.evenwhentheactionspaceisricher,theexpectedpayofftoan actionthatisnottakenistypicallynotobserved,makingthispredictionuntestable.hirschberg,levitt,and List(2009),ontheotherhand,areabletoaddressthispredictioninstudyingtheactionsofpokerplayers becauseoftherichnessoftheirdata.
4 penaltykicksinsoccer(chiappori,levitt,andgroseclose2002,palacios-huerta2003), orthedecisiontocall,check,fold,orraiseinlimitpoker(hirschbergetal.2009), equalizedpayoffsacrossactionsthatareincludedinthemixedstrategycannotbe rejected.theevidenceonserialindependenceinthefieldismoremixed,buthasfound somesupport(e.g.,hsu,huang,andtang2007). Thereareanumberofpossibleexplanationsforthesharpdifferencesobservedin priorstudiesdoneinthelaboratoryversusthefield.failureofminimaxinthelabmay betheresultofalackoffamiliaritywiththegamesthatareplayed,lowstakes,or selectionofparticipantsintothestudieswhodonothaveexperienceortalentformixing. Averydifferentexplanationforthecontrastingconclusionsoflabandfieldstudiesisthat fieldstudiestendtohaveverylowpowertoreject(levittandlist2007).forinstance, inchiappori,levitt,andgroseclose(2002),thetotalnumberofpenaltykicksisonly 459,spreadovermorethan100shooters.WalkerandWooders(2001)observe approximately3,000servesspreadoverfortygrandslamtennismatches. Inthispaper,weaddtotheexistingliteraturebystudyingbehaviorintwonew fieldsettings:thepitcher schoiceofpitchtype(e.g.,fastballversuscurveball)in professionalbaseball,andtheoffense schoiceofrunversuspassinprofessionalfootball. Ineachofthesesettings,weareabletoanalyzefarmoredatathanhaspreviouslybeen availableinfieldstudiesofmixedstrategybehavior.inthecaseofbaseball,weobserve everypitchthrowninthemajorleaguesovertheperiod atotalofmorethan 3millionpitches.Forfootball,weobserveeveryplayintheNationalFootballLeague fortheyears over125,000plays.inbothsettings,thechoicesbeingmade haveveryhighstakesassociatedwiththem.
5 Theresultsobtainedfromanalyzingthefootballandbaseballdataarequite similar.inbothcases,wefindcleardeviationsfromminimaxplay,asevidencedbya failuretoequalizeexpectedpayoffsacrossdifferentactionsplayedaspartofmixed strategies,andwithrespecttonegativeserialcorrelationinactions.inthenfl,wefind thatoffensesonaveragedosystematicallybetterbypassingtheballratherthanrunning. Inbaseball,pitchersappeartothrowtoomanyfastballs,i.e.,batterssystematicallyhave betteroutcomeswhenthrownfastballsversusanyothertypeofpitch. Infootball,teamsaremorelikelytorunifthepreviousplaywasapass,andvice versa.thispatternisespeciallypronouncedwhenthepreviousplaywasunsuccessful. Negativeserialcorrelationinactionsisconsistentwithalargebodyofpriorlaboratory evidence(e.g.,brownandrosenthal1990).pitchersalsoexhibitsomenegativeserial correlation,particularlywithfastballs,i.e.,theyaremorelikelytothrowanon-fastballif thepreviouspitchwasafastball,andviceversa. Themagnitudeofthesedeviationsisnottrivial.Back-of-the-envelope calculationssuggestthattheaveragenflteamsacrificesonepointagameonoffense (4.5percentofcurrentscoring)asaconsequenceofthesemistakes.Inbaseball,we estimatethattheaverageteamgivesupanextra20runsaseason(abouta1.3percent increase).iftheseestimatesarecorrect,thenthevaluetoimprovingthesedecisionsison theorderof$4millionayearforthetypicalbaseballteamand$5millionayearforan NFLfranchise. Theremainderofthepaperisstructuredasfollows.SectionIreportsourresults formajorleaguebaseball.sectioniianalyzesnflfootball.sectioniiiconcludes.
6 SectionI:Ananalysisofpitchchoiceinmajorleaguebaseball OurdataonpitchchoiceinmajorleaguebaseballwerepurchasedfromBaseball InfoSolutions,whichemploysdatatrackersatallgamesandcompilestheinformationin ordertosellittomajorleagueteamsandotherinterestedparties.thedatasetincludesa wealthofinformationforeachpitchthrowninthemajorleaguesovertheperiod :theidentityofthepitcherandbatter,thecurrentgamesituation(inning,count, numberofouts,currentscore,etc.),thetypeofpitchthrown,andtheoutcomeofthe pitch(e.g.,homerun,foul,sacrificebunt). Therearemultipledimensionsalongwhichpitchesvary:thetypeofpitch(e.g., changeup,slider,fastball),thelocationofthepitch,thevelocity,etc.welimitour attentiontojustoneofthesedimensions:pitchtype. 3 Therawdatacontain12different typesofpitches.afterconsultationwithmajorleagueteams,weconsolidatedtheseinto fivecategories:fastball,curveball,slider,changeup,andother. 4 Inouranalysis,wedrop allpitchesclassifiedas other. 5 Ourprimaryoutcomemeasureforanatbatisthebaseballstatisticknownas OPS, whichisthesumofabatter son-basepercentageandhissluggingpercentage.in 3 Unlikevelocityorlocation,pitchtypeisnotaffectedbyfaultyexecutiononthepartofthepitcher;a pitchermightintendtolocateapitchovertheinsidecorneroftheplate,butmistakenlythrowittothe outsideoftheplate. 4 Thesepitchtypes,withnumberofcasesinparentheses,areasfollows:Fastballs(2,083,248)andcut fastballs(39,830)arecombinedas fastball. Changeups(362,387)andsplitfingers(50,818)are combinedas changeup. Forkballs(430),knuckleballs(18,905),pitchouts(4,379),screwballs(766), sinkers(84),andunknowns(188,927)arecombinedas other. Sliders(449,378)andcurveballs(314,633) composetherestofthedata. 5 Theaccuracywithwhichpitchtypeiscodediscriticaltoourstudy.Asacheckonthisissue,weobtained thecodingforanoverlappingsampleofpitchescollectedbystatsinc.,whichcompeteswithbaseball InfoSolutionsinprovidinginformationtobaseballteams.Thesetwoindependentassessmentsofpitch typematchonover90percentofallpitches.thecodingmatchesespeciallywellonfastballs,withmore variationoccurringwhenthetwodatasetscodeanoff-speedpitchdifferently.ifwelimitourcomparison tofastballversusnon-fastball,approximately94percentofallpitchtypesmatchacrossthetwodatasets. Importantly,thedegreetowhichthetwodatasetsmatchdoesnotappeartobeafunctionoftheoutcomeof theatbat.matchratesarenearlyidenticalregardlessofwhetherthepitchisnotputinplay,isputinplay forahit,orisputinplayforanout.
7 priorempiricalresearch,opshasbeenshowntobeastrongpredictorofthenumberof runsateamscores(fox2006).ifabattermakesanout,hisopsforthatatbatiszero.if thebatterwalks,hisopsisone.asingleearnsanopsoftwo,adoubleanopsof three,atripleanopsoffour,andahomerunanopsoffive. Ourrawdatacoverseveryregularseasonpitchthrowninthemajorleaguesover theperiod aftergeneratingseasonlevelstatisticssuchasbatterops,we excludeanyatbatincludinganypitchcategorizedas other, aswellasdatafromextra innings.aftertheseexclusions,wehave3,110,429totalpitchesthrown.table1 presentssummarystatisticsforthesepitches.asshownincolumn1,fastballsarethe mostcommontypeofpitch,accountingforapproximately65percentofallthrows. Slidersarethesecondmostcommonpitchtype,followedbychangeupsandcurveballs. Columns2-5ofTable1reportthedistributionofoutcomesforeachpitchtype. Wereportfourmutuallyexclusiveandexhaustivepitchoutcomes:aball,astrike,theball isputintoplayandthebatterisout,andtheballisputintoplayandthebattergetsahit. 6 Pitchersareslightlymorelikelytorecordstrikeswhenthrowingfastballsrelativetoall non-fastballs,andslightlylesslikelytoregisteraball.changeupsaremostlikelytolead tobothanout(15.29percent)andahit(7.08percent);curveballsareleastlikelytoyield bothoutsandhits.column6oftable1showstheops(ourpreferredoutcomemetric) bypitchtypewhenthepitchendstheatbat.foreshadowingtheresultsfromthe regressionanalysis,theopsonfastballsishigherthanfornon-fastballs:.753versus.620.onepotentialexplanationforthatgap,however,isthatfastballsaremorelikelyto bethrowninhitters counts,asdemonstratedinthefinalthreecolumnsoftable1. 6 Afoulballthatisnotcaughtforanoutisclassifiedasastrikeinthiscategorization.
8 Tofurtherexploretheroleofthecount,Table2reportsresultsforpitchesthrown oneachpossiblecount,e.g.,1-0,3-1,etc.ascolumn3demonstrates,thelikelihoodofa fastballvarieswidelyacrosscounts.ona3-0count,almost95percentofallpitchesare fastballs;whenthecountis1-2theshareoffastballsisonly52percent.columns3-6 showopscomparisonsforfastballsandnon-fastballsbycount,forpitchesthatendtheat bat.thedifferencesinoutcomesforfastballsversusnon-fastballstendtobesmallwhen therearefewerthantwostrikes.ontwostrikecounts,however,non-fastballsgenerate anopsthatismorethan100pointslowerthanforfastballs,andthisgapishighly statisticallysignificant.thelastfourcolumnsoftable2reportthefinaloutcomeofthe at-batasafunctionofwhichpitchwasthrownateachcount,whenthatpitchdoesnot actuallyendtheat-bat.iftherearenospilloversacrosspitches,thereshouldbeno differenceinoutcomesacrosspitchtypesifthepitchdoesnotendtheatbat.tothe extent,however,thatfastballsareslightlymorelikelytogeneratestrikesthannon- fastballs,throwingafastballmayprovidesomebenefittothepitcherwhentheat-bat doesnotendwiththecurrentpitch. 7 TheresultsinthelastfourcolumnsofTable2 suggest,however,that,ifanything,throwingafastballonthecurrentpitchleadsto slightlyworseoutcomeswithinthisat-batifthepitchdoesnotterminatetheat-bat.for mostcounts,theeventualat-batopsiscloseforfastballsandnon-fastballs,butwithtwo strikesthenon-fastballsyieldlowerops. 7 Thereareotherchannels,aswell,viawhichafastballmightprovidedeferredbenefits.First,itmaybe thatitishardertohitapitchiftheprecedingpitchwasafastball.second,fastballsmightbelesslikelyto generateothernegativeresults,likewildpitches,passedballs,andstolenbases.third,fastballsmight causelesswearandtearonthepitcher sarm.backoftheenvelopecalculationssuggestthatnoneofthese channelsislikelytobeevenclosetoamagnitudetooffsettheobservedopsdifferencesbetweenfastballs andotherpitches.
9 Table3analyzesmoreformallythelinkbetweenpitchtypeandOPSusing regressionspecificationsofthefollowingform: (1) OPS = β Pitchtype + X ' Γ + λ + θ + ε apb k apb apb p b apb wherea,p,b,andkindexat-bats,pitchers,batters,andpitchtypesrespectively.opsis ourmeasureofhowsuccessfulthebatterisintheat-bat.pitchtypedenoteswhetherthe pitchthatendstheat-batisafastball,curveball,slider,orchangeup.alsoincludedinthe regressionisasetofcovariatesxthatincludesindicatorsforthecountpriortothefinal pitchoftheat-bat,theinningofthegame,thenumberofouts,andthenumberofrunners onbase.insomespecificationspitcherandbatterfixed-effectsareincluded,pitcher- batterinteractions,andinourmostfullysaturatedmodels,pitcher*batter*count interactions.intheseregressions,welimitthesampletopitchesthatendtheat-bat. 8 Column1ofTable3includespitchtype,butnoothercontrols.Changeupisthe omittedpitchcategory,soallcoefficientsshouldbeinterpretedasrelativetotheoutcome ifachangeupisthrown.withnocovariatesatall,asalreadynotedinthesummary statistics,theoutcomeswhenfastballsarethrownarequitebadforpitchers:anopsgap of.094(standarderror=.004)relativetochangeups.curveballsandslidershavethebest pitcheroutcomes.asdemonstratedincolumn(2),however,asubstantialfractionofthe gapacrosspitchtypeiseliminatedwiththeinclusionofcount-fixedeffects.after controllingforcount,thegapbetweenfastballsandchange-upsfallsto.041(se=.004). Slidersdoslightlybetterthanchangeups,curveballsslightlyworse. Column3ofTable3addsarangeofcontrolscorrespondingtothegamesituation: theinning,numberofouts,andnumberofrunnersonbase.includingthesecovariates 8 Wehavealsorunthesespecificationsforpitchesthatdonotendtheat-bat.Theresultsforspecifications matchingcolumns1,2,and3showchangeupsunder-performingallotherpitchesbyasmall,but significantamountwhilethosematchingcolumns4and5intable3aresmallandinsignificant.
10 haslittleimpactonthecoefficientsonpitchtype.opsislowestintheninthinningand highestwithnooutsandwiththebasesloaded.thelikelyexplanationforloweropsin theninthinningisthatonaveragethequalityofthepitcherishigherbecausespecialist closers arebroughtinduringthefinalinningsofclosegames.theinclusionofpitcher- batterinteractionsconfirmsthisintuitionincolumn(4).inthisspecification,itisthe earlyinningsinwhichopsislow.controllingforpitcher-batterinteractionsincreases theopsgapbetweenfastballsandotherpitches,whichimplieseitherthatbetterthan averagepitcherstendtothrowmorefastballs,orthatbetterthanaveragehittersseefewer fastballsthanotherhitters.column(5)addspitcher*batter*countinteractions.thus,the identificationincolumn(5)comesonlyfromcaseswherethesamepitcherandbatterare facingeachother,withthesamecount,andinoneinstancethepitcherthrowsaparticular pitch,andonanothersuchoccasion,adifferenttypeofpitch.addingthesethree-way interactionshaslittleimpactonthecoefficients. TheOPSgapsonfastballsinTable3aresubstantialinmagnitude.Fox(2006) estimatesthateach.001pointofopsoverthecourseofaseasontranslatesinto2.16 additionalruns.ifapitchingstaffwereabletoreducetheshareoffastballsthrownby10 percentagepointswhilemaintainingtheobservedopsgaponfastballs,thiswould reducethenumberofrunsallowedbyroughly15perseason,ortwopercentofateam s totalrunsallowed.becauseofbehavioralresponsesbybatters,thisislikelytobean upperboundonthecostofteamsthrowingtoomanyfastballs. Table4exploresthesensitivityofthecoefficientonpitchtypestoavarietyof subsetsofthedata,usingthespecificationreportedincolumn5oftable3asabaseline. Thethreecolumnsofthetablecorrespondtotheestimateforfastballs,curveballs,and
11 slidersrespectively,inallcasesrelativetotheomittedcategory,whichischangeups. Eachrowofthetablerepresentsestimatesfromoneregression;onlythecoefficientson thepitchtypevariablesarepresentedinthetable. Thetoprowofthetableshowsthebaselineestimatesfortheentiresample.The nextthreerowsoftable4dividethesampleaccordingtowhetheritisahitter scount,a neutralcount,orapitcher scount. 9 Interestingly,oncewecontrolforotherfactors, fastballsnotonlydoworseonpitcher scounts(aswasapparentinthelargeopsgapin therawdatafortwo-strikecounts),butonneutralandhitter scountsaswell.across thesethreeclassifications,thecoefficientonfastballrangesfrom.064to.087. Thenextthreerowsdividethesampleofpitcherswithatleast200plate appearancesagainstthemintothreeequal-sizedcategoriesaccordingtotheirops. Good pitchershavethelowestthirdofopsagainst,and bad pitchershavethe highestopsagainst.theopsgapassociatedwithfastballsissmallestforthegood pitchers.defininggoodandbadhittersinaparallelfashion,wefindthattheopsgapfor fastballsispresentonlyforgoodandmediumhitters.badhittersdobestwithchangeups andworstwithcurveballs. Thepitcherswhothrowthefewestfastballsgenerallydoworsewithfastballs thanpitcherswhothrowmorefastballs.fastballsdobestwhentherearerunnersin scoringposition;inthatcircumstance,fastballshaveworseoutcomesthanchange-ups, butsimilaroutcomestocurveballsandsliders.thereislittlesystematicdifferenceinthe coefficientonfastballasafunctionofthenumberofouts. 9 Specifically,wedefinehitter scountas1-0,2-0,3-0,and3-1counts,neutralcountas0-0,1-1,2-1,and3-2counts,andpitcher scountas0-1,0-2,1-2,and2-2counts.
12 Serialcorrelationinpitchchoice Minimaxtheorypredictsthatequilibriumactionswillbeseriallyuncorrelated.In thecontextofbaseball,testingthispredictioniscomplicatedbythefactthatthepayoff matrixchangesbothacrossat-bats,andevenwithinanat-bat.thepayofftoafastball,at leastaccordingtothechoicespitchersactuallymake,ishigherwitha3-0countthanan 0-2count.Theempiricalchallengeistoconvincinglycontrolfortheheterogeneityin payoffs,knowingthatthesepayoffsarepotentiallyafunctionofmanyvariablesthatare notinourdataset(e.g.,howfatiguedthepitcheris,whichwaythewindisblowing,etc.) Consistentwithourestimationstrategyabove,onemeansofcontrollingfor unobservablesistoincludepitcher*batter*countinteractions.insuchaspecification,the identifyingvariationcomesonlyfrominstanceswhenthesamepitcherandbatterreach thesamecountonmultipleoccasions,butthepitcherchoosestothrowdifferentpitches. Eventhisstrategy,however,issubjecttocriticismwhentryingtomeasureserial correlation:ifunobservablefactorsledthepitchertochooseafastballontheprevious pitchofthisat-bat,perhapsthosesamefactorsarealsorelevantwhenchoosingthenext pitchtothrow.onadaywhenapitcherhashiscurveballworkingeffectively,hewill tendtothrowmorecurveballs. Toaddressthispotentialcriticism,weconditionnotonlyonpitcher*batter*count, butalsoonthenumberofpitchesofeachpitchtypethathavebeenthrownthusfarinthe at-bat.thus,ouridentificationcomesonlyfromcaseswherethesamepitcherandbatter meetonmultipleoccasions,reachthesamecount,andprogressthroughtheexactsame numberoffastballs,curveballs,changeups,andslidersinreachingthatcount,butthe orderinwhichthosepitcheswerethrowndiffers.minimaxtheorywouldpredictthatfor
13 thesamebatterandpitcher,ifthecountis2-1,andthusfarintheat-battherehavebeen twofastballsandoneslider,itshouldnotmatterwhethertheslidercameonthefirst, second,orthirdpitchoftheat-bat. Formally,theregressionspecificationweestimatetakestheform: (2) Pitchtypeabpt = β Pitchtypeabpt 1 + X abpγ + δbpcn f ncu ns n + ε ch abpt Wherea,b,p,andtindexanat-bat,pitcher,batter,andthenumberofthepitchwithinan at-batrespectively.pitchtype,asbefore,correspondstowhetherthepitchisafastball, curveball,etc.thecontrolvariablesxincludethepercentofpitchesbypitchtypethat havebeenthrowntothisbatteronthiscountuptothispointduringtheseason(excluding thisobservation),thesamevariabledefinedforthepitcher,andtheshareofeachpitch typethrownbythepitcherthusfarinthisgame.theδ termrepresentsa pitcher*batter*count*numberofpitchesofeachpitchtypethrownthusfarintheat-bat, withcreflectingthecountandn f capturingthenumberoffastballsthrownthusfarinthe at-batandsimilarlywiththeothersubscripts. Table5reportsourestimatesofserialcorrelationusingvariationsonequation(2). Thedependentvariableineachregressionislistedatthetopofthecolumn.Incolumns 1-4,alongwiththeinteractionsandcontrols,weincludeanindicatorvariableequalto oneiftheprecedingpitchisthesameasthedependentvariable.thesespecifications measurewhether,conditionalonthecontrolsdescribedabove(e.g.,thecountandthe numberofpitchesbytypeinthisatbat),knowingthepitcherthrewaparticularpitchon thelastpitchhelpspredictwhetherhewillthrowitasthecurrentpitch.forthreeofthe fourpitchtypes,weobservestatisticallysignificantnegativeserialcorrelation.the largestcoefficientisforfastballs.ifthepitcherthrewafastballonthelastpitch,allelse
14 equal,itlowersthelikelihoodthispitchwillbeafastballby4.1percentagepoints.in relativeterms,thenegativeserialcorrelationforslidersisgreater,sinceslidersrepresent onlyabout10percentofallpitches.ifthelastpitchwasaslider,thelikelihoodthatthis currentpitchisasliderfallsbytwopercentagepoints,ortwentypercent.thenegative serialcorrelationisroughlyhalfaslargeforcurveballs,andnotpresentforchangeups. Columns5-8ofTable5addtheindicatorsforonce-laggedvaluesofeachpitch type,whichallowsustolearnnotjustwhetherpitchersrepeatthesamepitchmoreorless thanwouldbeexpected,butalsowhetherothertransitionalsequencesfrompitchtopitch appearmoreorlessfrequentlythanpredictedbytheory.ineachofthesecolumns,oneof thelaggedpitchtypesisomitted,andallresultsarerelativetothatomittedcategory.the resultsthatemergeincolumns5-8demonstratethatthereisgreaternuanceassociated withtheorderingofpitchesthansimplythenegativeserialcorrelationobservedinthe firstfourcolumns.forinstance,incolumn5,notonlyisitthecasethatfastballsfollow fastballslessthanwouldbeexpected,butalso,fastballsaremorelikelytooccurafter changeupsthanafterothernon-fastballs.incontrast,curveballsareleastlikelytofollow changeups(andviceversa),andcurveballsaremostlikelytofollowfastballs. Changeupsaremorelikelytooccurifthelastpitchwasachangeupthanitwasanother non-fastball. Calibratingthevaluetoateam sbattersofexploitingthesecorrelationpatterns requiresmakingassumptionsastohowvaluableitistoamajorleaguehittertoknow whattypeofpitchiscoming.executivesofmajorleaguebaseballteamswithwhomwe spokeestimatedthattherewouldbea.150gapinopsbetweenabatterwhoknewfora certainafastballwascomingversusthatsamebatterwhomistakenlythoughtthatthere
15 wasa100percentchangethenextpitchwouldnotbeafastball,butinfactwassurprised andfacedafastball.ifonemakesthefurtherassumptionthattheopsgapislinearina hitter sexpectationsaboutwhattypeofpitchwillbecoming,thenknowingthatafastball is4.1percentagepointslesslikelyifthelastpitchwasafastball(andconverselymore likelyifthelastpitchwasnotafastball)isworthroughly.006opspointstoabatter. Thus,thepotentialbenefitfromexploitingthepatternsofserialcorrelationisthesame magnitudeasidentifiedearlierfrompitchersthrowingtoomanyfastballs about10-15 runsperyear. SectionII:AnanalysisofplayselectionintheNationalFootballLeague OurdataonplaychoiceintheNationalFootballLeaguewascompiledby STATS,Inc.STATS,Inc.maintainsanetworkofreporterstrackingeverysnapindetail toprovideexclusiveinformationfromtheirproprietarydatabasetothenflandother clients.thedatasetincludesextensiveinformationforeachplayinthenfloverthe period :dateofgame,offensiveteam,defensiveteam,generalgame description(e.g.,stadium,weather,etc.),currentgamesituation(quarter,locationon field,down,yardstogo,etc.),offensiveformation,thetypeofplay(run,pass,punt,field goal,etc.),andtheoutcomeoftheplay(e.g.,yardsgained). Aswasthecasewithbaseball,therearemanydimensionsonwhichplaytypes vary:runorpass,direction,distance,movementofplayers,etc.welimitourstudyto justonedimension:thechoiceofwhethertocallarunningplayorapassingplay. OurrawdatacoverseveryplayfromregularseasonNFLgamesoverfivefull seasons: weexcludefourthdownplays,aswellasallplaysthatoccurinthe
16 lasttwominutesofthehalf,duringovertime,orwhenateamkneelsdowntorunoutthe clock.becauseofdifficultiesinourdataofidentifyingwhethertheteam sintentionwas torunortopassonplayswherethequarterbackruns,andonpenaltiescalledbeforea playunfolds(e.g.,falsestart),theseplaysarealsoexcluded.aftertheseexclusions,we have127,885totalplays. Unlikebaseball,wheretherearewell-establishedsummarymetricsfor evaluatingthesuccessofanatbat(e.g.ops),thereisnoparallelstatisticinfootball. Consequently,weconstructourownmeasureofsuccessforaplayinfootballasfollows. First,weestimatethevaluetoateamofhavingpossessionoftheballasafunctionof distancefromtheendzone,whatdownitis,andyardstoachieveafirstdown,usinga regressiontakingtheform (3) Y = f (down, yardstofirstdown,distancetogoal) wheretheoutcomevariableyisthechangeinthegamescorebetweenthecurrenttime andtheendofthehalf.weallowforaflexiblefunctionformwithrespecttotheright- hand-sidevariables,includingfullyinteractedquinticsofeachofthevariables.the valuesgeneratedfromequation(3)appearsensible.forinstance,thethreelinesin Figure1showtheestimatedvaluetoateamofhavingtheballfirstdownandtenyardsto go,seconddownandtenyardstogo,andthirddownandtenyardstogorespectivelyasa functionofthedistancetotheendzone. 10 Ifateamhastheballfirstandtenatthe opponent stenyardline,thatteamwillexpecttogainmorethanfourpointsrelativeto theotherteambytheendofthehalf.thevalueofhavingtheballfirstandtendeclines 10 OneothermeasureofperformanceinNFLfootballisNetExpectedScoring(NES),developedbyCitizen SportsNetwork.NESisnotasflexibleasoursuccessmetric,butishighlycorrelated(ρ= )withour successmetric.
17 nearlylinearlywithfieldposition;havingtheballfirstandtenonone sowntenyardline isassociatedwithessentiallynoexpectedchangeinthehalf-timescore.havingtheball secondandtencostsateamaboutone-halfapointrelativetohavingtheballfirstandten fromthesamefieldposition.movingfromsecondandtentothirdandtenisevenmore costlyforateam. Tocomputehowsuccessfulaparticularplayis,wecalculatethechangein expectedpointsscoredbeforeandaftertheplay(e.g.,lookingatfigure1,ifateamgains 20yardswhenitisfirstandtenfromitsown20yardline,expectedpointsscoredjump byroughlyone)andsubtracttheaveragechangeinexpectedpointsforallplaysinthe datasetthatbeganatthesamedown,distance,andyardstothegoal. 11 Theresulting statistic,whichwecallour successmetric, ismeanzero.thesuccessmetrictellsus,in unitsofexpectedpointsscored,howmuchthisplayexceedsorunderperformsthe averageplayrunonthisdown,distance,andyardstogoal. Inadditiontothisconstructedmeasureofsuccess,wealsoreportresultsformore traditional,buthighlyimperfectoutcomemeasures:yardsgained, 12 whetherafirstdown ismade,whetherpointsarescored,andwhetheraturnoveroccurs. Table6reportssummarystatisticsforthefootballdataset.Column1shows outcomesforallplays;columns2and3dividethesampleintorunningplaysandpassing playsrespectively.column4reportsthet-statisticofthecomparisonofmeansbetween runningandpassingplays.overall,runsrepresent44percentofthecombinedpassesand runs.forallplaysoursuccessmetric,bydefinition,hasameanvalueofzero,sinceitis 11 Whentheplayresultsinpointsbeingscored,thosepointsareincludedinourcalculationofthemetric. 12 Weadjustedyardsgainedperplaytocapturecertaincircumstances:penaltyyardageforpenalties occurringwithinplayswereincorporated,touchdownswithin10yardsofendzonewerecreditedwitha full10yardsgained,andinterceptionswereadjustedto-45yardsgained.
18 definedasthedeviationfromtheexpectedoutcomeonaplay.note,however, comparingthetoprowofcolumns2and3,thatpassingplayssystematicallyoutperform runningplays.themeangapbetweenthetwotypesofplaysisroughly.066,implying thatonaverage,apassingplaygenerates.066morepointsthanarun.thisdifferencein meansishighlystatisticallysignificant.consistentwiththisresult,passesonaverage yieldanextra.55yardsgained,andareninepercentagepointsmorelikelytoyieldafirst down.passesdo,however,producemoreturnovers,andthushavehighervariance.runs resultinscoringplays2.8percentofthetime;passesleadtoscoreswitha3.8percent probability. Tofurtheranalyzethedifferenceacrossrunsandpasses,weestimateregressions oftheform: (4) Outcome pij = α + βpass + X Γ + λ + ε pij pij ij pij wherep,i,andjindexaparticularplay,offensiveteam,anddefendingteamrespectively. Outcomeisourmeasureofsuccessforanoffensiveplay.Passisanindicatorvariable equaltooneiftheteamcallsapassingplay,andzeroiftheplayisdesignedtobearun. Xisavectorofcontrols,suchasthescoredifferentialatthetimeoftheplay,whetherthe gameisplayedongrass,whethertheoffensiveteamisthehometeam,theyearofthe game,etc.insomespecifications,wealsoaddteam-fixedeffectsfortheoffenseandthe defense,orinteractionsbetweentheoffensiveanddefensiveteams. Theresultsfromestimatingequation(4)arepresentedinTable7.Thefour columnsrepresentfourdifferentspecifications,withthenumberofcontrolsincreasing
19 movingfromlefttorightinthetable.thefirstcolumn,whichincludesnocovariates, confirmstherawdifferenceinmeansbetweenpassingandrunning;apassgeneratesan additional.066pointsinexpectation.column2addscontrols,butdoesnotincludeteam- fixedeffects.therelativevalueofapassincreasesto.083pointsinthisspecification. ThisspecificationalsohighlightsthesizablehomefieldadvantageintheNFL:an offensiveplayrunbythehometeamgenerateanextra.041points,orroughlyhalfthe differencebetweenapassandarun.offensesperformslightlyworseinthecold.the playingsurfacedoesnothavealargeimpactontheoffense seffectiveness. Column3addsfixed-effectsfortheoffensiveanddefensiveteams.Thesefixed effectswillabsorbanysystematicdifferencesacrossteamsinoffensiveanddefensive prowess.therelativevalueofapassdecreasesslightlyto.077pointsinthis specification.thelastcolumnofthetableincludesinteractioneffectsfortheoffensive anddefensiveteams.therelativevalueofapassisessentiallyunchangedat.075points. Table8examinesthesensitivityoftheresultsofrunningversuspassingtoa varietyofsubsetsofthedata,aswellasreportingresultsforanexpandedsetofoutcome measures(yardsgained,achievingafirstdownontheplay,turnovers,andwhethera touchdownisscoredontheplay).thecolumnsofthetablecorrespondtodifferent outcomemeasures,e.g.ourconstructedsuccessmetric,yardsgained,etc.eachrowof thetablerepresentsadifferentsubsetofthedata.inallcases,weincludeteam-fixed effectsandcontrolsmirroringthoseincolumn4oftable7.onlythecoefficientonthe passindicatorvariableisreportedinthetable. Focusingfirstonthecolumn1ofTable8,thetoprowofthetablereportsour baselinespecification.thus,theentryinthetoprowincolumn1matchesthecoefficient
20 wereportintable6,column4:a.075gapbetweenpassesandrunsonoursuccess metric.consistentwiththisresult,passesdobetteronyardsgained,firstdownsmade, andscoring,butalsoleadtomoreturnovers.movingdownthroughthetable,passes outperformrunsinallquartersofthegame,butbyagreatermargininthefirsthalfthan thesecondhalfofthegame.thebenefitsofpassingaccrueequallytohometeamsand visitors.thebestoffensesexhibitthegreatestgapbetweenpassesandruns;fortheworst offensesthedifferentialisnotstatisticallysignificant.teamsthatpassthemosthave slightlysmalleredgeswhenpassingthanotherteams. TheresultsinTables7and8demonstratethattheexpectedoutcomeofapass systematicallyexceedsthatofarun aresultthatisinconsistentwithminimaxtheory. 13 Accordingtothetheory,defensesshouldadjusttobetterdefendthepass.Absentthat adjustment,offensesshouldbepassingmoreoften.themagnitudeofthedeviationsin payoffsthatareobservedaresubstantial.thetypicaloffenserunsabout60playsagame, 56percentofwhicharepasses.Iftheoffensecouldincreaseitsshareofpassesto70 percentwithoutinducinganoffsettingresponseonthepartofthedefense,itwould generateanadditional0.63pointspergameinexpectation(8.4additionalpasses*0.075 expectedpoints),oranextratenpointsoverthecourseofaseason,orroughly3percent ofateam stotalscoring.becausedefensesarelikelytorespond,thatestimateislikely anupperboundonhowmuchanoffensecouldgainbyexploitingthedeviationsfrom minimaxplaythatarepresentinthedata. 13 Alamar(2006)showspassingplaysashavinganoutcomeadvantageofnearly1.8yardsperplay (adjustedyards)forthe2005nflseason.however,thesourceofthatdata-- reference.com/years/2005/--showsadifferenceof0.5yardswhenconsideringbothsacksandinterceptions aspassingplays( AdjustedNetYardsgainedperpassattempt ),whichisconsistentwithourfindings.
21 SerialcorrelationinNFLplaycalling Toassesswhetherthereisserialcorrelationinthechoiceofrunsversuspasseson thepartofnfloffenses,werunregressionsoftheform (5) Pass pij = α + βpass 1 + X Γ + ε p ij pij pij wherepassisanindicatorforwhethertheplaycalledwasintendedtobeapass.the coefficientβcapturesthedegreeofserialcorrelationinplaycalling.includedinthe vectorofcontrolsarethesamesetofcovariatesintheearlierfootballanalysis,alongwith threeadditionalvariables:thepercentageoftimethattheoffensiveteampassedoverthe courseoftheentireseason,thepercentageoftimethatthedefensiveteamwaspassed againstoverthecourseoftheentireseason,andtheshareofpassesbytheoffenseinthis game,upuntilthetimethisplayiscalled.becausewecontrolfordownanddistancein theseregressions,aswellasanoffense soveralltendenciestowardspassingversus running,ouridentificationcomesfromacomparisonof,forexample,whetheron2 nd downand10,apassismorelikelyifthepreviousplaywasarunthatwentfornogain,or thepreviousplaywasanincompletepass.minimaxplaypredictsnoserialcorrelation, implyingazerocoefficientonwhetherornotthelastplaywasapass. Thebasicestimationresultscorrespondingtoequation5arepresentedincolumn 1ofTable9. 14 Offensiveplaycallingrevealssubstantialnegativeserialcorrelation,with acoefficientof-.100(se=.003).inotherwords,conditionalonotherfactors,ateamis 14 ThenumberofobservationsinTable9issmallerthanintheearlieranalysisfortworeasons.First,the firstplayofeachdriveisnotincludedintheserialcorrelationanalysis.second,playsforwhichthe precedingplaycouldnotbecatalogedasarunorapass(e.g.becauseofapenaltyoraquarterbackrun)are alsoexcluded.
22 almost10percentagepointslesslikelytopassonthiscurrentplayiftheypassedonthe previousplay. Tofurtherexplorethequestionofserialcorrelation,wedividethesampleinto thirdsaccordingtohowsuccessfulthepreviousplaywas,withsuccessdefinedbyour constructedsuccessmetrictheresultsforthesethreesubsetsofthedata(i.e.previous playwasinthebottom-third/middle-third/upper-thirdsuccess-wise)areshownin columns2through4oftable9.negativeserialcorrelationismostpronouncedwhenthe precedingplaywasunsuccessful.experiencingapoorresultonthelastplayincreases thelikelihoodtheteamwillswitchfromaruntoapassorvice-versaby14.5percentage points.incontrast,whenthelastplayisintheupper-thirdofsuccessfuloutcomes,the tendencytoswitchawayfromthatplaytypeisgreatlymitigated(serialcorrelation coefficientofonly-.025). Thesecoefficientsimplytheopportunityfornon-trivialgainsforteamsthat successfullyexploitserialcorrelationonthepartofopposingoffenses.assume,for instance,thatifadefenseknewwith100percentcertaintywhetheraplaywouldbearun orapass,itcouldcuttheaverageyardagegainedinhalfbyadjustingdefensivepersonnel orpositioning.assume,aswell,thatifthedefensewas100percentcertainapasswas coming,butinsteadtheoffenserantheball,theexpectedyardagegainedwouldbe50 percentgreaterthantheaverage,andsimilarlyifthedefenseexpectedarunandthe offensepassed. 15 Finally,letusassumethatthevalueofknowingwhatplayiscomingis linearintheprobabilities,i.e.goingfrom50percentlikelihoodofarunto60percent likelihoodyieldsone-tenthofthebenefitofgoingfromzeropercentto100percent.take 15 BasedondiscussionswithNFLteams,theseassumptionsarelikelytobeconservative,understatingthe potentialvaluetodefensesofexploitingserialcorrelatedoffensiveplay.
23 thecasewhereabsentserialcorrelation,adefenseexpectsanequalmixofrunsand passes,witheachtypeofplaygaining4.5yardsonaverage.withserialcorrelation, however,thetruemixofplaysafterapasswillberoughly60percentrunsand40percent passes.undertheassumptionsabove,ifthedefenseadjuststothisinformation,the averagerunningplaywillyield3.6yardsandtheaveragepassingplay5.4yards,yielding anoverallaveragegainfortheoffenseof4.32yards.18yardslessthanifthedefense ignorestheserialcorrelation.thereareroughly60offensiveplaycallspergamethatare precededbyanotheroffensiveplay.iftheaveragereductioninyardsgainedperplayis.18yards,thenthisamountstoanoverallreductionof10.8yardspergame,which translatesintoroughly1pointpergame.onepointpergameisworthapproximatelya halfvictoryperyear considerablegiventhenflregularseasonincludesjustsixteen games. 16 Thepotentialbenefitfromexploitingthepatternsofserialcorrelationin footballisslightlylargerthanthebenefitfromcallingfewerrunningplaysanalyzed earlier. 16 Webasethiscalculationonthechangeinexpectedwinningpercentagebyscoring16additionalseason pointsasestimatedbythepythagoreanwinningpercentage[expectedwinningpercentage=(points scored^2.64)/(pointsscored^2.64+pointsallowed^2.64)].takinganaveragenflteamwith350season pointsscoredand350pointsallowed,increasingtheirpointsscoredto366increasestheirexpected winningpercentagefrom.500to.529.
24 SectionIV:Conclusion Inthispaper,weutilizetwoenormousdatasetsgeneratedbyprofessionalsina highstakesenvironmenttoprovidethemostpowerfultesttodateofminimaxbehaviorin anaturalsetting.incontrasttomostpriorstudiesusingfielddata,wefindsubstantial deviationsfromminimaxbehavior,bothwithrespecttoequalizingpayoffsandserially correlatedactions.thesedeviationsarenotenormousinmagnitude meaningthatthey mightplausiblynothavebeendetectedinthesmallerdatasetsthathavebeenavailablein mostpriorfieldresearchonthetopic butarelargeenoughthatateamthatsuccessfully exploitedthesepatternscouldaddoneortwoseasonwinsandmillionsofdollarsin associatedrevenue. OurfindingsreinforcetheresultsofRomer(2006),Levitt(2006),andPopeand Schweitzer(2009)indemonstratingthathighstakesalonearenotsufficienttoensurethat optimaldecision-makingwillensue,evenamongprofessionalsoperatingintheirnatural environments.
25 Figure1:ValueofPossessingtheBallbyDownandDistance ExpectedPoints FirstDown SecondDown ThirdDown YardsfromGoal
26 Table 1: Major League Baseball Summary Statistics by Pitch Type DistributionofOutcomes PercentThrownIn OPSifAB PitchType Numberof observations Ball Strike/ foul InPlay Out InPlay Hit Endson thispitch Hitter's Counts Neutral Counts Pitcher's Counts Fastball % 43.37% 13.43% 6.79% % 66.49% 55.52% AllNon-Fastball % 42.60% 13.09% 6.24% % 33.51% 44.48% Change-Up % 40.51% 15.29% 7.08% % 11.84% 14.16% Slider % 44.22% 12.14% 5.83% % 12.58% 17.38% Curveball % 43.07% 11.54% 5.70% % 9.10% 12.94% Notes:Datacoverpitchesfrom Pitchtypesbasedonclassificationsbythedataprovider,BaseballInfoSolutions,withsomeaggregationofcategoriesbythe authors.columns2-5reporttheoutcomeonthepitchinquestion.opsreferstothestatisticon-basepercentageplussluggingpercentage.hitter scountsaredefinedas 1-0,2-0,3-0,and3-1counts;neutralcountsare0-0,1-1,2-1,and3-2counts,pitcher scountsare0-1,0-2,1-2,and2-2counts.seethetextfordetailsonsample exclusions.
27 Table 2: Outcomes for Fastballs versus Non-Fastballs by Count OPSifpitchendsat-bat Number Count ofpitches %Fastballs All Fastball Non- Fastball P-value All Fastball OPSofat-batifthispitch doesnotendtheat-bat Non- Fastball P-value % % % % % % % % % % % % Strike Counts % Other Counts % Notes:DatacoverpitchesthrowninMajorLeagueBaseballbetween2002and2006.Countreferstothenumbersofballsandstrikespriortothepitchthatisthrown. ThemiddlefourcolumnsofthetablereportmeanOPS(onbaseplussluggingpercentage)forpitchesthatendtheat-bat,bypitchtype.Thep-valuecolumnreportsthe statisticalsignificanceofat-testoffastballsversusnon-fastballs.thelastfourcolumnsreporttheopsoftheat-batifthispitchdoesnotendtheat-bat.if,conditional ontheat-batnotending,throwingafastballonthispitchbenefitsthepitcherlateronintheat-bat,thentheopsonfastballinthethird-to-lastcolumnshouldbelessthan theopsfornon-fastballsinthepenultimatecolumn.
28 Table 3: Regression Analysis of Outcomes of Fastballs vs. Other Pitches (1) (2) (3) (4) (5) Fastball *** *** *** *** *** (0.004) (0.004) (0.004) (0.005) (0.008) Slider *** ** * * (0.005) (0.005) (0.005) (0.006) (0.011) Curveball *** (0.006) (0.006) (0.006) (0.007) (0.011) 1stInning *** *** * (0.006) (0.009) (0.016) 2ndInning *** *** * (0.006) (0.009) (0.017) 3rdInning *** *** (0.006) (0.009) (0.016) 4thInning *** (0.006) (0.009) (0.016) 5thInning *** (0.006) (0.009) (0.017) 6thInning *** * (0.006) (0.009) (0.016) 7thInning *** (0.006) (0.009) (0.016) 8thInning * (0.006) (0.008) (0.016) 0Outs *** *** *** (0.003) (0.004) (0.006) 1Out *** *** *** (0.003) (0.004) (0.006) 0Runners (0.008) (0.010) (0.018) 1Runners (0.008) (0.010) (0.018) 2Runners (0.009) (0.011) (0.019) R² CountFEs No Yes Yes Yes Yes PitcherFEs No No No Yes Yes BatterFEs No No No Yes Yes PitcherxBatter No No No Yes Yes FEs PitcherxBatter No No No No Yes xcountfes Notes:ThedependentvariableistheOPSoftheat-bat.Onlypitchesthatendtheat-batareincludedintheanalysis. Standarderrorsareshowninparentheses.Theomittedpitchtypeischange-up,sothepitchtypecoefficientsare relativetochange-ups.thenumberofobservationsisequalto834,345inallcolumns. * p<0.05, ** p<0.01, *** p<0.001
29 Table 4: Sensitivity Analysis of the Pitch-type Coefficients Fastball Slider Curveball Baseline 0.073*** 0.024* (0.008) (0.011) (0.011) Hitter'sCount 0.087*** (0.025) (0.04) (0.05) NeutralCount 0.064*** (0.013) (0.018) (0.020) Pitcher'sCount 0.077*** (0.010) (0.013) (0.014) GoodPitcher 0.055*** (0.016) (0.020) (0.022) MediumPitcher 0.085*** (0.014) (0.020) (0.020) BadPitcher 0.085*** (0.017) (0.024) (0.025) GoodBatter 0.080*** 0.030* 0.027* (0.009) (0.012) (0.013) MediumBatter 0.051*** (0.015) (0.021) (0.022) BadBatter (0.063) (0.075) (0.077) MostFasballs 0.059*** 0.064** (0.016) (0.022 (0.023) MediumFastballs 0.040** (0.013) (0.018 (0.018) FewestFastballs 0.111*** 0.034* 0.061*** (0.012) (0.016 (0.018) RISP 0.030** 0.053*** (0.011) (0.014 (0.015) Manon1st 0.090*** (0.023) (0.030) (0.033) BasesEmpty 0.091*** (0.013) (0.017) (0.018) 2Outs 0.085*** (0.021) (0.027) (0.029) 1Out 0.072*** (0.020) (0.027) (0.029) 0Outs 0.064** (0.020) (0.027) (0.028) Notes:ThedependentvariableinallcasesistheOPSofanat-bat,forpitchesthatendtheat-bat.Valuesinthetable arethecoefficientsonthepitch-typeindicatorsfromspecificationsthatparallelthoseshownintable3,column5.in allcases,theomittedpitchtypeisachange-up,soallcoefficientsarerelativetochange-ups.eachrowofthetable reportstheresultsfromadifferentregression.standarderrorsareshowninparentheses.thetoprowreproducesthe resultsforthebaselinesampleintable3.theremainingrowsreportresultsforarangeofsubsetsofthedata. * p<0.05, ** p<0.01, *** p<0.001
30 Table 5: Serial Correlation in Pitch Type (1) (2) (3) (4) (5) (6) (7) (8) Fastball Curveball Changeup Slider Fastball Curveball Changeup Slider PreviousFastball *** *** *** *** *** (0.001) (0.002) (0.001) (0.001) (0.001) PreviousCurveball *** *** *** (0.001) (0.002) (0.001) (0.002) PreviousChangeup *** (0.001) (0.002) PreviousSlider *** *** *** *** *** (0.001) (0.002) (0.001) (0.002) (0.002) Observations R² Notes:Dependentvariableisanindicatorvariableequaltooneifthepitchthrownisnamedatthetopofthecolumn,andzerootherwise.Inallcasesthevaluesreported inthetablearethecoefficientonanindicatorvariablecorrespondingtowhetherthepreviouspitchintheat-batwasthepitchtypenamedintherightmostcolumn.each columnrepresentsadifferentregression.columns(1)-(4)includepitch-typesoneatatime;columns(5)-(8)includeallpitch-typessimultaneously.allspecifications includeinteractionsforpitcher*batter*count*numberofpitchesofeachpitchtypethrownthusfarintheat-bat,soidentificationcomesonlyfromcaseswherethesame pitcherandbatterhavereachedthesamecountwiththesamedistributionofpitchtypes,butindifferingordersofpitchtypesthrown.standarderrorsareshownin parentheses. * p<0.05, ** p<0.01, *** p<0.001
31 Table 6: Summary Statistics for NFL Football Allplays Runsonly Passesonly P-valueofruns versuspasses SuccessMetric (1.233) (0.903) (1.441) YardsGained (11.51) (7.636) (13.82) FirstDownMade (0.441) (0.407) (0.462) FumbleorInterception (0.180) (0.122) (0.214) ScoringPlay (0.181) (0.166) (0.192) FarfromGoal (0.478) (0.475) (0.480) MediumfromGoal (0.485) (0.480) (0.489) ClosetoGoal (0.442) (0.456) (0.430) (0.396) (0.393) (0.398) * (0.404) (0.398) (0.408) * (0.402) (0.405) (0.399) (0.402) (0.405) (0.400) (0.396) (0.398) (0.395) Temperature40orBelow (0.325) (0.330) (0.320) HomeTeam (0.500) (0.500) (0.500) Grass (0.480) (0.480) (0.481) Numberofobservations Notes:Theunitofobservationisanoffensiveplay.Dataincludesplaysfrom fortheNationalFootball League,excludingfourth-downplays,playsinthelasttwominutesofahalf,overtime,andquarterbackruns(whichwe cannotaccuratelycategorizeintermsofintentionsintorunsversuspasses).thevariable successmetric isour estimateofagivenplay scontributiontotheoffensiveteam sscorerelativetotheaverageplayfromthisdown, distance,andyardstogoal.thefinalcolumnofthetablereportsp-valuesfromat-testofequalityofmeansfor runningandpassingplays.standarddeviationsareshowninparentheses. * p<0.05, ** p<0.01, *** p<0.001
32 Table 7: Regression Estimates of the Determinants of an Offensive Play s Success (1) (2) (3) (4) Pass *** *** *** *** (0.007) (0.008) (0.008) (0.008) * * * (0.011) (0.011) (0.013) (0.011) (0.011) (0.013) ** ** *** (0.011) (0.011) (0.013) (0.011) (0.011) (0.012) Temperature40orbelow (0.011) (0.011) (0.014) HomeTeam *** *** *** (0.007) (0.007) (0.008) Grass ** (0.007) (0.009) (0.011) R² DownxDistance No Yes Yes Yes QuarterxScoreDifferential No Yes Yes Yes OffensiveTeamFEs No No Yes Yes DefensiveTeamFEs No No Yes Yes OffensiveTeamxDefensiveTeamFEs No No No Yes Notes:Thedependentvariableisour successmetric, whichisourbestestimateofthemarginalcontributionofthis offensiveplaytotheoutcomeofthegame,measuredinunitsofpointsscored.thesuccessmetricismeasuredrelative totheexpectedoutcomeforaplayatagivendown,distance,andyardstothegoal.thevariable pass isanindicator variableequaltooneiftheplayisdesignedtobeapassandzerootherwise.standarderrorsareshowninparentheses. Thenumberofobservationsisequalto127,885inallcolumns. * p<0.05, ** p<0.01, *** p<0.001
33 Table 8: Sensitivity Analysis of the Gap between Runs and Passes Coefficientonpass: SuccessMetric YardsGained FirstDown Made Turnover Scoring Baseline *** *** *** *** *** (0.008) (0.073) (0.003) (0.001) (0.001) 1stQuarter *** *** *** *** *** (0.014) (0.133) (0.005) (0.002) (0.002) 2ndQuarter *** *** *** *** *** (0.017) (0.159) (0.006) (0.002) (0.003) 3rdQuarter *** *** *** *** *** (0.015) (0.141) (0.005) (0.002) (0.002) 4thQuarter ** ** *** *** *** (0.018) (0.167) (0.006) (0.003) (0.003) Visitor *** *** *** *** *** (0.011) (0.105) (0.004) (0.002) (0.002) Home *** *** *** *** *** (0.011) (0.103) (0.004) (0.002) (0.002) Top1/3Offenses *** *** *** *** *** (0.013) (0.124) (0.005) (0.002) (0.002) Middle1/3Offenses *** *** *** *** *** (0.014) (0.129) (0.005) (0.002) (0.002) Bottom1/3Offenses * *** *** *** (0.014) (0.130) (0.004) (0.002) (0.002) PassingOffenses *** *** *** *** *** (0.014) (0.132) (0.005) (0.002) (0.002) BalancedOffenses *** *** *** *** *** (0.013) (0.127) (0.005) (0.002) (0.002) RunningOffenses *** *** *** *** *** (0.013) (0.124) (0.004) (0.002) (0.002) Notes:Dependentvariableislistedattheheadofeachcolumn.Eachentryinthetableisfromadifferentregression paralleltothespecificationshownincolumn(4)oftable7.controlsincludedintheregressionareinteractionsfor team*opponent,down*distance,quarter*scoredifferential,andindicatorsforgrassversusturf,temperaturebelow40 degrees,andhometeam.thetoprowofthetableincludesthewholesample.theotherrowsofthetabledividethe datasetintosubsamples.standarderrorsareshowninparentheses. * p<0.05, ** p<0.01, *** p<0.001
Gasoline Empirical Analysis: Competition Bureau March 2005
Gasoline Empirical Analysis: Update of Four Elements of the January 2001 Conference Board study: "The Final Fifteen Feet of Hose: The Canadian Gasoline Industry in the Year 2000" Competition Bureau March
More informationRelationships Among Wine Prices, Ratings, Advertising, and Production: Examining a Giffen Good
Relationships Among Wine Prices, Ratings, Advertising, and Production: Examining a Giffen Good Carol Miu Massachusetts Institute of Technology Abstract It has become increasingly popular for statistics
More informationFair Trade and Free Entry: Can a Disequilibrium Market Serve as a Development Tool? Online Appendix September 2014
Fair Trade and Free Entry: Can a Disequilibrium Market Serve as a Development Tool? 1. Data Construction Online Appendix September 2014 The data consist of the Association s records on all coffee acquisitions
More informationTable A.1: Use of funds by frequency of ROSCA meetings in 9 research sites (Note multiple answers are allowed per respondent)
Appendix Table A.1: Use of funds by frequency of ROSCA meetings in 9 research sites (Note multiple answers are allowed per respondent) Daily Weekly Every 2 weeks Monthly Every 3 months Every 6 months Total
More informationThe R&D-patent relationship: An industry perspective
Université Libre de Bruxelles (ULB) Solvay Brussels School of Economics and Management (SBS-EM) European Center for Advanced Research in Economics and Statistics (ECARES) The R&D-patent relationship: An
More informationWhich of your fingernails comes closest to 1 cm in width? What is the length between your thumb tip and extended index finger tip? If no, why not?
wrong 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 right 66 65 64 63 62 61 60 59 58 57 56 55 54 53 52 51 50 49 score 100 98.5 97.0 95.5 93.9 92.4 90.9 89.4 87.9 86.4 84.8 83.3 81.8 80.3 78.8 77.3 75.8 74.2
More informationBiologist at Work! Experiment: Width across knuckles of: left hand. cm... right hand. cm. Analysis: Decision: /13 cm. Name
wrong 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 right 72 71 70 69 68 67 66 65 64 63 62 61 60 59 58 57 56 55 54 53 52 score 100 98.6 97.2 95.8 94.4 93.1 91.7 90.3 88.9 87.5 86.1 84.7 83.3 81.9
More informationGail E. Potter, Timo Smieszek, and Kerstin Sailer. April 24, 2015
Supplementary Material to Modelling workplace contact networks: the effects of organizational structure, architecture, and reporting errors on epidemic predictions, published in Network Science Gail E.
More informationVolume 30, Issue 1. Gender and firm-size: Evidence from Africa
Volume 30, Issue 1 Gender and firm-size: Evidence from Africa Mohammad Amin World Bank Abstract A number of studies show that relative to male owned businesses, female owned businesses are smaller in size.
More informationAppendix A. Table A.1: Logit Estimates for Elasticities
Estimates from historical sales data Appendix A Table A.1. reports the estimates from the discrete choice model for the historical sales data. Table A.1: Logit Estimates for Elasticities Dependent Variable:
More informationThe premium for organic wines
Enometrics XV Collioure May 29-31, 2008 Estimating a hedonic price equation from the producer side Points of interest: - assessing whether there is a premium for organic wines, and which one - estimating
More informationGender and Firm-size: Evidence from Africa
World Bank From the SelectedWorks of Mohammad Amin March, 2010 Gender and Firm-size: Evidence from Africa Mohammad Amin Available at: https://works.bepress.com/mohammad_amin/20/ Gender and Firm size: Evidence
More informationGrowth in early yyears: statistical and clinical insights
Growth in early yyears: statistical and clinical insights Tim Cole Population, Policy and Practice Programme UCL Great Ormond Street Institute of Child Health London WC1N 1EH UK Child growth Growth is
More informationTHE STATISTICAL SOMMELIER
THE STATISTICAL SOMMELIER An Introduction to Linear Regression 15.071 The Analytics Edge Bordeaux Wine Large differences in price and quality between years, although wine is produced in a similar way Meant
More informationInternet Appendix. For. Birds of a feather: Value implications of political alignment between top management and directors
Internet Appendix For Birds of a feather: Value implications of political alignment between top management and directors Jongsub Lee *, Kwang J. Lee, and Nandu J. Nagarajan This Internet Appendix reports
More informationCointegration Analysis of Commodity Prices: Much Ado about the Wrong Thing? Mindy L. Mallory and Sergio H. Lence September 17, 2010
Cointegration Analysis of Commodity Prices: Much Ado about the Wrong Thing? Mindy L. Mallory and Sergio H. Lence September 17, 2010 Cointegration Analysis, Commodity Prices What is cointegration analysis?
More informationBORDEAUX WINE VINTAGE QUALITY AND THE WEATHER ECONOMETRIC ANALYSIS
BORDEAUX WINE VINTAGE QUALITY AND THE WEATHER ECONOMETRIC ANALYSIS WINE PRICES OVER VINTAGES DATA The data sheet contains market prices for a collection of 13 high quality Bordeaux wines (not including
More informationOnline Appendix to. Are Two heads Better Than One: Team versus Individual Play in Signaling Games. David C. Cooper and John H.
Online Appendix to Are Two heads Better Than One: Team versus Individual Play in Signaling Games David C. Cooper and John H. Kagel This appendix contains a discussion of the robustness of the regression
More informationTariff vs non tariff barriers in seafood trade
(Debi Bishop/iStockphoto) Tariff vs non tariff barriers in seafood trade Kth Kathy Baylis, Lia Nogueira and Kathryn Kth Pace CATPRN workshop, Toronto, May 28, 2011 Objective To explore the effect and the
More informationShould We Put Ice in Wine? A Difference-in-Differences Approach from Switzerland
Should We Put Ice in Wine? A Difference-in-Differences Approach from Switzerland Alexandre Mondoux KOF Swiss Economic Institute, ETH Zurich 11th Annual AAWE Conference - Padua 2017 Thursday, 29 June 2017
More informationPanel A: Treated firm matched to one control firm. t + 1 t + 2 t + 3 Total CFO Compensation 5.03% 0.84% 10.27% [0.384] [0.892] [0.
Online Appendix 1 Table O1: Determinants of CMO Compensation: Selection based on both number of other firms in industry that have CMOs and number of other firms in industry with MBA educated executives
More informationInternet Appendix for Does Stock Liquidity Enhance or Impede Firm Innovation? *
Internet Appendix for Does Stock Liquidity Enhance or Impede Firm Innovation? * This Internet Appendix provides supplemental analyses and robustness tests to the main results presented in Does Stock Liquidity
More informationOnline Appendix to The Effect of Liquidity on Governance
Online Appendix to The Effect of Liquidity on Governance Table OA1: Conditional correlations of liquidity for the subsample of firms targeted by hedge funds This table reports Pearson and Spearman correlations
More informationInstruction (Manual) Document
Instruction (Manual) Document This part should be filled by author before your submission. 1. Information about Author Your Surname Your First Name Your Country Your Email Address Your ID on our website
More informationCoffee Price Volatility and Intra-household Labour Supply: Evidence from Vietnam
Coffee Price Volatility and Intra-household Labour Supply: Evidence from Vietnam Ulrik Beck U. Copenhagen Saurabh Singhal UNU-WIDER Finn Tarp UNU-WIDER June, 2016 Introduction Volatility in commodity markets
More informationSyndication, Interconnectedness, and Systemic Risk
Syndication, Interconnectedness, and Systemic Risk Jian Cai (WashU) Anthony Saunders (NYU Stern) Sascha Steffen (ESMT) Global Financial Interconnectedness Basel, 1-2 Oct 2015 1 This paper addresses the
More informationAppendix Table A1 Number of years since deregulation
Appendix Table A1 Number of years since deregulation This table presents the results of -in-s models incorporating the number of years since deregulation and using data for s with trade flows are above
More informationQuality Competition in Restaurants Industry: How Restaurants Respond to Fluctuating of Consumers Review. Ratings of Rivals
Quality Competition in Restaurants Industry: How Restaurants Respond to Fluctuating of Consumers Review Ratings of Rivals Mohammad Movahed October 25, 2018 Abstract: The goal of businesses is to maximize
More informationCredit Supply and Monetary Policy: Identifying the Bank Balance-Sheet Channel with Loan Applications. Web Appendix
Credit Supply and Monetary Policy: Identifying the Bank Balance-Sheet Channel with Loan Applications By GABRIEL JIMÉNEZ, STEVEN ONGENA, JOSÉ-LUIS PEYDRÓ, AND JESÚS SAURINA Web Appendix APPENDIX A -- NUMBER
More informationInside the Family Firm: The Role of Families in Succession Decisions and Performance *
Inside the Family Firm: The Role of Families in Succession Decisions and Performance * Morten Bennedsen Copenhagen Business School and CEBR Francisco Pérez-González Columbia University, GSB Kasper M. Nielsen
More informationUPPER MIDWEST MARKETING AREA THE BUTTER MARKET AND BEYOND
UPPER MIDWEST MARKETING AREA THE BUTTER MARKET 1987-2000 AND BEYOND STAFF PAPER 00-01 Prepared by: Henry H. Schaefer July 2000 Federal Milk Market Administrator s Office 4570 West 77th Street Suite 210
More informationInternet Appendix to. The Price of Street Friends: Social Networks, Informed Trading, and Shareholder Costs. Jie Cai Ralph A.
Internet Appendix to The Price of Street Friends: Social Networks, Informed Trading, and Shareholder Costs Jie Cai Ralph A. Walkling Ke Yang October 2014 1 A11. Controlling for s Logically Associated with
More informationPitfalls for the Construction of a Welfare Indicator: An Experimental Analysis of the Better Life Index
Clemens Hetschko, Louisa von Reumont & Ronnie Schöb Pitfalls for the Construction of a Welfare Indicator: An Experimental Analysis of the Better Life Index University Alliance of Sustainability Spring
More informationOnline Appendix. for. Female Leadership and Gender Equity: Evidence from Plant Closure
Online Appendix for Female Leadership and Gender Equity: Evidence from Plant Closure Geoffrey Tate and Liu Yang In this appendix, we provide additional robustness checks to supplement the evidence in the
More informationPEEL RIVER HEALTH ASSESSMENT
PEEL RIVER HEALTH ASSESSMENT CONTENTS SUMMARY... 2 Overall River Health Scoring... 2 Overall Data Sufficiency Scoring... 2 HYDROLOGY... 3 Overall Hydrology River Health Scoring... 3 Hydrology Data Sufficiency...
More informationPSYC 6140 November 16, 2005 ANOVA output in R
PSYC 6140 November 16, 2005 ANOVA output in R Type I, Type II and Type III Sums of Squares are displayed in ANOVA tables in a mumber of packages. The car library in R makes these available in R. This handout
More informationThe Bank Lending Channel of Conventional and Unconventional Monetary Policy: A Euro-area bank-level Analysis
The Bank Lending Channel of Conventional and Unconventional Monetary Policy: A Euro-area bank-level Analysis by U. Albertazzi, A. Nobili and F. Signoretti (Banca d Italia) Workshop : Effectiveness and
More informationFlexible Working Arrangements, Collaboration, ICT and Innovation
Flexible Working Arrangements, Collaboration, ICT and Innovation A Panel Data Analysis Cristian Rotaru and Franklin Soriano Analytical Services Unit Economic Measurement Group (EMG) Workshop, Sydney 28-29
More informationDATA AND ASSUMPTIONS (TAX CALCULATOR REVISION, MARCH 2017)
DATA AND ASSUMPTIONS (TAX CALCULATOR REVISION, MARCH 2017) Taxes on sugary drinks can generate considerable revenue for states, cities, and the nation. The revised Revenue Calculator for Sugary Drink Taxes
More informationGoing Strong. Comparing Ratios. to Solve Problems
Going Strong Comparing Ratios 2 to Solve Problems WARM UP Use reasoning to compare each pair of fractions. 1. 6 7 and 8 9 2. 7 13 and 5 11 LEARNING GOALS Apply qualitative ratio reasoning to compare ratios
More informationAJAE Appendix: Testing Household-Specific Explanations for the Inverse Productivity Relationship
AJAE Appendix: Testing Household-Specific Explanations for the Inverse Productivity Relationship Juliano Assunção Department of Economics PUC-Rio Luis H. B. Braido Graduate School of Economics Getulio
More informationThe Financing and Growth of Firms in China and India: Evidence from Capital Markets
The Financing and Growth of Firms in China and India: Evidence from Capital Markets Tatiana Didier Sergio Schmukler Dec. 12-13, 2012 NIPFP-DEA-JIMF Conference Macro and Financial Challenges of Emerging
More informationThe Effects of Presidential Politics on CEO Compensation
The Effects of Presidential Politics on CEO Compensation Humnath Panta, Ph.D. Assistant Professor and Finance Program Director Brenau University, Gainesville, GA 30501 Salil K. Sarkar, Ph.D., CFA Coordinator,
More informationNuclear reactors construction costs: The role of lead-time, standardization and technological progress
Nuclear reactors construction costs: The role of lead-time, standardization and technological progress Lina Escobar Rangel and Michel Berthélemy Mines ParisTech - Centre for Industrial Economics CERNA
More informationNotes on the Philadelphia Fed s Real-Time Data Set for Macroeconomists (RTDSM) Capacity Utilization. Last Updated: December 21, 2016
1 Notes on the Philadelphia Fed s Real-Time Data Set for Macroeconomists (RTDSM) Capacity Utilization Last Updated: December 21, 2016 I. General Comments This file provides documentation for the Philadelphia
More informationProblem Set #3 Key. Forecasting
Problem Set #3 Key Sonoma State University Business 581E Dr. Cuellar The data set bus581e_ps3.dta is a Stata data set containing annual sales (cases) and revenue from December 18, 2004 to April 2 2011.
More informationUpdate to A Comprehensive Look at the Empirical Performance of Equity Premium Prediction
Update to A Comprehensive Look at the Empirical Performance of Equity Premium Prediction Amit Goyal UNIL Ivo Welch UCLA September 17, 2014 Abstract This file contains updates, one correction, and links
More informationwine 1 wine 2 wine 3 person person person person person
1. A trendy wine bar set up an experiment to evaluate the quality of 3 different wines. Five fine connoisseurs of wine were asked to taste each of the wine and give it a rating between 0 and 10. The order
More informationMARKET ANALYSIS REPORT NO 1 OF 2015: TABLE GRAPES
MARKET ANALYSIS REPORT NO 1 OF 215: TABLE GRAPES 1. INTRODUCTION The following text is a review of the table grapes marketing environment. This analysis is updated on a quarterly 1 basis. The interval
More informationStatistics: Final Project Report Chipotle Water Cup: Water or Soda?
Statistics: Final Project Report Chipotle Water Cup: Water or Soda? Introduction: For our experiment, we wanted to find out how many customers at Chipotle actually get water when they order a water cup.
More informationSTAT 5302 Applied Regression Analysis. Hawkins
Homework 3 sample solution 1. MinnLand data STAT 5302 Applied Regression Analysis. Hawkins newdata
More informationImputation of multivariate continuous data with non-ignorable missingness
Imputation of multivariate continuous data with non-ignorable missingness Thais Paiva Jerry Reiter Department of Statistical Science Duke University NCRN Meeting Spring 2014 May 23, 2014 Thais Paiva, Jerry
More informationEffects of Election Results on Stock Price Performance: Evidence from 1976 to 2008
Effects of Election Results on Stock Price Performance: Evidence from 1976 to 2008 Andreas Oehler, Bamberg University Thomas J. Walker, Concordia University Stefan Wendt, Bamberg University 2012 FMA Annual
More informationInternet Appendix for CEO Personal Risk-taking and Corporate Policies TABLE IA.1 Pilot CEOs and Firm Risk (Controlling for High Performance Pay)
TABLE IA.1 Pilot CEOs and Firm Risk (Controlling for High Performance Pay) OLS regressions with annualized standard deviation of firm-level monthly stock returns as the dependent variable. A constant is
More informationRISK ASSESSMENT DEPT. OF AGROINDUSTRIAL TECHNOLOGY FACULTY OF AGRICULTURAL TECHNOLOGY UNIVERSITAS BRAWIJAYA
RISK ASSESSMENT DEPT. OF AGROINDUSTRIAL TECHNOLOGY FACULTY OF AGRICULTURAL TECHNOLOGY UNIVERSITAS BRAWIJAYA RISK ASSESSMENT To determine relative priority and get information to solve it Risk that should
More informationThe Development of a Weather-based Crop Disaster Program
The Development of a Weather-based Crop Disaster Program Eric Belasco Montana State University 2016 SCC-76 Conference Pensacola, FL March 19, 2016. Belasco March 2016 1 / 18 Motivation Recent efforts to
More informationFrom VOC to IPA: This Beer s For You!
From VOC to IPA: This Beer s For You! Joel Smith Statistician Minitab Inc. jsmith@minitab.com 2013 Minitab, Inc. Image courtesy of amazon.com The Data Online beer reviews Evaluated overall and: Appearance
More informationTo: Professor Roger Bohn & Hyeonsu Kang Subject: Big Data, Assignment April 13th. From: xxxx (anonymized) Date: 4/11/2016
To: Professor Roger Bohn & Hyeonsu Kang Subject: Big Data, Assignment April 13th. From: xxxx (anonymized) Date: 4/11/2016 Data Preparation: 1. Separate trany variable into Manual which takes value of 1
More informationLabor Supply of Married Couples in the Formal and Informal Sectors in Thailand
Southeast Asian Journal of Economics 2(2), December 2014: 77-102 Labor Supply of Married Couples in the Formal and Informal Sectors in Thailand Chairat Aemkulwat 1 Faculty of Economics, Chulalongkorn University
More informationBusiness Statistics /82 Spring 2011 Booth School of Business The University of Chicago Final Exam
Business Statistics 41000-81/82 Spring 2011 Booth School of Business The University of Chicago Final Exam Name You may use a calculator and two cheat sheets. You have 3 hours. I pledge my honor that I
More informationNapa County Planning Commission Board Agenda Letter
Agenda Date: 7/1/2015 Agenda Placement: 10A Continued From: May 20, 2015 Napa County Planning Commission Board Agenda Letter TO: FROM: Napa County Planning Commission John McDowell for David Morrison -
More informationTrip Generation at Fast Food Restaurants
Trip Generation at Fast Food Restaurants in Saudi Arabia This study developed trip generation rates at fast food restaurants in Jeddah, Saudi Arabia. The results showed that vehicle trip rates were not
More informationLollapalooza Did Not Attend (n = 800) Attended (n = 438)
D SDS H F 1, 16 ( ) Warm-ups (A) Which bands come to ACL Fest? Is it true that if a band plays at Lollapalooza, then it is more likely to play at Austin City Limits (ACL) that year? To be able to provide
More informationFigure 1: Quartely milk production and gross value
Million Litres Million Rands QUARTERLY DAIRY MARKET ANALYSIS BULLETIN 1 OF 215 1. INTRODUCTION The following discussion is a review of the dairy market environment. The analysis is updated on a quarterly
More informationWine-Tasting by Numbers: Using Binary Logistic Regression to Reveal the Preferences of Experts
Wine-Tasting by Numbers: Using Binary Logistic Regression to Reveal the Preferences of Experts When you need to understand situations that seem to defy data analysis, you may be able to use techniques
More informationThis appendix tabulates results summarized in Section IV of our paper, and also reports the results of additional tests.
Internet Appendix for Mutual Fund Trading Pressure: Firm-level Stock Price Impact and Timing of SEOs, by Mozaffar Khan, Leonid Kogan and George Serafeim. * This appendix tabulates results summarized in
More informationWeb Appendix to Identifying Sibling Inuence on Teenage Substance Use. Joseph G. Altonji, Sarah Cattan, and Iain Ware
Web Appendix to Identifying Sibling Inuence on Teenage Substance Use Joseph G. Altonji, Sarah Cattan, and Iain Ware A Data The paper uses data from the rst eight rounds of the National Longitudinal Survey
More informationThought: The Great Coffee Experiment
Thought: The Great Coffee Experiment 7/7/16 By Kevin DeLuca ThoughtBurner Opportunity Cost of Reading this ThoughtBurner post: $1.97 about 8.95 minutes I drink a lot of coffee. In fact, I m drinking a
More informationA Note on a Test for the Sum of Ranksums*
Journal of Wine Economics, Volume 2, Number 1, Spring 2007, Pages 98 102 A Note on a Test for the Sum of Ranksums* Richard E. Quandt a I. Introduction In wine tastings, in which several tasters (judges)
More informationWine Futures: Pricing and Allocation as Levers against Quality Uncertainty
Padua 2017 Abstract Submission I want to submit an abstract for: Conference Presentation Corresponding Author Burak Kazaz E-Mail bkazaz@syr.edu Affiliation Syracuse University, Whitman School of Management
More informationReligion and Innovation
Religion and Innovation Roland Bénabou Davide Ticchi Andrea Vindigni Princeton University IMT Lucca IMT Lucca & NBER & CIFAR Collegio Carlo Alberto American Economic Review P&P (2015) Introduction Economics
More informationValuing Health Risk Reductions from Air Quality Improvement: Evidence from a New Discrete Choice Experiment (DCE) in China
Valuing Health Risk Reductions from Air Quality Improvement: Evidence from a New Discrete Choice Experiment (DCE) in China Yana Jin Peking University jin.yana@pku.edu.cn (Presenter, PhD obtained in 2017,
More informationLONDON S WARMEST WELCOME
LONDON S WARMEST WELCOME M A T C H D AY H O S P I T A L I T Y A T WA T F O R D F C PREMIER LEAGUE HOSPITALITY AT VICARAGE ROAD FOOTBALL AND HOSPITALITY AT ITS FINEST COME AND EXPERIENCE LONDON S WARMEST
More informationData Science and Service Research Discussion Paper
Discussion Paper No. 69 A Network Analysis of International Financial Flows Hongwei Chuang and Navruzbek Karamatov Graduate School of Economics and Management, Tohoku University August, 2017 Data Science
More information7 th Annual Conference AAWE, Stellenbosch, Jun 2013
The Impact of the Legal System and Incomplete Contracts on Grape Sourcing Strategies: A Comparative Analysis of the South African and New Zealand Wine Industries * Corresponding Author Monnane, M. Monnane,
More informationThe 2006 Economic Impact of Nebraska Wineries and Grape Growers
A Bureau of Business Economic Impact Analysis From the University of Nebraska Lincoln The 2006 Economic Impact of Nebraska Wineries and Grape Growers Dr. Eric Thompson Seth Freudenburg Prepared for The
More informationAn Empirical Analysis of the U.S. Import Demand for Nuts
An Empirical Analysis of the U.S. Import Demand for Nuts by Sona Grigoryan M.S. Student, College of Agricultural Sciences and Natural Resources, Texas A&M University-Commerce sgrigoryan@leomail.tamuc.edu
More informationLecture 13. We continue our discussion of the economic causes of conflict, but now we work with detailed data on a single conflict.
EC3320 2016-2017 Michael Spagat Lecture 13. We continue our discussion of the economic causes of conflict, but now we work with detailed data on a single conflict. The units of analysis for Collier-Hoefller-Fearon-Laitin
More informationInvestment Wines. - Risk Analysis. Prepared by: Michael Shortell & Adiam Woldetensae Date: 06/09/2015
Investment Wines - Risk Analysis Prepared by: Michael Shortell & Adiam Woldetensae Date: 06/09/2015 Purpose Look at investment wines & examine factors that affect wine prices over time We will identify
More informationThe Sources of Risk Spillovers among REITs: Asset Similarities and Regional Proximity
The Sources of Risk Spillovers among REITs: Asset Similarities and Regional Proximity Zeno Adams EBS Business School Roland Füss EBS Business School ZEW Mannheim Felix Schinder ZEW Mannheim Steinbeis University
More informationThe multivariate piecewise linear growth model for ZHeight and zbmi can be expressed as:
Bi-directional relationships between body mass index and height from three to seven years of age: an analysis of children in the United Kingdom Millennium Cohort Study Supplementary material The multivariate
More informationReturn to wine: A comparison of the hedonic, repeat sales, and hybrid approaches
Return to wine: A comparison of the hedonic, repeat sales, and hybrid approaches James J. Fogarty a* and Callum Jones b a School of Agricultural and Resource Economics, The University of Western Australia,
More information*p <.05. **p <.01. ***p <.001.
Table 1 Weighted Descriptive Statistics and Zero-Order Correlations with Fatherhood Timing (N = 1114) Variables Mean SD Min Max Correlation Interaction time 280.70 225.47 0 1095 0.05 Interaction time with
More informationMBA 503 Final Project Guidelines and Rubric
MBA 503 Final Project Guidelines and Rubric Overview There are two summative assessments for this course. For your first assessment, you will be objectively assessed by your completion of a series of MyAccountingLab
More informationMultiple Imputation for Missing Data in KLoSA
Multiple Imputation for Missing Data in KLoSA Juwon Song Korea University and UCLA Contents 1. Missing Data and Missing Data Mechanisms 2. Imputation 3. Missing Data and Multiple Imputation in Baseline
More informationThe Effect of Fast Food Restaurants on Obesity
The Effect of Fast Food Restaurants on Obesity Janet Currie, Columbia University and NBER Stefano DellaVigna, UC Berkeley and NBER Enrico Moretti, UC Berkeley and NBER Vikram Pathania, UC Berkeley December
More informationScreening Natural Corks by Their Releasable TCA: Does it Work?
Screening Natural Corks by Their Releasable TCA: Does it Work? Eric Hervé Ph.D. E T S Laboratories, St. Helena, CA 1 1. What is Releasable TCA?. RTCA from bulk vs. individual cork soaks 3. Bulk Soaks as
More informationTHE LOMBARDI RULES PDF
THE LOMBARDI RULES PDF ==> Download: THE LOMBARDI RULES PDF THE LOMBARDI RULES PDF - Are you searching for The Lombardi Rules Books? Now, you will be happy that at this time The Lombardi Rules PDF is available
More informationClimate change may alter human physical activity patterns
In the format provided by the authors and unedited. SUPPLEMENTARY INFORMATION VOLUME: 1 ARTICLE NUMBER: 0097 Climate change may alter human physical activity patterns Nick Obradovich and James H. Fowler
More informationPlease sign and date here to indicate that you have read and agree to abide by the above mentioned stipulations. Student Name #4
The following group project is to be worked on by no more than four students. You may use any materials you think may be useful in solving the problems but you may not ask anyone for help other than the
More informationQUARTELY MAIZE MARKET ANALYSIS & OUTLOOK BULLETIN 1 OF 2015
QUARTELY MAIZE MARKET ANALYSIS & OUTLOOK BULLETIN 1 OF 2015 INTRODUCTION The following discussion is a review of the maize market environment. The analysis is updated on a quarterly 1 basis and the interval
More informationNovember K. J. Martijn Cremers Lubomir P. Litov Simone M. Sepe
ONLINE APPENDIX TABLES OF STAGGERED BOARDS AND LONG-TERM FIRM VALUE, REVISITED November 016 K. J. Martijn Cremers Lubomir P. Litov Simone M. Sepe The paper itself is available at https://papers.ssrn.com/sol3/papers.cfm?abstract-id=364165.
More informationModeling Regional Endogenous Growth
Modeling Regional Endogenous Growth A Structural Equation Model Approach Patricio Aroca 1, Robert Stimson 2 and Roger Stough 3 1 Universidad Católica del Norte 2 University of Queensland 3 George Mason
More informationAMERICAN ASSOCIATION OF WINE ECONOMISTS
AMERICAN ASSOCIATION OF WINE ECONOMISTS AAWE WORKING PAPER No. 195 Economics FACTORS IMPACTING WINE PRICE MARK-UPS IN RESTAURANTS Florine Livat and Hervé Remaud Apr 2016 www.wine-economics.org AAWE Working
More informationDietary Diversity in Urban and Rural China: An Endogenous Variety Approach
Dietary Diversity in Urban and Rural China: An Endogenous Variety Approach Jing Liu September 6, 2011 Road Map What is endogenous variety? Why is it? A structural framework illustrating this idea An application
More informationMicroanalytical Quality of Ground and Unground Marjoram, Sage and Thyme, Ground Allspice, Black Pepper and Paprika
Journal of Food Protection, Vol., No., Pages - (March ) Copyright International Association of Milk, Food, and Environmental Sanitarians Microanalytical Quality of and Marjoram, Sage and Thyme, Allspice,
More informationRegression Models for Saffron Yields in Iran
Regression Models for Saffron ields in Iran Sanaeinejad, S.H., Hosseini, S.N 1 Faculty of Agriculture, Ferdowsi University of Mashhad, Iran sanaei_h@yahoo.co.uk, nasir_nbm@yahoo.com, Abstract: Saffron
More informationMATCHDAY HOSPITALITY AT WATFORD FC LONDON S WARMEST WELCOME
MATCHDAY HOSPITALITY AT WATFORD FC LONDON S WARMEST WELCOME PREMIER LEAGUE HOSPITALITY AT VICARAGE ROAD COME AND EXPERIENCE LONDON S WARMEST WELCOME WITH ONE OF OUR EXCLUSIVE MATCHDAY HOSPITALITY PACKAGES.
More informationIntegration of Major Agricultural Product Markets of China
Integration of Major Agricultural Product Markets of China Wu Laping College of Economics and Management, China Agricultural University Abstract China s accession to the World Trade Organisation (WTO)
More informationTrade Facilitation and Supply Chain Security:
Trade Facilitation and Supply Chain Security: Evaluation of Integrated Cargo Containers Control (IC3) Program between Pakistan and the United States WCO-PICARD Conference 27-29 September 2016 Presenter:
More information