Q Pajita

Prejudice as a Response to Perceived Group Threat: Population Composition and Anti-
Immigrant and Racial Prejudice in Europe

Author(s): Lincoln Quillian
Source: American Sociological Review, Vol. 60, No. 4 (Aug., 1995), pp. 586-611
Published by: American Sociological Association
Stable URL: http://www.jstor.org/stable/2096296
Accessed: 23-02-2017 14:40 UTC
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.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access to
American Sociological Review
This content downloaded from 141.117.125.76 on Thu, 23 Feb 2017 14:40:14 UTC
All use subject to http://about.jstor.org/terms
PREJUDICE AS A RESPONSE TO PERCEIVED GROUP THREAT:
POPULATION COMPOSITION AND ANTI-IMMIGRANT AND
RACIAL PREJUDICE IN EUROPE*
Lincoln Quillian
Harvard University
Following the work of Blumer (1958), I extend and test a theory of prejudice
based on perceived threats to dominant racial or national groups by subor-
dinate groups. Perceived threat is hypothesized to be a function of economic
conditions and of the size of the subordinate group relative to the dominant
group. I test the group-threat theory using a multilevel model that combines
population data with survey results on attitudes towards immigrants and ra-
cial minorities from Eurobarometer Survey 30. "Group threat" explains most
of the variation in average prejudice scores across the 12 countries in the
sample and has a small but statistically significant effect on the influence of
certain individual-level variables on prejudice. These results demonstrate the
importance of perceived intergroup threat in the formation of prejudicial at-
titudes and suggest a re-interpretation of past findings on the relations be-
tween individual characteristics and expressions of prejudice.
L argely because racial prejudice mea- I aygue that the focus on individual char-
sures are widely available in sample acteristics and relations omits an important
surveys, the predominant research on preju- source of dominant-group prejudice: the per-
dice has focused on the relations between ception by the dominant group that an out-
demographic, social, and psychological char- side group threatens their group's preroga-
acteristics and prejudicial attitudes. These tives. As described by Blumer (1958), preju-
studies usually view racial prejudice as re- dice is a response to threats to established
sulting from individual propensities or expe- group privileges, which are not necessarily
riences, or as an outcome of the relation be- linked to the individual interests of group
tween individuals in the dominant and sub- members. I expand on Blumer's observations
ordinate groups. Although these studies have to develop a theory of prejudice toward out-
increased our understanding of individual- groups based on collective threat. I propose
level predictors of prejudice, they are limited that collective threat is a function of two fac-
in their ability to explore group-level sources tors: the numerical size of the subordinate
of prejudice. group relative to the dominant group, and
economic circumstances. This group-threat
theory conceptualizes prejudice as a largely
* Direct all correspondence to Lincoln Quillian,
Department of Sociology, Harvard University, collective phenomenon in which individual
Cambridge, MA 02138 (Internet: QUILLIAN@ISR. attitudes are crucially affected by intergroup
HARVARD.EDU). My thanks to Stanley Lieberson, relations. I test this theory using a cross-na-
Peter Marsden, James Davis, Tony Tam, Aage tional data set that includes both population
S0rensen, the participants in the Sociology 306 and survey data from more than 11,000 re-
seminar at Harvard University, two anonymous spondents in 12 European countries.
ASR reviewers, and the ASR Editor for their valu-
In this study, I build upon previous work
able comments. Peter Marsden provided addi-
both empirically and theoretically. First, I
tional help with ML3. I gratefully acknowledge
the financial assistance of a National Science
develop a more explicit and complete model
Foundation Graduate Fellowship, which sup- of the group-level causes of prejudice and
ported me during the writing of this paper. [Re- relate prejudice to measures of the demo-
viewers acknowledged by the author include graphic and economic positions of the domi-
Reeve Vanneman. -ED.] nant and subordinate groups. Second, I gen-
586 American Sociological Review, 1995, Vol. 60 (August:586-61 1)
PREJUDICE AS A RESPONSE TO PERCEIVED GROUP THREAT 587
eralize the model by examining the associa- Mowrer, and Sears 1939), a form of psycho-
tion between prejudice and threat in a multi- pathology resulting from personality orien-
national context and by measuring two types tations developed in childhood (Adorno,
of prejudice: anti-immigrant prejudice and Frankel-Brunswik, Levinson, and Sanford
racial prejudice. Finally, I test the importance 1950), or an expression of stereotypical be-
of group threat by employing representative liefs resulting from cognitive limitations and
survey data and aggregate demographic sta- distortions in attribution (Hoffman and Hurst
tistics. The data are analyzed using multi- 1990). Numerous social-psychological ex-
level models that disentangle individual and periments support the operation of these pro-
group-level influences on prejudice and that cesses (for a brief review, see Olson and
provide more accurate inference statistics Zanna 1993:141-45).
than do the regression models more typically A second stream of research focuses on in-
employed. dividual correlates of racial prejudice in at-
titude surveys. Typically, this approach uses
statistical techniques to associate the answers
THEORIES OF PREJUDICE
to questions about racial attitudes with indi-
For my purposes, prejudice against a group vidual characteristics such as education level
is "antipathy accompanied by a faulty gener- or age. The results of these studies are gen-
alization" (Pettigrew 1980:821). Prejudice is erally consistent: People from the working
characterized by irrationality (a faulty gen- class, from older cohorts, and who have less
eralization) and emotional evaluation (an- education 2 express more prejudice (Mayko-
tipathy). There are both individual-level and vich 1975; Hyman, Wright, and Reed 1975;
group-level theories of prejudice. Case, Greely, and Fuchs 1989). A related lit-
erature investigates the correlates of the re-
lated concepts of tolerance and support for
Individual-Level Theories of Racial
civil liberties (Stouffer 1955; Sullivan,
Prejudice
Pierson, and Marcus 1982).
The literature on causes of racial prejudice The final individual-level theory of racial
branches into three streams: social-psycho- prejudice, self-interest theory, postulates that
logical approaches, studies of individual cor- individuals develop negative affects and rigid
relates of prejudice, and self-interest based stereotypes toward individuals with whom
theories.1 I do not suggest that these ap- they are in competition and conflict. Because
proaches are wrong; nor do I intend to pro- individuals are seen to develop prejudices
vide a comprehensive review. But I do wish that further their own self interest, self-inter-
to point out that none of these theories pro- est theories are related to rational-choice per-
vides a complete explanation of prejudice, spectives. Probably the most familiar ex-
and thus, to lay a foundation for distinguish- ample for sociologists is split labor-market
ing individual and group-level theories of theory (Bonacich 1972). The empirical evi-
prejudice. dence on prejudice, however, demonstrates
Social-psychological approaches view ra- only a weak link between individual interests
cial prejudice as resulting from individual and prejudice. Several studies document that
emotional and/or cognitive processes that are dominant group members whose individual
beyond completely conscious control. This interests (economic or otherwise) are not di-
research has explored causes of prejudice as rectly threatened by an opposing group are
the psychological displacement of fear or often as likely to express racial prejudice as
anxiety onto others (Dollard, Doob, Miller, are those whose economic interests are di-
rectly threatened (for a review, see Sears and
i An additional stream of prominent research

on prejudice in the United States argues that, 2 There is considerable debate over the inter-
since the 1960s, prejudice has been increasingly pretation of the link between education and preju-
expressed symbolically rather than directly dicial attitudes toward racial minorities (Jackman
(Kinder and Sears 1981). It is not clear if these 1978; Jackman and Muha 1984), but for my pur-
arguments apply to Europe. I deal here with mani- poses the precise cause of the education-prejudice
fest prejudice. link is unimportant.
588 AMERICAN SOCIOLOGICAL REVIEW
Funk 1991). For instance, Bobo (1983) finds result of collective threat. Prejudice emerges
that parents with school-aged children are as groups develop a sense of their social po-
not any more likely to oppose busing than sition relative to one another. Blumer out-
people without school-aged children. Other lines four "feelings" among members of the
studies document a weak link between indi- dominant group that lead to racial prejudice:
vidual interests and support for government (1) a feeling of superiority, (2) a feeling that
policies, such as bilingual education, taxes the subordinate race is intrinsically different
and spending, guaranteed government em- and alien, (3) a feeling of proprietary claim
ployment, programs to promote gender to certain areas of privilege and advantage,
equality in the workplace, and national and (4) a fear and suspicion that the subordi-
health insurance (Sears and Funk 1991:23- nate race harbors designs on the prerogatives
25). Dominant group members may express of the dominant race.
a desire for immigration restrictions or es- Blumer argues that the dominant group de-
pouse prejudicial attitudes, even though they velops the view that certain resources belong
may profit from the lower production costs exclusively to them-a sense of group posi-
resulting from the employment of low-wage tion, in Blumer's terms. A sense of group po-
immigrants. Thus, self-interest theories can- sition grows out of a history of unequal
not explain the existence of prejudice among power relations between groups. Prejudice is
individuals whose interests are not directly a defensive reaction against explicit or (usu-
in conflict with the subordinate group. ally) implicit challenges to the dominant
A problem with all three individual-level group's exclusive claim to privileges. This is
explanations of racial prejudice is that they not to say that all members of the dominant
do little to explain the extensive variation in group respond to a challenge with prejudice.
prejudice across different regions and time Members of the dominant group do share a
periods. Middleton (1976), for instance, sense of group membership and of their
finds that after controlling for a host of indi- group's position relative to the subordinate
vidual-level variables, the large observed dif- group. The greater the sense of threat to their
ference in racial prejudice between the north- prerogatives, the more likely are members of
ern and southern United States remains un- the dominant group to express prejudice
explained. Yet these cross-regional variations against threatening outsiders. I refer to this
in prejudice are probably at least as impor- theory as group-threat theory, because it em-
tant to study as the sources of prejudice phasizes the perception of threat to dominant
among individuals. Prejudice causes the most group prerogatives.3
damage when a dominant national or racial A modified version of Blumer's theory, re-
group institutionalizes discrimination. While cently put forth by Bobo (1983, 1988) under
all individual-level theories provide interest- the name of realistic conflict theory, posits
ing insights and often correctly identify vari- that the subordinate group is "a threat to real
ables that influence prejudice, none offers a resources and accepted practices" of the
complete explanation because group-level dominant group (Bobo 1983:1197). This
variables are ignored. theory shares materialist assumptions with
Marxist theory. In contrast to Blumer, realis-
tic conflict theory emphasizes that there
Group-Level Theories of Racial Prejudice
tends to be a connection between real domi-
A few theories of prejudice emphasize nant group interests and subjective group
group-level causes, although they are less perceptions of their own interests. As a re-
explored than individual-level theories. In
particular, several theories explore threats to 3 An almost identical theory was developed in-
dependently by Vanneman and Pettigrew (1972),
the dominant group by the subordinate group
which is sometimes referred to as fraternal-dep-
as a cause of prejudice. One of the earliest
rivation. Vanneman and Pettigrew's research
versions of this theory is Blumer's (1958,
found that survey questions asking about feelings
henceforward Blumer). of fraternal deprivation (or threats) against
In his paper, "Racial Prejudice as a Func- Whites as a group were stronger predictors of
tion of Group Position," Blumer claims that voting behavior in the 1972 Los Angeles mayoral
racial prejudice by the dominant group is the election than were feelings of individual threat.
sult, prejudice in the dominant group is a re- Blalock (1956, 1957, 1967), who stresses the
sponse to collective threats against the real consequences of minority group size for dis-
interests of the dominant racial group, rather crimination and prejudice. Blalock (1967)
than to perceptions of group interests that outlines two reasons for a connection be-
may have no connection to any real group tween intergroup threat and prejudice: First,
interest. competition for scarce resources increases
In support of group-threat theory, Bobo with the relative size of the minority group
(1983) and Bobo and Kluegel (1993) argue to the dominant group; second, since num-
that Whites support the principle of equality bers are a potential resource for political mo-
more than they do race-targeted policies to bilization, group size can increase the poten-
achieve racial equality because race-targeted tial for political mobilization and result in a
polices are particularly threatening to greater threat to the dominant group.
Whites. Although I follow Blumer's descrip- Using U.S. data, a number of studies dem-
tion of the forces underlying collective onstrate a link between the percent of a city
threat, I do not attempt to adjudicate between or county's population that is Black and ra-
realistic and perceived group threats.4 cial inequality (Blalock 1956; Frisbie and
Neidert 1977; Wilcox and Roof 1978). Yet
these studies do not conclusively support the
SOURCES OF THREAT
relation between increased size of the subor-
The group-level theories discussed above re- dinate group and increased discrimination.
late perceived threat and prejudice, but they As pointed out by Semyonov, Hoyt, and
do not explore the causes of perceived threat. Scott (1984), the level of observed inequal-
I explore two causes of prejudice that are of- ity between groups is a function of several
ten thought to act through perceived group factors, only one of which is discrimination.
threat: relative size of the subordinate group Thus, a positive correlation between subor-
and economic circumstances. dinate group population size and racial in-
equality does not necessarily indicate that
subordinate group size and discrimination by
Relative Size of the Subordinate Group
the majority group are linked. For example,
and Perceived Threat
if a labor market is racially split so that
Sociologists have long speculated that preju- Blacks hold only peripheral sector jobs, then
dice and discrimination increase as the rela- an increase in percent Black could lead to
tive size of the subordinate group increases. greater competition for peripheral sector
This position is thoroughly developed by jobs, and cause a decline in wages and a cor-
responding increase in racial inequality-
4 In my opinion, Blumer's theory, which em- with no increase in discrimination. Even if
phasizes perceived group threats, is more plau- the link between discrimination and percent
sible than realistic conflict theory, which empha-
sizes actual group threats. First, as Sears and
Funk (1991:66) point out, realistic conflict pre- Third, realistic conflict tends to imply an "in-
supposes that the interests of Blacks and Whites tentional" interpretation to group action. Racial
are a zero-sum game. This is by no means a fore- prejudice based on group interests may have the
gone conclusion. Many economists, for instance, effect of furthering the interests of the dominant
would argue that greater competition across ra- group, but without a formal organization coordi-
cial groups would increase economic efficiency nating actions of racial group members it is mis-
and hence increase the pool of resources for both leading to describe this action as intentional.
groups. Dominant group members may well per- Blalock (1967) makes this point well:
ceive that their interests and interests of the sub-
Discrimination may therefore be maintained by
ordinate group are in zero-sum conflict, but it is
a series of uncoordinated though similar indi-
debatable whether this reflects an actual zero-sum
vidual acts no one of which has more than
racial conflict or merely the perception of such.
short-run significance. . . . In most Northern
Second, group interests and the interests of in- cities the Marxian thesis of a conscious and
dividual group members often conflict. Racial highly rational plot on the part of entire White
and national groups are often diverse enough in group would simply not apply. In a very real
social or economic position that it is inaccurate sense, the dominant Whites do not act as a
to describe the group as having a single interest. group at all. (P. 160)
Black is established, this does not necessar- ans opposition to immigration grew during
ily imply a link to racial prejudice: Discrimi- periods of economic recession. Schissel,
nation can be based on nonprejudicial ration- Wanner, and Frideres (1989), on the other
ales, such as statistical discrimination by hand, find only a weak relation between city-
employers based on productivity-related level rates of unemployment and individual
characteristics. The relation between preju- attitudes toward immigrants.6
dice and discrimination is too complex to It is usually argued that the link between
draw conclusions about prejudice from stud- economic circumstances and prejudice re-
ies of discriminatory behavior that do not sults from either blaming the subordinate
measure prejudice separately. group for economic hardship (scapegoating),
Few studies, however, have attempted to or from competition with the subordinate
directly relate relative size of the minority group for scarce resources. Although either
group and levels of prejudice. A handful of scapegoating or competition can occur be-
studies using aggregate data find that anti- tween individuals, the collective threat inter-
Black attitudes are more pronounced in cit- pretation implies that a worsening of eco-
ies or regions with larger Black populations nomic circumstances among some dominant
(Pettigrew 1957; Giles 1977). These studies, group members should increase prejudice
however, suffer from a lack of individual- among all group members, not only among
level controls in their analyses that make those directly in competition with immi-
ambiguous their results regarding group or grants. When dominant group members per-
individual-level processes. To the best of my ceive their economic circumstances as pre-
knowledge, only one study includes indi- carious, they fear they will lose their eco-
vidual-level controls and percent Black in nomic advantages over the subordinate
predicting prejudice. Fossett and Kiecolt group; when economic circumstances im-
(1989) regress measures of anti-Black atti- prove, the corresponding reduction in per-
tudes on percent Black in the local area and ceived competition decreases group feelings
on individual-level variables.5 They find that of threat.
anti-Black prejudice is positively related to
the percent of the population that is Black.
Threat in the Context of European
Immigrants
Economic Conditions and Perceived
Although most of the theories presented
Threat
above were proposed in the context of U.S.
A second hypothesis sometimes advanced re- race relations, they can be generalized to ex-
garding racial prejudice is that racial dis- plain prejudice against racial groups else-
crimination and racial prejudice are closely where and minority groups based on charac-
related to economic conditions. Many have teristics other than race. I apply these ideas
speculated on this connection, although little to both racial and anti-immigrant prejudice
research has been done. Blalock (1967:184- in Europe.
86) sees improved economic conditions as a I study immigrants and racial minorities in
possible source of improved relations be- the 12 countries of the European Economic
tween Whites and Blacks in the United Community (EEC). There is considerable di-
States. Similarly, Kinloch (1974) describes versity among these 12 countries in their ra-
the importance of economic improvement to cial, economic, and social makeup. Immi-
improved race relations in his historical de- grants from non-EEC countries into the EEC,
scription of racial progress in the United however, are fairly homogeneous in their
States. Tienhaara (1974) finds that Canadi- motives for immigration: The majority mi-
grate to improve their economic circum-
5 I am indebted to an ASR reviewer for bring-

ing this study to my attention. Fossett and 6 This hypothesis has also been explored in the
Kiecolt' s (1989) regression specification is literature on causes of ethnic conflict. Olzak
equivalent, as shall be discussed more below, to (1992), for instance, finds that any sudden change
an entirely fixed specification of group effects in economic conditions increases ethnic group
(see note 17). mobilization.
stances; some come to stay, while others in- Threat and Individual Characteristics
tend to return to their home country (Castles
and Kosack 1973; Kramer 1980; Castles Individual characteristics can interact with
1989). As a result, (1) migrants go in larger macrolevel variables. Lieberson (1980:374-
numbers to the wealthier economies, and (2) 76) postulates that latent dispositions carrie
migrants are usually in direct economic com- by individuals are increasingly expressed as
petition with low-wage earners and manual discrimination when the subordinate group
laborers, as most immigrants are employed grows in size. Similarly, when a racial or na-
in manual labor (Castles 1989). tional group is threatened, certain character-
The wealthier the population of a country, istics indicate which individuals are more
then, the smaller the percentage of its popu- likely to respond with prejudice. Threat and
lation that is likely to be in competition with the effect of individual characteristics on ex-
immigrants, because wealthy countries have pressions of prejudice, then, are not com-
fewer citizens employed in manual-labor oc- pletely separate.
cupations.7 The jobs most immigrants from Individual-level characteristics indicate in
non-EEC countries compete for are likely to part which individuals are most vulnerable to
be low-wage, low-skill jobs that most citi- expressing prejudice when they perceive that
zens, especially in the wealthier European their group is threatened. Individuals with
countries, prefer to avoid. Immigrants to particular characteristics may feel threats
Greece or Portugal, say, are likely to have the particularly acutely because the threats affect
skills to compete for jobs with many citizens. them more directly, or they may be under
In West Germany, on the other hand, few citi- psychological influences that make them
zens are likely to be in direct economic com- more likely to express prejudice when they
petition with immigrants, because few immi- perceive threats. Thus, I also investigate
grants have the skills to compete for the jobs whether the strength of relations between in-
held by most German citizens. Thus, eco- dividual-level characteristics and the expres-
nomic circumstances affect perceived threat sion of prejudice varies with the level of per-
by influencing the degree of potential eco- ceived threat faced by the dominant group.
nomic conflict between the dominant and
subordinate groups.
Both economic circumstances and subor- THEORY AND HYPOTHESES
dinate group size, then, are expected to af-
Theoretical Framework
fect perceived threats to the dominant
groups. The group-threat theory offers no My approach combines theoretical ideas
clear prediction about whether the effect of from Blumer (1958), Blalock (1956, 1957,
subordinate group size and economic condi- 1967), and Lieberson (1980) and rests on
tions will be additive or multiplicative. A four main tenets:
multiplicative effect is entirely possible, be- (1) Following Blumer, I conceptualize ra-
cause a dominant group facing both large cial prejudice as the result of a feeling by the
numbers of immigrants and difficult eco- dominant group that their prerogatives are
nomic circumstances might be particularly threatened by the subordinate group. Preju-
hostile toward immigrants. The degree of dice is, then, a response to perceived threat.
hostility expressed could then be more than (2) Following Blalock, I see the size of the
a linear combination of these two factors; subordinate group relative to the dominant
one factor may tend to intensify the effect group
of as a major demographic cause of per-
the other. I test this possibility below. ceived threat. A large subordinate group is
perceived as a greater threat to the group pre-
7 For the 12 countries of the EEC, the correla- rogatives of the dominant group than is a
tion between gross domestic product and percent small subordinate group.
of the population employed in occupations likely
(3) Again following Blalock (and others),
to be affected by immigrant labor is -.73. Occu-
out-groups are more threatening when the
pations I consider affected by immigrant labor are
those labeled by the Eurobarometer as "manual," economic situation of a host country is pre-
"skilled manual," "farming," or "fishing" occu- carious. The worse the economic circum-
pations. stances are, the more threatened is the domi-
nant group and the more prejudice will be strongly associated with prejudice in set-
expressed. This is because dominant group tings where a higher proportion of the
members fear that their group's economic population belongs to the subordinate
advantage will be damaged through compe- group and where the economic circum-
tition with the subordinate group. This fear stances are worse.
increases when poor economic conditions in-
crease competition for scarce resources. In A word is due about the unit of measure:
Europe, wealth tends to decrease the compe- countries. Why not cities or even smaller
tition for jobs and resources between immi- geographic regions? Countries are conve-
grants and the host society. nient; many more comparable statistics are
(4) Specific individual-level characteristics available for countries than for cities or other
predict which individuals are most at risk for collectivities. But I have a theoretical justifi-
developing prejudicial attitudes when their cation as well-countries are important cul-
group prerogatives are threatened. The more tural, political, and economic units. People
the dominant group is threatened, then, the develop their senses of threat with, as
stronger the association between some spe- Blumer says, a history of their own group
cific characteristics and prejudice. and of the other group within national
In contrast to individual-level theories of boundaries. Given the predominance of na-
prejudice, this four-part model does not con- tional product brands, media, labor pools,
ceptualize prejudice as the result of competi- and languages, the "people of a nation" is the
tion between individuals in the dominant most likely reference group when evaluating
group and individuals in the subordinate intergroup relations in Europe.
group. As described above, prior research
demonstrates only a weak link between
MEASURES AND MODELS
threats to individuals and prejudice. Rather,
it is the collective feeling that the dominant I use data from Eurobarometer Survey 30,
group is threatened that leads to prejudice which includes supplemental questions on
against the subordinate group. This group- immigrants and out-groups in Western Eu-
threat theory highlights the crucial impor- rope (Reif and Melich 1991b; Commission
tance of a notion of "our" race or nationality of the European Community 1989). In Octo-
and the "other" race or nationality in the for- ber and November of 1988, approximately
mation of prejudice. To borrow Anderson's 1,000 interviews were conducted in each of
(1981) term, the "imagined community" of the 12 countries in the EEC, except in Lux-
the nation or the racial group are primary so- embourg, where only 300 interviews were
cial identities through which prejudices conducted. The sample was representative of
against outsiders are formed. the population of citizens aged 15 and over.8
Hypotheses The Dependent Variables: Anti-Immigrant

and Racial Prejudice
The group-threat theory suggests two hy-
potheses that I test in this paper: Two indexes comprise the dependent vari-
ables in this study: one measures racial
HI: Prejudice is a function of the perceived
threat the subordinate group poses to the
8 The Eurobarometer data do not distinguish
dominant group. Perceived threat is in-
respondents who are members of a subordinate
fluenced both by the economic situation
racial group from those who are naturalized citi-
faced by the dominant group and by the
zens from a subordinate nationality (the Euro-
size of the subordinate group relative to barometer only surveyed citizens within each
the dominant group. The effects may be country, so no foreign nationals were included).
additive or interactive. In most countries, the number of such respon-
dents was probably small. This problem creates a
H2: Individual-level prejudice emerges when
bias that weakens the results postulated by the
the perceived threat posed by the subor- group-threat theory. For further details of Euro-
dinate group is greater. That is, specific barometer sampling methodology and survey de-
individual-level characteristics are more sign, see Reif and Melich (1991a).
prejudice and another prejudice against im- Responses of "too many" were coded 1 and
migrant groups. These indexes are con- other responses are coded 0.
structed from seven questions asking about
(7) Do you personally, in your daily life, find
prejudice toward people of other races and disturbing the presence of people of an-
nationalities. For each question, respondents other (nationality/race)?"
were read the statement: "I am going to read
you out opinions. For each opinion I read Responses were "disturbing" (coded 1), or
out, please tell me to which, if any, kinds of "not disturbing" (coded 0).
people it applies." They were then shown a A principal component analysis of the
card listing groups of people to whom the seven questions revealed that all the items in
statement could apply: "people of another each index loaded almost equally on a single
nationality," "people of another race,"9 principal component for both anti-immigrant
"people of another religion," "people of an- prejudice and racial prejudice; the other
other culture," "people of another social components had eigenvalues of less than one.
class," "none of the categories," and "all of As a result, I consider each set of scores as
the categories." Respondents could choose being a unidimensional measure of anti-im-
more than one. migrant or racial prejudice. The scores cre-
The seven statements used to construct the ated from the first principal component cor-
indexes are: related .994 for anti-immigrant prejudice and
.999 for racial prejudice with simple sums of
(1) They exploit social security benefits.
the seven questions. Since the principal com-
(2) Their presence is one of the causes of de- ponent scores were almost identical to the
linquency and violence.
simple sums, I used the more interpretable
(3) Marrying into one of these groups always sums of scores as the dependent variable.
ends badly. To create the two indexes, one for anti-im-
(4) To have them as neighbors creates prob- migrant and one for racial prejudice, I
lems. summed the scores on the seven items in each
scale and scaled the index to range in values
(5) If there are a lot of their children in school,
from 0 to 1.10 Reliability (Cronbach's alpha)
it reduces the level of education.
of the anti-immigrant prejudice index is .725;
Responses to these five questions were
coded 1 for the anti-immigrant prejudice in-
10 It can be argued that scores on some of these
dex if the respondent indicated that the state- survey items would tend to increase with the size
ment applied to the "people of another na- of the minority group, almost by construction (for
tionality," 1 for the racial prejudice index if instance, are there "too many" people of another
the respondent indicated that the statement race or nationality in the country). First, it is
applied to "people of another race," and 0 worth keeping in mind that this in no way makes
otherwise. less surprising that the interactive index of threat
is such a strong predictor (see results). Second,
The last two questions in the index follow
most of the questions in the index imply a value
a different format:
judgement against a group (i.e., the respondent
(6) Generally speaking (show card), how do chooses "too many" rather than "many"). I tried
you feel about the number of people of an- to avoid statements of fact for which responses
other (nationality/race) living in our coun- would increase with the size of the subordinate
try? Are there too many, a lot but not too group. Third, I ran an exploratory logistic regres-
many, or not many?" sion analysis using survey item 3 (which is least
vulnerable to this argument) as the dependent
9 Respondents could be thinking of any race variable. Using only this variable, the relation be-
when they answer the questions. However, evi- tween prejudice and threat is not markedly less
dence from other questions in the Eurbarometer strong. Finally, item 3 did not load any less
survey indicate that most are likely to be thinking strongly on the first principle component than the
of Blacks, except in Britain where most think of other items did. These checks support the conclu-
South Asian immigrants, in France, where most sion that the relation between percent non-EEC
think of Arabs, and in Germany, where most nationality in the country and prejudice is not se-
think of Blacks and a substantial fraction think of riously inflated by the design of the questions in
Turks. the index.
for the racial prejudice index reliability is Theories of migration predict a positive
.744. Means for the dependent and key inde- correlation between the percentage of immi-
pendent variables are shown in Appendix A. grants and national wealth because most mi-
gration is economically motivated (Findley
1982). Since these two measures are corre-
Index for Perceived Threat
lated, omitting one would bias the estimate
The other important variables for this analy- of the other's effect. Both to sort out their re-
sis are country-level measures of perceived spective effects and as a control, then, eco-
threat posed by the subordinate group to the nomic circumstances and percent of the
dominant group. Recall that I hypothesize population from non-EEC countries are in-
perceived threat to increase with the size of cluded together.
the subordinate group relative to the domi- Finally, I test the possibility that there is
nant group and with the worsening of eco- an interaction between the two indicators of
nomic circumstances in the host country. The threat. The measures that I take to indicate
size of the subordinate group relative to the perceived threat are:
dominant group is indexed by the percentage
Pk = Relative size of the subordinate group
of the population resident in the country that
to the dominant group in country k
is not a citizen of a country in the EEC." I Al-
(percent non-EEC immigrants);
though this may seem like an appropriate
measure for the relative size of an immigrant Gk= Economic conditions in country k;
community, a more direct measure, such as larger score means worse situation
percent non-White, might be a better indica- (Inverse of average GDP per capita,
tor of the proportion of racial minorities in a 1984-1988; more specifically,
country. Unfortunately, no such data are 1/GDP8488 x 1000);14
available, and so I used percent of immi-
Pk X Gk= Interaction of Pk and Gk.
grants from non-EEC countries to index the
presence of both racial minorities and immi- For notational convenience, I sometimes
grants. Exploratory analysis revealed that the use Tk to refer to the perceived group threat
percent of non-EEC immigrants in the popu- in country k. In practice different combina-
lation was more strongly related to expres- tions of the three variables above were tried
sions of prejudice than were several alterna- to measure threat, so that perceived threat
tive measures.12 To index economic condi- (Tk) is measured by Pk. Gk, and Pk X Gk.
tions, I use five-year average GDP per capita
(1984-88).13
Individual-Level Independent Variables
I include independent individual-level vari-

I These data are not available yearly for most
ables to test predictions of the group-threat
EEC countries. I used data for each country from
the years indicated in parentheses: West Germany theory of prejudice, not to test individual-
(1987), Belgium (1987), Denmark (1987), Spain level theories of prejudice. Hypothesis 2
(1987), France (1982), Greece (1987), Ireland
(1988), Italy (1987), Luxembourg (1981), Neth- to be comparable. This provides a measure that is
erlands (1987), Portugal (1987), United Kingdom more closely related to standard of living and is
(average of 1984-1986). The data are from Lebon less volatile than GDP adjusted by exchange rate.
(1990). GDP is in billions of 1991 dollars and 1991 PPPs
12The alternative measures I tried are total per-
(OECD 1991:145). Initially, I used average five-
cent of the resident population that are not citi- year unemployment rate (1984-1988) to index
zens and percent of the population who are not economic health. I switched to use GDP because
citizens and are immigrants from Africa. I also GDP statistics are more comparable across coun-
tried adjusting the percent of immigrants from tries and less volatile. Generally, average country
non-EEC countries to exclude individuals who unemployment rate 1984-1988 produced similar
were immigrants from North America. However, results.
data on immigrants from North America were not 14 The inverse of GDP (1/GDP) leads to a more
available for all countries, and in most cases in- straightforward interpretation and a better fit than
cluding these data made little difference. other transformations to reverse the orientation of
13 I used purchasing power parities (PPP) rather the GDP variable (such as multiplying by nega-
than exchange rates to adjust the GDP statistics tive 1).
claims that regardless of the precise reason a sudden decline in a person's economic sta-
for the link between an individual-level vari- tus is likely to increase frustration and thus
able and prejudice, the association should in- lead to increased prejudice against members
crease as the degree of threat to the domi- of the minority group (Dollard et al. 1939).
nant group increases. In addition, the indi- This dummy variable is coded 1 if the re-
vidual variables serve as controls for indi- spondent agrees that her or his household
vidual-level causes of prejudice that might economic situation has worsened in the past
otherwise confound the estimates of group- 12 months.
level variables on prejudice. Alienation score. Adorno et al. (1950) ar-
Each individual-level independent variable gue that alienation should be positively re-
is centered by the grand mean across all lated to racial prejudice. The alienation in-
countries when used in the regressions (the dex is based on a sum of five agree/disagree
centered variable is transformed by subtract- items that indicate alienation from others and
ing the grand mean, so the mean of the cen- from the political process.'5 These items
tered variable across all 11,706 cases is 0). were then resealed on a 0 to 1 scale.
This allows country-level effects to be evalu- Income in lowest quartile. This dummy
ated for a hypothetical individual with "aver- variable indicates whether or not the
age" European characteristics, thus holding respondent's family is in the lowest quartile
compositional differences in individual char- of the family income distribution for his or
acteristics constant. The country-level inde- her home country. Self-interest theories pre-
pendent variables (perceived threat and its dict that these individuals are the ones most
components) on the other hand, are not cen- likely to be in competition with immigrants
tered. The variables I include are: for jobs and thus are most likely to express
Education. Stouffer (1955) was the first prejudice.
to find that education had a major effect on Sex. Adorno et al.'s (1950) framework sug-
levels of tolerance. Since then others have gests that men are more likely to exhibit the
found that more educated respondents ex- traits of the "authoritarian personality" and
press less racial prejudice (Maykovich thus are more likely to express prejudice than
1975). Education is measured in years. women. Men are coded 1 and women 0.
Those with less than 8 years of education Life satisfaction. Psychological theories of
are coded as 8; those with more than 14 prejudice, such as the frustration-aggression
years are coded as 14. view, argue that frustrated individuals are
Age. Age captures the effect, reported by more likely to project their frustrations onto
Maykovich (1975) and others, that older in- an out-group. To operationalize this idea, I
dividuals and cohorts will tend to be more ask about life satisfaction. The less satisfied
prejudiced. No attempt is made here to sepa- an individual is, the more frustrated and
rate life cycle from cohort effects. therefore the more prejudiced that person is
Working-class status. Following rational- likely to be. This variable is measured on a
choice theory, the individuals most likely to scale of 1 to 10.16
be prejudiced against minorities are those
who are most likely to compete with them for
15 The reliability coefficient for the alienation
jobs. The majority of migrants to Europe measure is .6689. The five questions (agree/dis-
work in the low-wage manual-labor sector agree) are: "Most people in positions of power try
(Castles 1984). As a result, I expect that to gain something out of people like you";
members of the working class are more "People who run the country are not really con-
likely to express prejudice against racial mi- cerned with what happens to you"; "You feel left
norities or immigrants than are members of out of what is happening around you"; "The rich
get richer and the poor get poorer"; and "What
other classes. The Eurobarometer codes oc-
you think doesn't count very much."
cupation in 12 categories; I counted those
16 Following a procedure suggested by Winship
who responded "skilled manual" or "other
and Mare (1984), in some exploratory regressions
manual" as working class. Working-class I tried assigning to the 10 ordinal categories a
members are coded 1; others are coded 0. value based on a procedure that assumes normal-
Change in economic status. Psychological ity of the disturbance term and uses the condi-
theories of frustration-aggression predict thattional distribution of the dependent variable. The
Contact with members of other races/na- dominant group is greater. To test this possi-
tionalities. I include two variables to index bility I add interaction terms between the
likelihood of contract with people from other threat variables and individual-level vari-
races/nationalities: "Are there people of an- ables.
other nationality/race who live in your neigh- I accomplish the modeling using the sta-
borhood?" and "Are there people of another tistical software program ML3 (Prosser,
nationality/race at your work?" Yes is coded Rasbash, and Goldstein 1991). ML3 uses an
1 and no is coded 0. Although these variables iterative generalized least-squares algorithm
do not have clear theoretical predictions, I to estimate multilevel models. ML3 differs
include them to reduce the possibility that from standard regression techniques in its
the group-level indicator of threat, percent ability to partition variance among multiple
non-EEC immigrants in the country, is cap- levels and derive appropriate point estimates
turing the effect of individual-level contact and confidence intervals accordingly."7 For
with members of the subordinate group. my purposes, individuals are level one of the
I included other individual-level variables analysis and countries are level two.
initially, but dropped them later because their More formally, an illustrative model is one
coefficients were statistically nonsignificant that includes only individual-level variables
and substantively small. These include em- and allows their slopes to vary across coun-
ployment status, religion, union status, mari- tries. This could be considered as a series of
tal status, and work sector (public versus pri- k regression equations, such that:
vate).
Note that several of the individual-level in-
Pik = ak + XikPk + eik, (1)
dependent variables described here (includ- where Pik is the level of racial prejudice for
ing those tried and excluded) are similar to individual i in country k. Xik is a matrix of 10
economic condition (Gk) and percent non- individual level variables associated with ra-
EEC (Pk). Including several individual-level cial prejudice country k, Pk is the correspond-
measures similar to the country-level vari- ing vector of coefficients estimated, ak is the
ables decreases the chance that the macro- intercept for country k, and eik is an error
level variables are simply capturing the in- term. Note that Pk indexes k different vectors
fluence of individual-level variables. of coefficients, one for each of the 12 coun-
tries. For instance, Pi3' = (b,1 , b1,2, . . . b1,1o)
The Basic Models
where b1,10 is the coefficient corresponding
to the 10th individual variable for country 1.
To test the two basic hypotheses of group- The group-threat theory hypothesizes that
threat theory, I use a multilevel model that the country-level threat variables are related
allows the simultaneous modeling of indi- both to the average level of prejudice and to
vidual-level and group-level variables and the strength of the relation between indi-
their interactions. Hypothesis 1 postulates a
relation between perceived threat and expres- 17 For an introduction to multi-level models,
see Mason, Wong, and Entwisle (1984). The it-
sion of prejudice. This is tested by entering
erative GLS estimator used in ML3 is equivalent
the threat variables in the multilevel model
to the maximum-likelihood estimate and has
together with controls for individual-level properties similar to those discussed by Mason,
variables. Wong, and Entwisle. The most often used tech-
Hypothesis 2 asks if individual variables nique for modeling contextual effects is to enter
are more strongly associated with prejudice the contextual variable into an individual-level
in countries where perceived threat to the regression model as if the context effect were just
another individual-level characteristic (e.g., Fos-
sett and Kiecolt 1979). The problem with this
results in pooled regressions were little different technique is that, unless the contextual effect per-
than those with the measure used here. Generally, fectly (nonstochastically) determines the effect of
I think treating this variable as an interval intro- region, these models are incorrectly specified.
duces little distortion in the results, as the vari- The consequence is standard errors that are too
able is measured in fine intervals (10 categories) small; and that can cause misleading conclusions
and is roughly normally-distributed among these (Goldstein 1987; Bryk and Raudenbush 1992:98-
categories. 102).
vidual-level variables and prejudice. One Pik = b + i7Tk + XikC + (Xik X Tk)y
way to conceptualize this is as regressions + (Uk + eik). (4)
where countries are cases and the parameters
estimated in equation 1 are the dependent This model adds (Xik
variables. Saying that the average level of tiplicative interaction terms of individual
prejudice in a country is related to perceived variables and the variables that measure
threat is the same as saying that the intercept threat. Following the usual assumptions
from equation 1 varies across countries and made in the multilevel modeling literature,
is related to threat, as in: the error terms for each level, Uk and eik, are
assumed to be distributed multivariate nor-
ak = b + fTk + Uk, (2)
mal and independently of one another.
where b is the grand across-country intercept Models following the form of equation 4
term, Tk is perceived threat for country k are the basic models estimated in the results
(measured by one or more of Pk, Gk or Pksection.
X Issues of robustness in the face of
the small number of nations are discussed in
Gk), q is the estimate of the relation between
perceived threat and the intercept across Appendix B.
countries, and Uk is an error term referring to
country differences in prejudice that are not
RESULTS
attributable to threat. Similarly, Hypothesis
2 postulates that the slopes from equation I The basic results are presented Table 1 for
are related to threat, as in the country-level racial prejudice and in Table 2 for anti-im-
model: migrant prejudice.'8
Pk = C + Yk- (3)
Regressions Using Individual-Level
Pk is again a vector of coefficients estim
Variables Only
in equation 1, y is the corresponding vector
of coefficients, and c is a vector of constants. Model 1 in Tables 1 and 2 shows the results
I discuss the lack of an error term in equa- of an ML3 regression of individual-level
tion 3 in Appendix B. variables on prejudice for the pooled sample
Hypothesis 1, which states that the percent of 11,676 individuals from 12 countries. The
of non-EEC immigrants, GDP, and their in- slopes of the independent variables in Model
teraction affect prejudice, suggests that rj > I in both Tables 1 and 2 are constrained to
0. Hypothesis 2 suggests that when an ele- be identical across the 12 countries. As the
tables show, most of the independent vari-
ment of the vector Pk > 0, the corresponding
element of the vector y> 0; and when an ele- ables have statistically significant effects on
ment of the vector Pk < 0, the correspondingracial and anti-immigrant prejudice in the di-
rections hypothesized. The statistically and
element of the vector y< 0. That is, for all of
the independent variables that are signifi- substantively strongest effects are that preju-
cantly associated with prejudice at the indi- dice decreases with increased education and
vidual level, national indicators of perceived prejudice increases with increased age. Two
threat should have a statistically significant of the 10 individual-level variables, "other
effect on increasing the steepness of the races in neighborhood" and "other races at
slope of that variable. Thus the slope of indi- workplace," do not have predictions on the
vidual-level causes of prejudice should vary
18 Estimates from multilevel models are distrib-
positively with perceived threat if the micro-
uted asymptotically normal. Since there are only
level variable has a positive effect on preju-
12 level-two cases (countries), z-scores are prob-
dice; the slope should vary negatively with ably overgenerous. For tests of coefficients in
threat if the microlevel variable has a nega- models with a small sample of level-two units and
tive effect on prejudice. level-two variables, the more conservative t-dis-
ML3 estimates models, like those in equa- tribution usually provides more reliable hypoth-
esis tests (Bryk and Raudenbush 1992:222). Hy-
tions 1, 2, and 3, by estimating a single re-
potheses involving country-level variables (in-
duced-form equation. Substituting equations cluding interactions of the products of country-
2 and 3 into equation 1, we get the model and individual-level variables) in Tables 1 and 2
estimated by ML3: are evaluated with the t-distribution (d.f. = 10).
Table 1. Coefficients from the Bi-Level Regression of Racial Prejudice Index Scores on Individual-
Level and Country-Level Variables: 12 EEC Countries, 1988
Models
Independent Variables (1) (2) (3) (4) (5)
Individual-Level Variables and Interaction Terms
Education -.0083*** -.0084*** -.0083*** -.0l77*** -.0033*

(.0009) (.0009) (.0009) (.0048) (.0017)
Education x Percent non-EEC -.0009

immigrants (.0007)
Education x Economic .1 079*

condition (.0346)
Education x Percent non-EEC -.0274**

immigrants x Economic condition (.0079)
Age .00 1 3*** .00 1 3*** .0013*** .0029*** .0006*

(.0001) (.0001) (.0001) (.0007) (.0002)
Age x Percent non-EEC immigrants .0001

(.000 1)
Age x Economic condition -.0174**

(.0053)
Age x Percent non-EEC immigrants x .0043**

Economic condition (.0012)
Workinng class .0200** .0200** .0203** .0256 .0002

(.0069) (.0069) (.0069) (.0359) (.0125)
Working class x Percent .0059

non-EEC immigrants (.0057)
Working class x Economic -.1667

condition (.2609)
Working class x Percent non-EEC .1141

Change in economic status .0092 .0093 .0094 -.0781 ** -.0079

(.0054) (.0054) (.0054) (.0303) (.0102)
Change in economic status x Percent .0149*

Change in economic status x .5717*

Change in economic status x .1099

Percent non-EEC immigrants x (.0505)
Economic condition
Alienation score .0706*** .0706*** .0701*** -.0057 .0321*

(.0074) (.0074) (.0074) (.0401) (.0137)
Alienation score x Percent .0186*

Alienation score x Economic condition .3154

(.2872)
Alienation score x Percent non-EEC .1963*

Sex (1 = Male) .0187*** .0187*** .0187*** .0187*** .0189***

(.0045) (.0045) (.0045) (.0045) (.0045)
(Table I continued on next page)
(Table I continued)
Models
Individual-Level Variables and Interaction Terms (Continued)

Life satisfaction -.0023* -.0024* -.0024* -.0026** -.0025**
(.0011) (.0011) (.0011) (.0011) (.0011)
Income in lowest quartile -.0174** -.0175** -.0176** -.0144** -.0170**

(.0057) (.0057) (.0057) (.0057) (.0056)
Contact with Members of Other Races
Other races .0292*** .0291*** .0290*** -.0727** .0104

in neighborhood (.0046) (.0046) (.0046) (.0254) (.0088)
Other races in neighborhood x .0143**
Percent non-EEC immigrants (.0039)
Other races in neighborhood x .6907**
Other races in neighborhood x .0971*
Economic condition
Other races at workplace -.0160* -.0163* -.0165* -.0157* -.0165*

(.0078) (.0078) (.0078) (.0078) (.0078)
Country-Level Variables
Country percent non-EEC immigrants .0344* -.3479** -.3442** 3449**
(.0146) (.1020) (.1031) (.1000)
Economic condition -.3176 -3.534** -3.397** -3.514**

(.6586) (.9612) (.9716) (.9423)
Interaction (Percent non-EEC 4.662** 4.632** 4.627**

immigrants x Economic condition) (1.238) (1.251) (1.214)
.05 < .01 < .001 (two-tailed tests)
Note: Numbers in parentheses are standard errors. Test of hypotheses for country-level variables (includ-
ing interaction effects of the products of country- and individual-level variables) use t-distribution (d.f. =
10). All models are estimated with intercepts, although the intercepts are not shown. The total N for each
model is 11,676. The constants are random at the first and second levels of analysis.
basis of the individual-level theories dis- microlevel theories of prejudice: change in

cussed in the methods and data section. Of economic status in past 12 months, life satis-
the remaining 8 individual-level variables, 7 faction (for anti-immigrant prejudice), and
are in the hypothesized direction and are sta- income in the lowest quartile.
tistically significant for predicting racial How much of the difference in prejudice
prejudice; 5 are in the hypothesized direction scores between countries is explained by in-
and are statistically significant for predicting dividual characteristics? Almost none. The
anti-immigrant prejudice. Examining the size microlevel variables (variables in Model 1)
of these effects, however, most of the indi- explain none of the variance among countries
vidual characteristics have small effects on in racial prejudice, and actually increase the
prejudice. On the racial prejudice scale of 0 dispersion in the country-level means of anti-
to 1, a 5-year difference in education is ex- immigrant prejudice from an estimated vari-
pected to decrease prejudice by only .042, ance of .0058 to .0067 (a table with variances
while a 20-year difference in age is expected for all models in Tables 1 and 2 is available
to increase prejudice by only .026. from the author on request). Compositional
Three of the variables, however, are statis- differences in individual characteristics
tically nonsignificant or have effects in the across countries do not explain variations of
opposite direction as those predicted by prejudice at national levels.
Table 2. Coefficients from the Bi-Level Regression of Anti-Immigrant Prejudice Index Scores on In-
dividual-Level and Country-Level Variables: 12 EEC Countries, 1988
Models
Individual-Level Variables and Interaction Terms
Education -.0111*** -.0111*** -.0111*** -.0118** -.0034*

(.0009) (.0009) (.0008) (.0045) (.0016)
Education x Percent non-EEC - -.0025**

immigrants (.0007)
Education x Economic condition .0575

(.0329)
Education x Percent non-EEC - -.0420***

Age .0008*** .0008*** .0008*** .0023** .0002

(.0001) (.0001) (.0001) (.0007) (.0002)
Age x Percent non-EEC immigrants .0001

(.0001)
Age x Economic condition -.0157*

(.0051)
Age x Percent non-EEC immigrants x .0035**

Working class .0203** .0204** .0205** .0335 .0036

(.0066) (.0066) (.0066) (.0338) (.0118)
Working class x Percent non-EEC - .0040

immigrants (.0054)
Working class x Economic condition - -.2124

(.2459)
Working class x Percent non-EEC - - .0978
Change in economic situation .0078 .0079 .0079 -.0215 -.0071

(.0051) (.0051) (.0051) (.0285) (.0096)


Change in economic situation x - .0970

Economic condition
Alienation score .0755*** .0752*** .0748*** .0449 .0186

(.0070) (.0070) (.0070) (.0377) (.0129)
Alienation score x Percent .0205***

Alienation score x Economic - -.1496

condition (.2704)
Alienation score x Percent non-EEC - - .2966***

Income in lowest quartile -.0043 -.0044 -.0044 .0442 .0232**

(.0053) (.0053) (.0053) (.0288) (.0099)
(Table 2 contin ued on n
(Table 2 continued)
Models
Individual-Level Variables and Interaction Terms (Continued)

Income in lowest quartile x - -.013l1**
Income in lowest quartile x -.1739

Income in lowest quartile x -.1433**

Economic condition
Sex (male = 1) .0206*** .0206*** .0206*** .0207*** .0208***

(.0042) (.0042) (.0042) (.0042) (.0042)
Life satisfaction .0001 .0001 .0001 -.0003 -.0001

(.0010) (.0010) (.0010) (.0010) (.0010)
Contact with Immigrants

Immigrants in .0224*** .0223*** .0223*** -.0757** .0188**
neighborhood (.0044) (.0044) (.0044) (.0239) (.0080)
Immigrants in neighborhood x .0094*

Immigrants in neighborhood .7453**

Immigrants in neighborhood x .0154

Economic condition
Immigrants at workplace -.0187** -.0189** -.0188** -.0173** -.0190**

(.0066) (.0066) (.0066) (.0066) (.0066)
Country-Level Variables
Country percent non-EEC immigrants - .0333* -.1660 -.1578 -.1632
(.0114) (.1028) (.1073) (.0968)
Economic condition - -.7611 -2.429* -2.295* -2.409*

(.5177) (.9687) (1.012) (.9123)
Interaction (Percent non-EEC - 2.430 2.336 2.396

immigrants x Economic condition) (1.247) (1.302) (1.175)
<.05 < .01 < .001 (two-tailed tests)
Note: Numbers in parentheses are standard errors. Test of hypotheses country-level variables (including
interaction effects of the products of country- and individual-level variables) use t-distribution (d.f. = 10).
All models are estimated with intercepts, although the intercepts are not shown. The total N for each model
is 11,676. The constants are random at the first and second levels of analysis.
Results with Group-Level Variables subordinate group poses to the dominant

group; that perceived threat is influenced by
The basic results that test Hypothesis 1 are the economic situation faced by the dominant
shown in the in the second and third models group and by the size of the subordinate
of Tables 1 and 2. These models correspond group relative to the dominant group; and
in form to equation 4. The results generally that the effects may be additive or interac-
confirm the group-threat theory and the im- tive. Models 2 and 3 of Tables 1 and 2 test
portance of both subordinate group concen- this hypothesis. Model 2 includes the indi-
tration and economic circumstances to levels vidual-level variables and the group-level
of prejudice. Hypothesis 1 stated that preju- variables: percent of immigrants in the coun-
dice is a function of the perceived threat the try non-EEC (Pk) and economic conditions
as inverse of the average GDP for 1984-1988 tion term, Pk and Gk are positively related to
(Gk). Model three adds the interaction of prejudice across most of the existing range
these two factors (Pk x Gk)- of each variable.
For both racial and anti-immigrant preju- The substantive effect of perceived group
dice, in Model 2 Pk is statistically significant threat on racial and anti-immigrant prejudice
while Gk is not. Model 3, however, suggests within countries is moderately large. Using
that Model 2 is misspecified. Model 3 in the estimate from Model 3, shifting from Ire-
Tables 1 and 2 adds the interaction of eco- land, the country just below the 25th percen-
nomic condition and relative subordinate tile on the interaction term (Pk X Gk) to Bel-
group size (percent non-EEC immigrants). gium, the country just above the 75th percen-
tile on Pk X Gk, causes a change in expected
For racial prejudice, the results indicate that
average racial prejudice of about .23 and a
Pk X Gk is substantively and statistically sig-
nificant. For anti-immigrant prejudice, Pk X change in anti-immigrant prejudice of .19.
Gk is substantively significant and ap- On the 0 to 1 prejudice index scale, this
proaches statistical significance (p = .08). change is substantively large, particularly
The effects of percent non-EEC immigrants when considering that the standard deviation
and economic conditions intensify each of the mean prejudice scores for the 12 coun-
other, so that prejudice is more likely when tries (for both anti-immigrant and racial
there is both a large foreign presence and prejudice) is about .08.
poor economic conditions than would be ex- The measures of perceived group threat
pected by the sum of their additive effects. describe well the average level of prejudice
When examining the Model 3 results in across countries. The estimated variance of
Tables 1 and 2, it is worth keeping in mind the racial prejudice intercepts across coun-
the meaning of the negative signs of vari- tries in Model 1 (the model with no group-
ables average economic condition (Gk) and level predictors of threat) is .0068, while in
percent of the population that are non-EEC Model 3 it is .0015. For anti-immigrant preju-
immigrants (Pk). Since these variables are in- dice, the estimated variance among prejudice
cluded in the same model with their interac- intercepts is .0067 in Model 1 and .0015 in
tion term (Pk X Gk), the coefficient of Gk is Model 3. After controlling for group-level
the effect of economic conditions on preju- variables, the estimated variance in the preju-
dice conditional on 0 percent non-EEC im- dice intercepts across countries for both anti-
migrants (Pk = 0). Similarly, the coefficient immigrant prejudice and racial prejudice is
of Pk is the effect of percent of the popula- reduced by more than 70 percent.
tion non-EEC on prejudice conditional on The general result is that perceived threat
inverse of GDP at zero (Gk = 0). Using the is strongly related to the average country
results from Model 3 for racial prejudice level prejudice for both anti-immigrant and
(Table 1), for instance, I calculated that for
countries with a score on Gk greater than or of the interaction (a change of the coefficient of
equal to .075, the effect of Pk is predicted to Pk X Gk of less than 1.5 standard errors could
make the expected slope of Gk positive for all
be positive (not shown). A similar calculation
EEC countries). The predicted negative slope of
for Gk indicates it is expected to have a posi-
Gk for countries with low Pk, then, could easily
tive effect on prejudice for countries with be the product of sampling error or could be the
.759 percent or more non-EEC foreigners in result of a small nonlinearity. These results are
residence.19 When I account for the interac- consistent with the possibility that the effect of
Gk on prejudice in countries with low Pk is zero.
19 Using the Model 3 specification, for some Finally, a negative slope for Gk is not necessarily
countries with low Pk the expected slope of Gk inconsistent with the group-threat theory. Poor
computed from Tables 1 and 2 is negative (not economic circumstances could increase sympathy
shown). This implies what may seem like an for immigrants (who may be seen as marginalized
anomalous result: In countries with low percent by citizens of wealthy nations) as long as there
non-EEC immigrants, attitudes toward immi- are so few of them so that they pose little threat.
grants worsen as the country's economic condi- Thus, a negative estimated slope of Gk for some
tions improve. But the few expected negative countries does not deny that Pk and Gk intensify
slopes are well within the range that is possible each other's effect in causing prejudice, a key
due to sampling error in estimating the coefficient prediction of the theory.
racial prejudice. Perceived threat far outper- workplace, chi-square tests of variation in
forms individual-level variables in predicting the slopes across countries (not shown) were
country level of prejudice. In fact, the indi- not significant at the p < .25 level 20 for both
vidual-level variables explain none of the racial and anti-immigrant prejudice (for tech-
differences between countries at all, while nical details of these tests, see Goldstein
the two measures of perceived threat (Pk and 1987, appendix 2.1). In addition, the effect
Gk) and their interaction (Pk X Gk) strongly of having an income in the lowest quartile of
predict the degree of prejudice in the coun- one's country did not vary significantly
tries studied here. across countries for the racial prejudice de-
pendent variable only. These results suggest
that the effects of these individual-level vari-
Results with Interactions of Individual-
ables are the same for each of the 12 EEC
Level and Group-Level Variables
countries. Therefore, interactions between
Hypothesis 2 states that individual-level the group-level variables and these indi-
prejudice toward outsiders emerges when the vidual-level variables are not included in
perceived threat posed by the subordinate Models 4 and 5. These individual variables
group is greater (that is, individual-level do not vary as Hypothesis 2 predicts because
characteristics are more strongly associated their effect on prejudice is roughly the same
with prejudice in settings where a higher pro- for all countries, regardless of the degree of
portion of the population belongs to the sub- threat.
ordinate group and where the economic cir- Second, I estimated a model in which the
cumstances are worse). Results are shown in slopes of the individual-level variables are
Models 4 and 5 of Tables 1 and 2. When ex- predicted by Gk, Pk, and Pk X Gk (excluding
amining the tables, remember that the signs variables found to be constant across coun-
of the interaction terms vary because the tries in the step). But with so many highly-
signs of the effects in the individual regres- correlated two- and three-way interactions in
sions vary. Education, for instance, is nega- the model to be estimated, multicollinearity
tively related to prejudice, so the negative among the predictors made it impossible. es-
timate all the coefficients simultaneously.
sign of the interaction term for education and
percent non-EEC immigrants indicates that There was not enough independent informa-
education reduces prejudice more in coun- tion in the sample to separate the effects of
tries where perceived threat to the dominant these highly correlated terms, so some con-
group is greater. Similarly, the positive sign straints were necessary.
of the interaction between scores on the All three terms could not be included to-
alienation index and percent non-EEC immi- gether because of multicollinearity, so in-
grants indicates that alienation increases stead I tried models with different combina-
prejudice more in countries where perceived tions of Pk, Gk, and Pk X Gk. Two models are
threat is greater. estimated: one with Pk and Gk, and one with
The sample provides relatively limited in- only Pk x Gk21
formation on cross-country variations in ef-
fects (12 countries), so it is difficult to inves- 2( As these tests were preliminary to a more
tigate all hypotheses precisely. In particular, complete test of these patterns of interaction, I set
it is difficult to separate the effects of differ- the significance level conservatively at p < .25
ent predictors: Do the slopes of individual (rather than p < .05) to make sure I did not- ex-
variables vary with percent non-EEC (Pk), clude interaction terms that might turn out to be
economic conditions (Gk), and/or the inter- statistically significant in Models 4 or 5 of Tables
action between these (Pk x Gk)? 1 and 2.
21 1 tried several additional specifications using
As an initial step, I attempted to determine
different combinations of Pk, Gk , and Pk X Gk for
which individual-level variables have no sig-
each individual-level variable. The conclusion as
nificant variation in their effects, across
to which terms are statistically significant proved
countries, so as to reduce the number of in- to be fragile to the particular specification em-
teraction terms to be included in the model. ployed. The results suggested that although some
For sex, life satisfaction, and the presence of of these interactions were significant for some
immigrants and people of other races in the variables, there was not enough information in
Initially, Model 4 in Table 1 predicts the support the conclusion that 3 of the interac-
coefficient of individual-level variables for tion variables have constant effects across
each country based on percent non-EEC im- countries (sex, life satisfaction, immigrants
migrants (Pk) and economic condition (Gk) at workplace). There are, again, 20 (5 + 1 +
entered additively. Of the 12 interactions of 8 + 6) possible interactions between indi-
group-level and individual-level variables in vidual-level and group-level variables for the
Model 4 of Table 1, 5 interactions (change in anti-immigrant prejudice equation. Overall,
economic status X Pk, change in economic only 5 of these 20 potential interactions are
significant and in the direction predicted by
status x Gk, alienation score x Pk, other races
in neighborhood x Pk, other races in neigh- Hypothesis 2. For anti-immigrant prejudice,
the effects of less than half of the individual-
borhood x GO) are statistically significant and
in the direction predicted by Hypothesis 2, 2 level variables are intensified by either of the
are statistically significant and in the direc- threat variables.
tion opposite that postulated by Hypothesis Finally, Model 5 predicts the slopes of the
individual variables on racial and anti-immi-
2 (age x Gk and education x GO),22 and 5 are
not significant. Because tests at the prior step grant prejudice using the interaction of Pk X
supported the conclusion that four of the in- Gk. The main effect terms (Pk and Gk), are
dividual variables have a constant effect constrained to 0. This constraint is probably
across countries (sex, life satisfaction, in- incorrect, but as discussed above there is not
come in lowest quartile, and immigrants at sufficient independent variation in the slopes
workplace) 8 possible interaction terms were across countries to estimate a model includ-
not entered into the model. There are, then, ing both the interaction terms and main ef-
20 (5 + 2 + 5 + 8) possible interactions be- fect terms for the model. For country-level
tween individual-level and group-level vari- variables, the interaction term proved to be
ables for the racial prejudice equation. Over- quite important in predicting the intercept,
all, only 5 of these 20 possible interactions and Model 5 assess if it is similarly impor-
are statistically significant and in the direc- tant in predicting the slopes of the indi-
tion predicted by Hypothesis 2. For racial vidual-level variables.
prejudice, the effects of less than half of the There are 10 individual-level variables pre-
individual-level variables are intensified by dicting racial prejudice in Table 1, Model 5.
either threat variable. Because tests in the prior step support the
The results are similar for anti-immigrant conclusion that they have a constant effect
prejudice, as shown in Model 4 in Table 2. across countries, 4 of the individual-level
Of the 14 interactions in Model 4, 5 interac- variables do not have interaction terms in
tions (education x Pk, alienation x Pk, in- Model 5. Of the variables with interaction
come in lowest quartile x Pk, immigrants terms
in in Model 5, coefficients for 4 are sig-
neighborhood x Pk, immigrants in neighbor- nificant and in the direction predicted by Hy-
hood x Gk) have significant coefficients and pothesis 2 (education x Pk x Gk, age x Pk x
are in the direction predicted by Hypothesis Gk, alienation x Pk x Gk, other races in the
2, 1 (age x Gk) is significant and opposite the neighborhood x Pk x Gk), and coefficients for
direction predicted by Hypothesis 2, and 8 2 are not statistically significant. This means
are statistically nonsignificant. In addition, 6 that 4 of the 10 individual-level variables in
possible interaction terms were not included the racial prejudice equation have a statisti-
in the model because tests at the prior step cally significant interactions with the group-
level variable as Hypothesis 2 predicts.
Again, less than half of the effects of the in-
the sample to draw a conclusion about which of dividual-level variables are intensified by
the country-level predictor variables were signifi- threat.
cant for each dependent variable.
The results for anti-immigrant prejudice
22 Note, however, that the result for the effect
shown in Model 5 of Table 2 are similar.
for education is due to a single outlier country,
Portugal, and if I trim by dropping the highest and
Once again there are 10 individual-level vari-
lowest country slopes, then percent non-EEC im- ables. Because tests in the prior step support
migrants influences prejudice just as Hypothesis the conclusion that they have a constant ef-
2 suggests. fect across countries, 4 of the individual-
level variables do not have interaction terms a high-threat country (just above the 75th per-
in Model 5. Of the variables with interactions centile of countries on Pk X Gk). The right-
in Model 5, coefficients of 4 interactions most column in the table shows the impact of
(education x Pk x Gk, age x Pk x Gk, alien- different levels of perceived threat on preju-
ation x Pk x Gk, and income in lowest dice for a movement from the 25th to the 75th
quartile X Pk x Gk) are statistically signifi- percentile on each of the individual-level
cant and in the direction predicted by Hy- variables. The effects often change consider-
pothesis 2, and coefficients of 3 interactions ably with changes in the level of perceived
are statistically nonsignificant. Overall, threat. The estimated slope for Belgium often
group-level interactions with 4 of the 10 in- differs from the slope for Ireland by a factor
dividual-level variables in the anti-immigrant of 2 or more. On the other hand, these
prejudice equation are statistically signifi- changes in the slopes associated with changes
cant and in the direction predicted by Hy- in perceived threat between Belgium and Ire-
pothesis 2. The Model 5 results do not lead land has only small effects on prejudice. A
to markedly different conclusions than the 30-year difference in age is predicted to have
additive model. the effect of increasing the expression of ra-
For both racial and anti-immigrant preju- cial prejudice by .0342 more in Belgium than
dice, then, the results from Models 4 and 5 in Ireland. Similarly, a 5-year change in years
suggest that less than half of the individual- of education is expected to decrease racial
level variables that influence prejudice vary prejudice by .0391 more in Belgium than in
with at least one of the hypothesized causes Ireland. Relative to the 0 to 1 prejudice index
of perceived threat as predicted by Hypoth- scale, these are small effects. Compared to
esis 2. For these specific variables, perceived the size of the slopes of individual-level vari-
threat to the dominant group increases the ables for Ireland and Belgium, the changes in
strength of the relation between individual the slopes resulting from changes in Pk X Gk
characteristics and prejudice. However, re- are large. But as most of the individual-level
sults do not support Hypothesis 2 for all indi- variables make only small impacts on racial
vidual-level variables, and for some variables or anti-immigrant prejudice to begin with,
it appears that the interaction works with only these changes in the slopes still have only
one of the two macrolevel causes of group small effects on prejudice.
threat rather than with both. The estimates of The effects of some (but not all) indi-
individual terms are imprecise, and a model vidual-level characteristics on racial and
anti-immigrant prejudice, then, varies with
with all three terms (Pk, Gk and Pk X Gk) could
not be estimated, so future research could the degree of perceived threat the subordi-
usefully test this hypothesis with a sample nate group presents to the dominant group.
including more level-two units. For certain individual characteristics, the re-
So far I have dealt only with the statistical sults support the idea that latent dispositions
significance and direction of these estimates. among dominant group members influence
Since it is conventional (all else being equal) the formation of prejudicial attitudes when
to accept additive over multiplicative mod- their group prerogatives are threatened. The
els, I use the Model 4 specification to assess impact of changes on the slopes of indi-
the substantive size of the effects.23 vidual-level variables on prejudice is small,
The effects of changes in the level of per- however, because individual characteristics
ceived threat on the effects of individual-level generally have little effect on prejudice.
variables are estimated in Table 3. (The ef-
fects are estimated using Model 4 from Tables
DISCUSSION
1 and 2.) Table 3 shows the expected effect of
perceived threat on the slopes in Ireland, a The central result of this paper is that the av-
low-threat country (just below the 25th per- erage degree of prejudice in an EEC country
centile of countries on Pk X Gk) and Belgium, is strongly related to the threat perceived by
the dominant group resident there. Although
the historical and cultural details of the sub-
23 The estimates of the size of effects are very
similar using Model 5, however, and in no way ordinate group and the dominant society are
change my conclusions. no doubt important to understand the form
Table 3. Impact of Perceived Group Threat on the Effects of Individual-Level Variables for Ireland
and Belgium a
Impact of
Expected Effect forb Inter- Difference

quartile in Slope on
Variable Ireland Belgium Difference Rangec Prejudiced
Racial Prejudice
Education -.0033 -.0111 -.0078 5.0 -.0391
Age .0006 .0017 .0011 30.0 .0342

Working class .0055 .0298 .0243 1.0 .0243
Other races in neighborhood .0298 .0337 .0039 1.0 .0039
Change in economic situation .0083 .0201 .0118 1.0 .0118
Alienation score .0471 .0816 .0345 .4 .0138
Intercept .1101 .3372 .2271 .2271
Anti-Immigrant Prejudice
Education -.0051 -.0147 -.0096 5.0 -.0481

Age .0002 .0012 .0010 30.0 .0317
Workinng class status .0062 .0276 .0214 1.0 .0214
Immigrants in neighborhood .0319 .0200 -.0119 1.0 -.0119
Change in economic situation .0028 .0181 .0152 1.0 .0152
Alienation Score .0345 .0974 .0628 .4 .0251
Income lowest quartile .0137 -.0130 -.0267 1.0 -.0267
Intercept .1108 .2970 .1863 .1863
a Ireland and Belgium were chosen b

25th and just below the 75th percentile, respectively (Ireland = .069; Belgium = .282).
b The expected effects are computed from the estimates in Tables 1 and 2, Model 4.
C The interquartile range (IQR) is the difference between the 25th and 75th percentiles. The number in the
column is the IQR for the pooled sample of 11,676 individuals, except for dichotomous variables, where the
number in the column is 1. The effect of the difference in slopes is evaluated for a change in individual
characteristics from the 25th to the 75th percentile for ordinal and interval variables, and from 0 to 1 for
binary variables.
d This is the change in expected prejudice associated with a change in an individual characteristic list
under the IQR column for an individual in Ireland as compared to Belgium. For instance, a five-year chan
in education is predicted to decrease racial prejudice by .0391 more in Belgium than in Ireland because of
the greater threat presented to the dominant group in Belgium.
that prejudice takes and how it is expressed, onstrating the influence of perceived group
the results here indicate that the relative size threat, this study provides a useful counter-
of the subordinate group and the economic point to individual-level explanations that are
situation of the particular country can prominent in the contemporary empirical lit-
strongly influence the degree of prejudice erature on prejudice. The economic condi-
expressed by dominant group members. tions in a country and the size of the racial
Consistent with past research, individual- or immigrant group influence people's views
level prejudice variables alone are not suffi- of group relations, and in so doing influence
cient to explain cross-national variation in prejudicial attitudes. Threat is perceived by
prejudice. In fact, in this study individual individuals, but its relationship to prejudice
characteristics had little impact on prejudice depends on a comparison of the relations be-
and explained none of the difference in lev- tween dominant and subordinate social
els of prejudice between countries. By dem- groups.
Further, the results suggest that the effects mined by the population composition of a
of some individual-level characteristics on country and the economic conditions there,
racial and anti-immigrant prejudice are influ- suggests an interesting perspective on a num-
enced by the context of intergroup relations. ber of difficult problems in the areas of race
Most individual characteristics in this study, and ethnic relations. Why, for instance,
however, did not follow this pattern. More should prejudice against African-Americans
research is needed to better establish the be now, and historically have been, so much
presence or absence of interactions between worse than the prejudice faced by Asian-
individual-level and group-level factors that Americans? The group-threat theory and the
may influence expressions of prejudice. conclusions of this paper would suggest that
Taken with the work of Smith (1981) on the greater number of African-Americans as
racial attitudes in the United States, this compared to Asian-Americans may be one
study lends further credence to the idea that reason why Blacks face worse prejudice than
prejudice is largely a function of group posi- do Asians. Historical phenomena such as
tion. What this paper contributes, borrowing these result, no doubt, from complex and
especially from Blumer (1958), is a more multiple causes, and explanations of such
complete theory of the mechanisms and vari- phenomena will be equally complex. But re-
ables through which inter-group perceptions gardless of these complexities, in most situa-
of group position are formed, and it offers tions a few simple collective variables, such
empirical evidence to support this mecha- as population composition and economic cir-
nism. Additional research could contribute to cumstances, will profoundly influence the
understanding group causes of prejudice by feelings of members of the dominant group.
attempting to directly measure perceived
threat by asking questions about threats
Lincoln Quillian is a Ph.D. candidate in sociol-
against one's ethnic or national group (fol-
ogy at Harvard University. His dissertation fo-
lowing the lead of Vanneman and Pettigrew
cuses on mobility into and out of areas of high
[1972]), or by isolating other likely group- poverty concentration and the effects of nonran-
level determinants of prejudice and measur- dom selection into neighborhoods on estimates of
ing their effects (following procedures simi- neighborhood effects. His research interests in-
lar to those pursued here). clude stratification, race and ethnic relations,
The importance of group threat, as deter- and statistics.
Appendix A. Means and Standard Deviations for Dependent Variables and for Threat Variables, 12
Countries in the EEC, 1988
Anti- Percent Non- Inverse GDP

Racial Immigrant EEC Immigrants per Capita Interaction Number of
Country Prejudice Prejudice (Pk) (Gk) (Pk X Gk) Respondents
France .297 .278 3.9 .081 .317 998

(.294) (.268)
Belgium .389 .310 3.2 .088 .282 1,009

(.321) (.300)
Netherlands .175 .190 2.8 .085 .238 990
(.216) (.213)
West Germany .250 .277 5.3 .079 .416 1,045
(.283) (.281)
Italy .185 .158 .7 .086 .061 1,055
(.211) (.196)
Luxembourg .130 .249 2.0 .072 .144 298

(.187) (.238)
Denmark .233 .257 2.0 .079 .157 1,002

(.263) (.255)
(Appendix A continued on next page)
(Appendix A continued)
Anti- Percent Non- Inverse GDP

Racial Immigrant EEC Immigrants per Capita Interaction Number of
Country Prejudice Prejudice (Wk) (Gk) (Pk X Gk) Respondents
Ireland .141 .075 .5 .138 .069 994
(.193) (.124)
United Kingdom .248 .206 1.75 .084 .148 1,308

(.258) (.241)
Greece .184 .190 .8 .161 .129 987

(.218) (.218)
Spain .085 .090 .4 .121 .048 994

(.151) (.149)
Portugal .115 .099 .6 .169 .101 996

(.193) (.176)
Overall Meanb .209 .196 1.99 .104 .176 Total = 11,676

(.254) (.240) (1.55) (.034) (.114)
a As for all models in this paper, the means

For the racial prejudice and anti-immigrant prejudice variables, the overall mean is computed across all
11,676 individuals. For the country-level variables, the overall mean is computed across the 12 countries.
Appendix B. Robustness and the Problem of a Small Number of Cases
Given that there are only 12 EEC countries, my abil- Second, I used an alternate methodology that does
ity to control for confounding influences at the coun- not rely upon the assumption of fixed effects, al-
try level is extremely limited, and I must make some though it does make other assumptions.
assumptions to make the model at all identifiable. The ideal solution would be to include country-
First, since there are so few level-two cases, no level error terms for all of the individual-level vari-
additional second-level variables are included as ables when the final model (equation 4) is estimat-
controls. Even with more cases, however, it is not at ed. Unfortunately, with only 12 level-two units (the
all certain that including more controls for level-two EEC countries) it is impossible to estimate varianc-
variables would be constructive. Theories of group- es for so many error terms. In fact, I found that when
level causes of prejudice are poorly developed, and using ML3 there is only sufficient information to
so there is little theoretical or empirical basis on estimate three random level-two terms (and their co-
which to base choices for additional controls. As variances) in a single model simultaneously.
Lieberson (1985, chaps. 2, 10) explains, adding con- Accordingly, I ran several additional models for
trols with little understanding of their relation to the racial prejudice and anti-immigrant prejudice in
variables of interest is as likely to distort as to im- ML3. In each model the constant term and two of
prove estimates. My investigation is limited to two the individual characteristics were treated as random
variables: relative size of the subordinate group to effects (like adding an error term to equation 3).
the dominant group and country-level economic cir- When I included the two random effects together,
cumstances, but I do not claim that these are the only the point estimates of both the constant and of the
group-level causes of prejudice. interaction term never changed more than a trivial
Second, I make some assumptions in the models amount. The only difference is that in some cases
estimated in Tables 1 and 2. One assumption is that the standard errors of some interactions increased,
threat has a fixed effect on slopes of individual-lev- sometimes making them statistically nonsignificant.
el variables. Put another way, the variation in the The fact that the point estimates changed so little
slopes of individual variables across countries is as- when interactions were treated as fixed effects pro-
sumed to be described perfectly by the level of per- vides some evidence that there is little distortion in-
ceived threat plus an intercept (shown in equation troduced by considering the interactions as fixed.
3). This is unlikely to be true-the relationship be- My second approach to addressing the problem of
tween threat and the effect of most individual-level random effects at the second level of analysis is to
variances is probably not completely determined by estimate equations 1, 2, and 3 directly, using a two-
the two threat measures I use. stage regression procedure instead of ML3. In the
I use two strategies to evaluate the extent to which first stage, I first run 12 individual regressions (one
these assumptions might influence results. First, I for each country) relating racial prejudice and a se-
added additional random error terms to equation 4. ries of individual-level characteristics. Then I use the
intercept and the country-level slopes in a second- estimates, but the general pattern of both point esti-
stage regression, in which I estimate one equation mates and significance levels is very similar.a One
with the intercept as the dependent variable and a difference is that the hypotheses tests associated
series of equations in which the dependent variables with the slopes are not usually as significant as the
are the slopes from each of the first-stage regressions ML3 results, but most of the slopes are still signifi-
(like equations 2 and 3]). To reflect the differential cant at p < .1. This is not surprising, since the ML3
reliability of the estimated parameters used as de- estimates should be more efficient than estimates
pendent variables, I estimate the models with weight- from the WLS models.b In any event, using this al-
ed least squares (WLS), where the weight is the in- ternative methodology does not change the sub-
verse sampling variance of the dependent variable stantive results of this paper. This further increases
estimated in the level-one regression. my confidence that the results I present here are sub-
The main advantage of this procedure is that, in stantively correct, and the constraints made to iden-
contrast to the ML3 estimates, it treats the effect of tify the model have not created misleading results.
perceived threat on each of the slopes and the inter-
cept as a random effect. The major problem with this
approach, however, is that it cannot adjust for the a Tables showing the results mentioned in this
likely correlation among the slopes and the inter- section are available from the author upon request.
cept: It assumes that the intercept and each of the For both the procedures here, I use a specification
interaction terms are independent of one another. corresponding to equation 4 in Tables 1 and 2.
Additionally, estimates derived using the WLS esti- b See Bryk and Raudenbush (1992, chap. 5) for a
mates in the second-stage regression tend to be less comparison of multilevel modeling with correspond-
efficient than those derived using multilevel modeling models estimated at the person level and the
ing. Nevertheless, this procedure provides an addi- group level. In general, multilevel models estimated
tional check on the results using ML3, especially on at the individual level tend to have too generous
how it deals with the random coefficient problem. standard errors, while those estimated at the group
Using the second-stage regressions, the magni- level will tend to be much less efficient than esti-
tude of effects often differs somewhat from the ML3 mates from a multilevel modeling program.
REFERENCES
Adorno, Theodore W., Else Frankel-Brunswik, nation in the South." American Sociological
Daniel Levinson, and R. Sanford. 1950. The Review 22:677-82.
Authoritarian Personality. New York: Harper . 1967. Toward a Theory of Minority-
and Brothers. Group Relations. New York: John Wiley and
Anderson, Perry. 1981. Imagined Communities: Sons.
Reflections on the Origin and Spread of Na- Bryk, Anthony and Stephen Raudenbush. 1992.
tionalism. London, England: Verso. Hierarchical Linear Models: Application and
Blumer, Herbert. 1958. "Race Prejudice as a Data Analysis Methods. Newbury Park, CA:
Sense of Group Position." Pacific Sociological Sage Publications.
Review 1:3-7. Case, Charles, Andrew Greely, and Stephen
Bobo, Lawrence. 1983. "Whites' Opposition to Fuchs. 1989. "Social Determinants of Racial
Busing: Symbolic Racism or Realistic Group Prejudice." Sociological Perspectives 32:469-
Conflict?" Journal of Personality and Social 83.
Psychology 45:1196-1210. Castles, Stephen. 1984. Here for Good: Western
. 1988. "Group Conflict, Prejudice, and the Europe's New Ethnic Minorities. London, En-
Paradox of Contemporary Racial Attitudes." gland: Pluto Press.
Pp. 85-109 in Eliminating Racism: Profiles in . 1989. "Migrant Workers and the Trans-
Controversy, edited by P. Katz and D. Taylor. formation of Western Societies" (Cornell Stud-
New York: Plenum Press. ies in International Affairs Western Societies
Bobo, Lawrence and James R. Kluegel. 1993. Papers No. 22). Center for International Stud-
"Opposition to Race-Targeting: Self-Interest, ies, Cornell University, Ithaca, NY.
Stratification Ideology, or Racial Attitudes?" Castles, Stephen, and Gouda Kosack. 1973. Im-
American Sociological Review 58:443-64. migrant Workers and Class Structure in West-
Bonacich, Edna. 1972. "A Theory of Ethnic An- ern Europe. London, England: Oxford Univer-
tagonism: The Split Labor Market." American sity Press.
Sociological Review 37:447-559. Commission of the European Community. 1989.
Blalock, Hubert M. 1956. "Economic Discrimi- Eurobarometer: Public Opinion in the Euro-
nation and Negro Increase." American Socio- pean Community Special Issue (November).
logical Review 21:548-88. Dollard, John, Leonard Doob, Neal E. Miller, 0.
. 1957. "Percent Non-White and Discrimi- H. Mowrer, and Robert R. Sears. 1939. Frus-
tration and Aggression. New Haven, CT: Yale the Multilevel Linear Model." Sociological
University Press. Methodology 13:72-103.
Fossett, Mark and K. Jill Kiecolt. 1989. "The Maykovich, Minako K. 1975. "Correlates of Ra-
Relative Size of Minority Populations and cial Prejudice." Journal of Personality and So-
White Racial Attitudes." Social Science Quar- cial Psychology 32:1014-20.
terly 70:820-35. Middleton, Russell. 1976. "Regional Differences
Findley, Sally. 1982. "Internal Migration: Deter- in Prejudice." American Sociological Review
minants." Pp. 344-51 in International Encyclo- 41:94-117.
pedia of Population, edited by J. Ross. New Organization for Economic Cooperation and De-
York: Free Press. velopment. 1991. National Accounts 1960-
Frisbie, W. Parker and Lisa Neidert. 1977. "In- 1989. Vol 1. Paris, France: Organization for
equality and the Relative Size of Minority Economic Cooperation and Development.
Populations: A Comparative Analysis." Ameri- Olson, James M. and Mark P. Zanna. 1993. "At-
can Journal of Sociology 32:1007-30. titudes and Attitude Change." Annual Review
Giles, Michael W. 1977. "Percent Black and Ra- of Psychology 44:117-54.
cial Hostility: An Old Assumption Reexam- Olzak, Susan. 1992. The Dynamics of Ethnic
ined." Social Science Quarterly 58:412-17. Competition and Conflict. Palo Alto, CA:
Goldstein, Harvey. 1987. Multilevel Models in Stanford University Press.
Educational and Social Research. New York: Pettigrew, Thomas. 1957. "Demographic Corre-
Oxford University Press. lates of Border State Desegregation." American
Hoffman, Curt and Nancy Hurst. 1990. "Gender Sociological Review 22:683-89.
Stereotypes: Perception or Rationalization?" . 1980. "Prejudice." Pp. 820-29 in The
Journal of Personality and Social Psychology Harvard Encyclopedia of American Ethnic
58:197-208. Groups, edited by S. Themstrom, A. Orlov, and
Hyman, Herbert, Charles Wright, and John Reed. 0. Handlin. Cambridge, MA: The Belknap
1975. The Enduring Effects of Education. Chi- Press.
cago, IL: University of Chicago Press. Prosser, Robert, Jon Rasbash and Harvey Gold-
Jackman, Mary. 1978. "General and Applied Tol- stein. 1991. ML3: Software for Three-Level
erance. Does Education Increase Commitment Analysis. Users Guide for V.2. London, En-
to Racial Education?" American Journal of Po- gland: University of London Institute of Edu-
litical Science 22:302-24. cation.
Jackman, Mary and Michael Muha. 1984. "Edu- Reif, Karlheinz, and Anna Melich. 1991a. Euro-
cation and Intergroup Attitudes, Moral Enlight- barometer 30 Codebook [MRDF]. Ann Arbor,
enment, Superficial Democratic Commitment, MI: Inter-University Consortium for Political
or Ideological Refinement?" American Socio- and Social Research [producer and distributor].
logical Review 49:751-69. Reif, Karlheinz and Anna Melich. 1991b. Euro-
Kinder, Donald R. and David 0. Sears. 1981. barometer 30: Immigrants and Out-Groups in
"Symbolic Racism Versus Racial Threat to the Western Europe, October-November 1988
Good life." Journal of Personality and Social [MRDF]. Survey Conducted by Faits et Opin-
Psychology 40:414-31. ions, Paris, France. Ann Arbor, MI: Inter-Uni-
Kinloch, Graham. 1974. The Dynamics of Race versity Consortium for Political and Social Re-
Relations: A Sociological Analysis. New York: search [producer and distributor].
McGraw-Hill. Schissel, Bernard, Richard Wanner, and James S.
Kramer, Jane. 1980. Unsettling Europe. New Frideres. 1989. "Social and Economic Context
York: Vintage Books. and Attitudes toward Immigrants in Canadian
Lebon, Andre. 1990. "Ressortissants commun- Cities." International Migration Review 23:
autaires et strangers originaires des pays tiers 289-308.
dans l'Europe des douze" (Citizens from the Sears, David 0. and Carolyn L. Funk. 1991. "The
EEC Countries and other Foreigners in the Eu- Role of Self-Interest in Social and Political At-
ropean Economic Community). Revue Europe'- titudes." Advances in Experimental Social Psy-
enne des Migrations Internationales 6(1):185- chology 24:1-91.
202. Semyonov, Moshe, Danny R. Hoyt and Richard
Lieberson, Stanley. 1980. A Piece of the Pie: I. Scott. 1984. "Place, Race, and Differential
Black and White Immigrants Since 1880. Ber- Occupational Opportunities." Demography
keley, CA: University of California Press. 21:259-70.
. 1985. Making It Count: Toward the Im- Smith, Wade A. 1981. "Racial Tolerance as a
provement of Social Research and Theory. Ber- Function of Group Position." American Socio-
keley, CA: University of California Press. logical Review 46:558-73.
Mason, William, George Y. Wong, and Barbara Stouffer, Samuel. 1955. Communism, Conformity,
Entwisle. 1984. "Contextual Analysis through and Civil Liberties. New York: Doubleday.
Sullivan, John, James Pierson, and George 1972. "Race and Relative Deprivation in the
Marcus. 1982. Political Tolerance and Ameri- Urban United States." Race 13:461-86.
can Democracy. Chicago, IL: University of Wilcox, Jerry and Wade C. Roof. 1978. "Percent
Chicago Press. Black and Black-White Status Inequality:
Tienhaara, Nancy 1974. "Canadian Views on Im- Southern Versus Non-Southern Patterns." So-
migration and Population: An Analysis of Post cial Science Quarterly 59:422-34.
War Gallup Polls." Ottawa, Canada: Informa- Winship, Christopher and Robert D. Mare. 1984.
tion Canada. "Regression Models with Ordinal Variables."
Vanneman, Reeve D. and Thomas Pettigrew. American Sociological Review 49:512-25.

Q Pajita

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Q Pajita

Diunggah oleh

Hak Cipta:

Format Tersedia

Prejudice as a Response to Perceived Group Threat: Population Composition and Anti-

Immigrant and Racial Prejudice in Europe

586 American Sociological Review, 1995, Vol. 60 (August:586-61 1)

i An additional stream of prominent research

5 I am indebted to an ASR reviewer for bring-

Hypotheses The Dependent Variables: Anti-Immigrant

I include independent individual-level vari-

Independent Variables (1) (2) (3) (4) (5)

Individual-Level Variables and Interaction Terms

Education -.0083*** -.0084*** -.0083*** -.0l77*** -.0033*

Education x Percent non-EEC -.0009

Education x Economic .1 079*

Education x Percent non-EEC -.0274**

Age .00 1 3*** .00 1 3*** .0013*** .0029*** .0006*

Age x Percent non-EEC immigrants .0001

Age x Economic condition -.0174**

Age x Percent non-EEC immigrants x .0043**

Workinng class .0200** .0200** .0203** .0256 .0002

Working class x Percent .0059

Working class x Economic -.1667

Working class x Percent non-EEC .1141

Change in economic status .0092 .0093 .0094 -.0781 ** -.0079

Change in economic status x Percent .0149*

Change in economic status x .5717*

Change in economic status x .1099

Alienation score .0706*** .0706*** .0701*** -.0057 .0321*

Alienation score x Percent .0186*

Alienation score x Economic condition .3154

Alienation score x Percent non-EEC .1963*

Sex (1 = Male) .0187*** .0187*** .0187*** .0187*** .0189***

(Table I continued on next page)

Independent Variables (1) (2) (3) (4) (5)

Individual-Level Variables and Interaction Terms (Continued)

Income in lowest quartile -.0174** -.0175** -.0176** -.0144** -.0170**

Contact with Members of Other Races

Other races .0292*** .0291*** .0290*** -.0727** .0104

Other races at workplace -.0160* -.0163* -.0165* -.0157* -.0165*

Economic condition -.3176 -3.534** -3.397** -3.514**

Interaction (Percent non-EEC 4.662** 4.632** 4.627**

.05 < .01 < .001 (two-tailed tests)

basis of the individual-level theories dis- microlevel theories of prejudice: change in

Independent Variables (1) (2) (3) (4) (5)

Individual-Level Variables and Interaction Terms

Education -.0111*** -.0111*** -.0111*** -.0118** -.0034*

Education x Percent non-EEC - -.0025**

Education x Economic condition .0575

Education x Percent non-EEC - -.0420***

Age .0008*** .0008*** .0008*** .0023** .0002

Age x Percent non-EEC immigrants .0001

Age x Economic condition -.0157*

Age x Percent non-EEC immigrants x .0035**

Working class .0203** .0204** .0205** .0335 .0036

Working class x Percent non-EEC - .0040

Working class x Economic condition - -.2124

Change in economic situation .0078 .0079 .0079 -.0215 -.0071

Change in economic status x .0083

Change in economic status x .1459

Change in economic situation x - .0970

Alienation score .0755*** .0752*** .0748*** .0449 .0186

Alienation score x Percent .0205***

Alienation score x Economic - -.1496

Alienation score x Percent non-EEC - - .2966***

Income in lowest quartile -.0043 -.0044 -.0044 .0442 .0232**

Education -.0083* -.0084* -.0083* -.0l77* -.0033*

Age .00 1 3* .00 1 3* .0013* .0029* .0006*

Workinng class .0200 .0200 .0203** .0256 .0002

Alienation score .0706* .0706* .0701*** -.0057 .0321*

Sex (1 = Male) .0187* .0187* .0187* .0187* .0189***

Income in lowest quartile -.0174 -.0175 -.0176 -.0144 -.0170**

Other races .0292* .0291* .0290* -.0727 .0104

Economic condition -.3176 -3.534 -3.397 -3.514**

Interaction (Percent non-EEC 4.662 4.632 4.627**

Education -.0111* -.0111* -.0111* -.0118 -.0034*

Age .0008* .0008* .0008* .0023 .0002

Working class .0203 .0204 .0205** .0335 .0036

Alienation score .0755* .0752* .0748*** .0449 .0186

Sex (male = 1) .0206* .0206* .0206* .0207* .0208***

Immigrants at workplace -.0187 -.0189 -.0188 -.0173 -.0190**