SUPREME SPLIT: COMPARING THE ROBERTS AND REHNQUIST COURTS IDEOLOGICAL PREFERENCES TOWARD BUSINESS ACCOUNTING FOR CASE SELECTION

Transcription

1 SUPREME SPLIT: COMPARING THE ROBERTS AND REHNQUIST COURTS IDEOLOGICAL PREFERENCES TOWARD BUSINESS ACCOUNTING FOR CASE SELECTION Clfford Wnston Ja Yan Quentn Karplow Brookns Insttuton Washnton State Unversty Yale Unversty Abstract. There s rown concern that Supreme Court rulns are ncreasnly ben made alon partsan lnes. We develop a new methodoloy that jontly analyzes justces selecton of cases to hear and ther rulns whle allown for justce preference heteroenety to estmate the effects of ther deoloy on busness case rulns n the Roberts and Rehnqust Courts. In contrast to pror approaches that mute justces deolocal preferences by not controlln for case selecton we fnd stron evdence of rown poltczaton and polarzaton on the Supreme Court. We brefly dscuss explanatons for consequences of and possble reforms to mtate ths worrsome trend. May 2016

2 1 1. Introducton The Supreme Court of the Unted States (SCOTUS) has ultmate jursdcton over federal laws and plays a central role n determnn the naton s economc and socal polces. In contrast to Conress and the Executve Offce of the Presdent the Supreme Court s enerally thouht to be an ndependent and apoltcal body because justces are apponted for lfe are not ambtous for hher offce decde whch cases they wsh to hear and are not subject to reversal by a hher (Seal and Spaeth (2002)). Ths perceved aloofness from poltcs has bolstered publc trust n the Court and publc opnon polls have repeatedly shown that of the three branches of overnment Amercans hold the judcal branch n the hhest reard. But perceptons of the Supreme Court are chann. Chemernsky (2014) for nstance has characterzed the justces n the two most recent s the Roberts Court named after Chef Justce John G. Roberts and the Rehnqust Court named after Chef Justce Wllam H. Rehnqust as poltcans n fne robes who smply reflect the vews of the Presdent who apponted them. Althouh flashy such clams are not wthout evdence. Leal scholars for example have ponted to the recent and precptous rse n splt-decson rulns as an ndcator of the Court s rown polarzaton and poltczaton. 1 Speculaton that poltcal deoloy s ncreasnly nfluencn Court rulns fnds further support n the fact that the Roberts Court has made more major decsons alon partsan lnes than have prevous s. 2 Wth the recent 1 For example from 1801 to 1940 less than 2 percent of SCOTUS total rulns were resolved by 5-to-4 decsons. In contrast the Roberts Court has decded 21.5 percent of ts cases by a 5-to-4 ruln the hhest share of any and the Rehnqust Court has decded 20.5 percent of ts cases by that marn. The fures are reported n Davd Paul Kuhn The Incredble Polarzaton and Poltczaton of the Supreme Court The Atlantc June For example between 1937 and 2010 Devns and Baum (2014) report that one major case (n 1985) broke alon party lnes. But snce 2010 fve major decsons have been decded n that manner.

3 2 passn of Justce Antonn Scala there s concern that the Court wll be deadlocked on a number of major cases untl a nnth justce s apponted and confrmed. Fnally Rshkof Messener and Jo (2009) show that Justces recent hrn of law clerks s also occurrn much more alon party lnes than n the past. Because those clerks play crtcal roles n selectn the pettons that the Justces aree to hear and n draftn Justce opnons the authors arue that such partsan hrn practces have contrbuted to the Court s polarzed envronment. 3 Whle the crcumstantal evdence presented above alludes to the rown poltczaton of the Court justce deoloy s lkely to be a potentally mportant nfluence on to SCOTUS outcomes because SCOTUS justces are apponted for lfe and because ther decsons are not subject to reversal by a hher. As a result those polcymakers enjoy reater freedom than elected offcals to allow ther deolocal preferences to nfluence ther decsons. Moreover Posner (2008) arues that because Justces do not share a commtment to a local premse for makn a decson (for example cost-beneft analyss); they are deolocal because they cannot be anythn else. On the other hand Rosen (2013) arues that f justces based ther decsons solely on ther deoloes they could damae ther reputatons as far justces dscredt the Presdent who apponted them and possbly undermne the Court n the eyes of the publc whch could lead to 3 The rown polarty of the Supreme Court s further suested by Peppers and Gles (2012) fndns that the justces attendance at the Presdent s State of the Unon (SOTU) address has declned n recent decades and that a justce s more lkely to attend the SOTU address f he or she was apponted by the Presdent vn the address. Accordn to the authors from 1965 (Presdent Johnson s second address) to 1980 (Presdent Carter s last address) 84 percent of the justces attended the SOTU address on averae. From 1982 throuh 1999 the justces averae attendance declned to 56 percent. And snce 2000 only 32 percent of the justces attended the address wth Justce Clarence Thomas clamn that t has become so partsan and t s very uncomfortable for a jude to st there and Justce Antonn Scala descrbn the annual address as cheerleadn sessons.

4 3 certan actons such as term lmts or even retenton electons. It seems lkely then that deoloy s an mportant but not the only factor shapn SCOTUS decson-makn. Whle the vew that deoloy nfluences SCOTUS outcomes s compelln t s stll an emprcal queston as to whether stroner deolocal preferences are makn the Roberts Court more polarzed than prevous s. Chef Justce Roberts has hmself acknowleded the polcy mportance of ths queston clamn that f the Court were to adopt the extreme partsanshp that currently characterzes the other branches of the federal overnment t would serously damae SCOTUS s reputaton amon the publc as well as jeopardze the judcal nomnaton and confrmaton process. Accordnly Roberts has souht to assuae concerns that the Court s functonn as a poltcal entty by publcly pledn to embrace Chef Justce John Marshall s concepton of the judcal branch as a nonpartsan steward n a polarzed democracy. 4 Leal scholars who have attempted to determne systematcally whether the Supreme Court s becomn more polarzed under Chef Justce Roberts are dvded. For example Chemernsky s (2008) assessment of the Roberts Court after three years bemoans the new justces who have fored a sold conservatve votn majorty and concludes that the Court s exceednly pro-busness whle Adler (2008) dsarees and says the Court s only moderately more conservatve than t had been mmedately before and s not notably pro-busness. Adler ponts out that the smaller number of case pettons that the Roberts Court accepted durn the perod suests cauton should be exercsed about makn sweepn judments about ts deolocal trajectory. Lee Epsten Landes and Posner (2013a) performed a statstcal analyss of busness cases and concluded that the conservatves on the Roberts Court are extremely pro-busness and that the lberals are only 4 Justce Roberts concerns about polarzaton and partsanshp n the Supreme Court are summarzed by Rosen (2013) and by Professor Geoffrey Stone of the Unversty of Chcao Law School n Our Poltcally Polarzed Supreme Court? Huffnton Post Blo September

5 4 moderately lberal. Rchard Epsten (2013) challened ther fndn stressn that the authors do not control for potental selectvty bas n the case pettons that the Roberts Court accepts. In ths paper we explore the nfluences of justce deoloy n the Roberts Court by analyzn the determnants of ts rulns and the Rehnqust Court s rulns on busness cases. As noted by Epsten Landes and Posner (2013a) busness cases are useful to analyze because (1) justces tend to have well-defned preferences toward busness and the polces favored by most busness nterests and (2) justce poltcal deoloy (lberal or conservatve) can serve as a ood proxy for those preferences wth conservatve justces ben more lkely than lberal ones to make decsons that favor busness frms. Importantly we solate the effects of poltcal deoloy on judcal rulns by developn a jont mxed-lot model that accounts for both justce preference heteroenety and the selecton of pettons by SCOTUS. To the best of our knowlede ths s the frst applcaton of a new methodoloy to correct for selectvty bas when (1) dscrete choce probablty models are used to characterze both the selectvty behavor (petton selecton) and the outcome of nterest (justces ndvdual rulns for or aanst a busness entty) and (2) the selecton error s correlated wth the random coeffcents n the outcome model. Althouh t has lon been arued that falure to control for case selectvty can bas analyses of outcomes (Prest and Klen (1984)) prevous emprcal studes of SCOTUS rulns have not formally ncorporated case selecton decsons nto an econometrc analyss. 5 As a result ths paper contrbutes to the broader study of SCOTUS decson-makn by addressn a potentally crtcal but larely nored source of estmaton bas. In partcular a 5 For example see: Epsten Landes and Posner (2013a b); predcton models of Supreme Court outcomes developed by Ruer et al. (2004) and Katz Bommarto and Blackman (2014); and examnatons of chanes n ndvdual justces deolocal belefs by Iaryczower and Shum (2012) and Martn and Qunn (2007).

6 5 Court wth stron deolocal preferences may mask or exacerbate those preferences by areen to hear more or fewer cases of a certan type (e.. poltcally dvsve cases). Our methodoloy should also be of eneral nterest ven the ubquty of selecton problems and dscrete outcomes that economsts study. For example nstead of the classc Heckman selecton problem of the dscrete selecton choce of employment and a contnuous wae outcome one can thnk of the selecton problem of employment and a dscrete outcome of earnns above the mnmum wae. We fnd that justces deoloes have a statstcally and quanttatvely mportant nfluence on ther rulns on busness cases. Moreover when we compare justces deolocal preferences n the Rehnqust Court wth those n the Roberts Court and we mplement our methodoloy to control for case selecton we fnd stron evdence of rown poltczaton and polarty on the Supreme Court. Ths fndn holds even when we confne our analyses to those justces who have served on both Courts. In contrast we obtan much weaker evdence on rown poltczaton and polarty f we do not control for case selecton ndcatn that the Roberts Court justces mute ther deolocal preferences throuh the petton selecton process. Thus Supreme Court justces have not moderated ther deolocal preferences over the past two decades; nstead those preferences appear to have rown stroner and more dvsve. We conclude by brefly dscussn the explanatons for and adverse consequences of the Court s rown polarty as well as possble reforms to mtate ths worrsome trend. 2. An Overvew of the Determnants of Petton Selecton and Case Outcomes Two extreme theores about judes behavor exst (Edwards and Lvermore (2009)). The lealst theory asserts that judes mechancally apply the law to the facts whle the poltcal

7 6 scence theory arues that a jude s poltcal deoloy has a predomnant nfluence on hs or her rulns. Epsten Landes and Posner (2013b) suest a mddle round wheren deoloy s a component n judcal behavor that ncreases n nfluence as a jude moves up the judcal herarchy and ans more ndependence. In ths framework however deoloy does not rule out other possble nfluences on judes rulns. We draw on ths thrd approach to model two SCOTUS decson-makn processes: (1) the Court s selecton of pettons to hear (.e. the decson to rant a petton a wrt of certorar or rantn cert ); and (2) justce rulns on busness-related cases that are heard by SCOTUS (.e. case-outcome votes). We provde an overvew of the relevant nfluences and ther measurement on those two decson junctures. Case-Outcomes Model Data on U.S. Supreme Court busness case outcomes durn the Rehnqust Court and the Roberts Court up throuh 2011 are avalable n Epsten Landes and Posner s (2013a) Busness Ltant Dataset (BLD). Ths dataset conssts of all cases that were orally arued before the U.S. Supreme Court and that nvolved a busness entty as ether a respondent or a pettoner but not both. Because BLD was constructed usn Harold Spaeth s well-known U.S. Supreme Court Database t contans detaled nformaton on the leal characterstcs hstory and outcomes of Supreme Court dsputes as well as data on the votes of each Justce (Spaeth et al. (2014)). Due to petton data constrants that are dscussed n reater detal below we lmt our case outcomes analyses to the 198 busness cases that were submtted to and heard by SCOTUS between 1996 and Follown Epsten Landes and Posner (2013a) we classfy any vote n favor of the busness ltant as pro-busness and any vote n favor of the non-busness party as ant-

8 7 busness. Althouh t s possble for one company to wn at the expense of the nterests of the larer busness communty ths classfcaton scheme s superor to alternatve defntons of a busness wn because t s more transparent reles less on the subjectve codn judments of the researcher and allows for the analyss of a wder rane of busness-related cases. 6 We estmate separate models for the Roberts and Rehnqust s to determne the dstnct deolocal preferences of the justces on those s. For each we ntally follow Epsten Landes and Posner (2013a) by specfyn the probablty of a pro-busness vote as a functon of: (a) an ndcator for each justce servn on the whch captures the deolocal preferences of that justce; 7 (b) an ndcator of whether the lower from whch the case was appealed reached a pro-busness ruln; (c) a seres of dummy varables ndcatn whether the Solctor General fled an amcus bref on behalf of the busness ltant or the non-busness ltant; (d) an ndcator of whether the federal overnment was the non-busness party n the case; and (e) an ndcator varable of whether the substantve ssues of a case fall wthn the scope of Spaeth s economc actvty cateory. In an alternatve set of specfcatons we replace the ndvdual justce 6 One popular defnton of a busness wn focuses on cases that have been coded n the Spaeth Database as ben related to economc actvty and then uses Spaeth s deolocal classfcatons of a case as ether conservatve or lberal as a proxy for pro- and ant-busness decsons respectvely. Epsten Landes and Posner (2013a) however show that ths defnton excludes a substantal number of cases that do not fall nto the economc actvty cateory but nonetheless nvolved busness enttes. Even more telln the authors fnd that many of those excluded cases were accompaned by an amcus curae bref fled by the Chamber of Commerce a sure ndcator of the busness relevance of a case. In addton Epsten Landes and Posner call nto queston the accuracy of Spaeth s deolocal classfcaton of cases when they conduct ther own reanalyss of 147 busness-related cases. Strknly n about 39 percent of the 147 cases revewed the authors dsareed wth Spaeth s deolocal codn. Rates of dsareement were even hher for busnesses cases that dd not fall wthn the economc actvty classfcaton. 7 Note the dummy captures deolocal preferences and not just dsareement amon justces because we control for several other factors on case decsons and because our man analyses explore systematc dfferences amon justces preferences by assessn them relatve to the chef justce s preference.

9 8 ndcators wth a snle deoloy bnary measure that dentfes justces as ether Conservatve\Moderate or Lberal accordn to Epsten Landes and Posner s (2013b) classfcaton system whch they explan n detal n chapter 2 of ther 2013b book. Importantly ther methodoloy determnes justces deoloes based on ther actual votes n cases as opposed to the deoloy of the Presdent who apponted them. In addton the deolocal status of a justce does not chane over tme. Fnally as robustness checks we expanded the specfcaton of the case-outcomes model to nclude: (f) eleven topcal dummes that cateorze pettons to the Supreme Court accordn to the man leal questons and substantve matters of the dspute whch can be used to dfferentate between the subject matter of the busness cases heard by the Court; 8 and () a seres of lower dummes whch are descrbed n reater detal below. 8 We use Bloomber BNA s Supreme Court Today s dataset to obtan pettons to the Supreme Court for each year of our sample. That database cateorzes pettons accordn to 64 dfferent leal cateores. Due to sample sze ssues we areated those fne-raned classfcatons nto eleven broad topc cateores namely: (1) Crmnal law whch contans cases related to crmnal law prsons and racketeern; (2) Cvl rhts law whch s composed of Bloomber s cvl rhts educaton electon and employment dscrmnaton topcs; (3) Reulatory law whch contans pettons related to arculture anttrust enery envronment health care nternatonal trade securtes utltes telecommuncatons and transportaton; (4) Consumer and worker safety law whch contans Bloomber s consumer credt consumer protecton deceptve trade employee benefts mnn occupatonal safety product lablty unfar trade practces employment/labor attorney ethncs and torts topcs; (5) Prvate/state busness law whch contans pettons related to accountn bankn/fnance contracts corporate and busness law nsurance partnershps arbtraton alcoholc beveraes and ambln; (6) Intellectual property law whch s composed of Bloomber s copyrhts patents and trademarks topcs; (7) Government authorty ssues whch contans pettons related to admnstratve law aency law nternatonal law takns appellate procedure cvl procedure conressonal operatons evdence judcal appontments and consttutonal law; (8) Technoloy ssues whch contans pettons related to cyber law meda and prvacy; (9) Other overnment ssues whch s composed of Bloomber s bankruptcy overnment contracts mmraton nfrastructure martme mltary Natve Amercan postal servces socal securty veterans and freedom of nformaton topcs; (10) Tax law whch s composed only of Bloomber s tax topc; and (11) Prvate non-busness whch ncludes Bloomber s famly law trusts and estates and nonproft oranzatons topcs.

10 9 Petton Selecton Model In prncple there are two selecton decsons that may be relevant for our analyss: The SCOTUS justces selecton of cases from the pool of lower appeals and the selectvty behavor of ltants to nclude ther case n the pool of SCOTUS pettons. Assembln a data set and estmatn a model of the petton decsons by ltants would be a very formdable task of questonable value to our analyss ven: (1) all ltants who lose at lower s have the rht to petton the Supreme Court to hear ther case; and (2) many busness ltants take advantae of ths rht especally because the cost to those ltants of submttn a petton s not hh. Moreover we are not aware of any theoretcal or emprcal evdence that there s any systematc bas n the cases that busness ltants choose to appeal and ven that we are comparn the Roberts Court ( ) to the latter half of the Rehnqust Courts ( ) we lmt the possblty that dfferences n the types of cases that make t nto the SCOTUS pool of pettons are drvn our results. In partcular t seems hhly unlkely that the types of cases ben processed by the lower s durn the end of the Rehnqust Court are cateorcally dfferent than those crculatn throuh the lower s durn the Roberts Court. Thus we focus on estmatn the determnants of case selecton by the Roberts and Rehnqust Courts usn the onlne Bloomber BNA Supreme Court Today dataset. 9 The Bloomber database contans a myrad of nformaton on vrtually all pad cert. pettons fled wth the Supreme Court between 1996 and However because the format of the database s not Pad cert. pettons are also called non-pauper pettons. No one has collected data on pauper (non-pad) pettons whch represent only about 10% of the Supreme Court s docket and tend to be easly resolved by the justces (Posner (2008)). Note that some early-1996 pad cert. pettons

11 10 conducve to statstcal analyss and because data must be downloaded separately for each ndvdual petton athern nformaton on the more than pettons submtted durn ths tme perod proved to be lostcally nfeasble. We therefore constructed a sample of 2768 pettons that were randomly selected from the Bloomber database. 11 In addton the 198 cases n our Busness Ltant Database that are used to estmate the case-outcomes model can be treated as a choce-based sample n estmatn the petton-selecton model. Althouh the pettonselecton model can be estmated usn the random sample of 2768 pettons combnn the random sample wth the choce-based sample can mprove the effcency of the parameter estmates. Hence we estmate our model wth the random sample and then combne t wth the choce-based sample as a robustness check. We are not aware of a prevous Supreme Court petton selecton model that explans how the justces select pettons althouh t s known that a petton needs four justce votes to be selected the typcally hears about 1 percent of the nearly 9000 appeals that they receve each year and that t s customary for the s decson on whether to hear a case does not ndcate how ndvdual justces voted. 12 Thus we revewed the extant lterature on the factors are also mssn from the Bloomber database. The database does however contan the full populaton of non-pauper SCOTUS pettons submtted n all post-1996 years. 11 To construct our random sample of Bloomber pettons we frst sorted elble pettons by the date on whch the SCOTUS decded to hear (or not hear) the petton. Only pettons that were ether ranted or dened a hearn wth SCOTUS were elble for selecton. Then for each year we selected every tenth petton to nclude n our sample meann that on averae we sampled between 100 and 300 pettons for each SCOTUS term. Note that the 2768 randomly selected pettons and the 198 BLD pettons are not mutually exclusve because some randomly selected pettons were also ncluded n the Busness Ltant Dataset. 12 Jeffrey L. Fsher The Supreme Court s Secret Power New York Tmes September arues that justces votes on each petton should be announced to ncrease transparency on why justces pck the cases they do. The absence of ths votn nformaton makes t dffcult for us to assess whether justces enae n defensve denals or other forms of stratec behavor (Corday and Corday (2008)).

12 11 that are most relevant for explann the Justces case selecton decsons and specfed the probablty that a petton s ranted a hearn wth the Supreme Court as a functon of: (a) thrteen lower ndcators and a reference cateory that dentfy the system whose decson s ben appealed; 13 (b) a lawyer qualty dummy that ndcates whether any of the lawyers fln the petton are top Supreme Court advocates whch we defne to be ltators who have arued at least ten cases before the Supreme Court by 2010; 14 (c) a second lawyer performance dummy that dentfes top Supreme Court advocates who were workn as the U.S. Solctor General when the federal overnment fled the petton; (d) an ndcator for whether the Court nvted the Solctor General to submt a bref analyzn the petton and expressn the vews of the Unted States; 15 (e) a seres of Court-calendar dummes ndcatn whether the decson to rant or deny a petton was made durn the Court s summer recess the May or June lsts or one of the seven arument 13 More specfcally we nclude separate dummes for (a) each of the eleven crcut s of appeals; (b) the Dstrct of Columba crcut (ndcated n the estmatons as lower dummy12); and (c) the Federal Patent Court (ndcated n the estmatons as lower dummy 13). The reference cateory ncludes pettons from state and lower dstrct s as well as any not explctly dentfed by our set of ndcator varables. 14 Bskupc Roberts and Shffman (2012) and Lazarus (2008) arue that SCOTUS rants pettons fled by certan top advocate lawyers at a snfcantly hher rate than they rant pettons fled by other lawyers. Whle the Bloomber database provdes the names of the lawyers fln a petton we reled upon data publshed by Bhata (2012) to dentfy top Supreme Court Advocates. Specfcally Bhata ranks modern lawyers accordn to the number of SCOTUS cases they have arued between 2000 and 2010 as well as provdes lfetme numbers of Supreme Court appearances for each of those lawyers. Taken toether the Bloomber and Bhata data allowed us to dentfy pettons fled by lawyers who as of 2010 had arued at least 10 cases before the Supreme Court. 15 Thompson and Wachtell (2009) dscuss the close relatonshp that solctor enerals have wth the Supreme Court and the Court s requests for ther vews on pettons.

13 12 sessons; 16 and (f) a count varable measurn the sze of the Supreme Court s docket n a ven year Econometrc Model We now develop a new econometrc model to estmate the effects of the deolocal preferences of the Roberts and Rehnqust Court justces on the outcomes of busness-related cases accountn for the non-random selecton of pettons. We let ndex the two s wth 1 for the Rehnqust Court and 2 for the Roberts Court. For a ven case before there are up to nne votes by the justces on that. We let yj denote the vote of justce j on case y j wth equal to 1 f the vote s pro-busness and 0 otherwse. Thus the justce case outcomes vote under s ven by the follown bnary response model: y I y j * j D L j D C j X B e j 0 (1) 16 Corday and Corday (2004) dscuss how the volume of pettons affects the lkelhood of whether the Supreme Court wll aree to hear a case. The Supreme Court term bens the frst Monday n October and conssts of seven arument sessons durn whch the Justces hear cases and delver opnons; the May and June lsts durn whch the Court sts only to announce orders and opnons; and the summer recess durn whch the Court s only duty s to revew newly-submtted pettons and make preparatons for the fall arument sessons. Follown Cordray and Cordray (2004) petton decsons announced n July Auust September and the frst Monday of October were assned to the summer recess; subsequent petton decsons made n October (but after the frst Monday) were assned to the October sesson the frst of the seven arument sessons. Month dummes were then constructed to dentfy the sx remann arument sessons (November throuh Aprl) and the May and June lsts. 17 Annual docket sze was measured as the total number of pettons fled durn the year plus the total number of cases remann on the docket from the prevous year. Note that these counts nclude both the non-pauper pettons recorded n the Bloomber dataset and the much smaller number of pauper cases. Data on docket sze come from the Federal Judcal Center (

14 13 where I s an ndcator functon that s equal to 1 f the nsde arument s true and s equal to L C zero otherwse; D j s a lberal (or moderate) L deoloy and s a conservatve C deoloy dummy for justce j ; and X s a vector of exoenous nfluences on the case outcome dscussed prevously B s a vector of parameters and e s an error term. 18 The justces deolocal preferences toward a case captured by the parameters and would be expected to vary across cases because cases have dfferent characterstcs many of whch are unobserved. Gven that there are at most 9 votes for a case t s nfeasble for us to use fxed effects by nteractn case dummes wth deolocal dummes to control for the varaton n the deolocal preferences across cases. We therefore adopt a random-effects specfcaton and model and as two correlated random coeffcents; as noted later the jont dstrbuton of the two random coeffcents s allowed to be correlated wth the case selecton process because unobserved case characterstcs may affect both deolocal preferences and case selecton. Because we nclude an ntercept n the specfcaton the mean of s normalzed to zero and can be wrtten as where denotes a random term wth zero mean. Gven ths specfcaton the estmates of measure for a ven the averae pro-busness bas of lberal or moderate justces compared wth the pro-busness bas of conservatve justces. Selectvty Bas A probt or lot reresson to estmate equaton (1) suffers from sample-selecton bas because estmaton s based only on the cases that the Supreme Court areed to hear and nores D j 18 We ntally estmate separate dummes each justce and we then perform estmatons where we areate and specfy lberal/moderate and conservatve deoloy dummes for each justce. Our dervaton also holds for the frst approach.

15 14 the nformaton provded by the cases that the Court declned to hear. Let s denote the bnary ndcator for case selecton whch s equal to 1 f case s ranted a hearn by the Supreme Court and 0 otherwse. Thus the case selecton model under s specfed as: s I s * Zδ v 0 (2) where Z s a vector of observed exoenous nfluences on case selecton dscussed prevously (some of whch are not ncluded n X ) δ s a vector of parameters and ν s an error term. Formally the selectvty ssue arses because we can observe a case outcome yj only when s 1 : Pr * * yj 1s 1 Pryj 0 s 0. (3) The random coeffcents n equaton (1) and are lkely to be correlated wth the error term because unobserved case characterstcs that affect case selecton also affect how justces preferences are revealed n the merts-ruln stae of the process. Thus maxmzn the lo-lkelhood of the uncondtonal probablty random coeffcents leads to nconsstent parameter estmates. If we assume jont normalty of and such that μ as a probt or lot estmaton wth (4) where N 0 and v * ~ N 01 and f the latent varable yj s observable we can specfy ~ 22 the condtonal expectaton functon E Γ Pr yj * 0 θ v μ * L yj s 1 Dj XB D j θ Ev v Zδ (5)

16 where D j D D L j C j. It would appear that equaton (5) can be estmated to obtan consstent 15 parameter estmates by the usual two-step control-functon type approach (Heckman (1976)): frst estmate a probt selecton model to obtan the nverse-mlls rato and then reress * yj on the explanatory varables alon wth nteractons of the nverse-mlls rato and the deolocal dummes to control for the unobserved nfluences on selecton. However that approach cannot be appled here because we observe only the bnary case outcome y j. Gven the jont normalty assumpton n equaton (4) and the condton ven by equaton (3) we have P e e y j j j 1s v u I u I 1 L e D X B D v θ D u f u du f v v Z δ L e D X B D Ev v Z δ θ D u f u j j j j j j j j du de j dv de j (6) The last nequalty s the result of Jensen s Inequalty because the ndcator functon s nonlnear. Snce we cannot use the standard two-step approach to control for selectvty bas we nstead y derve the jont dstrbuton of the endoenous varables P s where y y1... y9 and we estmate the parameters n equaton (1) by Maxmum Smulated Lkelhood Estmaton (MSLE). Ths appears to be the frst applcaton to the best of our knowlede of a new methodoloy to correct for selectvty bas when both the selectvty and outcome equatons are dscrete. We extend ths applcaton to allow the selecton error to be correlated wth the random coeffcents n the outcome model.

17 16 We dentfy the case outcome model wth selectvty by assumn the dstrbuton of and s jont normal. 19 Droppn the superscrpt to smplfy the exposton yelds: 2 12 ~ MVN 0 MVN 0 MVN 0 Ω (7) e s assumed to have a lostc dstrbuton wth varance normalzed to 3. j To summarze then our econometrc model conssts of: two bnary response models a bnary probt model of petton selecton and a bnary mxed-lot model of pro-busness votn on cases selected by SCOTUS; a cross-model correlaton that s represented by equaton (7); correlatons amon dfferent justces votes on a case that are captured by random ndvdual effects and and ther correlaton; and justce deolocal preferences that nfluence case outcomes that are allowed to vary as ndcated by equaton (7) accordn to unobserved case characterstcs that affect both case selecton and case outcomes. Dervn the Data Lkelhood 1 Ω 12 Ω Ω 22 ; 19 Heckman and Sner (1984) propose a nonparametrc maxmum-lkelhood approach that uses a dscrete type dstrbuton for the random components n an econometrc model. Ther approach s dffcult to apply to our sample selecton model wth random coeffcents because we show n equaton (9) below that the data-lkelhood of the sample selecton model requres knowlede of both the jont dstrbuton and the condtonal and marnal dstrbutons of. Althouh we can use a common random component wth a dscrete type dstrbuton to capture the correlaton of t s not possble for us to derve marnal and condtonal dstrbutons from such a specfcaton. Thus assumn jont normalty provdes a tractable and plausble way to dentfy the sample selecton model.

18 17 After droppn the superscrpt we defne and. For any case n the Busness Ltaton Database the jont dstrbuton of and can be expressed as. (8) Note we assume that where ndcates the standard normal dstrbuton functon. The condtonal probablty s ven by j j j j j j j d f d f s y P d f d f s P d f s P d s f s P s P Z δ Γ Γ Θ D Γ W Z δ Γ Γ Θ D Γ W y Z δ Θ D W y Θ D W y Θ D W y Γ Γ (9) where ven our mxed lot specfcaton Θ D Γ W 1 j y j s P has a smple bnary lot functonal form. Accordn to the specfcaton (7) f Γ s a bvarate normal densty. The densty functon s a truncated normal that s truncated below (because ). When sample selecton s not correlated wth the case outcome equaton the nteraton n (9) does not depend on because. In that case the parameters n the case outcome equaton can be estmated consstently by maxmzn the lo of the lkelhood functon defned n (9). Estmaton Z X W B Ω δ Θ y y y s δ Z Θ D W y Θ D W y 1 1 j j s P s P s P Z δ δ Z s P 1 Θ D W y 1 j s P Z δ f δ Z 1 s f f Γ Γ

19 18 We combne the Busness Ltaton Database on case outcomes whch s denoted by wth the Bloomber data set on case pettons whch s denoted by to estmate the parameters n Θ. The lo of the data lkelhood takes the follown form: BNA BLD ln L Θ ln Py s 1 W D Θ ln Z δ BLD I s 1ln Zδ Is 0ln1 Zδ BNA j BLD (10) However as noted the parameters n the case selecton equaton can be estmated by an enrched sample that contans both and whch oversamples the share of busness cases that the BLD BNA Supreme Court selects to hear. We therefore take two approaches to estmaton. In the baselne estmatons we use only the Bloomber random sample to estmate the selecton parameters so that the baselne estmator s: Θˆ ar max ln L Θ Θ ln Py s 1 W D Θ BLD I s 1ln Zδ Is 0ln1 Zδ BNA j (11) As a robustness check we combne the Bloomber random sample wth the choce-based sample that we summarzed earler and use Mansk and Lerman s (1977) Wehted Exoenous Sample Maxmum Lkelhood Estmaton (WESML) as appled to jont models by McFadden Wnston and Boersch-Supan (1985) to estmate the parameters from ~ Θ ar max ln L Θ Θ ln Py s 1 W D Θ lnz δ BLD I s 1 1 ln Zδ Is 0 0 ln1 Zδ BNA j 1 BLD (12)

20 19 where % of cases ranted hearn n the Bloomber Sample 1 and 0 s defned smlarly to % of cases ranted hearn n the enrched Sample reweht the cases not selected by the Supreme Court. Ths estmator yelds consstent parameter estmates. 20 Estmaton of the model requres evaluatn the lo-lkelhood functon ven by equaton (11) or the pseudo-lo-lkelhood functon ven by equaton (12) at each teraton of the optmzaton procedure. 21 In those functons the condtonal probablty P y s 1 W D j Θ whch s ven n equaton (9) can be evaluated only numercally because the probablty nvolves nteratons wth respect to both Γ ven and tself. We use Monte-Carlo nteraton to approxmate the probablty where the Monte-Carlo approxmaton to equaton (9) s ven by P R R r r y W D Θ Py W Γ j j D j Θ R r 1 R r 1 j1 (13) r where represents the r th random draw of from the truncated normal dstrbuton TN δ01 Z r and Γ r represents the r th random draw of Γ from the bvarate normal dstrbuton condtonal on r N r Ω Ω Ω. R and R are the total number of random draws for and Γ respectvely Ω12 20 The WESML estmator s consstent but not effcent because the shares are determned from data nstead of ben estmated follown an approach developed by Cosslett (1981). However we found t computatonally dffcult to smultaneously estmate both the model parameters and the sampln wehts. 21 We use the term pseudo-lo-lkelhood functon because we are not takn Cosslett s (1981) approach to obtan effcent estmates by maxmum lkelhood estmaton. Vossmeyer (2015) suests a Bayesan approach that could be used to estmate the model.

21 20 The computatonal burden of evaluatn equaton (13) s consderable because of the nestn structure n the Monte-Carlo nteratons. To reduce computaton tme we used the Halton sequence (Tran (2009)) to smulate random draws when we mplemented the nteratons. The estmaton results we report are based on 150 Halton draws; we performed robustness checks on the base-lne results n table 4 by ncreasn the number of draws to 300 and we found that the results chaned very lttle. 4. Identfcaton We seek to estmate the jont probabltes of Supreme Court Justces votes on a case and whether a case s selected by SCOTUS. Because we cannot explot a natural or quas-natural experment to estmate the jont probabltes we dentfy the probabltes by excludn from the case-outcomes model predctors of petton selecton that are uncorrelated wth the justces rulns on the merts. We provde a careful nsttutonal justfcaton for the excluson restrctons that s nformed by the leal lterature and we perform robustness checks by estmatn alternatve models that relax some of those restrctons. Below we dscuss the factors that we assume are orthoonal to justce votn on the case merts. Monthly Calendar Dummes and Docket Sze Monthly calendar dummes and docket sze capture the mpact of chanes n the Justces workload on ther petton decsons. Durn the past several decades the volume of Supreme Court pettons has steadly rsen from about 2000 pettons n 1960 to a lttle under 9000 n 2012 (Federal Judcal Center (2012)). The task of processn thousands of pettons each year places a

22 21 clear and substantal burden on the Justces and ther law clerks and leal scholars have documented a varety of ways n whch petton volume has affected both the qualty of judcal revew and the composton of cases ultmately heard by the Supreme Court. 22 Partcularly relevant to our analyss are the fndns of Cordray and Cordray (2004) whch ndcate that cert-acceptance rates are dramatcally lower durn the summer recess and the months of February and March. Ths varaton cannot be explaned by systematc dfferences n case characterstcs because pettons must be fled wthn 90 days of the lower ruln and there s no reason to suspect that lower- decsons on more mertorous cases are more lkely to occur n certan months of the year. Furthermore once the appeal s fled the hhly structured and larely automated process of revewn and dsposn of pettons leaves lttle room for manpulaton (ether by the Justces or by the contestants nvolved) meann that the assnment of petton decsons to calendar months s larely a random process Qualtatve research for nstance suests that the Solctor General s Offce typcally drops a non-trval number of cert-worthy pettons based on ther perceptons of the Court s workload for that year (Grswold (1975)). 23 Once a petton s submtted the respondents have 30 days to fle a bref n opposton. If they choose to submt a bref n opposton the respondents can fle for an extenson (typcally another 30 days). If the respondents wave ther rht to fle a bref n opposton or f the 30-day response perod expres the petton s then crculated amon the Justces' chamber the next tme a conference lst of pettons oes out. New conference lsts are crculated on a weekly bass and the Court s scheduled to meet on pre-specfed dates to dscuss the pettons on each conference lst. If the respondents fle a bref n opposton the Court wats approxmately 10 days n order to allow the pettoner to fle a reply after whch tme the case s appended to the next conference lst. Occasonally when respondents fal to submt a bref n opposton the Justce wll nstruct the respondents to prepare and fle a bref n opposton wthn the next 30 days. On rare occason the Court wll also nvte the Solctor General to fle an amcus curae bref expressn the vews of the Unted States reardn whether the petton should be heard. Decsons on pettons may also be postponed untl the subsequent conference f Justces need more tme to consder the case. Such delays are rare however as the averae tme from submsson to dsposton of a petton s only sx weeks (Publc Informaton Offce of the Supreme Court of the Unted States (2014)). Nelson (2009) provdes more detals on the petton-revew process.

23 22 The observed monthly varaton n cert-acceptance rates appears to be drven by work pressures that arse from the structurn of the Court s calendar. For nstance the Justces workload peaks n February and March when they are scheduled to not only hear a full set of aruments but also to wrte ther opnons for all the cases heard n the frst four sessons of the term. At the same tme ncentves for rantn new pettons reach an annual low n February and March prmarly because January s the last sesson durn whch new cases can be added to the Court s docket for the current term and the Justces have ample tme to fll up next term s arument docket. Smlar nsttutonal forces can also explan the neatve downturn n rant rates durn the summer recess. 24 In sum the month dummes and docket sze varable prmarly capture the effects of exoenous resource constrants and dosyncrases n the Court s schedule on petton selecton. Clearly those nfluences do not affect the pro-busness votn tendences of justces on specfc cases and they are unlkely to be correlated wth unobserved nfluences on case-outcome votes. Top Advocate Dummy and Indcator for Whether the Solctor General s the Pettoner Court resource constrants also justfy our assumpton that the top advocate dummy and the ndcator of whether the Solctor General (SG) s the pettoner are purely exoenous nfluences on petton selecton. Because Justces must revew thousands of pettons each year 24 In partcular Cordray and Cordray (2004 p. 209) explan that Durn the summer recess the Court acts on almost no pettons for certorar and thus allows three months of flns to accumulate for dsposton en masse upon ther return n the fall. As a result n each of the nne terms we canvassed from 1996 to 2002 the Court dealt wth on averae more than 1600 pettons for certorar at the end of the summer recess as compared to an averae of 574 pettons durn each of the other perods. The authors hypothesze that the Justces may feel lke they are acceptn lare numbers of cases (even thouh ther acceptances rates are strknly low) or that the overwhelmn volume of cases has a numbn effect on the Justces and ther law clerks. A smlar overload effect may be contrbutn to the low rant rates n February. The authors note that the Justces consder about 50% more pettons n February than n the averae sesson an ncrease that they attrbute to the abnormally lon recess that precedes the February sesson.

24 23 they have lmted tme to devote to any ven petton. 25 Consequently Justces look to a varety of ndcators to determne the mportance of a petton and whether t should be approved. One marker of qualty and leal snfcance s whether the petton was submtted by the Solctor General SG (Cordray and Cordray (2008)). Due to the nature of the SG s clent (.e. the Unted States) wrts for certorar from the SG s Offce enerally mert the attenton of the Supreme Court. In addton as ponted by Cooper (1990) the Justces have confdence n the qualty of those pettons because the SG s Offce has a reputaton for rorously screenn and vettn ts pettons and because the Solctor General s a repeat player before the Court. Days (1994) provdes a hstory of the Offce and stresses that t enjoys a tradton of ndependence and that the Solctor General s not a hred un but s ultmately responsble for advancn the nterests of the Unted States whch may on occason conflct wth a overnmental entty. A smlar arument helps explan why SCOTUS rants pettons fled by certan seasoned Supreme Court advocates (many of whom were formerly employed n the SG s Offce) at a snfcantly hher rate than they rant pettons fled by other lawyers. Lazarus (2008) reports that the Justces clerks pay specal attenton to the pettons fled by promnent Supreme Court advocates because ther reputatons as top ltators convey an mportant messae about the snfcance of the leal ssues ben presented and the credblty of the assertons ben made. Thus much n the same way that the Solctor General s pettons carry an experence-based stamp of qualty assurance the name of a top Supreme Court advocate on a petton acts as a beacon of potental cert-worthness that helps dstnush t from the thousands of other competn appeals. 25 Hart (1959) for example estmated that SCOTUS Justces n the late 1950s could spend at most twenty mnutes revewn each nonfrvolous petton whle Chef Justce Huhes reported that hs Justces were unable to spend more than four mnutes dscussn each petton at conference (Cordray and Cordray (2004)). Fnally Lazarus (2008) explans that the Justces do not read a lare porton of pettons and at most read the memoranda prepared by the clerks.

25 24 Of course t s possble that the hh acceptance rates for SG and top-advocate pettons are drven n part by poltcal consderatons. If for example the Court has a pro-busness orentaton but the current U.S. admnstraton s enactn more prvate-sector reulatons then the Justces may actvely seek out SG pettons to lmt the scope of overnment reulaton. However because we nclude n our case-outcomes model an ndcator for whether the federal overnment was a party n the case we larely control for potental correlatons between the SGpettoner dummy n our petton-selecton model and the error term n our case-outcomes model. 26 Indcator for whether the Court Invted the Solctor General to Submt a Bref Analyzn the Petton and Expressn the Vews of the Unted States On rare occasons the Court wll solct the perspectve of the SG s Offce on the certworthness of a petton throuh a call for the vews of the Solctor General (CVSG). Extant research suests that the CVSG s another nformaton-athern tool that the Court uses to take advantae of the SG s ndependence and objectvty (Days (1994) and unmatched expertse n federal polcy and complex reulatory matters. For example Thompson and Wachtell (2009) fnd that 91% of the CVSG cases revewed between 1997 and 2004 nvolved complex reulatory and statutory schemes whle only 7% nvolved snfcant ssues of consttutonal law. Ths statstcal breakdown s consstent wth the Justces competence n dealn wth questons of consttutonalty ther relatve unfamlarty wth reulatory systems and the SG s obvous comparatve advantae n advsn the Court on ssues related to federal reulaton. The authors also observe that althouh the SG s response to a CVSG s almost always accompaned wth a 26 When we ncluded the top Supreme Court advocate dummy n the case outcomes model ts coeffcent was statstcally nsnfcant and the estmates of the model s other coeffcents were barely affected.

26 25 sueston for how the Court should rule on the case (assumn the petton s accepted) the Court s ultmate decson on the merts s uncorrelated wth the SG s recommendatons. Fnally Thompson and Wachtell pont to a plethora of anecdotal evdence that decouples the CSVG from any poltcal motvatons further supportn our treatment of the CSVG ndcator as exoenous to the fnal votes n our case-outcomes model. 27 Lower- Dummes It s well-known that the Supreme Court s partcularly prone to reverse the rulns of certan appellate s most notably the Nnth Crcut and the U.S. Court of Appeals for the Federal Crcut on patent dsputes. 28 Promnent explanatons for hh reversal rates nclude deolocal dfferences between lower- judes and Supreme Court Justces as well as varaton n crcut sze. 29 Notable dfferences n the subject matter and the number of cases dsposed by crcut s have also been documented (Hofer (2010)) and ths varaton may account for part of the varaton n cert-approval rates across appellate s. 27 Specfcally Thompson and Wachtell quote Justce Gnsber as sayn: The Solctor acts as a true frend of the Court when we call for ts vews on a case n whch the Unted States s not a party. They also pont to the follown quote from former Solctor General Lee: The Solctor General s offce provdes the Court from one admnstraton to another wth advocacy whch s more objectve more dspassonate more competent and more respectful of the Court as an nsttuton than t ets from any other lawyer or roup of lawyers. 28 The Nnth Crcut has been called the roue crcut due to the hh reversal rates of ts rulns by the Supreme Court. In addton Tony Mauro Key Patent Dspute Dvdes Supreme Court Justces Natonal Law Journal October reports that SCOTUS has an unspoken dstrust of the U.S. Court of Appeals for the Federal Crcut as the ultmate arbter of patent dsputes because ths crcut frequently reverses tral jude fndns; thus SCOTUS steps n to reset the balance. 29 Larer crcuts tend to have reater dffcultes n mantann unform law (aka consstency n decsons across judes) wthn the crcut whch would account for the observed postve correlaton between crcut sze and SCOTUS reversals of lower- rulns (Scott (2006)).

27 26 Althouh varaton n cert-ratn rates may n part be due to the polcy preferences of SCOTUS justces t s mportant to note that we control for whether the lower s ruln on a case was pro-busness n our case-outcomes model. Ths n turn enables the lower- dummes n our petton-selecton model to capture cert-rantn factors that are unrelated to the probusness nclnatons of the Supreme Court Justces. As a robustness check however we nclude the lower- dummes n an alternatve specfcaton of our case-outcomes model Estmaton Results Descrptve nformaton about the votn behavors of the Supreme Court justces provdes some ntal nshts nto ther deolocal preferences. In table 1 we break down the Rehnqust and Roberts Court justces votes on the 198 cases n the Busness Ltant Database that were ranted a hearn wth the Supreme Court between 1996 and There s a clear deolocal dvde n the Rehnqust Court justces preferences: the share of lberal justces (Souter Stevens Gnsber Breyer) votes that are pro-busness s almost unform at 37 percent whle the share of moderate and conservatve justces (Kennedy Scala Thomas O Connor Rehnqust) votes that are pro-busness s at least 48 percent. The deolocal dvde rows n the Roberts Court wth 30 It s worth notn that the petton-selecton model does not control for crcut splts whch occur when appellate s enerate conflctn rulns on a partcular leal ssue. Althouh the Supreme Court s substantally more lkely to rant cert n cases that create crcut splts data on crcut splts are sparse larely because dentfyn such cases would requre readn and comparn thousands of appellate judcal opnons. However the absence of a crcut splt varable does not pose a problem for our model because the lower-case dummes broadly capture a crcut s tendency to dvere from other appellate s on questons of law. For example the hh reversal rates of Nnth Crcut holdns partally reflect the Nnth Crcut s hehtened probablty of splttn wth other appellate s on leal ssues. In addton because nearly all cert-worthy cases nvolve crcut splts crcut splts are unlkely to be correlated n a meannful way wth Supreme Court rulns on the merts. Thus omttn crcut splts from the petton model should not affect the consstency of the estmates n the model.