Trade-Offs in Fairness and Preference Judgments

University f Pennsylvania SchlarlyCmmns Marketing Papers Whartn Faculty Research 1993 Trade-Offs in Fairness and Preference Judgments Lisa D. Ordñez Barbara A. Mellers University f Pennsylvania Fllw this and additinal wrks at: https://repsitry.upenn.edu/marketing_papers Part f the Behaviral Ecnmics Cmmns, Cgnitive Psychlgy Cmmns, Ecnmic Thery Cmmns, Incme Distributin Cmmns, Marketing Cmmns, and the Public Ecnmics Cmmns Recmmended Citatin (OVERRIDE) Mellers, B.A. & Ordñez, L.D. (1993). Trade-Offs in Fairness and Preference Judgments. In Psychlgical Perspectives n Justic: Thery and Applicatin, 138-4. Cambridge University Press. This paper is psted at SchlarlyCmmns. https://repsitry.upenn.edu/marketing_papers/393 Fr mre infrmatin, please cntact repsitry@pbx.upenn.edu.

Trade-Offs in Fairness and Preference Judgments Abstract At the heart f many debates abut distributive justice is the widely assumed trade-ff between equality and efficiency (Okun, 1975). In the present chapter, equality refers t the distributin f incme within a sciety. Equality increases whenever incme variability is reduced. Efficiency refers t the gds and services that result frm a given input prductin, physical capital, r human labr. Efficiency increases whenever sciety prduces mre frm the same input. Trade-ffs between equality and efficiency ccur because increases in ne ften lead t decreases in the ther. An egalitarian sciety satisfies basic needs by establishing prgrams that redistribute wealth. But thse prgrams can reduce efficiency when they intrduce bureaucratic waste r diminish financial incentives. A reductin in efficiency can lead t fewer investments, fewer jbs, and declining prductivity. Disciplines Behaviral Ecnmics Business Cgnitive Psychlgy Ecnmic Thery Incme Distributin Marketing Public Ecnmics This bk chapter is available at SchlarlyCmmns: https://repsitry.upenn.edu/marketing_papers/393

Trade-ffs in fairness and preference judgments Lisa D. Ordnez and Barbara A. Mellers At the heart f many debates abut distributive justice is the widely assumed trade-ff between equality and efficiency (Okun, 1975). In the present chapter, equality refers t the distributin f incme within a sciety. Equality increases whenever incme variability is reduced. Efficiency refers t the gds and services that result frm a given input - prductin, physical capital, r human labr. Efficiency increases whenever sciety prduces mre frm the same input. Trade-ffs between equality and efficiency ccur because increases in ne ften lead t decreases in the ther. An egalitarian sciety satisfies basic needs by establishing prgrams that redistribute wealth. But thse prgrams can reduce efficiency when they intrduce bureaucratic waste r diminish financial incentives. A reductin in efficiency can lead t fewer investments, fewer jbs, and declining prductivity. Theries f distributive justice Is there an ideal pint alng the equality-efficiency cntinuum? Plitical and mral philsphers have articulated a number f psitins abut what cnstitutes justice and hw t arrive at just distributins f resurces and rewards. These psitins place differential emphasis n equality and efficiency. Utilitarians believe that sciety shuld be arranged t maximize the ttal (r smetimes average) utility f all individuals (Bentham, 1961/1789). If utility is identical t prfit, then prfit maximizatin ccurs when efficiency We thank Jn Barn, Karen Biagini, Michael Birnbaum, Jhn Carrll, Shi-jie Chang, Alan Cke, Rbyn Dawes, and Philip Tetlck fr helpful discussins and cmments n drafts f this chapter. We als thank Kim Gallegus and Katty H fr help with data cllectin. This research was supprted by an NSF grant (BNS-90-16526) t the first authr and anther NSF grant (SES-89-08698) t the secnd authr. The first authr was als supprted by a Frd Fundatin Predctral Fellwship (880-0765).

Trade-ffs in fairness and preference judgments 139 is greatest. This versin f utilitarianism might emphasize efficiency ver equality. On the ther hand, if utility is a negatively accelerated functin f incme, the pr benefit mre frm any given dllar. This frm f utilitarianism might strike a cmprmise between equality and efficiency in an attempt t prvide bth assistance at the bttm and incentives at the tp. Egalitarians argue that sciety shuld be arranged t prvide fd, shelter, and essential medical care fr all cncerned. These basic needs are satisfied by means f redistributive plicies that minimize incme inequality. In this framewrk, the emphasis is n equality rather than efficiency. Mre recently, Rawls (1971) develped a philsphical thery f justice based n a scial cntract. Rawls asks what wuld happen if the framers f sciety perated behind a "veil f ignrance," s that they knew nthing abut their future sciecnmic psitin. Rawls cntends that tw principles f justice wuld emerge. The first is the principle f greatest equal liberty; each persn wuld have an equal right t the mst extensive system f liberties and freedms. The secnd is the difference principle, which asserts that scietal and ecnmic institutins shuld be arranged t benefit the wrst ff. This principle is a maximin rule. Disparities in wealth are assumed t be just when they imprve the welfare f the prest members f sciety. This thery fcuses n the incmes f the least advantaged members f sciety. Cmprmises between equality and efficiency Researchers have psited that, fr an assrtment f cgnitive and plitical reasns, peple have difficulty making trade-ffs between strngly held values such as envirnmental prtectin and ecnmic grwth (Abelsn & Levi, 1985; Einhrn & Hgarth, 1981; Tetlck, 1986). Hadari (1988) ntes that sme philsphical theries f distributive justice structure sciety t avid trade-ffs between basic values. Utilitarians escape trade-ffs by reducing all values t a single scale f utility r welfare. Rawls avids tradeffs by lexicgraphically rdering values. Only after a sciety has established basic liberties fr everyne des the maximin rule apply. The present chapter investigates hw peple make trade-ffs when judging scietal fairness and preference. It builds n earlier wrk by Mitchell, Tetlck, Mellers, and Ordnez (1992) that examined the judged fairness f hypthetical scieties. In thse experiments, scieties were displayed as incme distributins, that is, average incmes in each quartile. Equality and efficiency were manipulated using incme variance and average incme, respectively. Scieties with mre incme variance had greater incme inequality. Scieties with higher average incme had greater efficiency. Subjects were tld t assume that the distributin f inputs was held cnstant acrss all scieties, s greater prfits resulted frm greater efficiency. A fairness ranking ver scieties was derived fr each individual. When

140 LISA D. ORD6NEZ AND BARBARA A. MELLERS the effrt-reward crrelatin was lw, rankings crrelated higher with incme variance (equality). When the effrt-reward crrelatin was high, rankings were better predicted frm average incme (efficiency). Mitchell et al. (1992) cncluded that the effrt-reward crrelatin r level f meritcracy was a mderatr f equality-efficiency trade-ffs. In Mitchell et al. (1992), incme variance was cnfunded with minimum incme (i.e., average incme in the lwest quartile). We manipulate bth factrs and investigate whether the predictability f incme variance is due t a cncern fr the verall incme variability r the flr f the incme distributin. Furthermre, we manipulate the effrt-reward crrelatin in such a way that subjects can make trade-ffs between this factr and thers. We als investigate judged preferences t examine whether peple prefer t live in scieties they judge mre fair. Finally, d trade-ffs made at the scietal level als ccur at the individual level? We ask subjects t judge the extent t which an individual within different scieties is verbenefited r underbenefited. Of interest is whether the same factrs influence fairness judgments at the micr and macr levels (Brickman, Flger, Gde, & Schul, 1981). Macr fairness and preference Tw experiments examined judgments f scietal fairness and preference. Scieties were displayed as scatterplts with pltted against wrk index. Wrk index, ranging frm 0 (lwest cntributin) t 4 (highest cntributin), was said t reflect merit, senirity, effrt, prductivity, experience, etc. The distributin f wrk was held cnstant acrss scieties. In Experiment 1, eight scieties were cnstructed frm variatins alng three factrs: minimum, mean, and standard deviatin. 1 Figure 1 presents tw scieties that differ in mean and standard deviatin. Sciety A has a lwer mean, but less variability; Sciety B has a higher mean, but mre variability. In Experiment 2, sixteen scieties were frmed by manipulating the three factrs in Experiment 1 and the effrt-reward crrelatin. Figure 2 shws tw scieties that differ in crrelatin and minimum. Sciety A has a higher minimum but a lwer wrk- crrelatin, and Sciety B has a higher crrelatin but a lwer minimum. Table 1 gives infrmatin abut the eight scieties in Experiment 1. The crrelatin between wrk and ranged frm.19 t.77 acrss scieties. Table 2 prvides infrmatin abut the sixteen scieties in Experiment 2. In bth experiments, the distributin f wrk was held cnstant. Subjects were presented with all pssible pairs f scieties. In the fairness cnditins, subjects judged which sciety was mre fair. In the preference cnditins, subjects selected the sciety in which they preferred t live. Subjects were tld that the pverty level was $,000. 2 After cmple-

Trade-ffs in fairness and preference judgments 141 Y 80 1001........ r- Sciety A Sciety B O 60 -C. 40 20 0.0 1.0 2.0 3.0 4.0 0.0 1.0 WORK INDEX 2.0 3.0 4.0 Figure 1 Trade-ffs between measures f equality and efficiency in Experiment 1. Sciety A has a narrw standard deviatin and a lw mean. Sciety B has a wide standard deviatin and a high mean. 100 thus>ands (in SAI.ARY 80 60 40 20 0 Sciety A # Sciety B # 0.0 1.0 2.0 3.0 4.0 0.0 1.0 2.0 WORK INDEX 3.0 4.0 Figure 2 Crrelatin and minimum trade-ffs in Experiment 2. Sciety A has a high minimum but a lw crrelatin, and Sciety B has a high crrelatin but a lw minimum. tin f the tw tasks, subjects answered a questinnaire requesting infrmatin abut sex, sciecnmic status, plitical party affiliatin, and plitical views (rating n a seven-pint cnservative/liberal scale). In Experiment 2, subjects were als asked t state their anticipated sciecnmic status and rank the imprtance f each experimental factr in their fairness and preference judgments. Participants were undergraduates at the University f Califrnia at Berkeley wh received credit in a lwer divisin psychlgy curse. There were 80 and 70 subjects in Experiments 1 and 2, respectively.

142 LISA D. ORD6NEZ AND BARBARA A. MELLERS Table 1 Scieties in Experiment 1 (in thusands) Sciety Minimum Mean Salary std. dev. LLW LLN LHW LHN HLW HLN HHW HHN CJl CJl 5 5 31 30 41 40 34 33 43 42 23 18 25 18 19 12 21 18 Nte: The first letter in the sciety label refers t the minimum (L = lw and H = high). The secnd represents mean (L = lw and H = high). The third letter refers t the standard deviatin (W = wide and N = narrw). Table 2 Scieties in Experiment 2 (in thusands) Sciety Crrelatin Minimum Mean Salary std. dev. LLLW LLLN LLHW LLHN LHLW LHLN LHHW LHHN HLLW HLLN HLHW HLHN HHLW HHLN HHHW HHHN 0.22 0.27 0.19 0.25 0.19 0.21 0.22 0.18 0.87 0.79 0.87 0.83 0.79 0.88 0.87 0.86 5 5 5 CJl CJl CJl CJl CJl 31 30 41 40 34 33 43 42 31 30 41 40 34 33 43 42 23 18 25 18 19 12 21 18 23 18 25 18 19 12 21 18 Nte: The first letter in the sciety label refers t the crrelatin between wrk index and (L = lw and H = high). The secnd represents minimum (L = lw and H = high). The third letter refers t the mean (L = lw and H = high). The furth represents standard deviatin (W = wide and N = narrw).

Trade-ffs in fairness and preference judgments 143 Did subjects make trade-ffs? Fr each subject, tw scietal rankings were derived - ne fr fairness and ne fr preference. These rankings were cmputed by cunting the number f times a sciety was judged as mre fair (r mre preferred) than the thers. Scres culd range frm 0 t 7 in Experiment 1 and frm 0 t in Experiment 2. The higher the scre, the higher the rank f that sciety. If tw r mre scieties received the same scre, ranks were tied. Trade-ffs were assumed t ccur if scietal rankings were incnsistent with a lexicgraphic rder f any f the experimental measures. Fr example, a lexicgraphic rder f minimum wuld be ne in which all scieties with high minimum salaries were ranked abve thse with lw minimum salaries. Scieties with the same minimum might be distinguished n the basis f ther factrs, but this rdering wuld be secndary t cnsideratins f minimum. In Experiment 1, 57% and 52% f the subjects had fairness and preference rankings that were incnsistent with any f the lexicgraphic rders. In Experiment 2, 38% and 70% had fairness and preference rankings that were incnsistent with the lexicgraphic rders. These subjects appeared t make trade-ffs. It culd be argued that these percentages are t high, since they d nt allw fr errr r unreliability in the data. Therefre, subjects whse rankings were almst cnsistent with a lexicgraphic rder were reclassified. Whenever a subject's ranking was cnsistent with a lexicgraphic rder after the ranks fr tw scieties were reversed, that subject was assigned t the lexicgraphic grup. With this revised cunt, 40% and 39% f the subjects in Experiment 1 had fairness and preference rankings that were incnsistent with lexicgraphic rders. In Experiment 2, 29% and 60% had fairness and preference rankings that were incnsistent with lexicgraphic rders. Thus, an average f 42% f the subjects in each cnditin appeared t make trade-ffs. Which factrs accunted fr the mst variance? Individual fairness and preference rankings were crrelated with three rthgnal rankings, each based n the experimental measures (i.e., lw and high mean, lw and high minimum, r wide and narrw standard deviatin). Table 3 shws the percentage f individuals whse rankings crrelated highest with each experimental measure. In the fairness cnditin, the largest percentage f subjects had rankings that crrelated highest with standard deviatin. In the preference cnditin, the largest percentage f subjects had rankings that crrelated highest with minimum. Hwever, the mst striking result in Table 3 is that there are n clear winners. Subjects used a variety f different strategies in their judgments.

144 LISA D. ORD6NEZ AND BARBARA A. MELLERS Table 3 Percentage f subjects best fit by experimental measures Respnse Minimum Salary std. dev. Mean Fairness Preference 34 43 41 28 25 30 Table 4 Median crrelatins f subject shwn fr subject grups rankings with experimental measures Measures Minimum Fairness grups Salary std. dev. Mean Minimum Preference grups Salary std. dev. Mean Minimum 0.82 0.22 0.38 0.77 0.11 0.36 Salary 0.28 0.77 0.44 0.22 0.72 0.33 std. dev. Mean 0.35 0.28 0.66 0.44 0.44 0.78 Table 4 prvides infrmatin abut the average crrelatins f subjects in each f the three grups and additinal infrmatin abut the extent t which their rankings crrelate with ther measures. Rws are the three experimental measures, and clumns are the three grups f subjects whse fairness and preference rankings crrelated highest with each experimental measure. Entries are median crrelatins. Crrelatins shuld be high dwn the diagnals, since subjects in thse grups had rankings that were bestfit by that measure. Thse crrelatins ranged frm.66 t.82 fr fairness and.72 t.78 fr preference. 3 Crrelatins n the ff-diagnals shw average predictability f the measures. These crrelatins are lwer than thse n the diagnals, ranging frm.22 t.44 fr fairness and.11 t.44 fr preference. Althugh they are cnsiderably smaller, these ff-diagnal crrelatins differ significantly frm zer. In sum, ther factrs als cntribute t the predictability f the rankings. Hw d scietal rankings differ acrss the three grups? Figure 3 presents median fairness rankings fr subjects in each best-fitting grup. Numbers range frm 1 (least fair sciety) t 8 (mst fair sciety). In each bx, rws are means, and clumns are standard deviatins. A bx represents a set f scieties with the same minimum ; different bxes shw scieties with different minimum salaries.

Trade-ffs in fairness and preference judgments 145 Upper bxes, center bxes, and lwer bxes present median rankings fr the mean, the standard deviatin, and the minimum grup, respectively. Rankings differ cnsiderably acrss grups. Subjects whse rankings were best fit by mean assigned lwer ranks t lw-mean scieties and higher ranks t high-mean scieties. Subjects whse rankings were best fit by standard deviatin assigned lwer ranks t wide-standard-deviatin scieties and higher ranks t narrw-standarddeviatin scieties. Finally, subjects whse rankings were best fit by minimum tended t assign lwer ranks t lw-minimum- scieties and higher ranks t high-minimum- scieties. Despite these differences, certain patterns are cmmn acrss all three grups. Within a bx, the lw-mean, wide-standard-deviatin sciety is always ranked lwest. Furthermre, the high-mean, narrw-standarddeviatin sciety is always ranked highest. Of interest are the rankings in the ther tw cells. Subjects whse rankings were best predicted by mean (upper bxes) judged the high-mean, wide-standard-deviatin sciety as mre fair than the lw-mean, narrw-standard-deviatin sciety fr bth levels f minimum. Subjects whse rankings were best predicted by standard deviatin (center bxes) judged the lw-mean, narrw-standard-deviatin sciety as mre fair than the high-mean, widestandard-deviatin sciety. Subjects whse rankings were best described by minimum (lwer bxes) did nt have a cnsistent set f rankings. In sum, Figure 3 shws systematic differences in trade-ffs acrss grups. Similar patterns were fund fr preference rankings. Effects f the wrk- crrelatin Experiment 2 investigates scieties that vary in wrk- crrelatins as well as minimum salaries, means, and standard deviatins. Table 5 presents the percentage f subjects whse fairness and preference rankings crrelated highest with the fur experimental factrs. Apprximately 75% f the subjects had fairness rankings that crrelated highest with the wrk- crrelatin. The remaining 25% had rankings that were best described by minimum. Apprximately 50% f the subjects had preference rankings that were best described by the crrelatin and the ther 50% were best predicted by the minimum. Almst n subjects had fairness r preference rankings that were best predicted frm measures f equality r efficiency. Table 6 shws median crrelatins fr best-fitting grups, as in Table 4. Rankings f subjects in the crrelatin and minimum grups were crrelated with all fur experimental measures. Once again, crrelatins were highest fr the best-fitting measures. The ther crrelatins were lwer, but nt zer. These crrelatins ranged frm.16 t.52 fr fairness and.12 t.44 fr preference. Salary mean and standard deviatin als cntribute

146 LISA D. ORD6NEZ AND BARBARA A. MELLERS Fairness Ranks Mean Lw Mean Salary Grup (25%) Min. SaL=Lw Min. Sal.= Std. Dev. Std. Dev. Wide Nar. Wide Nar. 2 5 6 Lw 1 y i 8 Salary Standard Deviatin Grup (41%) Min. Sal.=Lw Min. Sal.= Std. Dev. Std. Dev. Wide Nar. Wide Nar. Mean Lw 1 y 6 Lw 2 ^ 5 3 i 4 8 Mean Lw Minimum Salary Grup (34%) Min. Sal.=Lw Std. Dev. Min. Sal.= Std. Dev. Wide Nar. Wide Nar. 1 y 3 ^ 2 4 Lw 5 y 1 1 Figure 3 Median rankings f scieties frm Experiment 1 fr each grup f subjects best fit by an experimental measure. Upper bxes, center bxes, and lwer bxes shw mean, standard deviatin, and minimum grups, respectively. Numbers range frm 1 (least fair sciety) t 8 (mst fair sciety). Table 5 Percentage f subjects best fit by experimental measures Respnse Crrelatin Minimum Salary std. dev. Mean Fairness Preference 74 50 26 44 0 0 0 6 t the predictability f fairness and preference rankings, althugh they are nt the best predictrs. Figure 4 presents median fairness rankings fr subjects in the crrelatin and minimum grups. Numbers range frm 1 (least fair sciety) t 16 (mst fair sciety). Within each bx, rws are minimum salaries and clumns are wrk- crrelatins. Different bxes reflect scieties with different levels f mean and standard deviatin. Once again, certain patterns are cnsistent acrss grups. Within a bx, the lwcrrelatin, lw-minimum- sciety is always ranked lwest. The highcrrelatin, high-minimum- sciety is always ranked highest. Subjects in the crrelatin grup (upper bxes) judged the high-crrelatin,

Trade-ffs in fairness and preference judgments 147 Table 6 Median crrelatins f subject rankings with experimental measures shwn fr subject grups Measures Crrelatin Minimum Salary std. dev. Mean Crrelatin 0.87 0.25 0.16 0.20 Fairness grups Minimum 0.40 0.77 0.19 0.52 Preference grups Crrelatin 0.80 0.44 0.12 0.25 Minimum 0.35 0.75 0.12 0.27 Nte: Mean and standard deviatin grups are nt presented because there were either nne r very few subjects in these grups. Fairness Ranks Min. Sal. Lw Crrelatin Grup (74%) Std. Dev. = Wide Std. Dev. = Narrw Mean = Lw Mean = Mean = Lw Mean = Crr. Crr. Crr. Crr. Lw Lw Lw Lw 3 A 1 Lw 1 A 2 Lw 2 Lw 4 9 7.5 14 6 16 5 A* Min. Sal. Lw Minimum Salary Grup (26%) Std. Dev. = Wide Std. Dev. = Narrw Mean = Lw Mean = Mean = Lw Mean = Crr. Crr. Crr. Crr. Lw Lw Lw Lw 1 Lw 3 Lw 2 Lw 4 9 13 10 14 y 11.5 12.5 Figure 4 Median rankings f scieties frm Experiment 2 fr subjects whse rankings were best fit by the crrelatin (upper bxes) and the minimum (lwer bxes). Numbers range frm 1 (least sciety) t 16 (mst fair sciety). lw-minimum- sciety as mre fair than the lw-crrelatin, highminimum- sciety in all fur bxes. Subjects in the minimum- grup (lwer bxes) shwed the ppsite pattern. Preference rankings were similar t fairness rankings. Discussin Results frm these experiments suggest that many peple make trade-ffs when judging scietal fairness and preference. When equality and effi-

148 LISA D. ORD6NEZ AND BARBARA A. MELLERS ciency are varied in cnjunctin with ther factrs, trade-ffs tend t ccur between minimum and wrk- crrelatin. These measures are better predictrs f fairness and preference judgments. Apprximately 75% and 50% f the subjects had fairness and preference rankings that were best described by wrk- crrelatin, respectively. Virtually nne f the subjects had fairness r preference rankings that were best predicted by equality and efficiency. Why might peple fcus n the minimum and the wrk- crrelatin? Minimum might be viewed as a safety net beneath which peple shuld nt fall. The wrk- crrelatin may reflect the incentive system within the sciety. A high crrelatin implies that greater effrt will lead t greater rewards. It als implies equal pay fr equal wrk. But is scietal fairness simply a high effrt-reward crrelatin? Crrelatins culd be high due t extreme utliers. Cnsider the salaries f sme chief executive fficers that are several hundred times mre than thse f the average wrkers. Effrt-reward crrelatins culd be lw fr mst f the wrkers, but the verall crrelatin culd be high due t a single pint, the CEO. It wuld be interesting t investigate the extent t which scieties with identical crrelatins but different utliers are judged fair. As far back as Aristtle, justice has been thught t depend n the relatinship between deservingness (effrt, wrk, r prductivity) and reward. Mellers (1982) examined fairness ratings f faculty members in hypthetical departments and fund quite different deservingness functins between and merit in departments with the same -merit crrelatins. Deservingness functins were psitively accelerated, linear, and negatively accelerated, depending n the jint distributin f and merit. Deservingness functins are nt captured by crrelatins alne. Hwever, in many situatins, crrelatins may be a gd first apprximatin. In fact, the crrelatin is a special case f Birnbaum's (1983) adjustment mdel f fair salaries. Birnbaum suggested that fair salaries might be determined by a cntext-dependent relatinship between deservingness and reward. Salaries are assumed t be fair t the extent that they are determined by this relatinship. Results frm these experiments als suggest that the scieties in which peple prefer t live are nt necessarily judged as mst fair. Many peple judge scieties with high crrelatins as fair, but they prefer t live in scieties with high minimum salaries. Differences between fairness and preference judgments have als been fund by Messick and Sentis (1979), wh investigated judged allcatins in situatins where subjects assigned payments t themselves and anther. In ne cnditin, subjects divided rewards fairly, and in the ther cnditin, they allcated rewards as they wished. Amunts that subjects preferred t give themselves were larger than amunts they judged t be fair. What mtivates scietal preferences? One pssibility is self-interest. Peple may be cncerned abut their wn psitin in sciety. When their status

Trade-ffs in fairness and preference judgments 149 CO O O 100 80 60. 40 20 Sciety A # # 0 (/) D O JZ qg 20 CO 0.0 1.0 2.0 3.0 WORK INDEX 4.0 1.0 2.0 3.0 WORK INDEX 4.0 Figure 5 Tw illustrative scieties frm Experiment 3. Cmmn members (slid circles) are nested within different scieties. Additinal members (pen circles) are varied acrss scieties. is uncertain, it may be prudent t prefer scieties that help the least advantaged. Rawls develps this pint in his thery f distributive justice. Anther pssibility is altruism. Peple may be genuinely cncerned abut the plight f the pr and prefer t live in scieties that care fr the needy. Research n scial dilemmas indicates that sme percentage f peple act in altruistic ways, even when it entails persnal sacrifice (Caprael, Dawes, Orbell, & van de Kragt, 1989). See Tyler and Dawes (this bk) fr a further discussin f this tpic. Micr fairness Experiment 3 examines whether factrs that describe fairness judgments at the individual level resemble thse at the scietal level. Scieties in Experiment 1 were cnstructed such that eight individuals (pints in the scatterplts) were cmmn t each sciety. Other pints were added t manipulate the minimum, mean, and standard deviatin. Figure 5 shws tw illustrative scieties. Slid pints are the cmmn members; pen pints are individuals wh differ acrss scieties. Sciety A has a high mean and a high minimum, but a wide standard deviatin. Sciety B has a narrw standard deviatin, but a lw mean and lw minimum. Subjects were asked t rate the extent t which an individual was underbenefited r verbenefited n an 80 t -80 scale (where 80 = very very verbenefited, 0 = fairly treated, and 80 = very very underbenefited). As in the ther experiments, they were tld that the pverty level was $,000. Subjects were given eight scieties and judged randmly rdered individuals in each sciety (the eight slid pints in each panel f Figure 5 and seven pints that differed fr each sciety). After cmpleting the experimental trials, subjects answered the same questinnaire used in Experi-

0 LISA D. ORD6NEZ AND BARBARA A. MELLERS 45 C O (T if) c/) 0) c 0) <D -Q L. CD - -30-45 13 25 38 50 63 Salary (in thusands) 75 Figure 6 Mean verbenefited ratings fr cmmn members averaged ver scieties. Ratings are pltted against with a separate curve fr each level f wrk. ment 1. Participants were 88 undergraduates at the University f Califrnia at Berkeley wh received credit in a lwer divisin psychlgy curse fr their participatin. Figure 6 presents ratings fr the eight cmmn individuals, averaged ver all eight scieties. Judged verbenefitedness increased with and decreased with wrk index. Similar means were btained fr each sciety. Average ratings fr the cmmn individuals never changed sign acrss scieties; an individual wh appeared verbenefited in ne sciety was never judged as underbenefited in anther. But despite the fact that the signs did nt change, the magnitudes f the ratings varied systematically acrss scieties. T illustrate the cntextual effects, Table 7 presents average ratings f the cmmn individuals in each f the eight scieties. Cmmn individuals were judged mre verbenefited as mean, standard deviatin, and minimum decreased. All three factrs were statistically significant. 4 Which factrs accunted fr mre variance? Fr each subject, eight ratings were cmputed, ne fr each sciety, by averaging respnses fr the cmmn individuals. These ratings were crrelated with three rthgnal rankings based n mean, standard deviatin, and minimum. Table 8 shws that fr the largest

Trade-ffs in fairness and preference judgments 1 Table 7 Mean verbenefitedness ratings fr cmmn members in different scieties Salary std. dev. Wide Narrw Lw Lw minimum Mean 0.51-3.53 5.76-1.92 Lw minimum Mean 0.67-6.58 2.54-3.28 Nte: Mean ratings are averaged ver cmmn members. The respnse scale ranged frm -80 t 80. Table 8 Percentage f subjects best fit by experimental measures Respnse Minimum Salary std. dev. Mean Rating 23 32 45 percentage f subjects, verbenefitedness ratings crrelated with mean, fllwed by standard deviatin, and then minimum. Crrelatins were als cmputed fr each cmmn member in all eight scieties, and results were similar t thse based n averages; fr the largest percentage f subjects, verbenefitedness crrelated with decreases in mean. Discussin Experiments 1 and 3 allw a cmparisn f scietal and individual fairness judgments. At the scietal level, the largest percentage f subjects had fairness rankings that crrelated highest with standard deviatin. Minimum came in secnd, and mean tk third place. At the individual level, the largest percentage f subjects had fairness rankings that crrelated highest with mean. Salary standard deviatin came in secnd place, fllwed by minimum. Differences between individual and scietal fairness might ccur because peple use different criteria t assess macr and micr justice. Subjects may make different trade-ffs between equality and efficiency at different levels. If s, trade-ffs culd cme int cnflict; it may be impssible t make ptimal trade-ffs simultaneusly.

2 LISA D. ORD6NEZ AND BARBARA A. MELLERS Individual differences The present experiments demnstrate individual differences in fairness and preference judgments. Several studies have examined individual differences in fairness. In a review n sex and gender differences, Kahn and Gaeddert (1985) nted that men tend t endrse distributins f rewards in which ratis f utcmes t inputs are equal acrss individuals, whereas wmen tend t allcate rewards in a mre unifrm fashin. Majr and Adams (1983) reprted that wmen tend t have less variability than men in their reward allcatins. Other studies have investigated individual differences in plitical idelgy. Rasinski (1987) fund that perceptins f fairness crrelated with plitical rientatin. Thse scring high n efficiency (prprtinality rules fr fair divisins) tended t be mre cnservative, whereas thse scring high n equality (egalitarian rules fr fair divisins) tended t be mre liberal. Subjects in the present experiments were UC Berkeley undergraduates, and ne might expect the vast majrity t be liberal Demcrats. Hwever, nly 60% f the subjects described themselves as Demcrats. The average cnservative/liberal rating was at the center f the scale. Rughly half were men, and half were wmen. The average self-reprt f sciecnmic status was 3.3 n a scale frm 1 (lwer class) t 5 (upper class). Membership in a best-fitting grup crrelated significantly with the judged imprtance f that measure. Hwever, grup membership did nt crrelate significantly with demgraphic variables and plitical pinins, with a few ntable exceptins. In Experiment 1, subjects with lwer sciecnmic status tended t have fairness rankings that were best predicted by minimum. In Experiment 2, men tended t have preference rankings that were best predicted frm the effrt-reward crrelatin. Demcrats, liberals, and wmen tended t have preference rankings that were best predicted frm the minimum. Cnclusins Results frm the present experiments suggest that sme peple make tradeffs between ecnmic measures f equality and efficiency when judging scietal fairness and preference. Hwever, trade-ffs tend t ccur mre ften between the minimum and the wrk- crrelatin. These measures predict fairness and preference judgments t a greater extent than equality and efficiency. Furthermre, trade-ffs tend t differ fr fairness and preference judgments. The percentage f subjects best fit by minimum was greater fr preference rankings than fairness rankings. Interestingly, preference rankings seemed mre in line with Rawls's thery f justice than fairness rankings. Fairness rankings were nt entirely cnsistent with Rawls's thery, egalitarianism, r utilitarianism. Finally, factrs that determine fairness at the individual level differ frm

Trade-ffs in fairness and preference judgments 3 thse that describe fairness at the scietal level. Micr and macrjustice can cme int cnflict whenever we think it is fair t treat individuals and grups differently. It may be ne thing t vte fr verall cuts in welfare prgrams, and anther thing t refuse assistance t a hmeless persn wh lst his jb due t illness. Understanding the nature f these differences may ultimately help us reach mre acceptable trade-ffs between ecnmic and psychlgical factrs. References Abelsn, R. P., & Levi, A. (1985). Decisin making and decisin thery. In G. Lindzey and E. Arnsn (Eds.), Handbk f scial psychlgy (3rd ed.). New Yrk: Randm Huse. Bentham, J. (1961/1789). Principles f mrals and legislatin. New Yrk: Dubleday. Birnbaum, M. H. (1983). Perceived equity f plicies. Jurnal f Applied Psychlgy, 68, 49-59. Brickman, P., Flger, R., Gde, E., & Schul, Y. (1981). Micrjustice and macrjustice. In M. J. Lerner and S. C. Lerner (Eds.), The justice mtive in scial behavir. New Yrk: Plenum. Caprael, L. R., Dawes, R. M., Orbell, J. M., & van de Kragt, A. (1989). Selfishness examined: Cperatin in the absence f egistic incentives. Behaviral and Brain Sciences, 12, 683-739. Einhrn, H. J., & Hgarth, R. M. (1981). Behaviral decisin thery: Prcesses f judgment and chice. Annual Review f Psychlgy, 32, 53-88. Hadari, S. A. (1988). Value trade-ff. Jurnal f Plitics, 50, 655-676. Kahn, A. S., & Gaeddert, W. P. (1985). Frm theries in equity t theries in justice: Deliberating cnsequences f studying wmen, in V. E. O'Leary, R. E. Unger, & B. S. Wailstn (Eds.), Wmen, gender, and scial psychlgy. Hillsdale, NJ: Erlbaum. Majr, B., & Adams, J. B. (1983). Rle f gender, interpersnal rientatin, and self-presentatin in distributive-justice behavir. Jurnal f Persnality and Scial Psychlgy, 45, 598-608. Mellers, B. A. (1982). Equity judgment: A revisin f Aristtelian views. Jurnal f Experimental Psychlgy: General, 111, 242-270. Messick, D. M., & Sentis, K. P. (1979). Fairness and preference. Jurnal f Experimental Scial Psychlgy,, 418-434. Mitchell, G., Tetlck, P. E., Mellers, B. A., & Ordnez, L. (1992). Judgments f scial justice: Cmprmises between equality and efficiency. Unpublished manuscript. Okun, A. (1975). Equality and efficiency: The big tradeff. Washingtn, DC: Brkings Institute. Rasinski, K. A. (1987). What's fair is fair - r is it? Value differences underlying public views abut scial justice. Jurnal f Persnality and Scial Psychlgy, 53, 201-211. Rawls, J. (1971). A thery f justice. Cambridge, MA: Harvard University Press. Tetlck, P. E. (1986). A value pluralism mdel f idelgical reasning. Jurnal f Persnality and Scial Psychlgy, 50, 819-827.

4 LISA D. ORD6NEZ AND BARBARA A. MELLERS Ntes 1 Subjects were given pairs f scieties and judged which sciety had: (1) a higher mean, (2) a larger variance, and (3) a higher crrelatin between and effrt. All subjects prperly detected differences in the manipulatins. 2 Half f the subjects were presented with scatterplts shwing the pverty line as a hrizntal line, and half were nt. There was n significant effect f pverty line, and all analyses were cllapsed ver the tw grups. 3 These crrelatins are pint-biserial crrelatins that cannt exceed.87. 4 All tests f significance were dne at an alpha level f.01.