You are on page 1of 46

Group Threat and Regional Change in Attitudes Toward African-Americans Author(s): Lincoln Quillian Source: American Journal of Sociology,

Vol. 102, No. 3 (Nov., 1996), pp. 816-860 Published by: The University of Chicago Press Stable URL: http://www.jstor.org/stable/2782464 . Accessed: 05/10/2013 15:59
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to American Journal of Sociology.

http://www.jstor.org

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

GroupThreatand Regional Change towardAfrican-Americans' in Attitudes


Lincoln Quillian Harvard University

the GSS 1972-91, the authorassesses sevUsing surveydata from changein and temporal eral possiblecauses of regionaldifferences forraceThe study concludesthat(1) support whiteracial attitudes. (2) changesin targeted policieshas not changedover birthcohorts, ofthedecline explainonlya smallportion individualcharacteristics ofeducacohorts, (3) theinfluence prejudiceoverbirth in traditional over birthcohorts, and has been increasing tionon racial attitudes black and average per capita incomeare good pre(4) percentage of racial attitudes of theNorth/South and explaina portion dictors gap and the changeover timein prejudice.

INTRODUCTION in the 1950sfirst a trend towarddeclinSinceattitude highlighted surveys timeand energy have spentconsiderable sociologists ingracial prejudice, usingsurveyitems.The resultsprocharting changesin racial attitudes ofwhiteAmeriofsomemajorchangesin theracialattitudes vide a record World War II, whiteattitudes cans. In the 40 years following changed to majorofsegregation and racialdiscrimination from open endorsement of race (Hyof equal treatment regardless ityacceptanceof theprinciple and Sheatsley1971).Whiteshave man and Sheatsley1956,1964;Greeley public policiesand, endingopenlydiscriminatory increasingly supported

Christopher My thanksto StanleyLieberson,Peter Marsden,Aage S0rensen, Peter for comments. AJSreviewers their helpful Cook,and several Cynthia Winship, I gratewiththeLISCOMP program. muchneededassistance also provided Marsden ofa NationalScienceFoundation Graduate support thefinancial fully acknowledge An of thisarticle. during the writing and a HarvardMeritFellowship Fellowship of the at the August 1994 meeting earlierversionof this articlewas presented to Lincoln Directcorrespondence in Los Angeles. Association Sociological American Massachusetts Cambridge, of Sociology, HarvardUniversity, Quillian,Department 02138.
1

? 1996 by The Universityof Chicago. All rightsreserved. 1.50 0002-9602/97/10203-0005$0

816

AJS Volume 102 Number3 (November1996):816-860

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat to a lesserextent, in relations laws banningdiscrimination betweenprivate citizens. Not all trendsin racial attitudes, however,are as favorableas the growthin supportforthe principleof equality.Supportforpolicies of government intervention designedto foster racial equalityhas remained are measured, there is no clear tenuous, and, as farback as theseattitudes trend. Othersurvey questions revealthecontinuing ofprivate importance resentments and fearsabout race. The disparity among these attitudes has led to a considerable of debate about the meaningand genuineness thesechanges(Jackman and Muha 1984;Schuman, and Bobo 1985; Steeh, Sears 1988). Despite thequalityof theworkto chartthesetrends in racial attitudes and an extensive debate about how to interpret thistrend, surprisingly littleresearch explainswhyattitudes changedas theydid. Many authors in acceptanceof the principle take forgranted of color-blind the growth equal treatment and thendebate the significance of this (or irrelevance) change forrace relationsmore generally. It is indicativeof this tendency that the leading work in the field,Schuman et al.'s Racial Attitudesin America(1985),devotesa chapterto interpretations in of trends whiteracial attitudes but has no parallelchapteron the sourceof these trends. This articletakes some steps toward explainingthese trendsby atto explainregionaland temporal tempting variationin whiteracial attitudes in the period 1972-91. The models here focus on two possible sourcesof change. First,following thelead ofpriorstudies(e.g.,Davis 1975),I investigate the extentto which individualcharacteristics, in particulareducation, could be one sourceof temporal and regionalpatterns of racial attitudes. Since the educatedare moretolerant than the less educated,the decline in antiblackprejudiceovertheperiod1972-91 mayreflect theexpansion of educationalopportunities duringthisperiod. Second,I argue that,to understand thesechanges,we need to look to collective as well as individual-level variables.I hypothesize thatchanges in prejudiceare in parttheresultof changesin thelevel of groupthreat presented by the subordinate groupto the dominant group. Finally,thearticleconsiders threeotherlikelysourcesofchangein racial attitudes: the impact of historical events such as the Civil Rights the influence of the incorporation movement, of social sciencein school and collegecurricula, and the legacyof racial oppressionin the South. Although precisemeasuresof thesefactors are not available, I includea set of controlvariables to capturetheireffects. Some of these theories also have implications examined withthedata here,providing someclues about theirinfluences on racial attitudes. 817

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology


TRENDS AND REGIONAL DIFFERENCES ATTITUDES, 1972-91 IN RACIAL

Racial attitudes after WorldWar II reflect growing support fortheprinciple of equal treatment but withlittleaccompanying increasein support formeasuresthatimplement racial equality(Schumanet al. 1985).White racial attitudes after1972continue to followthesetrends. Supportforthe principle of equal treatment has continued to increase,especiallyin the South. Althoughthe gap betweenthe South and the otherregionsof the United States has been shrinking, on average, Southerners remain moredisposedtowardracialprejudice thanthosefrom other regions (Firebaugh and Davis 1988). Supportforpoliciesof government intervention to achieve racial equality,on the otherhand, was mostly unchangedin the 1970s and 1980s(Schumanet al. 1985; Steeh and Schuman 1992). Although little systematic research tries to establish thesourcesofthese changesin racialattitudes, theliterature contains manyplausiblespeculationsabout whytraditional have declined.Thereare also prejudicemight several hypotheses about why the South continuesto be more inclined towardracial prejudicethanotherregions of the UnitedStates.In what followsI first discuss a numberof theories of racial attitudes and how thesetheories may be includedin a statistical model of attitude change, I concludewith thenI givedetailsofthemodeland measures, and finally results from estimating the models.

MEASURED CAUSES OF RACIAL ATTITUDES

One of the difficulties is finding withassessingexplanations statistically offactors thatmayinfluence suitablemeasures Directmeasures prejudice. oftwosourcesofattitudes are includedin thisstudy: individual characteristicsand groupthreat. IndividualCharacteristics variables are relatedto racial Many studiesargue that individual-level attitudes (e.g.,Adornoet al. 1950;Campbell 1971;Middleton1976;Tuch and Fuchs 1989).Although severalindividual-level 1987; Case, Greeley, to explainsubstantial variablesinfluence racial attitudes, variationover timeor amongregions, on a variable mustbothhave a strong influence racial attitudes and varyconsiderably amongregionsor over time.Few individualcharacteristics meetthesetwo criteria. Studiesrelating individual characteristics and racial-attitude changefocus almost exclusively on education(e.g., Hyman and Sheatsley1956). This is because oftheimportance ofeducationas a predictor ofprejudice 818

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat and the relatednotionof tolerance(Stouffer and 1955; Hyman,Wright, Reed 1975).Educationhas also increased overtimeas younger, moreeducated cohorts replaceolder,less educatedones. From 1972to 1985,mean to the General Social Survey years of education of white respondents (GSS) increasedby 1.2 years.2 Unless the associationbetweeneducation and tolerance thisperiod, we would expectgreater eduweakenedduring cationto cause a declinein antiblackattitudes. in educationmay also explainregionalvariation differences Similarly, in prejudice. Southerners are on averagelesseducatedthannon-Southernon averageto be less prejers.Because ofthis, we expectnon-Southerners udiced than Southerners. Studiesthathave examined theimportance ofindividual characteristics forchangeover timeand across regions, however,do not concludethat in prejudicebetween theycan explainall or even mostof thedifferences regionsor over time.Zaller (1992) findsthat,fortwo prejudice-related and agreequestions(opposition to laws thatforbid interracial marriage theincrease mentthatwhiteshave a right to segregated neighborhoods), in educationacross cohorts explainsabout one-third of the total change in attitudes from1972 to 1985. Likewise,Middleton(1976) regressed a measureof prejudiceon a Southern termand morethan30 indidummy vidual characteristics to be relatedto prejudice.Individualcharthought in prejudicialattiacteristics explainedonlya smallpartofthedifference tudesbetween theSouthand therestofthecountry. Middleton concluded thatmoregroup-level variablesmustbe responsible fortheSouthern tendencytowardracial prejudice. The results ofthesestudiessuggest thatindividualcharacteristics cannot explainmostof thechangeover timeor thegap in attitudes between the South and otherregions, althougheducationmay explainsome part in someimportant ofthesedifferences. Yet priorresults are limited ways. No studyhas attempted to assess ifindividualcharacteristics otherthan in racialattitudes. education couldexplainthetrend thesestudies Further, have examinedonlya fewquestionsrelating to supportforthe principle of racial equality. To the best of my knowledge,no studies examine sourcesof changeover timein attitudes towardmeasuresto implement racial equality. Group Threat If individualcharacteristics are not able to explainthesetrends, grouplevel factors may be responsible. One possible group-level explanation,
This usesmy1972-85sample, whichexcludes respondents thatdid notanswer one of the racial attitude or weremissing forany of the variablesin table 2 questions
2

below.

819

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology groupthreat theory, proposesthatprejudiceis in parta response to feelings thatcertainprerogatives believedto belongto the dominantracial groupare underthreat bymembers ofthesubordinate group.The central idea is thatattitudes towardthe otherrace are influenced by individual fearsthattheirown race will be put at systematic disadvantageto the other race-it is thusgroupthreat towardtheracialgroup, notindividual as a sourceofracialhostility.3 threat, thatthistheory emphasizes Although threats to individualsmay be a cause of prejudice,groupthreat emphasizes thefeelings individuals have ofbelonging to a racial groupand their view of the relations betweengroupsas a sourceof racial attitudes. Severalauthors have proposeddifferent versions ofthistheory (Blumer 1958;Blalock 1967;Vannemanand Pettigrew 1972;Bobo 1983;Bobo and Kluegel1993).Theyvaryin how they conceiveoftherelationship between individualand groupinterests, on whether or not groupthreatmustbe real or merely perceived, and on the terms and methodologies used. For threat need not be real with selfmy purposes, this nor need it coincide interest. Perception of threat to one's own groupis a sourceof prejudice thatis distinct from threats to one's self. What contributes to this feelingof group threat?One elementlong is thesize of thesubordinate to be relatedto thedegreeofthreat thought ofthis grouprelative to thedominant group.The bestknownformulation principle is foundin Blalock (1967). He argues that subordinate group size is relatedto perceivedthreatfortwo reasons.First,the largerthe proportion of the populationmade up by the subordinategroup,the greater is competition among racial groupsforjobs and othereconomic are potentially a resource thatcan be resources. Second,largernumbers used to engagein collective actionagainstthedominant group.The operaeconomicand political tionofthesesourcesofthreat dependson existing of relations betweenthe groups,and the politicaland economicsystem the hostcountry. The mostdirecttestof thishypothesis would be to examineifpercentmeasureof threatand thento age black is relatedto an individual-level see if that measureof threatis relatedto prejudiceand discrimination. to date have notasked questionsto meanationalsurveys Unfortunately, In the absence of this sure perceivedthreatfromotherracial groups.4
is a theory racial group 3As thisimplies, in thisarticle, groupthreat of dominant toprejudice prejudice toward a subordinate group. Threat probably has somerelation toward bysubordinate group members, butI suspect that subordinate group prejudice than members of thedominant groupresults morefrom past and present injustices ofwhatmight Thus a different threats occurin thefuture. theory maybe neededto accountforminority groupprejudice. of group 4 Giles and Evans (1985) and Fossett and Kiecolt(1989) createa measure in theNationalElection threat from measure, they survey questions Study. Usingthis
820

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat intervening measure,studieshave examinedthe relationbetweencityof the populationthat level or region-level measuresof the percentage against blacks (Blalock 1957), is black and measuresof discrimination 1957),racial inequality resistance to school racial integration (Pettigrew candidates(Heer (Frisbe and Niedert 1977), votingfor prosegregation surveys ofwhites(Petti1959),and racial prejudicemeasuredby attitude grew 1959; Giles 1977; Giles and Evans 1985; Fossettand Kiecolt 1989). All of thesestudiesfindthat percentage black is associatedwith an inordiscrimination. The mostfrequent creasein prejudice, racialinequality, black increases prejudice interpretation oftheseresults is thatpercentage black is relatedto degree of or discrimination because the proportion threatwhitesfeel fromblacks. Like past studies,I include percentage cause of prejudiceand accept the interpretation black as a group-level black increases racialhostility becauseit is relatedto perthatpercentage African-Americans ceived threat pose to whites.5 A secondgroup-level ofgroupthreat is ecofactor linkedto perception nomiccircumstances. Pettigrew (1957) and Blalock (1967)discussthisas a I (1995) givea moresystematic and elsewhere possiblecause ofprejudice, accountof the relationbetweenprejudiceand groupeconomicposition. their own groupto be more The idea is thatthedominant groupperceives threatenedby minoritypresence when economic circumstancesare worse.6 Economic prosperity reduces the feelingthat racial groupsare

find both that percentage ofpopulation blackincreases threat whites perceived among and thatperceived threat increases prejudice (Giles and Evans findthatthelatter proposition holdsonlyin theSouth,but Fossettand Kiecoltfinds it holdsforboth in modelswithperceived regions). However, ofpopulathreat controlled percentage topredict tionblackcontinues Thissuggests racialattitudes. that partoftheinfluence of percentage of population is some factor black on racial attitudes otherthanits A problem influence on threat. withthesefindings, however, is thatthemeasure of threat used by bothstudies is poor,consisting of questions sometimes lumpedwith prejudicemeasures. The threat measureis composedof two questions asking(1) civilrights whether arebeingpushedtoofast, at theright going speedorgoing slowly and (2) whether therespondent believesblackshave too muchinfluence, about the right amount ofinfluence, ornotenough influence. in the Randommeasurement error indexis likely to lead to underestimation oftheinfluence ofthreat on racialattitudes. 'As faras I am aware,groupthreat is theonlyexplanation in theliterature ofwhy blackshouldinfluence percentage whiteracialattitudes. 6 As this formulation I conceive suggests, oftheeffect ofeconomic conditions as absolute,notrelative amonggroups. That is, I do notpresume thatrelative increases of theincome or wealthof blacksto whites shouldmakewhites feelmorethreatened. In fact, as Wilson(1978,chap. 5) implies in discussing thedeclineofracialstrife on thejob, whites tendto feeltheleast threatened by middle-class or affluent blacks. Consistent withthisidea,Gilesand Evans (1985)find thathigher blacksocialstatus is negatively related to whites'perceptions ofblackthreat (on problems in Gilesand Evans measure ofperceived threat, see n. 4 above).
821

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology lockedin a zero-sum game over resources to supplybasic humanneeds. In this way improvedeconomicwell-being reducesperceivedthreatto thedominant As a result, group'seconomicposition. prejudiceand other antiblack attitudes whenthegroupeconomic increase amongwhites situationworsens. If thesegroup-level causes ofthreat are important, whatpatterns should An increasein economicprosperity we see in attitude trends? shouldbe relatedto reducedthreat and therefore to reducedracial animosities. On mostmeasures, economicprosperity increasedfrom1972to 1991.For inabout 19% from stance,averagepercapita incomeincreased 1970to 1980 and by another19% from1980to 1990 (U.S. Bureau oftheCensus 1977, 1992).7 This was a muchslowerrateofincreasethanin the 1960s,which GNP simisaw a better than35% increase, but it is growth nevertheless. larlyincreased. Groupthreat theory suggests thatthisincreasein prosperin the declineof racial prejudice.Income itymay have been important growth was particularly rapidin theSouthin the 1970s,leadingto a convergenceamong regionsin income-by 1980 the South was close to income paritywith the restof the country (Bishop, Formby,and Thistle thatthisshould be one factor be1992).Perceivedgroupthreat suggests hind the convergence in racial attitudes betweenthe South and the rest betweeneconomiccircumstances and ofthecountry. Finally,therelation in racial attitudes threatimpliesthat thereshould be some fluctuation to detectin with the businesscycle.These movements may be difficult the raw prejudicetimeseries,however, especiallysince any such movethatinfluence attitudes.8 mentcan easilybe offset by otherfactors
decline showlittle changeor even a slight of economic well-being 'Some measures contradict theclaimthatoveralleconomic Do thesemeasures overthistimeperiod. ofthese statistics shows inthe1970sand 1980s? An examination well-being improved or declined after medianwages of menstagnated thisis notthecase. For instance, as women increasingly wageearners however, there weremore 1970.Atthesametime, from1970to As a result incomeper capita rosenoticeably entered the workforce. by thecensusbureaudid notinmedianfamily incomeas reported 1990.Similarly, in thesizeoffamilies thedecline creasefrom 1972to 1991.But thisdoesnotadjustfor whilefamindividuals. Sincefamily sizedeclined amongunrelated orincome growth ily income remainedsteady,again income per capita increased.Danziger and incometo needsratio,whichthey a measureof family Gottschalk (1995) compute as a measure Theirmeasure income." has severaladvantages call "adjustedfamily economeasures: itadjustsfor family size,captures ofeconomic overcensus prosperity individuals. Theyfind that, unrelated families, and includes miesof scale forlarger incomeincreased by about 10.6%, between1973and 1991,medianadjustedfamily in twospurts from 1973to 1979and from 1982to withmostoftheincrease coming ofoverall increasing prosperity, that1973-91was a time 1989.On this basisI conclude off. was better economic also increased and noteveryone inequality although 8 There are severalreasons in the not appear verystrongly might whya recession in terms of per capita income, mostmodern recessions timeseries.First, prejudice mildeffects. Per capita incomedeclinedby about 4% from1980to have relatively
822

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat The connection betweenpercentage black and racial prejudiceimplies thatthe South shouldhave attitudes thatare the least favorabletoward African-Americans. Regionswitha low fraction of thepopulationblack, likethePacificCoast and New England,shouldbe least inclinedtoward prejudice.Percentage black by regiondid not changemorethanslightly after1972,so percentage of the trend black is not a possibleexplanation in racial attitudes from1972 to 1991. INCOMPLETELY MEASURED SOURCES OF RACIALATTITUDES on racial attiMany group-level variablesprobablyhave some influence causes of tudes. Since almost no researchestablisheswhat group-level racial attitudes might be important, decidingwithmuchconfidence what group-level controls shouldbe includedis difficult. Even moredifficult is decidinghow to measurethemproperly. My concern is notwithprecisely measuring theimpactofthesefactors on racial attitudes, but with whichthisarticleleaves forfuture research, thesevariablesdoes not bias estimates of the makingsure thatomitting effects of the includedvariables.The potentialproblemis that omitted group-level variablescouldbias theestimates ofthegroupthreat variables iftheyare correlated withthem.9 For instance, sincethe Southhas tradihad a largerproportion of the populationblack and because the tionally South has a legacyof slavery, the effect of proportion of the population black could in partjust reflect of slaveryin the theimpactof thehistory South. A similarspuriouscorrelation can arise froman omittedgrouplevel variablethatis trending in thesame direction as one oftheincluded foromitted group-level variables.To providesomecontrol variablesthat
1982,e.g.,a changesmallenoughto be difficult to detectin the overalltimeseries even ifper capita incomeweretheonlyvariableinfluencing Secondand prejudice. moreimportant, it shouldbe recalled thatthisis onlyone ofthevariables thatinfluencelevelsofprejudice. in thisstudy, Among themeasured factors percapitaincome is a fairly goodpredictor, butit is notmorepowerful thanother variables likecohort membership. Anyrecession effect couldeasilybe offset by effects ofother measured or unmeasured variablesmoving in theopposite direction. 'Omitting an individual-level variablethatinfluences thedependent variableand is correlated withan independent orindividual) variable(group willalso bias estimates oftheeffects oftheindependent is on omitted variables. My focus group-level rather thanindividual-level variables research becauseso little establishes likely group-level causes or how to measure A substantial them. bodyof priorresearch, on theother hand,has established a conventional listofindividual characteristics thatare importantpredictors ofracialattitudes. Thus I have moreconfidence thatmyindividuallevelcontrols are picking up themostimportant individual-level causesofprejudice. Detailsoftheindividual characteristics included arediscussed further in themethods and data section. 823

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology withthegroup-level might be correlated variablesin themodels, I include several additionalcontrols: a Southern dummyvariable,separatelinear fortheSouthand thenon-South, yeartrendterms a lineartermforbirth in whichtherespondent cohort, a variableforregion was socialized(lived in South at age 16 = 1; 0, otherwise), an interaction betweenbirthyear and regionof residenceat age 16, and an interaction of educationand birthyear.'0As I will argue below, these variables should capturethe most oftendiscussedcauses of trendsand regionaldifferences in racial prejudice,thusavoidingbiasingthe estimated model parameters. Threesourcesofracialattitudes in theracerelations discussed literature in the attitude are not includedexplicitly models:historical eventssuch as the Civil Rightsmovement, of education, and changesin the content in theSouth.Although thelegacyofsegregation and discrimination direct measures ofthesefactors are notavailable,indirect inferences abouttheir importance can sometimes be made from variablesincludedin themodel. The likely influence on each ofthesefactors on racialattitudes is discussed below. HistoricalEvents eventsmayhave influenced nationalattitudes about race Many historical in 1972-91,in particular ofthe1960s."1 theCivil Rights movement Survey in 1972-91 includemanywho wereyoungadults or adolesrespondents cents duringthe heightof civil rights and who may have had activity theirattitudes Historical eventscan influence shaped by theexperience.'2 forall age groups(a racial prejudiceeitherthrough changingattitudes the attitudes of younger, more periodeffect), by influencing presumably orboth.How do race-related impressionable, generations (a cohort effect), we observefrom1972 to 1991? the attitudes historical eventsinfluence in racial attitudes thatthesechangesare Surprisingly, patterns suggest the productof gradual, long-term changes ratherthan sudden shocks withthisresult is that caused by historical events.One finding consistent is extremely smooth and thedeclinein traditional prejudiceacrosscohorts linear (Firebaughand Davis 1988; see also fig. 1 below). This suggests in a radical way; such a thatno singleeventsuddenly changedattitudes
10 As discussed morein themethods and data section, including theyearand cohort thetrend intoperiodand cohort terms also provides a convenient wayto decompose effects. " For a brief listofcivilrights events that mayhaveinfluenced racialattitudes during thisperiod, see Page and Shapiro(1992,pp. 76-77). 12 Schuman to namehistorical and Scott(1989)asked respondents eventsthatwere significant for their ownlives."Civilrights" was thefifth most often mentioned event.
824

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat in a timetrend withsudden singleprecipitating force wouldhave resulted theeventwould be muchdifferent rises,so thatcohorts socializedduring from cohorts who were socializedafter theevent.Instead,each cohortis moresupportive ofequal treatment thanthenext. oftheprinciple steadily Hymanand Sheatsley (1964) reachedthe same conclusion by comparing eventsin race theresults ofsurveys conducted before and after significant relations. Theypointoutfrom surveys in the195 Osand 1960sthatpolitical such as the confrontation betweenfederal eventsinvolvingcivil rights, in LittleRock,Arkanoverschooldesegregation troopsand local officials to surveys a fewweekslater.As they sas,did notappear toaffect responses unbroken is thusone ofa massivetrend, "The overallpicture conclude, by theparticular newseventsoftheday"(Hymanand Sheatsley 1964,p. 17). Further evidenceof a gradual,unbroken trendcomes from investigation of the trendpatternforrecentcohorts.If historical eventsduring thenwe shouldsee one's youthimprint a certainattitudinal orientation, in youth. attitude based on eventsexperienced fluctuations acrosscohorts On thisbasis one mightexpectthat cohortssocialized duringthe Civil Rights movement's mostactiveperiod, the 1960s,shouldbe moretolerant thanthosesocializedin the 1980s.Steehand Schuman(1992) investigate and findthattolerance thispossibility has notdecreasedamonglatercohorts.For mostitemsthereare no cohorttrendsin the late 1980s,and, it is linearand in the direction when thereis change across cohorts, of increasedracial tolerance. This does not mean thatpublic eventshave had no consequencesfor racialattitudes. These results do imply, however, thatanyeffect they have had is partof an evolvingtrend, rather than a suddensourceof change. The evidencealso pointsto the factthatthecivil rights eventswere not theonlyreasonforchangesin attitudes aboutrace.Attitudes oftraditional prejudicebegan to declinein the 1940s,well beforeBrown v. Board of focusedpublicattention Education first on racialinequality (Schumanet al. 1985; Page and Shapiro 1992). Lipsetand Schneider (1978) proposeinsteadthatprejudicedeclinedbecause eventsduringand afterWorldWar II brought intoa sharp focus the inconsistency betweenAmericanideals of "equality"and "fairness" and the reality of racial segregation and discrimination. Attitude change resulted from forcing the public to recognize thisinconsistency between theirvalues and thepracticeof racial discrimination. Several eventscan be singledout as important in highlighting this inconsistency and thus a change in attitudes. The rhetoric of World War II and the furthering Cold War-strugglesjustified as fights for freedom-was clearly inconsistentwiththecontinued ofa substantial oftherights suppression minority in the United States (McAdam 1982). Even more so, the Civil Rights movement to bear on the inconsistency brought attention betweenegali825

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology tarian principles and the realityof legalized racial segregation and inequality.In thisway eventslike the Civil Rightsmovement and World War II may have contributed to a gradual shiftin Americanattitudes about race. This view has the advantage that it is moreconsistent withthe time trendof steadyand even changein racial attitudes. For the purposesof thisarticle, a gradualeffect ofthissortis likely to fallintothelineartrend and birthcohorttermsincludedas controls in the model. The Rise of Social Science and Beliefsabout Environmental Influences on Behavior Intellectual developments may also have playeda rolein changing ideas aboutrace.Page and Shapiro(1992)and Zaller(1992)arguethatan importantreasonforthedeclinein traditional in thebelief prejudiceis a growth that human behavioris attributable to cultureand environment rather thanheredity. Page and Shapiroattribute thechangeto theriseof social science,especiallytheworkof anthropologists such as Franz Boas. This on humanbehavior, combined withthesuccessofurbanblack perspective in theNorth, led sociologists and anthropologists migrants at Columfirst bia University and lateracrossthe country to the conclusion thatblacks werenot inherently in unequal. This viewpoint quicklyfoundadherents othersocial sciencedisciplinesas well. As late as the 1920s,manyacademicpsychologists maintained thatsome racial groupswere intellectuto others.By the end of the 1930s,a shift had occurredso ally superior thatfewacademicpsychologists thisview. Instead,prominent supported effects of discrimination on psychologists began to studythestigmatizing and theconnections African-Americans betweenmentaldisorders and racial prejudice(Zaller 1992).The view thatmostofthedifferences between blacks and whitesweretheresult ofenvironment rather thanbiology diffusedgraduallyfromsocial scientists, to college-educated elites,and finallyto thegeneralpublic(RichardP. Young, citedin Page and Shapiro 1992). Both socializationthroughchanges in the contentof education in thetreatment (Page and Shapiro 1992)and, at a laterdate,alterations ofrace by elitesin themedia (Zaller 1992)mayhave playeda rolein this diffusion. Withtheweakening ofrace-based racist biological determinism, beliefsabout blacks eventually beliefs. gave way to moreegalitarian influProbablythe mostimportant way that beliefin environmental from enceswas initially transmitted social scientists to thepublicat large was through social sciencescoursesat the collegelevel. At a later time social studiescoursesin secondary and elementary schoolsmayhave also been important. This suggeststhat the effect of education may have cohorts as thesocial sciencesbecame moreprominent changedoverbirth 826

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat in universities, as happenedin the decades afterWorldWar II. In fact, some evidenceforthis claim does exist.Campbell (1971) findsthat for in reducing cohorts educatedbefore WorldWar II, education has no effect antiblackprejudice.Amongcohorts educatedafter WorldWar II, on the other thanthenonhand,thecollegeeducatedare notably moretolerant collegeeducated.Like Page and Shapiro(1992),Campbell (1971,pp. 6869) pointsto theriseofthesocial scienceson collegecampusesas a likely in prejudiceamongeducational cause oftheincreasing differences groups. Since social sciencewas a standardpartofeducationalcurricula before in 1972,thistheory the GSS first interviewed respondents to is difficult testwiththedata here.Many GSS respondents, wereeducated however, fullinstitutionalization ofthesocial sciencesin educationalcurricbefore ula. Thus,we can look in theGSS data fortheinteraction between cohort (birth year)and educationfoundby Campbell (1971). It is not necessary thatit will be there:Campbell was usingdata fromoldersurveys (from 1964 to 1970) and was usingmeasuresof racial attitudes notin the GSS, such as feeling-thermometer interacscores.Even ifthereis a significant tionterm, thisdoes notruleout otherinterpretations. But a positiveand significant interaction betweenbirth yearand educationwould be consistentwiththeidea thattheincreasing of social sciencein educapresence tioncurricula may have reducedprejudice. CulturalContinuity A finallikely in racial attitudes sourceofregional disparities is thehistory in the South.Race relations in the South at of slaveryand race relations least untilthe 1960swere characterized by legalizedsegregation by race, in interracial ritualdeference contact, and a race-basedclass system (Wilson 1973). Such practicesprobablyinfluenced the attitudes of older cohorts ofwhiteswho lived underthemand possibly influence theattitudes of youngcohorts theircontactwitholdercohorts. through We can think ofthelegacyofslavery and discrimination as influencing respondents two ways:through Southern acculturation and through conto regional formity element normsabout race. The acculturation suggests thatindividualsraisedin thedistinctive racial environment of the South are likelyto hold moreantiblackattitudes than thoseraised elsewhere. To controlfordistinctively I include a dummy Southernsocialization, variableforSouthern residence at age 16. Given thedeclinein JimCrow in theSouth,theeffect segregation of Southern socialization may change over time.Thus I also includean interaction termbetweenyearof birth and Southern resident at age 16 to allow theeffect in the of socialization South to varyacrossyears. Distinctive Southern culture and norms can also operatebyinfluencing 827

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology in a moreimmediate attitudes way. Some respondents are likely to adjust theirattitudes towardconformity withtheprevailing groupnorms(Merton 1949;Pettigrew 1959).GiventheSouth'sconservative racialattitudes, especiallyamongoldercohorts, Southern normsshouldlead conformists towardmoreantiblack responses. This groupconformity effect shouldfall intotheSouthern dummy variable.This is a possibleinterpretation ofthe effect of the Southern dummy variableafter othercontrols are included. DATA AND METHODS This studyrelieson surveydata from whiterespondents interviewed by the GSS (Davis 1991). Using GSS questions,I create two measuresof racial attitudes: traditional prejudiceand supportforrace-targeted policies.Correspondingly, I constructed twosamples,choosing yearsin which the questionsthatmake up the dependent variablesare available.'3 The methodused to assess theinfluence of thefactors discussedabove in whiteracial attitudes on regional and temporal variation is straightforward. For each dependent variable,I first estimate modelspredicting an scoreon each attitude indexbased onlyon region expected (Southern resi= 0) a linearyearterm, dent= 1; otherwise and an interaction between the two.Then I introduce controls for birth cohort and Southern socialization, inindividualcharacteristics, and measuresrelatedto groupthreat. After each setofvariables, thechangeofthelinearyear, troducing byinspecting Southerndummyvariable,and interaction variables,I can assess how mucheach ofthesefactors to thetrend contributes and theSouthern/nonSouthern difference (thisis a methodsimilarto that used by Firebaugh and Davis [1988] to decomposeprejudiceintocohortand year effects). Since thegroupthreat measuresare group-level variables,theyneed to I be defined unit.To assess changesin groupthreat, over a group-level The nineregions dividethedata intocellsbyyearand region. correspond to the census division category:New England, Middle Atlantic,East East South Central, NorthCentral, West NorthCentral,SouthAtlantic, West SouthCentral, Mountain,and Pacific.For thetraditional prejudice measure,thisleaves me with63 cells (nineregionsX seven years),with an average of 115 cases per cell (minimum 40, maximum316). For the measureofopposition to race targeting, thisleaves me with81 cells (nine regionsX nineyears),withan averageof 61 cases per cell (minimum 36, maximum 318). This study uses censusdivisionas thegroup-level unit.The correct unit
For traditional variableis available for1972,1976-77, the dependent prejudice, thedependent variableis availablefor 1980,1982,and 1984-85.For race targeting, 1975,1983-84,and 1986-91.
13

828

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat is difficult to defineprecisely, since at which to measurethe influences people probablypicturerelationsbetweentheirown racial group and or nation) otherracial groupsat morethan one level (city, state,region, on thecontext. Distinctive depending cultures and regional-based regional mass media probablycontribute to people thinking about theirgroupat in thisarticleprobably a regional level. However,theresults imperfectly aboutraceusingother based refercapture peoplethinking geographically error flattens ence groupsas well. Since randommeasurement slopes of towardzero)any effect ofthe thegroupthreat variables(bias coefficients not begroupthreatvariablesis in spite of thismeasurement problem, cause of it. Measures of Racial Attitudes The The analysisherereliesmainlyon two indexesof racial attitudes.'4 and blatant discrimination first shows supportfor segregation against whichI call a measureof traditional African-Americans, prejudice.The secondindexis composedofquestions askingabout government spending to improveeconomicoutcomesforAfrican-Americans. and intervention I refer to the second index as a measureof oppositionto race targeting whenused as a strategy to redress between African-Americans inequality and whites.This divisionof attitudeitemsis similarto the distinction made by symbolic-racism theorists (e.g.,Sears 1988) betweentraditional of racism. and symbolic forms The measuresof traditional to race-targeted prejudiceand opposition to maintaincontinuity on withearlierliterature policiesare constructed The four-item indexof traditional racial attitudes. prejudiceis the same as that used by Firebaughand Davis (1988) to measureprejudice(see table 1). Like Firebaugh and Davis, I standardizedthese four items, and added six to make thescale consistently summedthem, The positive. reliability coefficient (Cronbach'salpha) forthe indexis .7394.15 The seconddependent invariablemeasuresopposition to government tervention to decrease racial inequality(see table 1). This measurealso followsthe precedent of earlierliterature: the itemsare similarto items
14 Theseareattitudinal measures andshould notbe confused with measures ofdisbehavior. On average, I expect criminatory that people whoaremore prejudiced are more todiscriminate. Therelationship andtheactof likely between these attitudes

is complicated, so thatactsofdiscrimination can occurwithdiscriminating however, outprejudice can exist without manifest inactsofdiscrimijustas prejudice becoming nation. 15I also tried theanalysis usinganother variablethathad onlytwoofthe dependent in thetraditional four questions prejudice indexand was availablefora longer time series. indexdid notalterthesubstantive results ofthisarticle. Usingthisalternative

829

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

ce u
0 .1.1 >
-4

biD

biD

biD

u (L) : cd 0 'o b 0 !;.-,4 bo o 4 .bD 0 (L)


;-4 ;-4 ;..4

> 0 ho

o bJD

b,O 0 I.., (u bD Cd
4b

bD bo -(L) Cd ho 0

(1) bID ce

r.

-ti
(1)
;-4

4-) bD bi)
C)

bo bD cd

cn

-o

-4

bi)

cd cd

bi)

-o -0 cd cd
m C:)

bJD bD Cd Cd -4

bD

bD

0-4 -q -q
1-4 m in

4
E-4 E-4

C,
biD

4 > u u 4

(L)
Cfd

t Ca %O 0
C'4 a 0 964 00 00 ON P

4-4

Cd ce En :

(L) 4

(3.) >

> cr% cd i::

OC)

>

biD
z --j

-+-)
ln En

En
-4 o

bO bD bb 4-" >
Zs Cd
En ;.

cd
t14

<u

4-4

bD
ce

Cd
4-4

1=4
z

cd

4-)

0
.

.1-1

(U
.,

Cd bJD

En
Cd (L)

bD 1=4+J Cd
4-i >

-4 u

Cd .0
0

En
O 134

:3
M

0 W

bD .0

En
u 1-4

>
bD
0 o

u
w

0
O

-4

bb bD En
;-4

0
cd 0

lc

O
0

Cd

0 4

0 ;. W)

Cd ,.j (L) w U cd

-4

CZ

'W
-4 4

tn bJD > biD Cd

-4.J

0
'S

En bi) 4 0 --,4 u
;. 0 cd 0S'
0

-Zi

cd

Cd
ce

> Cd 4-4

bD
Z

0 w

bID
4-4

.5 0 'A
-6-J
X44 4 1-4

En -Cd 1=4 bJD Cd

1:14

-+-j En
0

421
;-4 O

$=.. 'n

>,

-4
4'-4'

w bD
0

cd -4 1-4

cd

0
0

b.0

tDD 4-4 cd 0
-W'

0 O m
cd

4 -4

o
P:

4) tn En 'C$ cd u) +-4-,A En bID v cd W M


9 u

-4

r.

O r. .- .W .a

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat and Bobo and prejudice, used byMcClendon(1985) to measuresymbolic Kluegel (1993) use the second itemas part of an index of race-targeted outcomepolicies. Like the traditionalprejudice measure,the index is the two questions, summing them,and adding formed by standardizing coefficient positive.The reliability "3" to make the scale consistently (Cronbach'salpha) forthe indexis .5481. to racetargeting is low byconvenReliability for theindexofopposition Unfortutionalstandards because theindexis made up ofonlytwoitems. nately, including other itemsin theindexis notpossiblebecause doingso ofrespondents. Usingan unreliable indexwill greatly reducesthenumber in thedependent variable,reducing thegoodincreasemeasurement error or standard errors. nessoffit ofthemodels, butshouldnotbias coefficients The divisionof the questionsinto thesetwo indexesis supportedby matrixformed correlation by these my factoranalysisof the polychoric are also consistent withthefactor questions (see app. A).'6These measures similar to those and LISREL analysisofMcClendon(1985)in whichitems and "symabove are brokenintotwo measuresthathe calls "traditional" bolic" racism. of Group-LevelVariables Measures of Threat:Description group Group threattheory predictsthat as the size of the subordinate relativeto the dominantgroup increases,threatand thus antiminority sentiment increases. threat and therefore antiminority attitudes Similarly, of the dominant group. varywiththe economiccondition black in the group,I use percentage To measuresize of the minority oftheregional For someyears, a published estimate popregion byyear.'7 ulationby race was not available, so I imputedvalues based on fitting a Values are imputedfortheyears 1972, lineartrend termforeach region. 1976, 1977,1987, 1988, 1989,and 1991.18Since the changein percentage

the Census 1977, 1986, 1993).


1

16 Aretheitems in each indextrending overtime? As Firebaugh and Davis similarly (1988)report, thequestions making up thetraditional prejudice indexareall declining overtime.Similarly, fortheindexofopposition to racetargeting, bothquestions are The downward trend trending downward. is very modest for NATRACE so there is notmuchchangeovertimein thisvariable.HELPBLK is trending downward at a faster formostof theslight in theoverall pace and is responsible trend downward indexofopposition to race targeting. 17 Statistics weretakenfrom severalCPS reports (U.S. Bureau of the Census 1978, 1984,1988)and from theStatistical Abstract ofthe UnitedStates(U.S. Bureau of

data foryears 1970, 1973, 1975, 1980-85, and 1990.

The regression to impute valuesforpercentage ofpopulation blackbyregion used 831

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology black acrossyearsis verygradualand linearfortheyearsavailable, this procedure appears to have producedextremely reliableestimates.'9 To measureeconomicconditions, I use inverseper capita income(1/ in 1982-84 dollars)by regionbyyear.20 percapita incomeX 10,000, Note thatthismakestherelation betweensimplepercapita incomeand prejudice nonlinear, so thata changein per capita incomeof fixedsize is predictedto affect prejudicein a poor regionmorethan in a wealthyone. My analysisof residualsfoundthatthisspecification leads to a better fit thanothertransformations, and it has theadditionaladvantageof reorienting theindexso thatthehigher inverseper capita income, thegreater we expectthreat to be and the greater is expectedantiblackattitude. This articleinterprets percentage minority and economicconditions as important forracial attitudes because oftheir relation to feelings ofgroup in thetheory threat thisinterpretaamongwhites.As mentioned section, tion has not been validated by researchthatmeasuresperceivedthreat it Until data allow forfurther directly. checkingon this interpretation, remains possiblethatthesemeasurescould in partbe capturing unother To thebestofmyknowledge, theliteraidentified social forces. however, turedoes notoffer another oftheeffect ofthesetwovariables explanation on racial attitudes. Individualand Cohort-Level Variables Several individual-level variablesare includedto assess how changesin individualcharacteristics have influenced prejudice and to controlfor of the groupthe estimates of the effects variablesthat could confound level variables. The variablesincludedin the regressions and the means foreach are measures shownin table 2. Most ofthesevariablesare basic demographic relatedto prejudiceby priorstudies(Adornoet al. 1950; Campbell 1971; Tuch 1987;Case et al. 1989;Steehand Schuman1992).My reviewofthe and foundthatalmostall studiescontrol foreducation, literature income, gender.These are includedin all modelsin thisarticle.(For education, 12 years = 0; family income = thousandsof constant1982-84 dollars; forgender, male = 1.)
19 The estimated fitting themodelsis extremely small,less varianceofforecast from discussion and theformula than10- (see Goldberger [1991,pp. 175-76]fora brief forpredictive error). 20 It is notpossible (1995)measure (median adjusted to useDanziger and Gottschalk's in n. 8 above) in myregressions sinceit is notavailablefor family income, discussed the of per capita incomeare takenfrom all yearsor by censusdivision. Estimates Statistical Abstract ofthe UnitedStates(U.S. Bureau oftheCensus 1973,1977-78, 1985,1988,1990,1992).

832

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

TABLE 2
DESCRIPTIVE STATISTICS FOR VARIABLES INCLUDED IN REGRESSIONS MEAN VARIABLE 1972-85* 1975-91

Trend decomposition: Year surveyed............. Southern resident.................

........................... .......................

(4.37)

7.43 .33
(.47)

13.89
(4.61)

.33

Group level (regionby year): Inverseper capita incomet ........................................ %black ........ ................................

(.47)

.88 (.11) 10.67 63


(6.65)

.79 (.12) 10.86 81


(6.37)

N at grouplevel .............. .......................... Cohortand Southern socialization: Birthyear ......... ............................... Southern resident at age 16 ........................................ Individuallevel: Traditionalprejudice(2-12) ...................................... Oppositionto race-targeted policyindex(0-7) Male ........................................ Education ......... ...............................

(17.72)

34.74 .28 (.45)

(17.92)

39.39 .27 (.45)

6.01 (3.01) .47 (.50) .21


(3.12)

3.91 (1.69) .47 (.50) .67


(2.97)

Currently unemployed........................................ Currently working.................. Unhappyt......... Familyincome?............. Urban residence.............. . Suburbanresidence ......................

.03 (.17) .59 .11

.03 (.17) .61 .10

............................... ........................... ..........................

(.49)

(.49) (.30) 24.66

24.51 (16.38) .26


(.44)

(.32)

(15.38) .23
(.42) (.46)

........................................

.28

Otherraces in neighborhoodll ..................................... N at individuallevel ........................... .............

.42 (.49) 7,261

(.45)

.32

.44 (.50) 4,934

NOTE.-NOS. in parentheses are SDs. The sample consistsof all whiteGSS respondents presenton all variables.For traditional thedependent variableis available for1972,1976-77, 1980,and prejudice, 1984-85. For race targeting, the dependent variableis available for1975, 1983-84, and 1986-91. * This is theunweighted mean computed overall individuals (forindividualvariables)and all regionby-time groups(forgroup-level variables). to be in unitsbased on the averagevalue ofthedollarin 1982-84. t Dollars are adjustedforinflation could choosefrom "veryhappy," "pretty happy,"and "nottoo happy."This variableis * Respondents coded "1" if the respondent chose "nottoo happy,""O" otherwise. I createdthis variable by recodingrespondents' ? The GSS asks forrespondent's incomecategory. incometo be equal to themidpoint oftheir incomecategory and thenadjustingto thousands ofconstant 1982-84 dollars. 1I This is based on respondent's in the survey. self-report

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology Religionand occupationare also includedby someauthors. Religionis missing formanycases and is not available in a comparableformat for all of theyearsin thisanalysis,so it does not appear in the modelshere. Case et al. (1989) foundthat,when educationis controlled, occupational prestige has no statistically significant impacton threeof the prejudice measuresused to make up my prejudiceindexes.Thus I do not include occupation. Three othervariables that I consideredas controlvariables are residencein an urbanarea (yes = 1),residence in a suburb(yes = 1),and the presence ofother racialgroups in therespondent's neighborhood (yes= 1). As a reviewer of an earlierversionof thisarticlepointedout,thesethree variablesare likelyto be a function as well as a cause of racial attitudes: respondents may choose where to live in part based on theirattitudes towardblacks.Since I am trying to estimate thetotaleffect ofpercentage I of populationblack and inverseper capita incomeon racial attitudes, excludethesethreepotentially variablesfrom the modelsin endogenous the body of the article. To testthesensitivity of myresults to theexclusionof thesevariables, ofthefinal appendixB showsestimates modelfrom theresult section with urbanresidence, racesin theneighborhood suburbanresidence, and other The models in tables 3 and 5 below show resultsunderthe controlled. and thatthe variablesurban residence, suburbanresidence, assumption otherraces in the neighborhood racial attitudes, while do not influence the modelsin appendixB show results underthe assumption thatthese variablesinfluence racial attitudes but are not influenced by racial attitudes. Most likelythe true model is somewhere betweenthese two extremes.Includingthese variables reduces the coefficients of the group threatvariables, but the basic substantivestoryof the article is unchanged. I includebirth of cohort, To assess theinfluence year(1950 = 0). This and periodeffects and is allows forthe separation of cohortreplacement the basic methodFirebaughand Davis (1988) use to decomposechange in prejudicebetweencohort and period.The importance ofcohort change in priorworksuggests influthatearlyadult experiences have important ences on racial attitudes thattendto persist across the lifecourse.2' I cannotinclude Since birth year,age, and yearare linearly dependent, all threeof thesevariablesin a modelat once. I includecohortand year
21In their study ofcollective memories, Schuman and Scott(1989)find thatmemories

ofnational or worldevents from earlyadulthood are especially likely to be recalled fortheir byrespondents as important own lives.This suggests thatanygenerational imprinting ofpolitical events on cohorts is most likely to occur during lateadolescence or earlyadulthood. 834

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat surveyed (1972 = 0) in themodel,thusassumingthatage does notaffect racial attitudes. Steehand Schuman(1992,pp. 343-44) reviewtheliteraon racial attitudes and concludethatage does notsubtureon age effects influence racial attitudes netof cohort effects. indistantially (Additional vidual variablesare currently unemployed [yes = 1], currently working the respondent feelsunhappy[yes = 1]). [yes = 1],and whether Models and Methods:A MultilevelModel withCrossed VarianceTerms To testthetheory, thetwo attitude indexesin thestudyare regressed on individual-level variables like educationand group-level variables like One techpercentage of the populationblack in the regionof residence. nique sometimes used forthissortof analysisis to assigneach individual variable fortheirgroupand thento model the value of the group-level theoutcomewitha singleregression. least squares (OLS) Usingordinary to estimate theseequationsis notappropriate, because theerror however, terms amongindividualswithinthe same group-level unitwill be correlated, resulting in incorrect standarderrorsand inefficient model estiin underestimations ofthestanmates.This is particularly likely to result dard errors forthe group-level independent variables.22 Instead,I estimate the modelsusinga package forhierarchical linear estimate modelswitha hierarchical structure modelsthatcan efficiently and provide appropriately calculated standarderrors.23 Recall that the group-level unitsare the crossclassification of regionsand years.Correthe model divides the variance into threeseparate compospondingly, terms: one reflecting thevarianents, and the modelincludesthreeerror tion among individualswithinthe same group-level unit,one reflecting variationacross years,and one reflecting variationacross regions.Each group-level (regionby year)unitbelongsto two crosscutting clusters: region and year.This error structure best reflects the grouping of the data

22 In my case these arepercentage variables ofpopulation blackand inverse percapita income. Because group-level independent variables are bydefinition thesameforall members ofthesameunit, controlling forgroup-level variables inducesa correlation in theerror termamongindividuals within thesame group-level unit.This results in estimates oftheOLS standard errors ofthegroup-level variablesthatare biased downward. Whenthere are largenumbers ofindividuals in each higher levelunit(as is thecase inthis study), this can lead toa drastic underestimation ofthetrue standard error ofthegroup-level For a formula forthebias in OLS estimates variables. of a bivariate withcorrelated error see Goldstein regression terms, (1995,p. 25). 23 For an introduction to multilevel models,see Brykand Raudenbush(1992) or Goldstein (1995). Chap. 8 of Goldstein modelswith (1995) focuseson multilevel crossed suchas thoseused in thisstudy. variancestructures,

835

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology and willprovidea moreaccuratecalculation ofstandard errors thanmodIn addition,by dividingthe variance els witha simplererrorstructure. intobetweenregion, betweenyear,and betweenindividualcomponents, I can see how addingor deleting variablesexplainsthevariancein these threecomponents separately. Simplifying the actual computational details,the procedure is closely relatedto a two-stage model.In thefirst stagethe dependent variableis regressed on theindividual-level independent variables,allowinga separateintercept foreach regionby timeunit.In thesecondstage,theinterin thefirst ceptsestimated stageare regressed on thegroup-level independentvariables.The use of the intercepts is similarto usingthe adjusted meansin an ANCOVA whereeffects are linearand additive, exceptthat thegroupeffect is treated as randomrather to thanfixed (theconnection ANCOVA is discussedby Brykand Raudenbush[1992,pp. 18-20]). The actual modeland computations the are morecomplex.I estimated modelshere withthe program ML3E (Prosser, Rasbash, and Goldstein 1991),employing techniquesin Rasbash (1993) to estimatethe model's A few equationshelp to describethe procecrossedvariance structure. and dure.We can imaginethatthereare r = 1, 2, . . . , R - 1,R regions a totalofR X theseform y = 1, 2, ... , Y - 1, Y years.Cross classified, Y cells at the grouplevel. There are i = 1, 2, . . . , I individualswithin each of these groups(I varies across regionand year-but forheurisif one thinks tic purposesit makes no difference of it as identicalforall region-by-year units).The model consistsof a seriesof structural equations.First,thereis a set of equations,one foreach of theyear-by-region groupings:
Piry = Cry + XYryb + eiry (1)

r and yeary on theapprois thescoreforindividuali in region whereP,ry and priateracial attitude index;XzY. is a 1 X j row vectorof individualofindividualand cohort cohort-level variables(j is thenumber predictive predictive variables);b is aj X 1 vectorofthecorresponding coefficients; and cry is an intercept. The cohortvariables that are includedin X are at age 16, and an a dummy variableforplace of residence year of birth, termbetweenthe two. Group threat thatthe interaction theory suggests foreach region-by-year cell will vary accordingto percentage intercept black and inverseper capita income.Symbolically
Cry = Zryg + SrS +

Tyt+ (Sr * Ty)q + ur + vy.

(2)

The intercept of from the individual-level equations,then,is a function X 1 a vectorof group-level Z k k is Z. is a row where variables, vector, 836

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat ofvariablesat thegrouplevel.The group-level thenumber variablesincludedin Z are somecombination ofpercentage black,inverseper capita income, and an intercept depending on themodel.The corresponding vectorof estimated coefficients is g, whichis a k X 1 vector.The equation specifies how the intercept changes based on group-level variables,so thereis no individual-level error term. In thisarticle, I am particularly interested in explaining thetimetrend in prejudice.To do so, I include and Southern/non-Southern difference linear trendtermsforyear surveyedand year of birth,similarto the method used byFirebaughand Davis (1988)to decomposechangein prejIf the trendin racial attitudes is udice intoregionand year components. linear (as foundto be the case forprejudiceby Firebaugh and Davis [1988];see also figs.1-4 below),thenthisprovidesa good way to measure ofadditionalvariableson explaining thelinearchangein racial theeffect forpotential omitted variattitudes. These variablesalso help to control ables:they setup a minimum standard thattheindependent variablemust explainthe trendbetterthan a lineartrendforthe South and the nonSouth.In equation(2), T is a variable forsurveyyear (a scalar) and t is itscorresponding coefficient; variableS is a dummy variable(a scalar)for Southern residence withs as its corresponding coefficient; q is the coefficientof the interaction betweenthesevariables. The variancecomponent associatedwithregionis u, and the variance associatedwith timeis vy.Followingthe usual assumptions component made in multilevel and models,themodelassumesthattheerrors (ur, vy, are independent ofeach other and uncorrelated withtheindependent ei,y) variables.The modelthatML3E estimates is thereducedform equation, from substituting (2) into(1): = Z,yg+ SrS + Tyt+ (Sr X Ty)q + X,yb + ur + P,,y
vy

+ eiry (3)

ML3E estimates The results in thetablesfollowthisform. this presented model withiterative generalized least squares (see Goldstein[1995, sec. 2.5] fordetails).The parameter estimates are consistent, and ifur,vy,and are normally distributed, the estimates are equivalent to themaximum ei,y likelihood estimates and, therefore, efficient. This model constrains the effect of the individual-level controls to be the same across regionsand years.In some models,I relaxedthis conacrossregions straint, allowingtheslopesto varyrandomly and years(not are quite similarto thosethatconstrained theslopes shown).The results in to be fixed, withthemajor differences beingincreasedstandarderrors themodelthatletstheslopes vary.I also testedforinteractions between theindividual-level variablesand thegroup-level variables(as I hypothe837

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology sized earlier[1995]).For mostindividual-level variables,theinteractions were small and statistically insignificant.24 The group threattheory impliesthat coefficients in g corresponding to percentage black and inverseper capita incomeshould be positivein the structural equation(3). Concerning equations,the group-level threat indicators shouldsignificantly predict the intercept in equation(1). Finally,one additionalpotentialproblemmustbe considered. At the grouplevel,the data can be thought of as like a panel: nine timeseries, one foreach region. As DiPrete and Grusky(1988) pointout,the grouplevel errortermvymay be autocorrelated. This possibility is considered in the resultssectionbelow. RESULTS For each dependent variable,two graphsand a seriesof modelsin tables I startby discussing 3 and 5 below give the results. the graphs. in Two Indexes of Racial Attitudes Trends and RegionalDifferences To begin,it is illuminating to examinetrendsin racial attitudes forthe towardtraditional period 1972-91 usingthe GSS data. Attitudes prejudice and oppositionto race-targeted policies to achieve racial equality have bothchangedover time. A graphofmeantraditional prejudicebybirth 1. yearis shownin figure from thefigure, there is a steadyand evendeclinein traditional As evident raisedin theSouthand thoseraisedelsewhere. prejudicebothforcohorts The trendslowed a bit forcohorts who lived outsidetheSouthat age 16 and were bornafter1950. Figure 2 graphsaverage traditional prejudice For boththe South and non-Southern by yearforSouthern respondents. withthetrendslightly faster and non-South, prejudicedeclinessteadily, in recent The graphalso showsa yearsin the Souththanthenon-South. in traditional persistent regionaldifference prejudice. 3 and 4. This index is graphedin figures Oppositionto race targeting is available forfewer respondents perperiodthantheprejudiceindex,so 1 and 2 due to sampling thanin figures is moredispersion variation. there to race targeting. The downFigure3 graphscohort changesin opposition
24 Tables and a brief written summary justifying theseconclusions are availablefrom Theseresults a previous ofmine(Quiltheauthor uponrequest. contradict hypothesis 2). I also tested fora multiplicative relation between economic lian 1995,hypothesis in myearlier also hypothesized situation and subordinate groupconcentration, reof the interaction search(1995). The coefficient between percentage of population was consistently black and percapitaincome verycloseto zero.

838

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

o Lived outside

South at 16

A Lived

2n South at Age 16

00

-10 9
C)

10 40 30 20 Year of Birth, 1900=0

50

60

to theN used to computethe mean).


4m 0

is proportional prejudice(size ofsymbol FIG. 1.-Cohort changesin traditional

--

a, CD

CZ

A Southern Respondents Non-SouthernRespondents _______________________________________

a, a)
c0

C-6-

a), co

5-

72

76

80 Year of Survey

84

is proportional FIG. 2.-Period changesin traditional prejudice(size ofsymbol to theN used to computethe mean).

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

o Lived outside

of South at 16 A Lived in South at 16

4-1

6-

o~
at)

~~~ocJ
23

a)

4-1

41 -10 0 10

Year of Birth, 1900=0

20

30

40

50

60

70

to race targeting by cohort (size of symbolis proFIG. 3. Averageopposition portional to N used to computethe mean).
Q) 3

CD

o Non-SouthernRespondents cn 5-

Southern Respondents

0 a)

75

80

85 Year of Survey

90

is proportoracetargeting FIG. 4.-APeriod changesinopposition (sizeofsymbol tionalto the N used to computethe mean).

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat to race targeting is much weakerthan it is for ward trendin opposition traditionalprejudice. There is also convergence between respondents raised in the South and thoseraised elsewhere. Figure4 shows changes in oppositionto race targeting across years.There is littlechange over timefortheSouthand a modestdeclinein prejudiceacrossyearsoutside the South.As fortraditional in prejudice,however,regionaldifferences race targeting persist from1975 to 1991. These graphs, our expectations based on priorliterature. then,confirm Traditionalprejudicehas been declining from1972 to 1985,while race targeting has declinedonlyslightly from1975 to 1991. TraditionalPrejudice The analysisusingthetraditional-prejudice indexas thedependent variable is shown in table 3. Models 1-3 show a set of baseline resultsfor comparison. The variablesof theoretical interest are added in models4, 5, and 6. Model 1 is includedto provideestimates in prejuofthetotalvariation dice amongyears,regions, and individuals.As shownformodel 1 at the bottomof table 3, about .798 or 8.6% of the varianceat the grouplevel is between regions, about .303or3.2% is between years, and theremaining is betweenindividualswithinregionand year. Model 2 adds a lineartimetrend, a dummy variableforSouthern residence,and an interaction betweenthetwo. Since thereare no othervariables in thismodel,thecoefficients ofyear,Southern and their residence, in model 2 describethe overalltrendin prejudice. interaction in table 3, we can see that the traditionalFrom the coefficients prejudiceindexdeclinesby .093 per yearfornon-Southerners and about .139peryear(.093 + .046)for Southerners. Model 2 also showsthatSouthin 1972 score about 2.1 pointshigher erners on the traditional-prejudice indexthan do non-Southerners. At timesfiguring how muchof the trendis explainedfrom comparing coefficients across models can be confusing since thereare interactions in themodel.Supplementing (such as year X Southern residence) table 3 is table 4, whichcalculatestheresidualor unexplained trend in prejudice from table3 separately for non-Southerners and Southerners. The residual trendis the portion of the trendthatis not explainedby the substantive (groupthreat or individual-level) variablesin themodel.For model2, the first two lines show thatthereis an unexplaineddeclinein prejudiceof .093 peryearfornon-Southern respondents and .139 forSouthern respondents. Table 4 also calculatestheunexplained gap betweentheSouthand the restof the country in 1972 and 1985. Because prejudicedropsfaster for 841

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

TABLE 3
ESTIMATES OF EFFECT OF THREAT INDICATORS ON TRADITIONAL
MODEL VARIABLE

PREJUDICE

INDEX

Year surveyed Southernresident Year X Southern resident .-.046** Inverseper capita income %black Birthyear Southern resident at age 16 Birthyear X Southern residentat age 16 Education Education X birth year ......... Familyincome .................... Male M ae ......................................... Currently unemployed ........... Currently working ................. Unhappy .................... 6.004*** (.308) .798 .303 8.131

-.093*** -.041*** -.047*** (.013) (.011) (.012) 2.105*** 1.112*** 1.056** (.312) (.336) (.325) -.033 -.032 (.023) (.021) (.021)

-.051*** (.002) 1.028*** (.135) -.011** (.004)

-.046*** (.002) 1.004*** (.134) -.011* (.004)

-.002 -.032** (.011) (.014) 1.061*** -.218 (.294) (.300) -.029 -.011 (.019) (.023) 3.585*** (.722) .055*** (.013) -.037*** -.037*** (.002) (.002) .769*** (.128) -.011** (.004) -.374*** (.016) -.005***
(.001)

.757*** (.128) -.011* (.004) -.373*** (.016) -.005*


(.001)

.................... Intercept var(u) region .................... var(v) time.................... var(e) individuallevel ...........

6.093*** (.191) .152 .045 8.135

-.021*** -.008*** -.008*** (.002) (.002) (.002) * .245*** .251*** .254** 2 5 * .51 * .54 (.066) (.063) (.063) -.054 -.180 -.174 (.190) (.182) (.182) -.252*** -.037 -.036 (.075) (.072) (.072) .212* .213* .456*** (. 100) (.096) (.096) 4.879*** 5.532*** 5.343*** 1.755* (.183) (.191) (.181) (.685) .OOOa .135 .117 .108 .030 .033 .022 .060 7.139 6.957 6.352 6.349

terms. The models are SEs. These modelsincluderegionand year error NOTE.-Nos. in parentheses are estimated usingthe 1972-85 sample. a The value is assumed.I had trouble to converge the estimates whenI did not constrain this getting thisis frequently a problem whenvariance variancetobe zero.In myexperience, withtheML3E program estimates ofone or morevariancecomponents are close to zero.The suddenjump in the estimated varithe constraint of no residualvarianceacross regions. ance of timein model 6 is a resultof imposing
**P ***

*P < .05.
< .01.

P < .001.

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat
TABLE
RESIDUAL YEAR, REGION, AND COHORT

4
ON TRADITIONAL MODEL PREJUDICE

EFFECTS

2 Residual change per year: Non-Southern respondent ............................ -.093 -.139 Southern respondent .................................. Residual non-South/South gap: 1972 .................................... 2.105 1985 .......... 1.510 .......................... Coefficient of year of birth: Non-Southern resident at age 16 Southern resident at age 16 . .-.063 Gap between non-Southern and Southern . residents .1.028

3 -.041 -.074 1.112 .683 -.051

4 -.047 -.079 1.051 .636 -.046 -.057 1.004

5 -.032 -.061 1.061 .681 -.037 -.048 .769

6 -.002* -.013* -.218* -.362* -.037 -.048 .757

in table3. The modelnumbers NOTE.-Table values werecomputed from mainand interaction effects to the modelnumbers in table 3. Model 2 is yearand Southernresident correspond only,model3 adds the cohortvariables,model4 adds the individualvariables,model 5 adds education, and model 6 adds the groupvariables. * The null hypothesis = < thatcoefficient 0 at P .05 cannotbe rejected.

in the traditional Southerners than fornon-Southerners, the difference prejudicescale between theSouthand non-South narrows from about 2.1 in 1972 to about 1.5 in 1985. interUsinga lineartimetrendvariableand a Southern/non-Southern action variable are onlyvalid means to decomposethe trendif changes and regional in prejudiceare explainedwell by thesevariables. variation Does a lineartrendtermexplain the timetrendwell? Does a Southern variableexplainthebetween-region dummy variation well?A conciseanswerto thesequestionsis providedby comparing theestimates of region and timevariance in models 1 and 2 of table 3. The lineartrendterm and Southern dummy variablesexplainabout 81% ofthebetweenregion variance and 85% of the betweenyear variance.This suggests that the declineacrossyearsin prejudiceis linear, and thatthegap between Southernand non-Southern regions in prejudice capturesmostofthevariation For traditional betweenregions. between prejudice,the gap in attitudes the South and the restof the United States is stillthe most important regionalattitude cleavage. Model 3 adds threemeasuresto examinecohorteffects and regional socialization: a dummyvariable forSouthern year of birth, residenceat termbetweenthesetwo variables.The trend age 16, and an interaction is the difference explainedby cohortreplacement betweenthe residual trendin model 2 and model3. Table 4 showsthatthe residualtrendde843

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

of Sociology American Journal From - .09 in model2 to -.04 in model3 for non-Southerners. clinesfrom in prejuabout 55% ofthetrend amongnon-Southerners, theseestimates, 45% is due to pewhiletheremaining dice is due to cohortreplacement, declinesfrom ofyearforSoutherners Fromtable4 theeffect riodeffects.25 -.139 to -.074. So this suggeststhat less than half (about 47%) of the withthe remaining is due to cohortreplacement, trendforSoutherners are quitecomparable These results half(about 53%) due to periodeffects. to Firebaugh and Davis (1988), who found that for non-Southerners less than half the trendwas explainedby cohortreplacement, slightly more than half was explainedby cohort slightly while forSoutherners replacement. difference Model 3 also revealsthatalmosthalfthe non-South/South (thegreater prejudiceof in prejudiceresults Southern socialization from in those raised in the South). This may in part resultfromdifferences raisedand theNorthern betweenthe Southern individualcharacteristics raised. in table 3 is reof (year X Southern) residence Finally,the coefficient socializamodels3 to 4. This showsthatwhenSouthern duced a bitfrom in prejudiceover tionis controlled for,some of theregionalconvergence timeis explained.This is similarto resultsfoundby Wilson (1986) and in racial Kanagy (1994), who findthat one reason forthe convergence is the migration betweenthe South and the restof the country attitudes of people socializedoutsideof the South (who tend to be moreracially natives)intothe South in the 1970s and 1980s. liberalthan Southern to see how theyexplain Models 4 and 5 add individualcharacteristics the trendacross cohorts(the the trendacross years (the period effect), theSouthand thenon-South. and theregional gap between cohort effect), variablesexcepteducationand Model 4 adds all of the individual-level of educationand birth theinteraction year.Model 5 adds educationand of educationand birth the interaction year. explainthe trendtowarddeDo changesin individualcharacteristics thaneduother Model 4 adds individualcharacteristics clining prejudice? The most The coefficients changelittle. cationto assess their importance. other is that adding the individualcharacteristics noticeabledifference in prejudiceto increaseto -.046 thaneducationcauses theresidualtrend variablesother peryear,an increaseofabout 15%. So theindividual-level acrossyears(theperiod thaneducationdo notexplaintheprejudicetrend otherthanedueffect)-in fact,thechangesin individualcharacteristics overtime. cationare in a direction as to cause prejudiceto go up slightly residualtrend larger. Thus,addingthesevariablesmakestheunexplained
- theresidual = (prior 25 Percentage trend trend ofthetrend modelresidual explained trend. totalresidual withthemodelvariablesincluded)/the

844

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat Individual characteristics can also explain changesin prejudiceover table 4 also shows the residual cohorts.To investigate this possibility, trendacross birthcohorts. The individual-level variablesin model 4 do in prejudiceacrosscohorts a bitbetter at explaining differences thanthey did over time.As table 4 shows,theyexplainabout 10% of the change in prejudice across cohortsfor both non-Southerners and Southerners (from .051 to .046 fornon-Southerners and from .063 to .057 forSoutherners).Individualvariablesin model4 do notexplainanyofthedifferences in attitudes across regions(non-South/South). Model 5 adds themostimportant individual-level variables:education and theinteraction ofeducationand birth Fromtable4, education year.26 in racial prejuplus theother individualvariablesreducethelineartrend dice forSoutherners by about 20% (.041 in model 3 to .032 in model 5), and forSoutherners by about 17% (.074 to .061). Thus, the increasein for educationacrossyearsexplainsabout 15%-20% of the periodeffect both Southerners and non-Southerners. educationhas inUnexpectedly, creasedovertimeamongall age groups, and thisincreaseexplainsa small part of the periodeffect towardreducedprejudice.On the otherhand, educationexplainsalmostnoneofthe South/non-South gap in prejudice. educationhas often been credited withsome of thedecline Increasing in prejudiceacrosscohorts (e.g.,Hymanand Sheatsley 1956).By comparingthebirth yearand birth yearbySouthern resident at age 16 interaction frommodels 3 and 5, we can see how educationalchanges may have Once again,table playeda partin changing prejudiceacrossbirth cohorts. 4 has done some calculationswiththe interaction variables to examine trendacrossmodels.Comparing coefchangesin theunexplained cohort ficients leads to theconclusion thateducationexplainsabout 25% of the cohorteffect -.063 to -.048 for (-.051 to -.037 fornon-Southerners, Educationalso explainsslightly Southerners). less than25% ofthedifference in prejudicebetweenSouthern-raised individualsand individuals raisedoutsideof the South.One reasonrespondents who were Southern residents at age 16 are more prejudiced,then,is because theyare less educated. From models4 and 5, changein individualcharacteristics, principally is responsible forabout 25% of theeffect of cohort and about education, 15%-20% oftheperiodeffect for bothSoutherners and non-Southerners.27
26 I also tested for an interaction ofeducation and survey yeartoexamine thepossibilofeducation in thesurvey. itythattheeffect changed overyears The resulting coefficient was .0039(SE = .0024;thenullhypothesis cannot be rejected at P < .05).Thus there does notappearto be anysubstantively ofeducation largechangein theeffect overyearsofthesurvey. 27 Modelsthat excludetheinteraction between birth yearand education (notshown) trend to model5. giveresidual results almostidentical

845

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

of Sociology American Journal in prejalso explainabout 25% ofthedifference Individualcharacteristics thosesocializedin theSouthand thosesocializedelsewhere. udicebetween gap. Overall, then, They explain none of the residualnon-South/South are or the regioneffects theperiodeffect, less thanhalfthecohorttrend, of This is reminiscent explainedby changesin individualcharacteristics. in prejudicebetweenthe that the difference Middleton's(1976) finding cannotbe accountedforby differences Southand therestof the country in individualcharacteristics. betweeneducationand birthyear. Model 5 also adds the interaction ofeducation so thattheeffect is significant and negative, The interaction This suggests thatthe cohorts. on reducing overbirth prejudiceincreases attitudes ofeducationmayhave changedin a way thatinfluences content Campbell (1971) foundthesame reabout race. As discussedpreviously, and mine,is thatthis the 1960s.His interpretation, sult usingdata from in higher and,accomeducation likely reflects theriseofthesocialsciences inofenvironmental acceptanceof theimportance panying it,a growing thiseffect is roughly linearover on behavior. To myexamination, fluences thanconclusive, however, birth This evidenceis suggestive rather year.28 are possible. and otherinterpretations of populationblack and invariables,percentage The two group-level theory preverseper capita income,are added in model 6. Group threat in incomeoran increase in proportion ofthepopuladictsthata reduction increase tionblack shouldincreasethreat amongwhitesand, as a result, show thatbothinverseper capita incomeand perprejudice.The results consistent with prejudiceorientations, centageblack increasetraditional A changein per capita incomefrom the theory of groupthreat. $15,000 of to $20,000,then,is expectedto cause a regionalprejudicereduction in percentage of population about .6. Similarly, a 5% regionaldifference in prejudiceof .27. black is expectedto resultin a regionaldifference is onlymoderate, the the size of theseeffects Althoughsubstantively the that these null conclusion supports clear rejectionof the hypothesis in thedirections theory. suggested by groupthreat variablesexerteffects income term means that the the linear per capita The inclusionof year the factthatprejudicedeclinedfrom1972 variableis notjust capturing to 1991 whileper capita incomeincreased:regionalper capita incomeis of the trendthan fitting a separatelineartrendterm a betterpredictor is also sugfortheSouthand thenon-South. Finally,correct specification
28

Thissuggests that the and insignificant term was small statistically. substantively wasa gradual over time. inthecoefficient ofeducation process change 846

in theeffect ofbirth (notshown).The resulting yearon theslopeofeducation arity

X education) for some nonlinea (birth toallow term, I tried including year squared

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat ofyear,Southern gested byhow neatly thefinal modelleaves theestimates residence, and (year X Southern residence) close to zero. What effect have changesin thesecollective variableshad on the dein table clinein traditional prejudice? Extrapolating againfrom theresults 4, thegroup-level effect ofincreasing percapitaincomeovertimeexplains about 70% oftheperiodeffect towardreducedprejudicefornon-Southern respondents, and about80% oftheperiodeffect forSouthern respondents. ofthetotaldeclinein prejOverallthesevariablesexplainaboutone-third udice forboth non-Southerners and Southerners (thetotaldeclineis the sum of the period decline and the cohortdecline).Afterincludingthis variable,thereis no longera statistically trendacross years. significant in perceivedgroupthreat This analysissupports the idea thatreduction from increasedper capita incomehas been one factor responsible forreducingoveralllevels of prejudice. Likewise,regionalracial composition and per capita incomeexplain in prejudice betweenthe South and nonabout 50% of the difference South.This suggests thatthe South is moreprejudicedin large partbecause the largernumberof blacks in the South is more threatening to The other whites. halfoftheSouth/non-South gap is explainedby Southernsocialization so theresidualSouthterm and individual characteristics, is close to zero in model 6. Oppositionto Race Targeting The analysisof opposition to race-targeted policiesis shown in table 5. Table 6 givestheresidualtrends by region (non-South/South) after model I frequently switch controls, just as table 4 did fortable 3. As previously, back and forth from betweenresults tables 5 and 6. The downwardtrendin oppositionto race-targeted policiesis much more subtle than the downwardtrendin racial prejudice.Comparing model 1 and model 2 of table 5, the linearyearterm, a dummyvariable forSouthern and their interaction residence, explain45% of thevariance in prejudiceacross years and 65% of the variance across regions.The linearyeartermand Southern dummy variabledo not providenearlyas an of the good approximation towardrace-targeted changesin attitudes policyas theydo fortraditional prejudice.This reflects thelack ofa clear in opposition trend to race-targeted policy. For non-Southern respondents, to race targeting opposition declinedby .027 per yearfrom1972 to 1991, whileforSouthern whitesopposition to race targeting declinedby an almostimperceptible .002 per year. There is, then,a trendonlyfornonSouthern Because opposition to race targeting respondents. dropsoutside the South from1972 to 1991 while in the South oppositionto race tar847

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

TABLE
ESTIMATES OF EFFECT OF THREAT

5
ON INDEX OF OPPOSITION TO RACE

INDICATORS TARGETING

MODEL VARIABLE 1

Yearsurveyed Southern resident


Year X Southern resident

-.023*** (.007)
.129 (.206) .025 (.013)

-.023** (.007)
-.014 (.217) .032* (.013)

-.022** (.007)
-.020 (.219) .032* (.013)

-.019* (.007) .005


(.212) .029* (.013)

-.006 (.009) -.667**


(.218) .035** (.012)

Inverseper capita income


%black Birth year -.005** (.002) -.007*** (.002) .017 (.092) -.003 (.002) -.047 (.091)

1.871***
(.43 1) .031*** (.009) -.003 (.002) -.046 (.091)

Southern resident at
age 16 .018 (.092)

resiBirthyear X Southern dentat age 16


Education

-.008* (.003)

-.007* (.003)

-.007* (.003)
-.113***
(.0 1 1)

-.007* (.003)
-.113***
(.0 1 1)

Education X birth year Familyincome


Male Currently unemployed Currently working Unhappy Intercept ..................... var(u) region ..................... var(v) time ..................... var(e) individual level ........... 3.878*** (.095) .070 .039 2.761 4.080*** (.117) .024 .022 2.763 3.969*** (.121) .022 .023 2.744

-.002***
(.001)

-.002***
(.001)

.000
(.002) .054 (.050) -.014 (.147) .141* (.058) .145 (.080) 3.812*** (.135) .023 .023 2.737

.005**
(.002) .062 (.049) -.115 (.146) .178** (.058) .094 (.080) 3.398*** (.132) .019 .021 2.680

.005**
(.002) .059 (.049) -.118 (.146) .179** (.058) .092 (.079) 1.822*** (.444) .000 .016 2.682

The models terms. are SEs. These modelsincluderegionand yearerror NOTE.-Nos. in parentheses are estimated usingthe 1975-91 sample.
* P < .05. ** P < .01. *** P < .001.

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat
TABLE
RESIDUAL YEAR, REGION, AND COHORT EFFECTS

6
ON OPPOSITION TO RACE TARGETING

MODEL 2 3 4 5 6

Residual change per year: Non-Southern respondent .................... Southern respondent ............................. Residual non-South/South gap: 1972 .......... ...................... 1985 .......... ...................... Coefficient of year of birth: Non-Southern resident at age 16 .-.005 Southern resident at age 16 .-.012 Gap between non-Southern and Southern residents at age 16 (birth year = 1950) who lived in the South at age 16 ..018*

-.027 -.002* .129* .456

-.023 .009* -.014* .399

-.022 .010* -.020* .399 -.007 -.014

-.019 .010* .005* .386* -.003 -.011

.006* .040 -.667 -.218* -.003 -.010

.017*

-.047*

-.046*

NOTE.-Table values werecomputed from mainand interaction effects in table5. The modelnumbers in table 5. Model 2 is year and Southern correspond to the modelnumbers resident only,model3 adds the cohortvariables,model4 adds the individualvariables,model 5 adds education, and model 6 adds the groupvariables. * The null hypothesis = 0 at P < .05 cannotbe rejected. thatcoefficient

betweenthe non-Southand the getingdoes not change,the difference South growsfrom.129 in 1972 to .456 in 1985. at age 16, and Model 3 of table 5 adds birthyear,regionof residence an interaction in birthcohortfor betweenthe two. A 10-year difference non-Southern respondents translates intoa puny(but statistically signifi.048 forSoutherners is cant) change in the index. The estimatedeffect stronger but stillsubstantively small (10 yearstranslate intoa difference of.1). Thus cohort has had little on attitudes toward replacement influence race-targeted policies.A major difference betweensourcesof changein attitudes towardrace-targeted policiesand traditional prejudice, then,is thatwhiletraditional prejudiceis sharply differentiated by cohort, young cohortsare only slightly morefavorably disposed toward race-targeted policythan oldercohorts. Models 4 and 5 investigate of the individual-level variables. the effect As models4 and 5 in table5 show,mostvariableshave effects in thesame direction as thosetheyhave on traditional prejudicebut usuallyweaker. An exception is theincomevariable,whose effect is reversed: incomereduces traditional prejudicebut increasesopposition to race targeting. As we can Model 4 adds theindividual characteristics excepteducation. see from table 5, mostof thesevariableshave weak effects on prejudice. 849

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology Comparingthe residualtrendover yearsformodels 3 and 4 (easiest to see in table 6), we can see thatincluding individualcharacteristics other than educationexplainsalmostnone of the period,cohort, or regioneffects.Some of the coefficients includedto measurethe residualtrendin table 6 changea bit,but the changesare substantively small. Model 5 adds education and theinteraction ofeducationand birth year. Education explains about 17% of the period effect in attitudes toward race targeting amongnon-Southerners (explaining about 14% ofthetrend overall).29 Thus, educationcontributes onlymodestly towardexplaining an alreadymodesttrend.The otherindividual-level variablesdo not explain the trendat all. Model 5 of table 5 also includestheinteraction betweeneducationand birth theeffect of educationin reducing year.Like traditional prejudice, to race targeting is stronger cohorts thanit is opposition amongyounger This suggests thata changein thecontent ofeducaamongoldercohorts.30 suchas growing socialsciencecontent, tion, mayhave influenced attitudes towardrace targeting. Model 6 includesthe two group-level sourcesof prejudice, percentage ofthepopulationblack and inverseper capita income.As table 6 shows, the two group-level variables reducesthe periodeffect across including to close to zero. It does not explainthe trend yearsfornon-Southerners forthe South.After forthe groupthreat variwell,however, controlling lessopposedto race-targeted ables,theSouthin 1972is significantly policy thanthenon-South thattheSouthmovestowardpar(themodelsuggests over time).This unexplainedresidualtrendsugitywiththe non-South in attitudes sourcesof the trends togeststhatthereare otherimportant thatare not ward race targeting forthe South (and maybebothregions) includedin the model. In factthelack ofa trend in opposition remains to race targeting somein table 5. Most ofthevariablesthat of an enigmafrom theresults thing in table 5 are changing to race-targeting are good predictors ofopposition in a way that should cause a trendtoward reducedoppositionto race similarto traditional targeting prejudice(althoughthe trendshould be weaker than the trendfortraditional prejudicebecause thereis almost no cohorteffect). But the actual timetrendshows thatthereis a modest cause ofthedisparity downwardtrend onlyfortheNorth.The mostlikely
is .002 7 (SE = .0087, The coefficient foran interaction between yearand education notsignificant at P > .05). Thus theeffect of education does notappear to change overyearsin thesurvey. significantly 301 also tried an interaction between education and birth yearsquaredto allow for in theeffect The coefficient nonlinearities ofbirth of yearon education (notshown).
29

the interactiontermwas very small and statistically insignificant.

850

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat the between modelpredictors is variablesomitted from and overalltrend model thathave changedover timeforthe South in a way as to offset the trenddownwardfrom variablesin model 6. the substantive This does notinvalidate thefact, however, thatbothofthedemographic variablesworkas thegroupthreat circumtheory predicts: pooreconomic stancesand a largerblack populationare associatedwithgreater opposiis similarto their tion to race-targeted policy.The size of these effects effects on traditional prejudice-which is to say, moderately important but not huge. A change in average per capita incomefrom$15,000 to to race targeting $20,000is predicted to decreaseopposition by .3. A 5% in percentage of the populationblack betweenregionsis predifference of .16. In this to race targeting dictedto cause a decreasein opposition case the increasein incomefrom1975 to 1991 only resultedin a very In contrast, the to race targeting. modestdownwardtrendin opposition in results from a substantial cohort effect comtrend traditional prejudice bined witha stronger per capita incomeeffect. CheckingforAutocorrelation for Since thegrouplevel includesa timeseriescomponent, theerror term a regionat one periodmaybe correlated forthatregionat withtheerror the nextperiod.Unless corrected errorstructure for,thisautocorrelated in biased standarderrors. will result To assess theextent ofthisproblem, I computed estimates ofthedegreeofautocorrelation acrosstimeperiods fortwo modelsfrom modelthatis a variation tables3 and 5, and a third of model 6 in tables 3 and 5. in the errorterm(p) betweentime The estimated average correlation t and timet - 1 is shownin table 7. The autocorrelation estimates assume a first orderautocorrelative process.31 Since thereare onlyseven (traditionalprejudice)or nine(opposition to race targeting) timeperiodsin the I computed theestimated autocorrelation analysis, assumingan identical errorcorrelation over timeforeach region(see Greene[1993, chap. 16] fora discussionof consistent of autocorrelation estimators withmultiple repeatedcrosssections). In the traditional prejudiceindex,thereis substantial autocorrelation in the modelswithfewdependent variables,but in the fuller modelsthe is modest(r < .15). For opposition autocorrelation to race targeting, the results are similar exceptthatthedegreeofautocorrelation is evensmaller.
is characterized First-order autocorrelation bytherelation thatvy= pvYl, where p is thecorrelation between theerror at time y and time y - 1. Formsofautocorrelation other arealso likely thanfirst order ofp. Thus,a lowestimate tocausea high estimate ofp suggests no autocorrelation of anykind.
31

851

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology


TABLE
AUTOCORRELATION ESTIMATES

7
IN TABLES 3 AND

FOR MODELS

DEPENDENTVARIABLE MODEL Model 2: year, South, year X Southern resident ............................ ................ Model 6: all variables ..................................... Modified model 6: all variables except year, South, year X Southern resident ..... TraditionalPrejudice .424 .108 .142 Oppositionto Race Targeting .104 -.050 -.090

in thecell is theestimated correlation (p) between an identiNOTE.-The number vtand vt-lassuming foreach region(see Greene1993,pp. 456-57). The residualsused to form cal degreeof autocorrelation the estimates are the groupresiduals(see Goldstein1995,app. 2.2). Estimateswerecomputedusingthe matrixalgebracommandin SPSSx.

studiessuggest thatit is better notto correct forautocorSincesimulation relationwhenr < .2 (Hanushekand Jackson1977,p. 173),no stepsare forthisproblem. These results takento correct thatthestandard suggest in themodelswithfewvariablesmaybe somewhat errors inaccurate. But in themodelswithmorevariables, appearsto pose no real autocorrelation modelsare theonesofprimary problem, and theresults from thecomplete interest.
CONCLUSIONS

has been a considerable declinein support Over theperiod1972-91,there and traditional doctrines of racial forsegregation, open discrimination, Traditionalprejudicehas not completely disappeared,howsuperiority. in the South. The declinein traditional prejudicehas ever,particularly erodedsupportforgovernment policiesbased on blatantdiscrimination but has notbeen accompaniedby an increasein support and segregation intervention to promote racial equality. foractive government are some pieces toward a complexand What this articlecontributes have followedthesepatterns. incomplete pictureof why racial attitudes Priorworkby Firebaughand Davis (1988)showedthatperiodand cohort in the decline of prejudice.This factorswere about equally important of individual articleextendsthese resultsby assessingthe importance for explainingtrend and regional differences, characteristics adding and by comparing theimvariablesto measuregroupthreat, group-level to race tarpact of social forceson traditional prejudiceand opposition conclusions from this Here I highlight fiveof themoreimportant geting. analysis.
852

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat First,consistent withpriorresearch (Firebaughand Davis 1988),there is a sharp declineacross cohortsin traditional prejudice.There is little decline,however, acrosscohorts in opposition to race targeting. For nonSoutherners thedeclinehas been modest, whileforSoutherners there has been no declineat all. Whatever cohort relatedfactors caused thedecline in traditional prejudiceacross cohorts did not do the same forattitudes towardrace targeting. in traditional Second,about 25% ofthecohort gradient prejudiceis the resultof individualcharacteristics in particular changing across cohorts, thegreater educationoflatercohorts. Whiletheexpansionofeducational in thedeclineof prejudiceacrosscoopportunities is an important factor horts,then,it does not explain most of the cohort-related decline as is sometimes implied.The remaining trendin traditional 75% ofthecohort prejudiceis not explainedby thisarticleand could well be the resultof a trendtowardacceptanceof the principle of equal treatment caused by historical eventssuchas theCivil Rights movement. This is an area worth in future exploring research. Third,although theeffect ofchangein individual-level variablesis not negligible, it explainslittleof the overalltrendin prejudiceor difference in racial attitudes is consistent betweenregions.This finding withprior in individualcharacteristics findings thatmeasured differences do notexplain differences betweenregions(Middleton1976) or nations(Quillian 1995)and adds to priorfindings thatindividualcharacteristics do notexplain mostofthechangeovertime, either. The wide variety ofindividual characteristics employedand the facttheyexplainlittleof the betweenunitvariancesuggests variables(like theneed to look towardgroup-level groupthreat) to explainbetter theseregional gaps and temporal changes. forbothtraditional to race targeting, Fourth, prejudiceand opposition of educationon racial attitudes theeffect has grownacrossbirth cohorts. of thistrendis thateducationafter One interpretation WorldWar II has increasingly incorporated ideas fromand coursesin social science.Page and Shapiro(1992) arguethatsocial sciencechangesattitudes about race the importance of social and environmental sourcesof by emphasizing rather thanbiologicalones. Educationmay be increasingly imbehavior, in influencing portant attitudes about race,then,because of the increase in socialsciencecontent in education. ofthisinteracOtherinterpretations tion termare also possible,however,so further researchto explain the of educationand birth interaction year would be useful. Fifth, group-level averagepercapita incomeand percentage ofpopulationblack have substantial on racial attitudes. effects Consistent withmy on prejudicein Europe, regionalincomeper capita (1995) priorresearch is inversely relatedto traditional prejudiceand oppositionto race tarevenwhencontrolling at theindividual levelfor income.Similarly, geting, 853

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology thepercentage ofthepopulation thatis black is positively associatedwith traditional prejudiceand oppositionto race targeting. This resultreinforces theconclusion ofearlier empirical studies byPettigrew (1959),Giles and Evans (1985),Fossettand Kiecolt(1989),and Quillian(1995)thatfind prejudiceto be a function of therelativesize of thesubordinate groupto thedominant group.This investigation adds to theseresults by showing that the two group-level variables (percentage of populationblack and per capita income)are good predictors of opposition to race targeting as well as traditional prejudice, thattheassociation betweenthegroup-level variables and racial attitudesremainsstrongwhen a numberof controlvariablesare introduced, and thattheassociation betweenthegrouplevel variablesand the racial attitude indexesholds over timeas well as across space. In theU.S. context, theresults indicate thatincreased percapitaincome is one factorbehindthe reduction in prejudicefrom1972 to 1985. The group-threat theoryinterprets this association as indicatingthat the growthin incomein the period after1972 made whitesfeel that their group'seconomicpositionwas moresecure,resulting in less hostileattiin incomeexplainsabout tudestowardblacks.I estimate thatthegrowth 25% of the totaldeclinein prejudicefrom1972 to 1985. The growth in incomealso explainspart of the (verymodest)decline in oppositionto race-targeted policyfrom 1975to 1991.This factor also contribprobably utedto thedeclineoftraditional World prejudicein theentire periodafter War II, although the data used herecoversonly 1972-85. in percentage of populationblack are important Likewise,differences in explainingwhy thereremainsa large discrepancy in racial attitudes I estimate betweentheSouthand therestofthecountry. thatabout 50% in traditional of thedifference prejudicebetweenthe Southand thenonin per capita income South can be accountedforby regionaldifferences in the South and racial composition. This suggeststhat racial attitudes are morehostile towardblacksin partbecause ofthesenseofgroupthreat of the black populationin the South. createdby the concentration I have exploredseveralsourcesof changein racial attitudes Although in this article,thereare still manymore questionsabout these attitude I do not directly changesthan thereare answers.For instance, explore therelationship theCivil Rights movement and attitude between changes. My guessis thatthisfactor probably explainsmuchof theresidualeffect theconnections ofcohort on traditional Nor do I explore prejudice. among different bases ofstratification. As Page and Shapiro(1992) pointout,the declineof traditional prejudicecan be seen as part of a longertrendin ofall kindshas whichstratification on thebasis ofascribedcharacteristics The likelyconnection between increasingly been viewed as illegitimate. 854

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat declining prejudiceand changing attitudes about genderand class is yet anotherreason forfutureresearchto understand betterthe sourcesof theseattitude changes. APPENDIX A Index Construction This appendixpresents theresultofanalysesthatsupport thedecisionto construct the two attitude indexes,traditional prejudiceand opposition to racetargeting, used in thisarticle. To verify theavailable quesdividing tionsintotwo indexes,I carriedout a confirmatory factor analysisusing Methuen'sLISCOMP program. LISCOMP performs the factor analysis usinga matrix ofpolychoric whichbetter measuretheassocicorrelations, ation betweencontinuous variableswithdiscrete indicators (such as the factor survey questions used here)thanan ordinary analysisusingPearson variables. correlations, whichassume continuous The two indexesI construct here use six GSS surveyitems.Four of these items measure traditional prejudice: RACDIN, RACPUSH, ofthese RACMAR, and RACSEG (see table 1 for and description wording variables).The othertwo measuresindicateopposition to race-targeted policy (NATRACE and HELPBLK). Unfortunately, across the entire GSS samplefrom on all 1972to 1991,only354 cases from 1984are present sixvariables.Priorresearch indicates thatRACDIN, RACPUSH, strongly in one index (Firebaughand RACMAR, and RACSEG belongtogether Davis [1988],e.g.,defendgrouping theseitemstogether). Somewhatless existsforwhether NATRACE and HELPBLK shouldbe considsupport ereda separateindexor partofthissame index.To examineifthesecond set ofquestions I ran thefactor shouldbe includedwiththefirst, analysis usingonlythequestions NATRACE, HELPBLK, RACMAR, RACSEG, sincethesequestionsare present fora largersampleof 1,560respondents in theyears 1984 and 1987-91. A polychoric correlation matrix amongthefour items revealstheir basic structure (see tableAl). Note thatNATRACE and HELPBLK form one highlycorrelated pair, while RACMAR and RACSEG formanother.I a confirmatory factoranalysis of a two-factor performed model with NATRACE and HELPBLK on one factor and RACMAR and RACSEG on the other(assuminga positivecorrelation betweenfactors). This resultedin factor loadingsLISCOMP estimated as shownin table A2. The factthat the factorscores are rathersimilarforitemsmakingup both indexessupports creating an additiveindexcombining the (unweighted) sum of the variablesintoan index.The chi-squarevalues indicatesthat thisstructure fits well,thussupporting the two-factor interpretation. 855

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

TABLE Al
POLYCHORIC CORRELATION MATRIX FOR FOUR VARIABLES NATRACE HELPBLK RACMAR RACSEG

NATRACE .......... HELPBLK ........... RACMAR ............. . RACSEG ..............

... .493 .367 344

.295 .283

.637

...

TABLE A2
ESTIMATED FACTOR LOADINGS FOR TWO-FACTOR MODEL Variable 1 2

. NATRACE ....... HELPBLK ........ RACMAR ........ RACSEG ....... .

821 776 000 000

.000 .000 .634 .777

meansquare NOTE._X2 with1 df= .047; root residual = .002; P = .8288.

856

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

APPENDIX

B TABLE Bl

MODELS INCLUDING ADDITIONAL INDIVIDUAL-LEVEL CONTROL VARIABLES Oppositionto Race Targeting

Variable

TraditionalPrejudice

Year surveyed ...................


Southern resident ....................

...................
..................

-.004 (.013) (.278) -.005 (.021) 3.230*** (.676) .049*** (.013) -.036*** (.002) .753*** (.127) -.011*** (.004) -.362*** (.016) -.005*** -.008*** (.002) .259*** (.063) -.176 (.181) -.018 (.071)
-.229** (.001) -.141

.005 (.009)
-.673**

Year X Southern resident..................................... Inverseper capita income ..................................... %black .................... Birthyear ................... .................. ...................

Southern resident at age 16 .................................. Birthyear X resident at age 16 ........................... Education ................... ...................

Education X birth year ...................................... Familyincome ................... Male ........ ............ ................... ..................

(.216) .036** (.012) 1.803*** (.429) .031*** (.009) -.003 (.002) -.673** (.216) -.008* (.003) -.110*** (.011) -.002*** .005*** (.002) .061 (.050) -.112 (.146) .185** (.058)
-.139* (.001)

Currently unemployed ...................................... Currently working......................................


Urban residence .................. ....................

Suburbanresidence...................................... Unhappy .................... ..................

Otherraces in neighborhood ................................ Intercept


var(u) region

....................................
................... ...................

(.071) -.112 (.068) .262** (.095) -.511*** (.063) 2.432*** (.644)


.000

(.058) .018 (.052) .103 (.079) -.102* (.050) 1.968*** (.444)


.000

var(v)time .................... .................. var(e) individuallevel ......................................


are SEs. NOTE.-Nos. in parentheses
*P < .1. ** p < .05. *** P < .01.

.045 6.271

.015 2.676

857

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

AmericanJournal of Sociology
REFERENCES Adorno, TheodoreW., Else Frankel-Brunswik, Daniel Levinson, R. Sanford. 1950. The Authoritarian Personality. New York: Harper. Bishop,JohnA., JohnP. Formby, and Paul D. Thistle.1992."Convergence of the Southand Non-South IncomeDistributions, 1969-1979." American Economic Review 82 (1): 262-71. Blalock,HubertM. 1957."Per Cent Non-White and Discrimination in the South." American Sociological Review22 (December): 677-82. a Theory ofMinority Group Relations. New York:Wiley. . 1967.Towards Blumer, Herbert. 1958."Race Prejudice as a SenseofGroupPosition." PacificSociologicalReview 1:3-7. Bobo,Lawrence.1983."Whites' Opposition to Busing:Symbolic Racismor Realistic GroupConflict?" Journal ofPersonality and Social Psychology 45:1196-1210. Bobo, Lawrence, and JamesR. Kluegel.1993."Opposition to Race-Targeting: SelfInterest, Stratification Ideology, or Racial Attitudes?" American SociologicalReview58 (4): 443-64. Bryk, Anthony, and Stephen Raudenbush. 1992.Hierarchical LinearModels:Applicationand Data Analysis Methods. Newbury Park,Calif.:Sage. towardBlack People. Ann Arbor, Mich.: Campbell,Angus.1971. WhiteAttitudes Institute forSocial Research. AndrewGreeley, and StephanFuchs. 1989."Social Determinants of Case, Charles, Racial Prejudice." Sociological Perspectives 32:469-83. Danziger, Sheldon, and Peter Mass.: Gottschalk. 1995.America Unequal.Cambridge, HarvardUniversity Press. Davis,James A. 1975."Communism, and Categories: Conformity, Cohorts, American Tolerancein 1954and 1972-73."American Journal 81:491-513. ofSociology NORC ed. Chicago:NORC. . 1991.GSS (General Social Survey, 1972-1991). Distributed by RoperPublicOpinionResearch New Haven, Conn. Center, ofTrendswith DiPrete, Thomas,and David Grusky. 1988."The Multilevel Analysis RepeatedCross-Sectional 18:337-68. Data." Sociological Methodology Firebaugh, Glenn,and KennethE. Davis. 1988. "Trendsin Antiblack Prejudice, 1972-1984: Region and Cohort Effects." American Journal ofSociology 94 (2): 25172. 1989."The RelativeSize ofMinority Fossett, Mark,and K. JillKielcolt. Populations 70 (4): 820-35. and WhiteRacial Attitudes." Social Science Quarterly Size ofMinorW. Parker, 1977. andtheRelative Frisbie, andLisa Neidert. "Inequality A Comparative Journal American 32:1007ityPopulations: Analysis." ofSociology 30. An Old Assumption Black and Racial Hostility: Giles,Michael W. 1977."Percent Social Science Quarterly 58:412-17. Reexamined." Perceived S. Evans. 1985."External Giles,MichaelW., and Arthur Threat, Threat, 66:50-66. Social Science Quarterly and GroupIdentity." Arthur Mass.: Harvard S. 1991.A Coursein Econometrics. Goldberger, Cambridge, Press. University Statistical Models.New York:Halstead. Goldstein, Harvey.1995.Multilevel 1971."Attitudes towardRacial Integration." and Paul Sheatsley. Greeley, Andrew, American 225:13-20. Scientific 2d ed. New York: Macmillan. William.1993.Econometric Greene, Analysis, 1977.Statistical Methods Hanushek, Eric,and John Jackson. forSocial Scientists. New York:AcademicPress. of WhiteSupremacy: An EcologicalStudy." Heer,David M. 1959."The Sentiment Journal 64:592-98. American ofSociology 858

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

Group Threat
Hyman, Herbert, and Paul Sheatsley. 1956."Attitudes toward Desegregation." Scientific American 196:35-39. towardDesegregation." Scientific . 1964."Attitudes American 211:16-23. Hyman, Herbert, Charles Wright, andJohn Reed. 1975.TheEnduring Effects ofEducation.Chicago:University of ChicagoPress. Attitudes: Jackman, Mary,and Michael Muha. 1984. "Educationand Intergroup Moral Enlightenment, Superficial DemocraticCommitment, or IdeologicalRefinement?" American Sociological Review49:751-69. Kanagy,Conrad.1994."Is Racismin theSouthDisappearing? theRegional Testing oftheAmeriofAntiblack at themeetings Convergence Prejudice." Paperpresented can Sociological Los Angeles, Association, August. Lipset,Seymour, and WilliamSchneider. 1978."The Bakke Case: How WouldIt Be Decided at theBar ofPublicOpinion?" Public Opinion1 (1): 38-44. McAdam,Doug. 1982.PoliticalProcessand theDevelopment ofBlack Insurgency, of ChicagoPress. 1930-1970.Chicago:University to RaMcKee J. 1985."Racism, RationalChoice,and WhiteOpposition McClendon, Public OpinionQuarterly cial Change:A Case Studyof Busing." 49:214-33. Merton, Robert. 1949."Discrimination and theAmerican Creed."Pp. 99-128 in DisM. Maclver.New York: crimination and theNationalWelfare, editedby Robert Harper. in Prejudice." Middleton, Russell.1976."Regional Differences American Sociological Review41:94-117. Page,Benjamin, and Robert Shapiro.1992.TheRationalPublic.Chicago:University of ChicagoPress. ofBorder-State Pettigrew, Thomas.1957."Demographic Correlates Desegregation." American Review22:683-89. Sociological in Anti-Negro Differences Prejudice." . 1959."Regional Journal ofAbnormal and Social Psychology 59:28-36. Three-Level Prosser, R.,J.Rasbash,and H. Goldstein. 1991.ML3: Softwarefor Analysis Users'Guidefor Version 2. London:University ofLondonInstitute of Education. Lincoln.1995."Prejudice as a Response to Perceived Quillian, GroupThreat: Populain Europe."American tionComposition and Anti-Immigrant and Racial Prejudice Review60 (4): 586-611. Sociological ML3-E Version Rasbash, J.1993. 2.3ManualSupplement. London:University ofLonof Education. don Institute MemoSchuman, Howard,and Jacqueline Scott.1989."Generations and Collective ries." American Sociological Review54:359-81. in Schuman, Howard,CharlesSteeh,and LawrenceBobo. 1985.Racial Attitudes America. Mass.: HarvardUniversity Press. Cambridge, Racism." Racism:Profiles Sears,David 0. 1988."Symbolic Pp. 53-84 in Eliminating in Controversy, edited by P. A. Katz and D. A. Taylor. New York: Plenum and HowardSchuman. Did RacialAttiSteeh, Charlotte, 1992. "YoungWhite Adults: tudesChangein the 1980s?" American Journal ofSociology 98 (2): 340-67. Stouffer, SamuelA. 1955.Communism, Conformity, and CivilLiberties. New York: Doubleday. Tuch,Steven.1987."Urbanism, Region, and ToleranceRevisited." American SociologicalReview52:504-10.
U.S. Bureau of the Census. 1973, 1977-78, 1985-86, 1988, 1990, 1992-93. Statistical Press.

Abstract D.C.: Government Office. ofthe UnitedStates.Washington, Printing States: . 1978."Population Estimates, byRace,for July 1, 1973and 1975."Ser. no. 67. Washington, P-23,Current Population Survey Report D.C.

859

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

American Journal of Sociology


ofthePopulation ofStates,1970-1983." . 1984."Estimates Ser.P-25,Current no. 957. Washington, Population Survey Report D.C. and Household . 1988."StatePopulation Estimates with Age,Sex,and Components ofChange1981-1987." Ser.P-25,Current Population Survey Report no. 1024. Washington, D.C. ReeveD., and ThomasPettigrew. Vanneman, 1972."Race and Relative Deprivation in theUrbanUnitedStates." Race 13:461-86. Wilson, Thomas.1986."Interregional Migration and Racial Attitudes." Social Forces 65:177-86. in Theoretical Wilson, William J. 1973.Power, Racism, and Privilege: Race Relations and Sociohistorical Perspectives. New York: Free Press. ofRace. Chicago:University . 1978.The DecliningSignificance of Chicago Press. Zaller,JohnR. 1992.The Natureand Origins ofMass Opinion.Cambridge: CamPress. bridge University

860

This content downloaded from 192.12.88.213 on Sat, 5 Oct 2013 15:59:24 PM All use subject to JSTOR Terms and Conditions

You might also like