We develop an informational model of judicial decision-making in which deference to precedent

Transcription

1 American Poliical Science Review Vol. 96, No. 4 December 2002 Informaive Preceden and Inrajudicial Communicaion ETHAN BUENO DE MESQUITA AND MATTHEW STEPHENSON Harvard Universiy We develop an informaional model of judicial decision-making in which deference o preceden is useful o policy-oriened appellae judges because i improves he accuracy wih which hey can communicae legal rules o rial judges. Our simple model yields new implicaions and hypoheses regarding condiions under which judges will mainain or break wih preceden, he consraining effec ha preceden has on judicial decision-making, he voing behavior of Supreme Cour Jusices, he relaionship beween a preceden s age and is auhoriy, he effec of legal complexiy on he level of deference o preceden, he relaive sabiliy of rules and sandards, and long-erm paerns of legal evoluion. Perhaps mos imporanly, we demonsrae ha legalis feaures of judicial decision-making are consisen wih an assumpion of policy-oriened judges. P oliical scieniss have long recognized he imporance of cours as poliical acors. However, while an exensive lieraure examines he judiciary s sraegic ineracion wih he oher branches of governmen (e.g., Ferejohn and Weingas 1992; Gely and Spiller 1992), less aenion has been paid o he effecs of he insiuional srucure of he cours hemselves on paerns of judicial decision-making. Ye models of he judiciary s unique insiuional dynamics are essenial o undersanding he cours, jus as analyses of congressional commiees (e.g., Gilligan and Krehbiel 1990; Shepsle and Weingas 1987) or he bureaucracy (e.g., Moe 1982) are essenial o undersanding he legislaive and execuive branches. We develop a formal model ha demonsraes how he problem of communicaion beween differen levels of a hierarchical cour sysem, such as he one in he Unied Saes, can, under some condiions, give judges incenives o defer o precedens esablished in prior cases and how his consraining effec influences he policy decisions made by he cours. This approach sheds ligh on one of he mos perplexing and conroversial problems in he social-scienific sudy of judicial behavior: he debae over he relaive imporance of legal and policy concerns in judicial decisions. Scholars subscribing o policy-oriened models of judicial behavior argue ha judges are concerned wih he exernal effecs of heir rulings on allocaions of risk, wealh, power, or opporuniy. 1 Judges may be Ehan Bueno de Mesquia is a Ph.D. candidae, Deparmen of Governmen, Harvard Universiy, Liauer Cener Norh Yard, Cambridge, MA (mesquia@fas.harvard.edu). Mahew Sephenson is a Ph.D. candidae, Deparmen of Governmen, Harvard Universiy, and J.D. candidae, Harvard Law School, Cambridge, MA (msephen@fas.harvard.edu). We are indebed o Sco Ashworh, Bruce Bueno de Mesquia, Charles Cohen, Eric Dickson, Gilles Serra, Ken Shepsle, Judge Sephen Williams, several anonymous referees, and he paricipans in he Harvard Raional Choice Lunch Group, he Harvard Poliical Economy Research Workshop, he Texas A&M Conference on Insiuional Games and he U.S. Supreme Cour, and he William H. Riker Conference on Consiuions, Voing, and Democracy a Washingon Universiy for valuable commens. Sephenson graefully acknowledges suppor from he John M. Olin Cener for Law, Economics, and Business. Bueno de Mesquia graefully acknowledges suppor from he Cener for Basic Research in he Social Sciences. moivaed by policy concerns because of parisan loyalies, a sincere desire o effec paricular changes in he world, or pursui of promoion or reelecion. Whaever he reason, he policy-oriened judge cares abou acual judicial oupus more han any paricular mehod of arriving a hose oupus. Ohers, however, argue ha judges are concerned wih legalism, ha is, wih correcly following he rules and norms of proper judicial reasoning. A legalis judge maximizes uiliy by adhering faihfully o hese inernal rules, regardless of he exernal resul. The purely policy-oriened judge and he purely legalis judge are of course ideal ypes, and few scholars, if any, believe ha judges are moivaed solely by concern wih exernal effecs or by fideliy o inernal norms. Noneheless, hese wo ses of facors are ofen presened as compeing explanaions for judicial behavior, and heir relaive imporance is he subjec of heaed debae (e.g., Knigh and Epsein 1996; Segal and Spaeh 1996a, b; Songer and Lindquis 1996). We conribue o he discussion of he naure of judicial decision-making by demonsraing ha legalis principles are, a leas in some cases, compaible wih and in fac explained by judges concerns wih he exernal policy effecs of heir rulings. Focusing on one cenral principle of judicial decision-making in Anglo- American legal sysems sare decisis, or deference o preceden we show ha purely policy-oriened judges will ofen defer o legal preceden, even when doing so requires hem o issue decisions ha deviae from he rulings hey oherwise would prefer. The reason for his is ha appellae judges can use prior cases o increase he accuracy of heir communicaion wih rial judges. Ofen, a judge is willing o modify he subsanive ruling in order o purchase his increased accuracy in communicaion. Thus, in he model we presen, i is no he case ha policy-oriened judges ignore preceden, nor is i he case ha judges care abou preceden insead of, or in addiion o, caring abou policy. Raher, judges care abou preceden because hey care abou policy. 1 Boh he aiudinal model (Segal and Spaeh 1993) and he sraegic model (Knigh and Epsein 1998) of judicial behavior are policy-oriened, in ha boh models assume judges are primarily concerned wih subsanive oucomes. 755

2 Informaive Preceden and Inrajudicial Communicaion December 2002 In addiion o demonsraing his basic idea, our model has he advanage of being able o accoun sysemaically for boh adherence o and deparure from preceden. Specifically, our comparaive saic analysis shows how variaion in four parameers he disance beween he exising legal rule and he deciding judge s ideal, he age of he exising preceden, he difficuly of inelligibly inegraing exising preceden wih new rulings, and he precision or imprecision of communicaion beween judges affecs he relaive likelihoods ha a judge will adhere o or break wih exising legal preceden. Our model also has implicaions for how much judges are able o change he subsanive law wihou ever openly breaking from esablished preceden. Furher, our informaional model of sare decisis sheds ligh on oher imporan empirical puzzles in he sudy of judicial decision-making. We reconcile he seemingly conradicory observaions ha argumens from preceden play a major role in U.S. Supreme Cour deliberaion and adjudicaion, and ye jusices consisenly voe heir preferences raher han following esablished precedens. Our approach also offers an explanaion for why long lines of cases migh evoke boh deference and skepicism. Finally, our model yields several novel hypoheses, including predicions regarding he ypes of legal issues in which long lines of preceden will emerge, he relaive sabiliy of clear rules versus flexible sandards, and paerns of longerm legal evoluion. COMPETING THEORIES OF JUDICIAL DECISION-MAKING We focus on he principle of sare decisis, 2 which dicaes ha judges ough o apply rules and principles laid down in prior cases, because i is, or a leas is claimed o be, one of he mos imporan principles of judicial decision-making in he Anglo-American common law sysem (e.g., Schauer 1987). Judges are no, in ordinary circumsances, supposed o overurn seled law (Nelson 2001). 3 Our focus on sare decisis is also due o is salience for he more general quesion of he naure of judicial preferences. Judicial deference o esablished preceden is a focus of he debae beween he policy-oriened and he legalis models of judicial decision-making because hese models seem o offer such differen predicions for how imporan preceden will be in pracice. A legalis judge is expeced o place grea weigh on he sare decisis principle and consequenly is expeced o defer o prior decisions even when ha judge would have decided he precedenseing case differenly. 4 In conras, i is no clear why a policy-oriened judge would ever defer o consrains imposed by prior decisions. I may be ha judges ofen agree wih he principles laid down in old cases and follow hem for ha reason, bu if his is he case, hen sare decisis is merely a descripion of raher han a reason for paerns of judicial decision-making. To he exen ha a legal preceden exers a causal influence on a leas some judicial decisions, i mus be he case ha in hese decisions judges would prefer o issue a differen ruling if he preceden did no exis. Thus, a policy-oriened judge would be expeced o aach lile imporance o esablished preceden when making decisions. Scholars have offered a number of reasons why even a policy-oriened judge migh wan o respec esablished legal preceden. One argumen he one mos familiar o lawyers is ha sabiliy in he law is in iself a valued policy goal, and judges would herefore be willing o defer o an esablished legal rule because he ac of deference iself advances heir policy preference for sabiliy. Bu his explanaion has a difficul ime accouning for adherence o preceden in areas of law where sabiliy and he need for long-erm planning are less salien. Perhaps more imporanly, sabiliy in he law is a collecive good; for a given judge o sacrifice oher policy goals for he sake of sabiliy, ha judge mus believe ha oher judges will also value sabiliy sufficienly highly ha hey will no overurn preceden. Bu if a judge believes ha oher judges do place a high value on sabiliy, ha judge may be emped o break wih preceden and esablish a new legal rule, since i will be respeced by fuure cours wih lile overall loss in legal sabiliy. Addiionally, he value of legal sabiliy o a judge is considerably reduced if he sable legal rule is objecionable (Kornhauser 1989). While he need for sabiliy may be enough when he judge is indifferen beween legal rules, i is a less plausible basis for deference o preceden when he judge has srong subsanive preferences beween rules. Anoher suggesion as o why policy-oriened judges migh respec he principle of sare decisis is ha judges wan heir own precedens followed and, herefore, follow precedens se by ohers (Landes and Posner 1976; Rasmusen 1994). Because judges prefer a world in which all precedens, including heir own, are respeced o a world in which no precedens are respeced, hey will aemp o enforce universal respec for preceden by punishing judges who defec. 5 While here may be some of his i-for-a in he real world, his 2 Sare decisis e non quiea movere Sand by he hing decided and do no disurb he calm. 3 One aspec of his principle is he idea ha lower cours are supposed o follow he precedens se by higher cours (verical sare decisis). This is no much differen from he principle in many hierarchical organizaions ha subordinae unis are supposed o follow he direcions and guidelines laid down by heir superiors. A more ineresing aspec of he sare decisis principle and one more unique o judicial decision-making is he principle ha cours are supposed o follow heir own prior decisions (horizonal sare decisis). 4 This migh no always be he case since here migh be some oher legal principle, e.g., he proper inerpreaion of a saue, ha could rump he principle of deference o decided cases even for a legalis judge. Noneheless, as a general rule i is safe o assume ha legalis judges in a common law sysem would aach subsanial weigh o preceden. 5 If here is no specific, argeed realiaion agains paricular judges for failing o follow preceden, he general judicial ineres in preceden-following alone is insufficien because respec for precedens is a collecive good and individual judges dominan sraegy would be o free-ride (Macey 1989). 756

3 American Poliical Science Review Vol. 96, No. 4 explanaion has several problems as a general accoun of he pracice of sare decisis. Firs, i relies on he empirically dubious assumpion ha judges look o oher judges respec for legal preceden when deermining wheher o follow preceden se by hose judges. Second, i has rouble accouning for why he precedens of reired judges are ever followed wihou reference o even more complicaed, and empirically problemaic, punishmen mechanisms. Third, in he absence of such mechanisms, his explanaion also does no explain well why some judges break wih some precedens bu follow ohers. In a simple model where judges expec general realiaion for any break wih exising preceden, if hey break once, hey have no reason no o break always. A hird possibiliy is ha policy-oriened judges do no care abou preceden per se bu recognize he need o preserve heir insiuional power and legiimacy. Because his legiimacy derives in par from a public belief ha judges apply a specialized se of legal skills, including he abiliy o inerpre and apply esablished legal precedens, as well as a belief ha judges are poliicallyneural inerpreers of law, policy-oriened judges will wan o perpeuae he belief ha hey make decisions according o preceden (Cox 1976; Malz 1980). This may in urn lead hem o modify heir decisions someimes o show more respec for preceden in order o enhance heir insiuional power. However, he collecive acion objecion applies o his explanaion as well; he overall effec of any paricular decision on insiuional legiimacy is likely o be small, while he policy ramificaions of he case iself ofen are large. Moreover, he effec of a decision on public percepion of he cour s legiimacy may have more o do wih he conen of he decision han wheher i involved overurning a preceden (e.g., Hyde 1983; Nelson 2001). Finally, several sudies poin ou ha deference o preceden may be valuable even o policy-oriened judges because of he informaional funcion ha judicial precedens serve (e.g., Rehnquis 1986; Shapiro 1972). The informaional perspecive comprises wo ypes of explanaions for deference o preceden. Firs, reasoning from precedens may improve communicaion beween appeals cours, allowing for judicial specializaion and error correcion (e.g., Kornhauser 1989; Macey 1989). A more pessimisic version of he same basic argumen is ha he pracice of sare decisis is essenially an informaion cascade in which raional agens ignore heir own informaion and imiae he behavior of preceding decision-makers, ofen leading o uncorreced inefficien resuls (Talley 1999). We sugges a second informaional funcion ha he docrine of sare decisis migh serve. This funcion involves communicaion beween high cours and lower cours. The basic idea is ha he developmen of lines of cases can communicae a legal principle beer han any individual case could. An iniial case may invoke a general phrase or principle, such as due process, reasonable, or compelling ineres ; fuure cases develop and give meaning o hese inherenly vague phrases. Hence, a lower cour can learn more abou he appellae cour s view of he proper inerpreaion of, say, a due process balancing es by examining a line of 10 cases in which he same es was applied han by reading he firs (or he las) ruling he appeals cour issued. Similarly, an iniial case migh declare a brigh-line rule ha, hough clear, is boh over- and underinclusive. 6 Furher cases can carve ou excepions and make qualificaions so ha he line of cases applying he rule offers lower cours a much more nuanced es han ha announced by he original decision. Our model explores his second ype of informaional use of legal preceden, a use ha we believe has been negleced in he scholarly discussion of judicial decision-making and ha has no been formally developed or rigorously analyzed. In so doing, we provide an accoun of judicial decision-making ha explains why policy-oriened judges are expeced o be influenced by preceden, while also capuring cases where judges break from preceden. THE MODEL Consider a simple judicial sysem consising of an appellae judge and a rial judge. The appellae judge hears a limied number of cases, and, hrough he decisions in hese cases, he appellae judge can announce how he law ough o be inerpreed. The rial judge decides he vas majoriy of cases. We make he simplifying assumpion ha when making decisions he rial judge aemps o apply esablished law, wihou reference o he rial judge s own personal policy preferences. There are several possible subsanive defenses of his assumpion ha could ulimaely be modeled explicily. These include consideraions such as promoion being dependen on sric adherence o he law esablished by appellae decisions, a desire no o be overurned on appeal, and oher facors. For he purposes of his model, i suffices o assume ha some mechanism exiss ha leads he rial judge o behave as a faihful agen of he appellae judge. While we assume a nonsraegic rial judge, we do address he condiions under which a sraegic rial judge migh behave in his way laer in he paper. The rial judge s undersanding of appellae rulings is imperfec, and he rial judge ofen does no decide cases exacly as he appellae judge would have. The appellae judge canno correc all hese misakes by alering or reversing he rial cour s decision on appeal. This may be because he appellae cour s jurisdicion is discreionary and many appeals are no heard (e.g., U.S. Supreme Cour review of circui cour decisions) or because many cases are no appealed despie he exisence of an auomaic righ o an appeal (e.g., federal circui cour review of disric cour decisions). Thus, he appellae cour judge, o influence he applicaion of he law o he majoriy of cases successfully, needs o communicae a preferred inerpreaion of he law o he rial cour as accuraely as possible, subjec o ime and resource consrains. 6 A brigh line rule is a rule ha minimizes ambiguiy by seing welldefined and simple caegories of prohibied and permissible behavior. 757

4 Informaive Preceden and Inrajudicial Communicaion December 2002 The appellae judge has preferences over he legal rule defined on a unidimensional coninuum. 7 Denoe he appellae judge s ideal poin j R. The policy coninuum migh represen, for example, he level of care exercised by a defendan in a civil acion, wih j represening he minimum amoun of care he appellae judge believes a defendan mus exercise o avoid liabiliy. Tha is, he appellae judge in his example believes ha all defendans who exercise a level of care less han j ough o be liable, and all defendans who exercise a level of care above j ough o escape liabiliy. As such, j is he appellae judge s ideal legal rule. The rial judge has no preference over he policy dimension per se; raher, he rial judge ries o implemen whaever he appellae judge has declared o be he proper legal rule. When deciding a case, he appellae judge issues a ruling wih wo componens. Firs, he appellae judge announces a subsanive holding on he proper applicaion of he legal rule o he case a hand. This subsanive holding corresponds o a poin in he policy space, denoed r R. Second, he appellae judge declares wheher his subsanive holding is consisen wih exising preceden or wheher he appellae judge is breaking wih preceden. Preceden here means he line of appellae cases on he relevan legal issue ha have been decided prior o he presen case and ha have never been overruled by a subsequen appellae case. Prior decisions are no assumed all o have been made by he curren appellae judge. Raher, one can hink of he game beginning wih an exising line of preceden ha may have been esablished before he curren appellae judge ook office. If he appellae judge issues a holding and declares ha i is consisen wih preceden, hen he case is added o he relevan line of cases, and he rial judge will inerpre he appellae judge s subsanive holding in he conex of he oher subsanive holdings in he line o ascerain he legal rule he appellae judge wishes o enac. If, however, he appellae judge declares a break wih preceden, hen a new line of cases is esablished, and he rial judge will rea he mos recen appellae decision as he exclusive saemen of he legal rule. 8 The number of cases in he line of preceden, including he decision being made in he curren round, is denoed. Thus a line of preceden is a series of subsanive rulings (r 1, r 2,...,r 1 ), where r 1 is he mos recen ruling and he curren appellae judge issues ruling r. The declaraion ha he appellae judge is mainaining or breaking from preceden is communicaed perfecly o he rial judge. However, he communicaion of he subsanive holding of each ruling is inher- 7 We refer o he judge s preferences over his coninuum as policy preferences; however, we use policy broadly, o reflec public-policy preferences, normaive judgmens regarding fairness or jusice, or a weighed combinaion of various facors. 8 The discree binary choice beween mainaining and breaking wih preceden is a simplifying assumpion. Clearly, in he real world, muliple lines of preceden may be in play in any given decision and he judge may decide o mainain pars of exising preceden. Framing he issue in he manner ha we do capures in sarkes form he same essenial decision problem ha would arise in more complex adjudicaive choices. enly noisy. Thus he rial judge, when aemping o undersand an appellae ruling, observes signals, each drawn from a normal disribuion wih variance σ 2.The means of hese disribuions are he rs associaed wih he subsanive holdings of he differen cases in he line of preceden. The rial cour judge averages hese signals o esimae he legal rule. 9 Thus, he rial judge s esimae of he legal rule is a sum of a series of normally disribued random variables. Such a sum is iself a normally disribued random variable. 10 Consequenly, he rial cour s esimae can be reaed as if i were a single signal, denoed x, drawn from a normal disribuion wih mean µ = ( i=1 r )/ and variance σ 2 / (DeGroo 1970). I is imporan o sress ha he rial judge does no acually believe ha all of he observed signals are drawn from a single disribuion. If he rial judge did believe his, signals far from he esimaed mean would make he rial judge less confiden in his esimae. However, he rial judge recognizes ha he various observed signals are drawn from differen disribuions wih differen means, reflecing earlier decisions, poenially made by differen appellae judges. Given ha he rial judge aggregaes hese signals, each addiional signal, no maer how far i is from he mean, decreases he variance of he rial judge s esimae of he legal rule o be implemened. This variance is he variance of he rial judge s esimae, no he rial judge s esimae of he variance of a single underlying disribuion. If he appellae judge breaks wih preceden, here is only one decision in he line of cases, so = 1 and he rial judge receives a signal drawn from a normal disribuion wih mean r =1, he mos recen appellae holding, and variance σ 2. The mean of he disribuion 9 The rial judge could, in heory, aggregae hese signals in some oher way (for example, by aking a weighed average), and his would in urn change he opimal decision of he appellae judge. We focus aenion on a simple average for he following reasons. Firs, including he possibiliy of a weighed average would no aler he appellae judge s basic decision problem he rade-off beween greaer precision (if he appellae judge s holding is aggregaed wih hose of oher judges) and he opporuniy o announce he judge s mos preferred legal rule (if he rial judge looks only a he appellae judge s ruling). The unweighed average is hus he simples among a class of similar aggregaion rules, any of which migh be reasonable depending on oher background insiuional assumpions. Second, calculaing he opimal weighed average is iself a complex problem ha depends on he appellae judge s deriving he opimal weighs as a funcion of he appellae judge s preferences and exising preceden. Third, no only is his calculaion complex for he appellae judge, bu he proper weighs, once calculaed, would have o be communicaed o he rial judge. This communicaion, like he communicaion of he subsanive holding, would be inherenly noisy. Thus, he calculaion and ransmission of appropriae weighs essenially, a more refined and sophisicaed principle of sare decisis inroduces anoher level of complexiy ha, hough ineresing, is beyond he scope of his paper and is herefore lef o fuure research. 10 In paricular, if X 1,...,X n are independen random variables and X i N(µ i,σi 2), i=1,...,n and a 1,...,a n are consans such ha a i 0 for a leas one i, hen he random variable a 1 X 1 + +a n X n is normally disribued wih mean a 1 µ 1 + +a n µ n and variance a1 2σ a2 n σ2 n. (DeGroo 1970, 38). In our case, he rial judge is aking an average, a i = (1/), µ i = r i, and σi 2 = σ 2 for i = 1,...,. Consequenly he average of he signals is iself a random variable wih mean µ = ( i=1 r )/ and variance σ 2 /. 758

5 American Poliical Science Review Vol. 96, No. 4 of he signal in he period prior o he curren appellae judge s decision is denoed µ 1, and we refer o his value as exising preceden. Because x (he value of he draw from he disribuion of he rial cour s esimae) is he only informaion ha he rial judge has, i is he rial judge s bes guess as o wha he appellae judge wans. Consequenly he rial judge will rea x as he conrolling legal rule when deciding cases. The appellae judge s uiliy has wo componens. Firs, he appellae judge would like he decisions of he rial cour judges o be as close as possible o he appellae judge s own ideal poin. Tha is, he appellae judge wishes o minimize x j. The reason for his is ha any cases ha fall in his inerval are cases ha he rial judge will ge wrong from he appellae judge s poin of view. Again, consider a case in which he policy dimension represens he level of care aken by a or defendan. If he defendan exercised a level of care less han he minimum of x and j, he rial cour will correcly find he defendan liable. Similarly, if he defendan exercised a level of care above he maximum of x and j, he rial cour will correcly find he defendan no liable. However, when he defendan exercised a level of care in he inerval beween x and j, he rial cour will rule incorrecly. In his even, if x > j, he rial cour will misakenly find liabiliy, while if x < j, he rial cour will misakenly find no liabiliy. The larger he size of he inerval x j, he larger he number of cases ha will be decided incorrecly. We assume ha he appellae judge s uiliy funcion is quadraic ha is, he appellae judge wans o minimize he expeced value of he square of he disance beween x and j, which is equal o ((µ j) 2 + (σ 2 /)). Second, if he appellae judge does no announce a break wih preceden, i is cosly o offer a ruling, r, ha is subsanively differen from he exising preceden, µ 1. We are agnosic as o he relaive imporance of hese wo facors in he appellae judge s uiliy calculaion; we weigh he imporance of he laer by he parameer α 0 on which we perform comparaive saic analysis. 11 Thus, he uiliy cos associaed wih changing he law while claiming o adhere o preceden is α(µ 1 r ) 2. If he appellae judge breaks wih preceden, his cos does no apply. This cos arises because wriing an opinion ha inelligibly inegraes exising preceden wih a change in he subsance of he legal rule becomes increasingly difficul as he disance beween preceden and he subsanive holding grows. Thus, here are real coss in inellecual effor and research associaed wih such a decision. If hese coss are no invesed, ha is, if he appellae judge were o declare ha a decision was consisen wih preceden wihou explaining how he 11 If α>1, his implies ha he cos of wriing he decision is weighed more heavily han how close he rial cour s decision is o he appellae cour s ideal poin. If 0 α<1, he closeness of he rial cour decision o he appellae judge s ideal poin is weighed more heavily. If α = 0, adherence o preceden is cosless no maer how much he legal rule is changed subsanively; in his special case, he appellae judge would always claim o adhere o preceden, because doing so imposes no consrains whasoever on he disance he appellae judge can move he legal rule. subsance of he ruling and prior rulings could be inelligibly inegraed, hen he rial judge would be unable o make decisions wih reference o he full line of cases. This aspec of he model capures he inuiion ha rial judges would have difficuly simply averaging signals ha are quie disan from one anoher. Trial judges are able o aggregae such signals only if he appellae judge has invesed subsanially in explaining how o do so. This effor is cosly o he appellae judge and becomes more cosly he greaer he disance beween he appellae holding and exising preceden, reflecing he increasing difficuly of he rial judge s ask. In our model he inuiion ha i would be difficul for he rial judge o aggregae a sharply divergen holding ino exising preceden is capured by he cosliness o he appellae judge of making such a holding. One migh objec o he assumpion ha judges bear a cos when mainaining preceden bu do no bear a cos when breaking wih preceden. However, he assumpion is benign wih respec o he conclusions of our model. Our analyic resuls are unchanged by he addiion of a fixed cos for breaking wih preceden. Such a cos would change he acual poin a which appellae judges break bu would no affec he comparaive saics. There is a second, more concepual jusificaion for making his assumpion. Any cos from breaking wih preceden, beyond he loss of informaion already modeled, would arise from legalis values. We omi legalis values from he judicial uiliy funcion o see wheher he norm of sare decisis is consisen even wih purely policy-oriened judges and, indeed, wheher wha appear o be legalis values may emerge as a resul of policy moivaions. The expeced uiliy o he appellae judge is herefore ) EU = ((µ j) 2 + σ2 α(µ 1 r ) 2, where α = α if he appellae judge mainains preceden and α = 0 if he appellae judge depars from preceden. ANALYSIS The Appellae Judge s Decision The appellae judge has wo choices o make. The appellae judge mus selec a subsanive holding (r ) and mus also decide wheher o declare a break wih preceden. These wo decisions are inerrelaed. To deermine wheher or no o break wih preceden, he appellae judge mus firs know he expeced oucomes, which are a funcion of he differen subsanive rulings ha he appellae judge would issue depending on wheher or no preceden is mainained. Thus, we work backward, firs deermining he appellae judge s opimal subsanive holding coningen on wheher he judge has broken wih or mainained preceden. If he appellae judge breaks wih preceden, he second erm in he judge s expeced uiliy funcion is zero, because α = 0. Furher, because he appellae judge has broken wih exising preceden, revers o 1. Therefore, he line of cases considered by he rial judge 759

6 Informaive Preceden and Inrajudicial Communicaion December 2002 conains only he appellae judge s mos recen decision, ha is, µ =1 = r =1. I is obvious ha he appellae judge s expeced uiliy in his case is maximized when r = µ = j. Thus, if an appellae judge breaks wih preceden, he subsanive holding will be he judge s ideal legal rule, and he judge s expeced uiliy in his case, EU(break), is σ 2. If he appellae judge does no break wih preceden, he decision is more complicaed. The appellae judge would like o issue a ruling ha moves preceden oward he judge s own ideal poin bu is consrained by he cos of deviaing oo much from he line of preceden of which his decision becomes a par. The appellae judge hus chooses r by solving he following maximizaion problem: [(µ j) 2 + σ 2 max r { ] α(µ 1 r ) 2 }. (1) Noe ha µ, he mean of he disribuion of he rial judge s esimae of he legal rule, is a funcion of r. In paricular, µ = [( 1)µ 1 + r ]/. Thus, we can rewrie he maximizaion problem { [ (( ) ] 1)µ 1 + r 2 max j + σ 2 r α(µ 1 r ) 2 }. (2) Solving for he firs-order condiion yields r = µ 1 + ( j µ 1). 1 + α 2 Knowing he opimal subsanive holding, r, allows us o calculae he mean of he new disribuion from which he rial cour judge will draw an esimae of he proper legal rule. This new mean, µ, is given by µ = ( 1)µ 1 + r = j + α 2 µ α 2. (3) Now, o find he appellae judge s expeced uiliy from issuing a decision consisen wih preceden, EU(mainain), we subsiue r and µ ino he expeced uiliy funcion. This yields [ (j+α 2 ) ] µ 2 1 EU(mainain) = j + σ2 1+α 2 [ ( α µ 1 µ 1 + (j µ )] 2 1), 1+α 2 which simplifies o EU(mainain) = α2 1+α 2(j µ 1) 2 σ2. (4) In our model he cenral benefi o appellae judges of mainaining preceden is informaional. The more cases he rial judges have o refer o, he more accuraely hey will undersand he legal rule for which ha line of cases sands. Thus, he appellae judge wishes o mainain preceden because i makes communicaion wih rial cours less noisy. However, he use of preceden comes a a price. In paricular, appellae judges bear a cos for deviaing subsanively oo far from he line of preceden ha hey claim o uphold. This consrains judges who are mainaining preceden from implemening a legal rule ha maches heir personal ideal, as can be seen in he model. When preceden is mainained, µ = ( j + α 2 µ 1 )/(1 + α 2 ) j (unless α = 0 or µ 1 = j). The rule applied by he rial cour when preceden is mainained is biased away from he curren appellae judge s ideal poin ( j) in he direcion of he old preceden (µ 1 ). Furher, i is ineresing o noe ha he appellae judge, when mainaining preceden, chooses he holding r = µ 1 + [( j µ 1 )/(1 + α 2 )], which is no equal o exising preceden (µ 1 )orhe appellae judge s ideal poin ( j) unless, by chance, j = µ 1 or = 1 + α 2. This ruling is always in he direcion of he appellae judge s ideal poin, bu may be more or less exreme, relaive o exising preceden. If > 1 + α 2, he appellae judge will issue a ruling farher from exising preceden han he appellae judge s ideal poin. Oherwise, he appellae judge will issue a ruling beween he ideal poin and exising preceden. While he appellae judge can make sure, by announcing a break wih preceden, ha he mean of he rial cour s signal is equal o he appellae judge s ideal poin, he variance of his signal (ha is, he noisiness of communicaion) will be higher. Again, his can be seen clearly in he model, since σ 2 >σ 2 / as long as > 1. The appellae judge hus faces a rade-off beween he accuracy wih which he legal rule is communicaed o he rial cours and he proximiy of ha rule o he appellae judge s ideal. The appellae judge deermines wheher or no o break wih preceden by comparing he expeced uiliies associaed wih each choice, breaking wih preceden if and only if EU(break) EU(mainain) = σ α α 2 ( j µ 1) 2 > 0 (5) α 3 ( 1)(1 + α 2 ) ( j µ 1) 2 σ 2 > 0. The Trial Judge s Decision The rial judge in his model is no a sraegic acor. Raher, he rial judge simply follows he insrucions of he appellae judge, aggregaing all exising holdings in he line if preceden has been mainained bu considering only he appellae judge s mos recen ruling if preceden has been broken. Because he focus of our model is he appellae judge s decision, aking he rial judge s aggregaion rule as given, we do no formally analyze he ramificaions of sraegic behavior by he rial judge. Noneheless, because we jusified his simplifying assumpion in par by claiming ha rial judges wan o please heir superiors on he appellae 760

7 American Poliical Science Review Vol. 96, No. 4 bench or avoid being overurned on appeal, one migh wonder if he rial judge would be beer off aemping o deduce and implemen he curren appellae judge s ideal poin, raher han following exising preceden. This urns ou, however, no o be he case. The rial judge could, of course, form an unbiased approximaion of he appellae judge s ideal poin by invering he maximizaion problem ha he appellae judge solved in equaion (2), condiioning on he rial judge s esimae of preexising preceden, µ Tha is, he rial judge could observe he signals associaed wih he firs 1 rulings in he line of preceden and use his informaion o esimae he preceden he curren appellae judge faced. Then, condiioning on his preexising preceden, he rial judge could inver he appellae judge s maximizaion problem o esimae wha value of j would have caused he curren appellae judge o issue he ruling he rial judge observed. This esimae is unbiased because he expeced value of he rial judge s observaion of any holding, r i, in he line of preceden is r i. Noice, hough, ha he informaion regarding preexising preceden can be used only o help he rial judge figure ou wha he curren appellae judge s maximizaion problem was. Having done his, he only informaion he rial judge has regarding he appellae judge s ideal poin is he signal ha he rial judge observes, which is drawn from a normal disribuion wih mean equal o he curren appellae judge s ruling, r. The variance of his disribuion is σ 2, no σ 2 /. Consequenly, following his inversion procedure canno lead he rial judge o a beer esimae of he appellae judge s ideal poin han would be achieved if he appellae judge had simply broken wih preceden and issued a ruling exacly a he appellae judge s ideal poin. Indeed, because he rial judge approximaes preexising preceden wih error, he esimae of j deduced from he inversion procedure is less precise han he esimae he rial judge would have formed if he appellae judge had broken wih preceden. The appellae judge mainains preceden only when he ruling of a rial judge who follows ha preceden (ha is, averages all he signals) will, in expecaion, be closer o he appellae judge s ideal poin han i would have been had he appellae judge broken wih preceden. Since he rial judge s esimae of he appellae judge s ideal poin formed by he inversion procedure is even worse han ha produced by a break wih preceden, he rial judge is more likely o please he appellae judge (and less likely o be overurned) by following preceden han by rying o deduce he appellae judge s ideal poin. If his were no he case, he appellae judge would have broken wih preceden. Thus, he rial judge would be beer off acing as a faihful agen of he appellae judge, deciding cases according o preceden when insruced o do so This is possible only if he rial judge knows α and σ 2. If no, he rial judge is unable even o enerain he possibiliy of rying o deduce he appellae judge s ideal poin from he observed signals. 13 This is no o say ha an aggregaion rule oher han averaging migh no lead o superior oucomes in some circumsances. RESULTS Comparaive Saics Comparaive saic analysis on he parameers of his model yields a number of resuls regarding how hese parameers affec he relaive desirabiliy of mainaining or breaking wih exising preceden. I is clear from equaion (5) ha he desirabiliy of breaking wih preceden decreases as σ 2, he noisiness of each individual signal, increases. The inuiion is ha as communicaion beween appellae and rial cours becomes less precise, he exra informaion provided by siuaing a decision in a line of preceden becomes more valuable o he appellae judge. Equaion (5) also implies ha increasing j µ 1, he disance beween he appellae judge s ideal poin and exising preceden, increases he araciveness of breaking wih preceden. Because a large disance beween exising preceden and he appellae judge s ideal poin consrains how much he appellae judge can move he expeced rial cour decision, he appellae judge is less willing o rade off conrol over he subsanive rule for increased accuracy of ransmission. Equaion (5) addiionally shows ha breaking wih preceden becomes more aracive as α, he marginal cos of making a subsanively divergen decision wihin an exising line of preceden, increases. The reason for his is ha, when α is close o zero, he appellae judge can move µ very close o j, even while mainaining preceden. Thus, he informaional benefi of siuaing he curren decision in a long line of cases comes a very lile cos in erms of subsance. However, as α grows, he appellae judge s abiliy o move he legal rule close o j becomes more consrained, making adherence o exising preceden less aracive. One can see his by examining equaion (3). Assuming ha preceden is mainained, he disance ha he legal rule will be moved is given by µ µ 1 = (j µ 1 )/(1 + α 2 ), which is clearly decreasing in α. Deriving he comparaive saics on, he number of cases in he line of preceden, is more complicaed. Because is a discree variable, o calculae he change in he araciveness of breaking wih preceden as he age of he preceden increases, we ake he firs differences of equaion (5) wih respec o : [EU +1 (break) EU +1 (mainain)] [EU (break) EU (mainain)] FD() = (j µ 1 ) 2 (6) α 4 + 2α 3 + α α (1 + α 2 + 2α + α)(1 + α 2 )( 1). As discussed earlier, he opimal aggregaion rule is a funcion of a hos of complex insiuional facors beyond he scope of his model. Noneheless, he basic logic ha he appellae judge s choice o mainain preceden implies ha he rial judge will do beer by following preceden han by aemping o deduce he appellae judge s ideal poin holds under any aggregaion rule. 761

8 Informaive Preceden and Inrajudicial Communicaion December 2002 FIGURE 1. EU(Break) EU(Mainain) wih α>ᾱ(2) A lile algebra demonsraes ha his firs difference can be posiive or negaive, depending on α and. In paricular, FD() < 0 FD() > 0 if if ᾱ() <α, 2 (7) ᾱ() >α Increasing decreases he desirabiliy of breaking wih preceden when α>ᾱ(); oherwise increasing increases he desirabiliy of breaking wih preceden. I is imporan o noe ha his hreshold, ᾱ() = ( )/( ), is iself a decreasing funcion of. There are wo cases o consider in undersanding his resul. The firs is when α>ᾱ(2) [he lowes possible values of ᾱ() when he appellae judge chooses o mainain exising preceden]. In his case, for all values of, he araciveness of breaking wih preceden decreases as increases. Tha is, for a sufficienly large α, older precedens are always less vulnerable han younger precedens. This case is illusraed in Figure 1. If, however, α<ᾱ(2), hen increases in increase he araciveness of breaking wih preceden for a cerain number of periods. Specifically, increases in will increase he likelihood of breaking as long as α is below he hreshold value of ᾱ(). Bu as increases, his hreshold value decreases, meaning ha α will evenually be greaer han he hreshold. A ha poin, he effec of increasing swiches so ha increases in decrease he desirabiliy of breaking wih preceden. This case is illusraed in Figure 2. To undersand he inuiion behind hese comparaive saics, recall ha an increase in has wo effecs on he desirabiliy o he appellae judge of mainaining he line of preceden. On he one hand, an increase in increases he consrain on how much he judge can move he subsanive legal rule, making older precedens less aracive. On he oher hand, an increase in improves he accuracy wih which he legal rule is communicaed o he rial cour, making older precedens more aracive. When α is sufficienly high, he informaion effec always overwhelms he consraining effec. The reason for his is ha when α is high, he appellae judge s abiliy o move he legal rule is already so consrained ha he marginal effec of an increase in on his consrain is negligible. However, when α and are sufficienly low, he consraining effec of an increase in is more imporan o he judge han he effec on informaion, hereby making an increase in increase he araciveness of breaking wih preceden. As increases furher, he informaion effec of marginal increases in will evenually overwhelm he consraining effecs; no maer how low α is, when is sufficienly high he appellae judge is already so consrained ha he marginal consraining effec of an increase in is again negligible. 14 Limis of Legal Change An ineresing quesion in he conex of our model is, How far on he subsance dimension can a legal 14 Noe also ha as goes o infiniy, he appellae judge is compleely consrained by he old preceden. In his case, he value of α does no maer, and he payoff of an appellae judge who mainains preceden converges o ( j µ 1 )

9 American Poliical Science Review Vol. 96, No. 4 FIGURE 2. EU(Break) EU(Mainain) wih α<ᾱ(2) rule move wihou a judge ever claiming o have broken from preceden? Is his disance bounded or unbounded? Tha is, how far would preceden move if, over an infinie series of urns, each appellae judge s ideal poin was such ha ha judge was exacly indifferen beween moving exising preceden and breaking, leading hese judges o move µ as far from µ 1 as is possible always in he same direcion wihou declaring a break from exising preceden? Reurning o he or liabiliy example discussed earlier, imagine ha over ime appellae judges wan a progressively sricer sandard of care. We wan o know wheher i is possible for he sandard of care o become infiniely more sric han he original sandard announced by he firs judge o consider he quesion, wihou any judge ever openly breaking wih esablished preceden. Of course, he idea ha here are muliple appellae judges deciding cases in sequence implies a dynamic model, beyond he scope of his projec. Noneheless, our model does provide a framework ha allows us o gain some heoreical leverage on his problem. If an appellae judge does no know wha he ideal poins of fuure judges will be, i is reasonable for he curren judge o assume ha he probabiliy of a fuure judge having preferences any given disance o he lef of exising preceden is he same as he probabiliy of ha judge having preferences he same disance o he righ. Under such a symmery assumpion, he expeced ideal poin of a fuure judge lies a exising preceden. Consequenly, he appellae judge s bes guess of fuure preceden is curren preceden, and so he decision problem in he dynamic model reduces o he one-sho problem analyzed above Of course, his is also he case if each appellae judge is myopic, caring only abou he payoff in he curren round. Therefore, we can find he maximum disance he legal rule can move under hese condiions (i.e., max µ µ 1 ) by solving he following consrained infinie sum: { j+α {µ 2 } µ 1 µ 1 }= µ 1+α 2 1 =2 = =2 =2 { j µ 1 1+α 2 }, (8) s.. ( j µ 1) 2 α 3 (1 + α 2 )( 1) σ 2 = 0. By making he consrain an equaliy, we impose he condiion ha in each round he preferences of he appellae judge are such ha he legal rule moves as far as is possible wihou he appellae judge preferring o break wih preceden. Solving he consrain for ( j µ 1 ) and subsiuing, we find ha he sum is equal o σ 1 1 α. (9) 1 + α 2 =2 Noice ha if α 0, his can be bounded above as follows: σ 1 1 α < σ α 2 α α 2 =2 = σ α =2 =2 ( 1 π 2 ) σ = α. (10) Thus, he disance ha can be moved in one direcion wihou a break from preceden is finie, even over 763

10 Informaive Preceden and Inrajudicial Communicaion December 2002 an infinie number of urns. Specifically, he legal rule canno move a disance greaer han ((π 2 /6) 1)(σ/α) from he decision in he firs case in he line of preceden wihou a leas one judge deciding o break. We can find comparaive saics on he acual poin of convergence by referring direcly o equaion (9). Noice ha if α is very small, so ha judges are relaively unconsrained by he precedenial line in which hey are wriing, his disance is quie large, whereas as α ges large he disance shrinks. Similarly, as σ becomes large, he disance ha can be moved ges larger because judges are willing o wrie opinions in line wih preceden even when he subsanive holdings of hose decisions are very far away from he subsanive posiion of he exising preceden. DISCUSSION Resoluion of Some Empirical Puzzles Our model provides a new perspecive on when, why, and how judicial decisions are consrained by exising precedens. This perspecive helps accoun for empirical observaions of judicial behavior ha are oherwise difficul o reconcile. To ake a sriking example, consider he role of deference o preceden in Supreme Cour decision-making. On he one hand, a rigorous analysis of voing paerns on he Cour finds ha jusices who voe agains a landmark case ha is, a case ha esablishes an imporan preceden end no o swich heir voing paern in subsequen cases, even hough he landmark case decision ough o consiue a legal preceden (Segal and Spaeh 1996a). This seems o be srong evidence ha judges, a leas a he Supreme Cour level, do no feel consrained by legal precedens as such. On he oher hand, if i is really he case ha he jusices do no aach much imporance o preceden, hen i is hard o explain why hey devoe so much ime and inellecual energy o i in heir deliberaions and why hey place so much emphasis on i in mos of heir decisions. Analysis of he U.S. Supreme Cour decision-making process, afer all, reveals ha argumens from preceden vasly ounumber all oher kinds of argumens in aorneys wrien briefs, he Cour s wrien opinions, and he jusices argumens in conference discussions (Knigh and Epsein 1996; Phelps and Gaes 1991). If i were really he case ha judges cared abou case oucomes raher han preceden, hen all he emphasis on argumens from preceden would seem o be a wase of resources. Bu if preceden is really influencing jusices decisions, hen he persisen paerns of coninuous dissen from landmark decisions is difficul o explain. Our model accouns sraighforwardly for his seeming conradicion. Jusices care abou preceden precisely because hey care abou policy; if hey can sufficienly improve heir communicaion of he proper legal rule by inegraing heir decision wih an exising line of cases, hey will do so, even if i means somewha modifying he legal rule hey announce and expending some energy on wriing a compelling and coheren opinion ha inegraes seemingly divergen rulings. Thus, hough our model does no formalize he process of formulaing a judicial opinion, i is enirely consisen wih he observaion ha judges pu a lo of ime and aenion ino rying o inegrae heir preferred oucome ino an esablished line of cases. However, in our model, if an appellae judge decides ha he addiional accuracy is no worh he sacrifice in erms of subsanive policy, hen he judge will voe o break wih preceden. If a given judge dissens in a landmark case, herefore, ha judge will usually coninue o dissen in subsequen cases. Afer all, in our model judges do no care abou preceden per se, so here is no reason for a judge o voe o uphold a legal rule ha is far from ha judge s ideal simply because ha rule had been esablished in a prior case. An appellae judge will voe o adhere o preceden only if, in doing so, he legal rule can be moved sufficienly closer o he judge s ideal poin ha he gain in erms of accuracy is worh he cos in erms of subsance. A second apparen empirical anomaly ha our model illuminaes is he seemingly schizophrenic aiude of judges oward long-esablished precedens. On he one hand, many would quoe approvingly Jusice Holmes s (1897) quip ha i is revoling o have no beer reason for a rule of law han ha i was laid down in he ime of Henry IV. I is sill more revoling if he grounds upon which i was laid down have vanished long since, and he rule simply persiss from blind imiaion of he pas. On he oher hand, i is ofen hough ha a legal rule wih a long hisory is worhy of paricular deference. Thus, lawyers and judges someimes argue, implicily or explicily, agains ampering wih long-esablished legal rules even while disagreeing wih heir subsanive conen. Again, our model suggess a simple reconciliaion of hese apparenly conradicory noions. Recall from he comparaive saics ha increasing has wo effecs. I consrains he judge s abiliy o affec he subsance of he legal rule, leading o he frusraion embodied in Holmes s remark. However, an old preceden i.e., a long line of cases hough difficul o move, has a grea deal of informaional value. The consequence is ha old precedens become enrenched so ha even when judges disagree wih he subsanive rules hey are relucan o overrule hem. New Hypoheses In addiion o offering new insighs ino hese imporan empirical puzzles, our model also suggess a number of new hypoheses regarding paerns of judicial decision-making. Firs, recall he comparaive saics on he parameer σ, which measures he inheren difficuly in communicaion beween appellae and rial judges. The model demonsraed ha as σ increases, he araciveness of breaking wih preceden decreases. Subsanively, his implies ha areas of law ha are highly complex and no amenable o simple legal regulaion are more likely o develop long lines of cases, wih boh high levels of deference o preceden and evoluion and 764