Shaping the Development of Prejudice

Abstract

Social dominance orientation (SDO) has been theorized as a stable, early-emerging trait influencing outgroup evaluations, a view supported by evidence from cross-sectional and two-wave longitudinal research. Yet, the limitations of identifying causal paths with cross-sectional and two-wave designs are increasingly being acknowledged. This article presents the first use of multi-wave data to test the over-time relationship between SDO and outgroup affect among young people. We use cross-lagged and latent growth modeling (LGM) of a three-wave data set employing Norwegian adolescents (over 2 years, N = 453) and a five-wave data set with American university students (over 4 years, N = 748). Overall, SDO exhibits high temporal rank-order stability and predicts changes in outgroup affect. This research represents the strongest test to date of SDO’s role as a stable trait that influences the development of prejudice, while highlighting LGM as a valuable tool for social and political psychology.

Keywords

social dominance orientation outgroup affect prejudice longitudinal studies latent growth modeling

The idea that responses to social outgroups might be driven by an early-forming preference for group-based hierarchy continues to be counter-intuitive. As such, the role of social dominance orientation (SDO; Ho et al., 2015; Pratto, Sidanius, Stallworth, & Malle, 1994) as a stable, causal variable in the development of intergroup attitudes has attracted substantial attention and debate (Ekehammar, Akrami, Gylje, & Zakrisson, 2004; Kreindler, 2005; Kteily, Ho, & Sidanius, 2012; Kteily, Sidanius, & Levin, 2011; Lehmiller & Schmitt, 2007; Schmitt, Branscombe, & Kappen, 2003; Sibley & Duckitt, 2010; Sibley & Liu, 2010). Although several studies with longitudinal data have yielded results consistent with SDO’s status as a stable predictor of intergroup attitudes (Asbrock, Sibley, & Duckitt, 2010; Kteily et al., 2011; Sibley & Duckitt, 2010; Sibley & Liu, 2010; Sibley, Wilson, & Duckitt, 2007a; Thomsen et al., 2010), these studies have been restricted to cross-lagged designs over two waves, with adult samples. The present article reports, for the first time, studies of SDO and outgroup affect using multi-wave data collected from samples of young people in two different countries—a three-wave study of adolescents in Norway (Study 1) and a five-wave study of university students in the United States (Study 2). We used a combination of cross-lagged panel analysis and latent growth modeling (LGM) to examine the development of SDO and outgroup affect over multiple years, and the ability of each variable to predict over-time changes in the other.

The Nature of SDO

Since its introduction, the concept of SDO has been theorized to be an enduring, general stance toward intergroup inequality. SDO is defined as “expressing the value that people place on non-egalitarian and hierarchically structured relationships among social groups” (Sidanius & Pratto, 2001, p. 21), such that a person high in SDO will see a hierarchy of social groups as both natural and desirable (Pratto et al., 1994). Recent studies using factor analysis have indicated that SDO has a two-dimensional structure, and the literature now refers to intergroup dominance orientation (SDO-D) and intergroup anti-egalitarianism (SDO-E) as its correlated sub-dimensions (Ho et al., 2015; Ho et al., 2012; Jost & Thompson, 2000; Kugler, Cooper, & Nosek, 2010). SDO-D is one’s desire to see some groups actively dominate and oppress other groups, and is most strongly related to hostile attitudes such as old-fashioned racism, nationalism, support for the death penalty, militarism, and support for war. SDO-E indexes a preference for inequality between groups and is more related to subtle forms of racism, opposition to affirmative action, hierarchy-enhancing social ideologies, and career choice (Ho et al., 2015; Ho et al., 2012). Although apparently forming two sub-dimensions, SDO-D and SDO-E are strongly correlated, hanging together as a unified construct, and are analyzed as such in this article.

The origins of SDO were theorized to include both dispositional and situational factors. As identified by both social identity theorists and social dominance theorists, being socialized into a dominant group or an environment that encourages group hierarchy will tend to result in a stronger SDO (Dambrun, Kamiejski, Haddadi, & Duarte, 2009; Fischer, Hanke, & Sibley, 2012; Lehmiller & Schmitt, 2007; Pratto, Sidanius, & Levin, 2006; Schmitt et al., 2003; Sidanius & Pratto, 1999, 2001; Sinclair, Sidanius, & Levin, 1998; Turner & Reynolds, 2003), and one’s SDO increases even if one is temporarily associated with a high power position (Guimond, Dambrun, Michinov, & Duarte, 2003, Study 3).

Yet, social dominance theory does not posit SDO as a mere reflection of socialization, existing intergroup relations, and contextual variations. In fact, Sidanius and Pratto theorized that SDO is an early-emerging, stable trait with a status akin to personality, and with a causal impact on intergroup attitudes (Sidanius & Pratto, 1999, Chapter 3). Positing an individual’s “temperament” as a key source of SDO, social dominance theorists have argued that biological traits such as sex, and personality facets such as empathy, would be reliably linked to SDO (Pratto et al., 1994).

Evidence for the Stability of SDO

The first set of evidence in favor of SDO’s stability comes from data suggesting links to underlying temperament. A recent meta-analysis showed that the sex difference in SDO, in which men report higher levels than women, is robust across 22 countries (I. Lee, Pratto, & Johnson, 2011). Furthermore, SDO has consistently been found to be negatively correlated with core personality traits, such as agreeableness (Sibley & Duckitt, 2008), openness to experience (Duriez & Soenens, 2006; Ekehammar et al., 2004), and the honesty/humility component of the Honesty-Humility, Emotionality, eXtraversion, Agreeableness (versus Anger), Conscientiousness (HEXACO) model (K. Lee, Ashton, Ogunfowora, Bourdage, & Shin, 2010; Sibley, Harding, Perry, Asbrock, & Duckitt, 2010). Indeed, there is even evidence implying that one personality facet—empathy—can be causally influenced by SDO (Sidanius et al., 2013), again supportive of the predictive, trait-like status of the latter.

Aside from demonstrating robust links to personality and sex, SDO has been theorized as having trait-like features such as high rank-order stability across time and contexts. The term rank-order stability does not imply that SDO levels never meaningfully change over different social contexts. Rather, social dominance theorists argue that individuals’ relative SDO scores, if not their absolute scores, will be fairly stable over time and across social contexts (Levin, 1996; Pratto et al., 2006; Sidanius, Sinclair, & Pratto, 2006).

One source of evidence for SDO’s rank-order stability across time is found in the high autocorrelations (or test–retest reliability coefficients) between SDO scores administered across waves of panel surveys, even where such waves are as long as 4 years apart (Kteily et al., 2011; Sibley & Duckitt, 2010; Sibley & Liu, 2010; Sidanius et al., 2013; Sidanius, Levin, van Laar, & Sears, 2010; Thomsen et al., 2010). Even more notable is the observation of rank-order stability in SDO scores across different intergroup contexts, as demonstrated in a study conducted with Jewish Israelis by Levin (1996; see Pratto et al., 2006; Sidanius & Pratto, 1999, Chapter 3). This study showed that when a randomly selected half of the SDO scale items were administered in the context of the conflict between the high-status Ashkenazi and low-status Mizrachi Jews in Israel, SDO was higher among the former group than among the latter. However, when the other half of the SDO scale was measured after the same sample was primed to think about Israeli Arab–Jewish relations, in which both groups have much higher status than a third group (i.e., Arabs), the SDO scores of both Ashkenazi and Mizrachi Jews were higher, and were more similar. Despite this situational shift in the mean level of SDO scores, the correlation between the SDO scores across contexts was nonetheless robust (r = .56, or r = .72 when the Spearman–Brown split-half formula was used). This implies that among Israeli Jews, those with relatively high SDO scores in the Ashkenazi–Mizrachi ethnic context also had relatively high SDO scores in the Israeli Arab–Jewish national context (and the converse for those with relatively low SDO scores).

SDO and the Development of Outgroup Attitudes

In addition to positing SDO as an enduring individual difference trait, social dominance theorists argue that it will influence the development of a range of traits, attitudes, and behaviors within the intergroup domain (Sidanius, Pratto, & Mitchell, 1994). Indeed, two decades of research have shown that SDO has robust predictive power over a range of consequential intergroup attitudes and behaviors, encompassing both specific outgroups and generalized prejudice. These include phenomena such as individual levels of prejudice, discriminatory behaviors, support for war, hostility toward immigrants, the aggressive persecution of terrorists, opposition to affirmative action and wealth redistribution, physiological fear response to outgroup faces, and choice of careers that have been defined as hierarchy-enhancing (e.g., the police) versus hierarchy-attenuating careers (e.g., social work; for reviews, see Pratto et al., 2006; Sidanius, Cotterill, Sheehy-Skeffington, Kteily, & Carvacho, in press). It is through this long, robust, and theoretically coherent array of relationships that SDO has been suggested to be one of the most important individual difference variables in the fields of political psychology and intergroup relations (e.g., Kandler, Bleidorn, & Riemann, 2012; McFarland, 2010; Sibley & Liu, 2010).

Prominent critiques from the social identity tradition, however, have argued against the interpretation of these correlations as indicative of the causal influence of SDO (Lehmiller & Schmitt, 2007; Schmitt et al., 2003; Turner & Reynolds, 2003). For instance, Schmitt and colleagues (2003) argued that correlations between scores on SDO and expressed outgroup attitudes in surveys result from the fact that the respondent is thinking about his or her attitude toward a particular outgroup when filling out the SDO questionnaire. In response to this critique, evidence assessing SDO’s causal status comes, first, from cross-sectional studies, which have investigated SDO’s covariation with prejudice and discrimination even in the case of minimal groups or novel social categories and new social policies (Amiot & Bourhis, 2005; Ho, Sidanius, Cuddy, & Banaji, 2013; Ho et al., 2012, Sample 7; Krosch, Berntsen, Amodio, Jost, & Van Bavel, 2013; Pratto et al., 1994; Reynolds et al., 2007). Amiot and Bourhis (2005), for example, found that SDO was correlated with preferential allocation to the ingroup in a minimal group scenario, despite the fact that SDO was measured 1 month earlier. Thus, the correlation between SDO and intergroup discrimination could not have resulted from participants thinking of the minimal group scenario as they completed the SDO scale. Although this does not rule out the potential for other forms of prejudice to affect SDO, it does support the claim that when new attitudes are being formed in a minimal group setting, SDO is an important input variable to those attitudes.

A more persuasive way of demonstrating SDO’s causal power vis-à-vis intergroup attitudes is to analyze longitudinal data. This approach has been used by many recent studies of SDO and prejudice. These studies have analyzed responses to measures of SDO and outgroup attitudes obtained from the same sample twice, across time intervals varying from 5 months to 4 years (Asbrock et al., 2010; Dhont, Van Hiel, & Hewstone, 2013; Kteily et al., 2011; Sibley & Duckitt, 2010; Sibley, Wilson, & Duckitt, 2007b; Thomsen et al., 2010). Most of these studies converge on a picture of SDO as predicting prejudice and discrimination over time, occasionally supplemented by the reverse causal path. For example, Asbrock et al. (2010) conducted a longitudinal analysis of attitudes of undergraduate students at two time points, 6 months apart. The authors found that SDO among undergraduates at the first administration predicted prejudice 6 months later toward “derogated” groups (e.g., housewives and the unemployed) and “dissident” groups (e.g., protestors and feminists), controlling for the initial levels of the prejudice variables. Cross-lagged effects with SDO as a predictor have also been found in the case of hostile sexism (Sibley et al., 2007a), perceived ethnic victimization among Whites (Thomsen, Green, & Sidanius, 2008), outgroup friendships (Kteily et al., 2011), and support for ideologies that legitimize an unequal status quo (Sibley & Liu, 2010, Studies 2-4).

A recent paper by Dhont et al. (2013, Study 2) yielded a different conclusion. The authors first demonstrated that experiences of intergroup contact on a short-term school trip were associated with a decrease in SDO levels compared with before the trip (Dhont et al., 2013, Study 1). In a subsequent study, Dhont et al. used a two-wave longitudinal design with data collected over 3 months, and analyzed SDO and intergroup contact using structural equation modeling (SEM), testing cross-lagged effects. They also tested prejudice as a third variable. Using standardized estimates, they found that intergroup contact assessed at Wave 1 exerted a moderate (beta = −.17) effect on SDO assessed at Wave 2, whereas SDO had no effect on intergroup contact. When adding prejudice as a third variable, they found SDO to be affected primarily by prejudice (beta = .22) and to have only a minor (yet still statistically significant) downstream effect on prejudice (beta = .09). The conclusion from this research is that over short time periods, varying from 1 week to 3 months, SDO may be particularly malleable in response to intergroup experiences and related attitudes.

Limitations of Previous Longitudinal Research

Despite the important insights into SDO and outgroup evaluations gained from previous longitudinal research, it has some important limitations. First, the test–retest correlations used therein do not give decisive answers as to whether a psychological variable is trait-like, with rank-order stability across time and context. The issue is one of knowing how large a correlation should be to indicate rank-order stability. For example, a correlation of .50 may be high, but still shows a substantial proportion (75%) of non-common variance across measurements. A simple, non-statistical alternative might be to compare the rank order of all individuals at different time points. In other words, do some people change their position in the rank order? Nonetheless, this approach would result in a similar problem. A few people will probably change their positions, and because there is bound to be some fluctuation due simply to measurement error, it is not clear at what point there is too much “error” to reject the claim of rank-order stability. What is needed is a statistical test that does not rely on subjective judgments, but tells us whether there is a significant difference in “fit” between a model claiming trait-like rank-order stability and a model claiming significant changes across time. Second, the predominant use of only two waves limits analyses of longitudinal developments, both for tests of rank-order stability and for tests of relationships between variables. Changes in measured variables between two time points can be random, including representing a regression to the mean (Willett & Sayer, 1994). Consequently, one should use data from at least three measurement occasions, and a statistical analysis that integrates the repeated measurements into a single test of rank-order stability. Investigations into causality will also be strengthened by use of multiple measurement occasions, and integrating the repeated measurements into a single cross-lagged path representing the assumed causality. This is because the effect of SDO on outgroup attitudes will be a continuous process, one which is distorted by traditional cross-lagged panel analysis, relying as it does on paths between (usually only two) time-specific measurements (Rogosa, 1980). In addition, the time-specific measurements are often arbitrarily chosen, such that cross-lagged panel analysis may result in contradictory findings depending on the timing of observations (Oud, 2002). This is particularly an issue to the extent that the stability versus malleability of SDO and prejudice vary depending on the time intervals used (see above).

Longitudinal analysis with three or more waves, however, allows for more detailed and robust estimates of over-time developments. With three or more waves, cross-lagged panel analysis can be improved because it uses at least two cross-lagged paths for an assumed causal effect, the similarity in their effect sizes (or lack thereof) being informative. Nonetheless, because measurement points in cross-lagged designs are often arbitrary, the problem of reducing a continuous process to time-specific measurements remains. Thus, an even more important benefit of analyzing data from the same sample obtained at three or more waves is that one can use a more advanced statistical method, LGM, to investigate over-time trajectories (Curran & Hussong, 2003; Duncan, Duncan, & Strycker, 2006; Preacher, Wichman, MacCallum, & Briggs, 2008).

A further important limitation in most of the previous research on SDO, whether cross-sectional or longitudinal, is that it has been restricted to adult populations (including university students). Commentators on the debate about the causal status of SDO highlight the importance of obtaining developmental data among adolescents to gain a deeper understanding of SDO’s role in the formation of prejudice and political attitudes (Wilson & Liu, 2003). Intergroup attitudes, as other politically relevant attitudes, are likely to be still in formation in the early teenage years, making the study of younger populations a particularly ripe opportunity to examine SDO’s role in the formation and solidification of prejudice (Merelman, 1972; Sears, 1975, see also Torney-Purta, 2004). Indeed, a recent meta-analysis of the development of prejudice throughout childhood and young adulthood points to a severe lack of longitudinal studies in youth and in adolescence in particular (Raabe & Beelmann, 2011).

LGM in the Present Research

This article goes beyond previous research by using multi-wave longitudinal data and by introducing LGM in the analysis of SDO and outgroup affect. In exploring developments in SDO and outgroup affect, and a potential causal relationship between these variables, we use both cross-lagged panel analysis and LGM. We analyze data from two longitudinal samples, one assessed at three time points and one assessed at five time points. We also expand the lens of analysis from commonly studied American and Australasian university student and adult samples to a European sample of middle-school students, aged between 13 and 15 years.

LGM (Curran & Hussong, 2003; Duncan et al., 2006; Preacher et al., 2008), also known as “growth curve modeling,” is an advanced multivariate technique that can test different aspects of over-time change in single construct variables and associations between two longitudinally assessed constructs. Because LGM is a new approach in the analysis of intergroup relations, we explain our use of this method before continuing with our specific research questions.¹

Once provided with data on a variable measured more than twice over time, LGM can produce estimates of two factors for that variable: a latent intercept representing its initial level, and a latent growth factor, which represents its change over time. Each of these factors has a mean and a variance, the significance of which can be tested. In the top section of Figure 1, the variables and paths with thick lines depict a growth model for SDO, which allows us to assess its stability over time. If there is no growth (change) over time in SDO, then a model in which the factor loadings between the latent intercept for SDO and all of its time-specific measurements (labeled t₁, t₂, and t₃) are fixed to one would be sufficient to explain the three waves of SDO scores. If, however, the data indicate that levels of SDO change over time, then a growth factor is required for the model to explain scores beyond the initial measurement. Thus, in Figure 1, measurements at t₂ and t₃ are explained both by the latent intercept and the growth factor, the growth factor representing change over time (separating change at t₂ and t₃ from the overall stability at all three measurement occasions).

Figure 1.

An example of a multivariate growth model of SDO and outgroup affect.

A growth factor being necessary indicates either that there is an overall change over time in mean levels of SDO, or that there is significant variation between individuals in patterns of change over time. Whereas a significant mean for the growth factor indicates overall mean changes over time, a significant variance for the growth factor implies that SDO does not change in a uniform manner between individuals.

If both the latent intercept and the latent growth factor have significant variance (reflecting individual variations around the overall trend), then the covariance between the two factors is of interest. The covariance between the latent intercept and the latent growth factor may answer the question, for example, of whether individuals who originally have very low scores in SDO catch up with others and develop similarly high scores in SDO. Zero covariance between the two factors would give a very clear indication of no changes in rank order, whereas a positive covariance would indicate that rank order is maintained, but that the differences in SDO scores increase over time. A small-to-moderate negative covariance between the latent intercept for SDO and the growth factor would suggest that those with initially lower scores of SDO increase their SDO scores relative to others. Rank order may still be maintained, but the data would indicate a tendency toward reduced differences between individuals in SDO. A strong negative covariance, however, would be inconsistent with the assumption of rank-order stability in SDO.

Following an analysis of the growth models of SDO and outgroup affect separately, the best-fitting growth models for SDO and outgroup affect are combined into a multivariate growth model. Figure 1 shows a potential outcome of this model development, using growth factors for both constructs. The multivariate growth model can distinguish between different components of the overall correlation between SDO and outgroup affect. Of particular interest are the cross-lagged paths between the intercept of one variable and the growth factor of another, which give an estimated effect from the initial state of one construct to over-time changes in the other construct. Notably, the cross-lagged paths modeled by multivariate growth modeling are not dependent on time-specific measurements, as they are in traditional cross-lagged panel analysis.

One cross-lagged path in Figure 1 represents estimated effects from outgroup affect to SDO (a causal path acknowledged both by social identity theorists and social dominance theorists and thus labeled SIT (SDT) in the figure). The other cross-lagged path goes from SDO to outgroup affect (not acknowledged by social identity theorists but central to social dominance theory and thus labeled “SDT”). If both constructs need a growth factor, then both cross-lagged paths in Figure 1 can be tested. If, however, the initial tests indicate that no growth factor is required for one of the constructs, then neither the growth factor for this construct nor the causal paths toward it will be included in the model, as there would be no change in that construct to explain.

Another path of interest is the covariance between the two latent intercepts, which indicates the overall covariance between the constructs, and is derived from data at all measurement occasions. In addition, the analysis should test for covariances between the residuals for time-specific measurements of SDO and outgroup affect (in Figure 1, one of these two-headed paths is labeled Time-specific covariance). If the associations between SDO and outgroup affect are time-specific rather than representing a stable covariance (Schmitt et al., 2003), then covariances between time-specific measurements should be enough to explain the association between SDO and outgroup affect.

Finally, the model in Figure 1 includes the residual covariance between the two growth factors. If the growth factors have a high residual covariance (their disturbance/error variables have a strong covariance), then a reasonable interpretation would be that their common growth is not well explained by the initial scores for the two constructs, but rather, is dependent on one or more variables not included in the model.

Research Questions

This first application of LGM to the study of SDO, outgroup affect, and their interrelationship, enables us to ask two key research questions, each with related sub-questions:

Research Question 1: What is the nature of changes or stability in SDO and feelings toward ethnic outgroups?

Specifically, we can assess how growth in SDO and growth in outgroup affect compare, by testing alternative growth models for our two constructs separately. This also enables us to test the assumption of rank-order stability in SDO, which would be supported either if no growth factor is required for SDO or if the covariance between the growth factor and the intercept is positive or close to zero.

Research Question 2: What is the nature of the interrelationships between SDO and feelings toward ethnic outgroups?

In particular, we can test the claim by social dominance theorists that SDO explains over-time developments in outgroup affect, by conducting these tests of assumed causal paths with both LGM and cross-lagged panel analysis. LGM also allows us to examine whether the covariance between SDO and outgroup affect can be explained merely by time-specific associations, by seeing whether the covariances between the residuals for time-specific measurements are sufficient to explain the overall association between the two constructs. Support for the claim of the associations being merely time-specific would also require that both the cross-lagged paths and the covariance between intercepts are non-significant. Finally, LGM allows us to assess whether developments in SDO and outgroup affect are explained by these variables affecting each other, rather than by a third variable. A third causal variable is likely at play if the analysis suggests that both SDO and outgroup affect need a growth factor, and there is a substantial covariance between the two growth factors.

As far as we are aware, this is the first instance of the application of LGM to the intergroup relations literature. This approach allows us to conduct rigorous tests of some of the key assumptions of social dominance theory—whether SDO exhibits trait-like stability and whether it predicts the development of outgroup affect—in two samples of young people at a critical life stage.

Study 1

Sample

Data for Study 1 were collected in Drammen, a town with 63,000 inhabitants outside Oslo, Norway. About 25% of adolescents in Drammen have a non-Western immigrant background (the largest ethnic minority group being ethnic Turks, about 20% of the minority population, followed by ethnic Pakistanis). All of the town’s students in Grades 8 to 10 (and thus all of its six middle schools) were invited to participate in a longitudinal questionnaire study. The questionnaire assessed mental health and various aspects of intergroup relations, in addition to drug use. A measurement of SDO was introduced in 2008, and the present research used data from the cohort participating at all three waves between 2008 and 2010 (Grades 8 to 10), with measurements conducted in November each year. Participation required active parental consent and was voluntary. The questionnaires were completed online on personal laptop computers (routinely provided by the school) under teacher supervision (teachers were instructed not to look at the students’ answers). We used a prize lottery to motivate participation at all waves of the data collection.

Analyses were restricted to youth who self-categorized as being ethnic Norwegians, providing a sample size of 453 (54% girls). The overall response rate was 75% (for majority and minority students), and ethnic Norwegians had a higher response rate than minority youth (a common observation in this type of data collection, in particular when active consent from parents is required). There were no data available to compare responders with non-responders. Dropout was low: Sample size at t₁ (ethnic Norwegian students in Grade 8) was 380; the dropout rate at t₂ was 19%, and the dropout rate from t₂ to t₃ was 24%. Seventy-three students joined the study after t₁.

Measurements

Following Sidanius et al. (2010), SDO was assessed with four items: “It is a good thing that certain groups are at the top and other groups are at the bottom,” “Sometimes other groups must be kept in their place,” “We should do what we can to equalize conditions for groups,” and “We should do what we can to increase social equality.” We used a Norwegian translation, displayed in the Online Appendix Figure A1. Table 1 shows descriptive statistics and correlations for measured variables.

Table 1.

Variable Descriptive Statistics and Intercorrelations From All Three Waves in Study 1.

	M	SD	% missing	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23
SDO
t₁
1. Item 1	2.74	1.65	20.3	1.00
2. Item 2	2.82	1.73	20.5	.63*	1.00
3. Item 3	2.66	1.74	20.7	.16*	.19*	1.00
4. Item 4	3.54	1.88	21.4	−.01	.03	.43*	1.00
t₂
5. Item 1	2.80	1.58	21.6	.23*	.23*	.12*	.00	1.00
6. Item 2	2.63	1.57	21.6	.21*	.31*	.23*	.14*	.55*	1.00
7. Item 3	2.78	1.68	21.6	.03	.13*	.23*	.26*	.25*	.32*	1.00
8. Item 4	3.61	1.90	23.8	.06	.17*	.08	.24*	.19*	.19*	.57*	1.00
t₃
9. Item 1	3.00	1.80	30.0	.20*	.20*	.05	.03	.32*	.25*	.20*	.25*	1.00
10. Item 2	2.68	1.73	30.6	.21*	.28*	.15*	.07	.27*	.34*	.30*	.27*	.70*	1.00
11. Item 3	2.70	1.75	30.0	.09	.11	.20*	.11	.18*	.24*	.25*	.16*	.25*	.39*	1.00
12. Item 4	3.43	1.84	30.6	.00	.08	.11	.11	.14*	.17*	.24*	.24*	.24*	.27*	.54*	1.00
Outgroup Affect
t₁
13. Turks	5.17	2.97	26.0	.12*	.16*	.10	.07	.09	.10	.10	.05	.07	.11	.14*	−.04	1.00
14. Pakistanis	5.43	2.97	27.8	.12*	.21*	.09	.06	.10	.14*	.10	.06	.07	.08	.17*	−.05	.82*	1.00
15. Indians	5.01	3.05	27.3	.07	.15*	.08	.03	.10	.14*	.11	.00	.01	.06	.15*	−.04	.79*	.84*	1.00
16. Iraqis	5.49	3.06	29.7	.11	.20*	.13*	.12*	.12	.16*	.13*	.08	.06	.09	.16*	.00	.80*	.85*	.83*	1.00
t₂
17. Turks	4.43	2.97	23.4	.20*	.31*	.05	.03	.21*	.23*	.24*	.21*	.20*	.25*	.18*	.17*	.42*	.39*	.36*	.36*	1.00
18. Pakistanis	4.80	3.03	25.8	.16*	.29*	.05	.04	.20*	.26*	.20*	.22*	.22*	.23*	.18*	.17*	.39*	.44*	.37*	.37*	.87*	1.00
19. Indians	4.27	2.98	25.3	.11	.26*	.10	.10	.21*	.28*	.25*	.20*	.20*	.26*	.18*	.17*	.34*	.42*	.42*	.42*	.79*	.82*	1.00
20. Iraqis	5.04	3.17	26.4	.15*	.25*	.08	.09	.18*	.29*	.26*	.24*	.19*	.26*	.19*	.20*	.40*	.44*	.40*	.45*	.84*	.89*	.84*	1.00
t₃
21. Turks	4.46	2.80	29.7	.03	.19*	−.01	.00	.14*	.12	.15*	.22*	.21*	.27*	.23*	.19*	.36*	.39*	.27*	.33*	.48*	.45*	.36*	.42*	1.00
22. Pakistanis	4.70	2.91	30.8	.02	.17*	−.04	−.03	.15*	.17*	.16*	.20*	.18*	.28*	.23*	.16*	.33*	.41*	.30*	.34*	.43*	.47*	.36*	.43*	.86*	1.00
23. Indians	4.21	2.81	31.3	.00	.16*	.03	−.01	.14*	.16*	.17*	.20*	.22*	.35*	.20*	.18*	.24*	.31*	.30*	.31*	.39*	.40*	.46*	.43*	.80*	.85*	1.00
24. Iraqis	4.86	2.95	30.4	.00	.11	−.01	.03	.13*	.14*	.20*	.24*	.20*	.31*	.26*	.19*	.30*	.36*	.27*	.34*	.44*	.45*	.39*	.48*	.82*	.89*	.83*

Note. SDO = social dominance orientation.

p < .05.

Outgroup Affect was assessed with a Feeling Thermometer Scale (Alwin, 1997), using a drawing of a thermometer and temperatures ranging from zero to 100 degrees, with 50 degrees being a neutral midpoint (see the Online Appendix Figure A2). The adolescents were asked to indicate how coldly or warmly they felt toward girls and boys, respectively, from four ethnic outgroups: Turks, Pakistanis, Indians, and Iraqis (i.e., eight items in total for Outgroup Affect), all groups being represented among Drammen’s minority population. Expressed Outgroup Affect toward the various ethnic outgroups had strong intercorrelations, resulting in very high Cronbach’s alpha for the eight items (with values varying between .96 and .98 in the three school grades). We used an average of expressed Outgroup Affect toward girls and boys in each specific ethnic outgroup (which correlated at .81 to .84 in Grade 8 and equally or more highly in the following two school grades) to feed into our factor analysis (yielding four indicators at each measurement occasion, with Cronbach’s alpha values of .95 or .96). To simplify the presentation, we reversed the scale for Outgroup Affect, with high scores indicating less favorable affect.

Outgroup Affect and SDO were estimated as latent variables using confirmatory factor analysis. Model fit indicated that we needed to allow the residuals for either the SDO-D or SDO-E items to correlate. This was a decision that is in line with the bi-dimensional structure of SDO.² As it produced the more stable measurement model, we present results where residual variables for the two SDO-D items are correlated, thus defining SDO primarily by the two SDO-E items (due to the lower factor loadings for the SDO-D items produced by allowing their residuals to correlate).

Analysis

As described in the introduction, we used both cross-lagged panel analysis and LGM. By estimating time-specific measurements as latent variables, we applied a relatively advanced form of LGM: second-order LGM. Using second-order growth models enabled us to separate (a) measurement errors and (b) time-specific departure from the mean trajectory in the growth model. Measurement errors were modeled as residual variables for the indicators of time-specific measurements (similar to modeling measurement errors in ordinary confirmatory factor analysis). The time-specific departure from the mean trajectory was modeled as residual variances for the time-specific factors (see Preacher et al., 2008). This is shown in Figure 1, where the label “var1” refers to residual variances representing deviation from the mean trajectory for SDO, and “var2” refers to residual variances representing deviation from the mean trajectory for Outgroup Affect. We fixed the time-specific deviations for a specific construct to be invariant across measurements, consistent with recommendations in the literature (e.g., Preacher et al., 2008).

In addition, we tested measurement invariance across time. Because LGM incorporates not only covariances but also variable means, both factor loadings and intercepts of the indicators for SDO (or Outgroup Affect) should ideally be invariant across time. By having invariant factor loadings and invariant indicator intercepts, one achieves full scalar invariance (Vandenberg & Lance, 2000). Partial scalar invariance is seen as an adequate criterion to be satisfied (Byrne, Shavelson, & Muthén, 1989), meaning that very few factor loadings or indicator intercepts vary across measurements (this is also necessary for the second-order growth model to be mathematically identified). We tested models with invariant factor loadings and invariant indicator intercepts (using p < .05 in the scaled hierarchical chi-square test as indicating statistically significant differences). If the scaled hierarchical chi-square was significant, we freed single parameters on an exploratory basis, to allow variation across time until the chi-square value was above .05, achieving a model with partial scalar invariance. All these models allowed residuals for measured variables to be correlated across time, such correlations reflecting systematic measurement error due to item wording.

The data supported full scalar invariance (time-invariant factor loadings and time-invariant indicator intercepts) for Outgroup Affect (the scaled chi-square difference was non-significant, p = .084). For SDO, all factor loadings could be fixed to be time-invariant, and all but one indicator intercept (Item 2 at t₁) could be fixed to be time-invariant (p = .393), thus achieving almost full scalar invariance.

Estimation method

We used maximum likelihood estimations with robust standard errors. Robust standard errors lowered the risk of false positives (Type 1 error) that can arise with artificially reduced standard errors when analyzing variables with skewed distributions. Using robust standard errors implied that nested models (e.g., testing time-invariant factor loadings and indicator intercepts) had to be compared with the scaled difference chi-square test (Satorra & Bentler, 2001). All analyses with LGM used a sandwich estimator (Asparouhov, 2005) to account for students’ clustering in school classes, again to reduce the risk of false positives. Analyses with cross-lagged panel analysis could not use the sandwich estimator because the number of parameters was too high compared with the number of clusters.

Model fit

We evaluated models with fit indices commonly used in SEM. Although a non-significant chi-square is preferable, we followed common recommendations for the use of SEM models (Mueller & Hancock, 2010) and accepted models with approximate fit, if these had better fit than alternative models. We used commonly recommended cutoff values for indices of approximate fit (e.g., Mueller & Hancock, 2010), specifically, the root mean square error of approximation (RMSEA, with a cutoff value at .05) along with its 90% confidence interval (CI); the comparative fit index (CFI, with a cutoff value preferably at .95, or at least .90), and the standardized root mean square residual (SRMR). We used two-tailed p tests for individual parameters, but we also considered the one-tailed p test if we had an a priori prediction about whether associations should be positive or negative (see Hurlbert & Lombardi, 2009).

Due to having partly missing data (nonresponse to single items or dropouts), we used full information maximum likelihood estimations (Arbuckle, 1996; Enders, 2010). Dropouts were thus unlikely to represent any bias effect. The analyses used Mplus 7.2 (Muthén & Muthén, 2012).

Results

Cross-lagged panel analysis

Cross-lagged panel analysis with latent variables and full (Outgroup Affect) or nearly full (SDO) scalar invariance across time, χ²(241) = 331.197, p < .001, RMSEA = 0.029, CFI = 0.977, SRMR = 0.082, gave no clear conclusion. Both SDO and Outgroup Affect had one cross-lagged path that suggested an effect (two-tailed p = .064 for Outgroup Affect as a predictor and .048 for SDO as a predictor). Yet, both SDO and Outgroup Affect also had one cross-lagged path that was clearly non-significant (see the Online Appendix Table A1, for details).

LGM of SDO and outgroup affect

LGM tested whether SDO and Outgroup Affect had best fit with an intercept-only model, a linear growth model, or a model allowing for non-linear growth. The linear growth model pre-defined all factor loadings for the growth factor, with factor loadings fixed at 0 at t₁, at 1 at t₂, and at 2 at t₃. The non-linear growth model allowed one factor loading to be freely estimated by only fixing the first (t₁, fixed at 0) and the last factor loadings (t₃, fixed at 2).

For SDO, fit was best for the intercept-only model, χ²(53) = 106.837, p < 0.001, RMSEA = 0.048, CFI = 0.937, SRMR = 0.095. The linear growth model did not improve fit (e.g., RMSEA = 0.050), and its growth factor contained a non-significant mean and variance, corroborating the interpretation that the growth factor was redundant. The non-linear growth model failed to converge. Thus, the data indicated that SDO levels did not change over time.

The model of Outgroup Affect had best fit if it included a growth factor (linear or non-linear). We used the model with a linear growth factor, χ²(53) = 49.150, p = .62, RMSEA = 0.000, CFI = 1.000, SRMR = 0.040; the non-linear model did not improve fit significantly, p = .13. The growth factor had a significant and positive mean (M = 0.32, p < .001). Because higher scores on Outgroup Affect meant more negative evaluations, a positive mean implied a minor development toward less favorable Outgroup Affect. The variance of the growth factor was not statistically significant (var = 0.23, p = .31), and the latent intercept and the growth factor had no significant covariance. Thus, Outgroup Affect became more negative on average over time, the nature of this over-time change was not related to initial levels of Outgroup Affect, and the differences between individuals on their Outgroup Affect scores neither widened nor narrowed significantly.

We estimated a multivariate growth model of SDO and Outgroup Affect, using the two growth models that emerged, an intercept-only model for SDO and a linear growth model for Outgroup Affect with both an intercept and a growth factor. As shown in Figure 2 (see also Table A2 in the Online Appendix for further details), SDO predicted developments in Outgroup Affect: b = 0.34 (95% CI = [0.039, 0.638]), a small, but statistically significant effect on the 11-point scale of Outgroup Affect. This effect and the path from the intercept for Outgroup Affect (b = −0.13, p = .031) were sufficient to explain the variance in the growth factor for Outgroup Affect: The residual variance for the growth factor for Outgroup Affect was minor and not statistically significant (var = 0.15, p = .47).

Figure 2.

Multivariate growth model of SDO and Outgroup Affect in Study 1: Unstandardized estimates.

Time-specific covariances were not included because they did not improve model fit (p = .10). Even when (the redundant) time-specific covariances were included, the analysis indicated a cross-lagged effect from the SDO intercept to the growth factor for Outgroup Affect (b = .30, two-tailed p = .070, one-tailed p = .035). Also, the model in Figure 2 did not include a growth factor for SDO. We tried to add such a growth factor for SDO (at least theoretically allowing for a cross-lagged path from the intercept for Outgroup Affect), although the previous test of growth models for SDO had shown that no such growth factor should be estimated. The multivariate growth model with growth factors for both Outgroup Affect and SDO (linear or non-linear for SDO) failed to converge. Finally, we omitted the (required) growth factor for Outgroup Affect and introduced a (redundant) growth factor for SDO. Even this model failed to indicate any effect from Outgroup Affect to SDO; the cross-lagged path from the Outgroup Affect intercept to the SDO growth factor was non-significant.

Discussion of Study 1

This first application of LGM to the study of prejudice supported predictions from social dominance theory. SDO was relatively constant across time (making a growth factor redundant), compatible with the view that SDO reflects enduring individual temperament, and supporting its rank-order stability. In contrast, LGM indicated that feelings toward ethnic outgroups became less favorable over the 2 years assessed, as reflected by its growth factor for Outgroup Affect.

Furthermore, LGM suggested a significant over-time influence of SDO on feelings toward ethnic outgroups. Although the cross-lagged panel analysis yielded inconclusive results, the more reliable estimates of LGM suggested a cross-lagged effect from the SDO intercept, apparently driving changes in Outgroup Affect over the 2 years studied. The LGM analysis further suggested that the association between SDO and Outgroup Affect was more than time-specific covariances, as adding time-specific covariances to the model did not improve fit, and did not weaken SDO’s role as a predictor. Finally, there being no observed development in SDO means that the results are not consistent with the joint development of SDO and Outgroup Affect being explained by a third variable.

Study 2

Sample

The data for Study 2 were taken from a five-wave panel study of undergraduates from the University of California at Los Angeles. The study began during freshman orientation in the summer of 1996 and ended the spring of 2000 (for a comprehensive description of the sample, see Sidanius et al., 2010). Only White students were used in our analyses, of which the total number was 748 (54% female), with 196 providing data across all five waves. The participants ranged in age from 17 to 20, with a mean age of 17.9 (SD = 0.35). Total sample size at t₁ was 719. Dropout was moderate, 28% at t₂, increasing to 59% at t₅, the final measurement occasion. Twenty-nine students responded only after t₁.

Measurements

SDO was measured by use of the same four English language SDO items described in Study 1. Outgroup Affect was assessed by asking the White students how positively they felt toward the three major minority groups in the United States: (a) Latinos/Hispanics, (b) Asians/Asian Americans, and (c) African Americans/Blacks. All of these questions had a 7-point response scale from “very negative” to “very positive.” We reverse-scored the data for ease of presentation (high scores thus meaning less favorable Outgroup Affect).

Table 2 gives an overview of descriptive statistics and correlations for variables assessed at t₁, t₃, and t₅. Table A5 in the Online Appendix shows descriptive statistics for variables at all five measurement occasions, and Online Table A6 shows bivariate correlations for all five measurement occasions.

Table 2.

Variable Descriptive Statistics and Intercorrelations From Three of Five Waves in Study 2.

	M	SD	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20
SDO
t₁
1. Item 1	2.13	1.45	1.00
2. Item 2	1.62	1.13	.60*	1.00
3. Item 3	2.86	1.60	.32*	.31*	1.00
4. Item 4	2.53	1.44	.38*	.35*	.80*	1.00
t₃
5. Item 1	1.93	1.26	.36*	.18*	.06	.10*	1.00
6. Item 2	1.50	0.96	.22*	.20*	.14*	.08	.48*	1.00
7. Item 3	2.73	1.59	.18*	.13*	.40*	.40*	.30*	.26*	1.00
8. Item 4	2.19	1.30	.14*	.12*	.38*	.39*	.27*	.28*	.74*	1.00
t₅
9. Item 1	1.81	1.16	.26*	.13*	.06	.09	.55*	.25*	.21*	.26*	1.00
10. Item 2	1.38	0.80	.21*	.28*	.12	.13*	.28*	.50*	.23*	.25*	.38*	1.00
11. Item 3	2.48	1.45	.11	.24*	.46*	.44*	.22*	.20*	.67*	.55*	.29*	.25*	1.00
12. Item 4	2.04	1.27	.12*	.20*	.43*	.39*	.23*	.34*	.57*	.57*	.26*	.27*	.71*	1.00
Outgroup Affect
t₁
13. Latinos	2.51	1.32	.18*	.22*	.19*	.24*	.08	.03	.15*	.10	.06	.09	.20*	.18*	1.00
14. Asians	2.32	1.23	.10*	.13*	.14*	.20*	.03	.00	.09	.10*	.07	.08	.10	.08	.81*	1.00
15. African	2.42	1.31	.17*	.18*	.20*	.24*	.05	−.01	.16*	.11*	.02	.06	.16*	.17*	.87*	.79*	1.00
t₃
16. Latinos	2.09	1.07	.16*	.17*	.19*	.18*	.26*	.23*	.32*	.33*	.17*	.21*	.25*	.33*	.30*	.22*	.20*	1.00
17. Asian	2.19	1.20	.15*	.16*	.16*	.12*	.25*	.22*	.22*	.22*	.22*	.15*	.15*	.29*	.23*	.22*	.16*	.78*	1.00
18. African	2.08	1.03	.17*	.15*	.17*	.16*	.30*	.22*	.34*	.31*	.20*	.22*	.27*	.31*	.28*	.21*	.23*	.88*	.74*	1.00
t₅
19. Latinos	1.97	1.13	.14*	.17*	.24*	.20*	.16*	.29*	.23*	.19*	.19*	.23*	.32*	.38*	.22*	.11	.21*	.51*	.48*	.50*	1.00
20. Asians	2.07	1.21	.14*	.20*	.21*	.17*	.21*	.30*	.23*	.21*	.24*	.20*	.29*	.34*	.12	.13*	.10	.43*	.60*	.44*	.76*	1.00
21. African	2.03	1.08	.10	.12*	.27*	.21*	.13*	.23*	.25*	.25*	.18*	.19*	.32*	.38*	.17*	.12*	.21*	.50*	.47*	.52*	.89*	.76*

Note. SDO = social dominance orientation.

p < .05

To determine the degree to which the students participating in all waves of the data collection differed from those who did not, extensive attrition analyses were conducted on study “persisters” (those present for all waves of the panel study) and study “dropouts” (those not present for all waves of the study, see Sidanius et al., 2010, Appendix C). Attrition analyses showed that the “persisters” did not significantly differ from the “dropouts” in either demographic or ideological factors of interest. We also note that the present study applies full information analyses, thus including dropouts in the analysis.

Confirmatory factor analysis of all four SDO items showed that these items did not load on a single factor. The factor model without correlated error variances did not fit the data (e.g., RMSEA = .395, CFI = .807 at t₁; RMSEA = .218, CFI = .902 at t₅). We therefore allowed residuals for the two SDO-D items to correlate, thereby primarily defining the factor by the SDO-E indicators.

Analysis

The analyses in Study 2 were identical to those used in Study 1, with two exceptions. First, Study 2 extended cross-lagged panel models and latent growth models to include five-wave data. Second, as Study 2 did not involve data clustered in classrooms, we did not apply a sandwich estimator. We again used maximum likelihood estimations with robust standard errors because of skewed distributions in measured variables.

As with Study 1, Study 2 tested for measurement invariance across time prior to estimating growth and causal models. Three of the four SDO items (Items 1 to 3) had their factor loadings fixed to be invariant across all five measurement occasions. Item 4 was fixed to be invariant at three measurement occasions and to have a separate value at two measurement occasions (t₃ and t₅, with the same value at these two time points). The Online Appendix, Tables A7 and A8, describes factor loadings in detail and also shows the partially invariant indicator intercepts. The restrictions introduced to achieve partial scalar invariance for SDO did not result in a statistically significant drop in model fit according to the conventional cutoff, estimated with the scaled chi-square difference test (p = .051).

Outgroup Affect had one factor loading fixed to invariance across all measurement occasions (Item 2, Outgroup Affect toward Asians, the loading for which we fixed to one in subsequent analyses to identify the factor). Simultaneously, partially invariant factor loadings were supported for the remaining two items, and all three indicator intercepts could be fixed to be invariant at all measurement occasions except for t₁. The Online Appendix, Tables A7 and A8, describes invariance for factor loadings and indicator intercepts in detail. The partial measurement invariance introduced for Outgroup Affect did not result in a statistically significant drop in model fit (p = .061).

As in Study 1, Study 2 tested alternative single growth models before combining these into a multivariate latent growth model. The linear growth model used factor loadings increasing by one at each measurement occasion (from zero at t₁ to four at t₅). The non-linear growth model used growth factors with the factor loading for t₁ fixed to zero and the factor loading for t₅ fixed to four, with the remaining factor loadings allowed to be estimated freely based on the data.³

Results

Cross-lagged panel analysis

Our initial test of causal models applied cross-lagged panel analysis to the five-wave data for SDO and Outgroup Affect with latent variables, using the partial scalar invariance developed above.⁴ Figure 3 shows standardized results with all five measurement occasions. Further details of this analysis can be found in Online Appendix Table A7. The five-wave analysis indicated a stronger effect of SDO on Outgroup Affect than of Outgroup Affect on SDO: Three of four cross-lagged paths from SDO were statistically significant, but only one of the four cross-lagged paths from Outgroup Affect; effect sizes for cross-lagged paths from SDO were also substantially higher than those from Outgroup Affect. Unstandardized estimates (see online Table A7) showed that residual covariances between SDO and Outgroup Affect were substantially reduced in the cross-lagged model (from an initial covariance of 0.39 at t₁ to a residual covariance of 0.11 at the last two measurement occasions). This is compatible with the assumption that the statistical association between SDO and Outgroup Affect was largely due to causal effects between these two variables.

Figure 3.

Cross-lagged panel analysis of five-wave data of SDO and Outgroup Affect in Study 2: Standardized estimates.

LGM of SDO and Outgroup Affect

We used second-order LGM, that is, growth models with latent indicators of growth factors (again using the partial scalar invariance developed above).⁵ The five-wave data of SDO had the best fit with an intercept-only model, χ²(155) = 369.663, p < .001, RMSEA = 0.044, CFI = 0.923, SRMR = 0.099, indicating that SDO did not change over time. The linear growth model did not improve fit (e.g., RMSEA = 0.045) and its growth factor contained a non-significant mean (M = −0.002, p = .96) and variance (M = 0.018, p = .14). The non-linear growth model resulted in negative variance for the growth factor, and thus did not converge correctly. Moreover, even in this model, the growth factor for SDO had a non-significant mean and non-significant variance, further corroborating the conclusion that the intercept-only model had the best fit with the data.

Outgroup Affect, however, appeared to change over time. The model of Outgroup Affect had best fit when estimated as a non-linear growth model, χ²(71) = 94.495, p = 0.033, RMSEA = 0.021, CFI = 0.993, SRMR = 0.044, the intercept-only model giving lower fit (e.g., RMSEA = 0.045, with chi-square-based p < .0001). The scaled chi-square difference test also showed statistically significantly better fit for the non-linear growth model than for the linear growth model (p < .0001). Although there appeared to be little overall growth in Outgroup Affect (the mean of the growth factor was non-significant; M = −0.13, p = .16), the variance of the growth factor was statistically significant (var = 0.05, p < .001) indicating that some individuals developed more favorable and some developed more negative Outgroup Affect over the five measurement occasions. Specifically, the data suggested a moderate tendency toward students with more negative Outgroup Affect becoming more positive relative to the other students over the five measurement occasions, as indicated by the negative covariance (covar = −.12, p < .001) between the latent intercept and the growth factor.

We tested a multivariate growth model of SDO and Outgroup Affect with latent indicators of growth factors and partial measurement invariance as developed above. This multivariate model used the growth models developed previously (intercept-only model for SDO, non-linear growth model for Outgroup Affect). Covariances between residuals for time-specific measurements improved model fit (p < .001) and were thus included (as shown in Figure 4). The multivariate growth model applying this solution indicated that the latent intercept for SDO predicted developments in general Outgroup Affect (b = 0.08; 95% CI = [0.053, 0.100]; see Figure 4 and Table 3, with further details in Online Appendix Table A8). We also tested a model that included a latent linear growth factor for SDO (see beneath “With SDO Growth Factor” in Table 3 and further details in Online Appendix Table A8). A model with a non-linear growth factor for SDO failed to converge. Adding a linear growth factor for SDO did not change the findings. No cross-lagged effect by Outgroup Affect was suggested (b = 0.02 [−0.059, 0.098], two-tailed p = .63), but the model still resulted in a significant cross-lagged path from the SDO intercept to developing more negative Outgroup Affect (b = 0.07 [0.041, 0.097], two-tailed p < .001).

Figure 4.

Multivariate growth model of SDO and outgroup affect from Study 2: Unstandardized estimates.

Table 3.

Second-Order Growth Model of SDO and Outgroup Affect in Study 2: Unstandardized Estimates.

	No SDO growth factor		With SDO growth factor
	Estimate	p	Estimate	p
Causal paths (Intercept → Growth)
Outgroup Affect → Outgroup Affect	−0.18	<.001	−0.18	<.001
SDO → Outgroup Affect	0.08	<.001	0.07	<.001
SDO → SDO	—		−0.02	.511
Outgroup Affect → SDO	—		0.02	.629
Growth factors
SDO
t₂	—		1.00
t₃	—		2.00
t₄	—		3.00
t₅	—		4.00
Outgroup Affect
t₂	3.23	<.001	3.19	<.001
t₃	3.75	<.001	3.73	<.001
t₄	3.60	<.001	3.60	<.001
t₅	4.00		4.00
Model fit
χ²		864.539		860.174
df		519		514
p		<.0001		<.0001
RMSEA		0.030		0.030
CI RMSEA	[0.027, 0.034]		[0.027, 0.034]
CFI		0.957		0.957
SRMR		0.079		0.077

Note. See Table A8 in the Online Appendix for complete results. Latent intercepts were estimated with all factor loadings fixed to one (not shown in the table). Underlined parameters were fixed to the value shown. Both models included time-specific covariances between residuals for growth indicators; the scalar chi-square test showed a significant (p = .003) drop in model fit if these covariances were omitted. SDO = social dominance orientation; RMSEA = root mean square error of approximation; CI = confidence interval; CFI = comparative fit index; SRMR = standardized root mean square residual.

Discussion of Study 2

Study 2 provided even stronger evidence than Study 1 compatible with the conceptualization of SDO as a stable orientation that causes changes in feelings toward outgroup members. This time using five-wave data, LGM again uncovered a growth factor for Outgroup Affect, indicating that it changed over time. Meanwhile, the redundancy of a growth factor for SDO again indicated that SDO exhibited not only rank-order stability, but also stability in mean levels, this time over the longer time period of 4 years. Improving on Study 1, in Study 2, both the cross-lagged panel analysis and the multivariate LGM were consistent in their depiction of significant cross-lagged paths from SDO to the development of Outgroup Affect over five measured time points. With the exception of one of four of the cross-lagged paths from Outgroup Affect, neither cross-lagged analysis nor LGM indicated an effect in the opposite direction. Although time-specific associations between the constructs played a role, they were not enough to explain the interrelation of the constructs, and the cross-lagged paths in both analyses remained significant. Finally, and again consistent with Study 1, there being no growth factor for SDO suggested that the joint development of SDO and Outgroup Affect was not caused by a third, unmeasured variable.

General Discussion

As globalization and mass migration bring diverse populations into more contact than ever before, the pertinence of studying inter-ethnic prejudice will only continue to grow. At the core of such prejudice is a set of negative emotional reactions to members of groups other than one’s own: affective orientations that are thought to depend on upbringing, socialization, and inter-ethnic contact. Into this social psychological space, social dominance theory highlights that groups rarely come together on an equal footing, and thus that one’s attitude toward intergroup inequality is key to understanding the development of one’s attitude toward ethnic outgroups, particularly where the latter are lower in status. The present research stands as the most definitive test to date of the predictive power of the most widely used measure of attitudes toward intergroup inequality on the development of affect toward minority outgroups in youth.

In two multi-wave longitudinal studies conducted with young people in two different national contexts, we applied the statistical tool of LGM to the study of the development of SDO and Outgroup Affect over time, and tested causal paths between these constructs. LGM allowed us to avoid some of the limitations of the widely used two-wave cross-lagged panel analysis, and to assess the evolution of each variable over time, as opposed to merely comparing mean levels at one time with those at another. LGM and multi-wave cross-lag results from samples of adolescents in Norway assessed over three time points (spanning 2 years) and of undergraduates in the United States assessed over five time points (spanning 4 years) were broadly consistent, and in line with predictions from social dominance theory.

The first research question addressed was whether and how responses on measures of SDO and Outgroup Affect changed over the time periods assessed. By testing alternative growth models, we could investigate whether and how mean levels in either of our constructs changed over time, and also the nature of this change. It could have been, for example, that everyone increased in their SDO levels, but also (or instead), that there was a change in the size of the difference between individuals’ SDO levels, or indeed, a shift in individuals’ rank ordering on SDO. In fact, no such changes in mean levels, differentiation, or rank ordering, were observed for SDO in either study. That is, the data were clear that SDO levels across the three (Study 1) and five (Study 2) time periods measured were best explained by a growth model that contained just an intercept (indicating initial SDO level), and not a growth factor (indicating any kind of change over time). This finding provides further empirical evidence supporting conceptualization of SDO as an enduring trait,⁶ implying that it has remarkable stability at least over the 3 and 5 year periods assessed, in these youth samples.

Not only did SDO exhibit mean-level and rank-order stability, it was also more stable than Outgroup Affect. In contrast to the data on SDO, the data on Outgroup Affect in both studies were best explained by models that included a growth factor. In the Norwegian sample of adolescents, mean levels of Outgroup Affect became slightly more negative over time, and did so in roughly equal measure for teenagers of high versus low levels of outgroup liking. This finding qualifies the conclusion of Raabe and Beelmann (2011) from a meta-analysis of primarily cross-sectional studies of the development of prejudice, which was that unlike in childhood, prejudice does not change systematically in the adolescent years. Our finding is also concerning, given the high levels of contact our Norwegian participants had with ethnic outgroup students in their schools. Intergroup contact (although usually of the more involved kind than mere school attendance together) has been found to attenuate the growth of prejudice in children (Raabe & Beelmann, 2011). Previous research on older teenagers and young adults indicates that prejudice decreases with exposure to university (Dhont, Roets, & Van Hiel, 2011; Dhont et al., 2013; Sidanius, Pratto, Martin, & Stallworth, 1991; Wodtke, 2012), a trend that is only somewhat corroborated by our data. Specifically, scores on Outgroup Affect from our American university sample did not decrease on average over time, although those initially more prejudiced became less prejudiced over time, in relation to other participants. By illuminating details such as changes in variance over multiple time points, this application of LGM to the study of the development of Outgroup Affect thus challenges existing assumptions regarding the developmental trajectory of a key social psychological construct (see Dyck & Pearson-Merkowitz, 2014).

The second broad question addressed by this research concerns the nature of the interrelationship between SDO and Outgroup Affect. The now widely observed association between SDO and intergroup prejudice has been dismissed by some critics of social dominance theory as a mere product of the fact that one has a particular group context in mind when completing the SDO scale (Schmitt et al., 2003). Although this possibility is rendered less likely by evidence for SDO’s generality found in cross-sectional samples (see Kteily et al., 2012) and its over-time predictive power in two-wave longitudinal samples (e.g., Kteily et al., 2011; Sibley & Duckitt, 2010), it has hitherto been impossible to test directly the explanatory power of the posited time-specific associations. LGM is able to separate two origins of the time-specific correlations observed in cross-lagged panel analysis: the overall correlation between SDO and Outgroup Affect across time, and the time-specific deviations from the overall trend. Specifically, we were able to examine the impact on model fit of including covariances among the residuals of SDO and Outgroup Affect at each time point, and found them either to have no role in the model (Study 1), or to be small enough to have no impact on SDO’s predictive power vis-à-vis the development in Outgroup Affect (Study 2). Thus, these results tend to cast doubt on the plausibility of the interpretation by Schmitt and colleagues of the connection between SDO and negative Outgroup Affect.

Rather than being explained by time-specific covariances, any similarities in the development of SDO and Outgroup Affect could instead be spurious, caused by their joint relationship with some third, unmeasured variable. LGM can test for this possibility by analyzing whether the residual variables for the two growth factors were correlated (a possibility that would indicate the existence of such a third variable). As our analyses consistently yielded no evidence for changes in SDO, there was no growth factor with which the residuals of the growth in Outgroup Affect could be correlated, and thus, there was no evidence for a third variable causing joint developments in SDO and Outgroup Affect.

In fact, the results were compatible with the idea that the relationship between SDO and Outgroup Affect is one of over-time causation of Outgroup Affect by SDO. This conclusion can be drawn, first, based on the results of the multi-wave cross-lagged panel analyses, which, although unclear in the first study, were clear in the more informative second study, in suggesting SDO’s predictive power regarding later levels of Outgroup Affect. This application of cross-lagged panel analysis is more robust than previous published uses of cross-lagged analysis with SDO (e.g., Kteily et al., 2011), as it is based on more than two waves (and thus multiple sampling of cross-lag paths), and involves more robust estimations. Even more informative than the multi-wave cross-lagged analysis, however, was the multivariate LGM analysis of SDO and Outgroup Affect. Here, both studies were consistent in indicating a cross-lagged effect from initial levels of SDO to changes in Outgroup Affect over the time periods studied. This effect, however, seems moderate. Unstandardized coefficients for the suggestive prediction of the growth in Outgroup Affect by the SDO intercept were fairly small (one third of a point on an 11-point scale in Study 1, and one eighth on a 7-point scale in Study 2). Thus, these estimates reflect the limited change in Outgroup Affect.

The consistency in SDO’s predictive power as observed across two national contexts and educational stages, in such a uniquely robust analysis, which is simultaneously supported by cross-lagged panel analysis with five waves, is an important finding for the field of intergroup relations. It supports the conclusion that in the critical age periods of adolescence and young adulthood (13 to 20 years), SDO appears to influence the development of attitudes about ethnic outgroup members. As SDO is a general orientation shown to predict evaluations of many outgroups, one implication of these results might be that efforts to reduce prejudice should focus less on specific outgroups than on the more basic and psychologically upstream phenomenon of orientation toward intergroup hierarchy. Educating young people concerning general principles such as egalitarianism and universal rights might thus be more effective than educating them about racism specifically, especially given that explicit manifestations of the latter can more and more easily be hidden as children age (Banaji, Baron, Dunham, & Olson, 2008).

That said, caution should be exercised in inferring insights about prejudice in general from a study of just one component of it. The psychological distinction between “ingroup love” and “outgroup hate” (Brewer, 1999), for example, implies that evaluations of ingroup members, and intergroup bias in evaluations, may not follow the same pattern as do evaluations of outgroup members—an interesting question for follow-up studies.

We intend that the lessons to be taken from this research go beyond the theoretical and applied, to the methodological. Indeed, we hope that this first demonstration of the use of multi-wave cross-lagged panel analysis and LGM in intergroup relations research might inspire fellow researchers to apply advanced and robust analytic techniques in collecting and analyzing their data. We have demonstrated how LGM goes further than existing methods in assessing change in a psychological construct over time. This is because it can explicitly test for violation of rank-order stability. LGM can also convey the extent to which changes in test scores on a construct are driven primarily by people who had relatively low or high scores on the construct in the first place. Even more alluringly, LGM improves analyses of assumed causalities in longitudinal data, through its ability to distinguish between the different origins of overall correlations between two or more variables, and between overall trajectories versus individual variations around these trajectories.

Methodological Caveats

Highlighting the benefits of the advanced statistical approaches used here should not come without acknowledging the limitations of the particular use we made of them in this research. One issue concerns the SDO scale used, which, in four-item form, was not as psychometrically sound as it would have been had we been able to use the latest version of the 16-item scale (see Ho et al., 2015). Using the full SDO₇ scale would also enable us to analyze the role of intergroup dominance orientation and intergroup anti-egalitarianism separately, as opposed to privileging one sub-dimension, as we do here.

One shortcoming of all correlational analyses, including the current longitudinal analyses, is that they cannot speak to the issue of causality as well as can experimental research. Although discovery of the successful experimental manipulation of SDO levels would be intriguing, thus far, the literature shows relatively little evidence that the general component of SDO can be meaningfully shifted, even if its group-specific components have been found to be malleable (see, for example, Guimond et al., 2003; Schmitt et al., 2003). This is consistent with the recent suggestion (Bergh, Akrami, Sidanius, & Sibley, in press), that besides tracking reactions to specific groups within specific contexts, SDO will be sensitive to low-status groups in general and across a variety of social contexts.

One recurrent issue regarding causality that this set of correlational data can address, however, is the third variable problem. By searching for and failing to find a correlation between the residuals of the two growth factors (in our case, because SDO had no growth factor), we were able to gain confidence that the link between changes in the variables over measurements was unlikely to be a spurious one.

Finally, a more general constraint on the conclusions to be drawn from this article is the, as-yet, unknown generalizability of the results beyond the specific age groups, countries, and cultures studied here. Indeed, part of our enthusiasm in suggesting the use of advanced data-analytic techniques for multi-wave data to the intergroup relations literature is that this article be followed by many more attempts to investigate the evolution of critical social attitudes across a range of temperaments, ages, nations, cultures, and intergroup contexts.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Data collection and work by the first author for Study 1 were funded by the Norwegian Research Council, project number 185731. Data collection for Study 2 was supported by the Russel Sage Foundation.

Supplemental Material

The online supplemental material is available at .

Notes

References

Alwin

D. F.

(1997). Feeling thermometers versus 7-point scales which are better? Sociological Methods & Research, 25, 318-340.

Amiot

C. E.

Bourhis

R. Y.

(2005). Discrimination between dominant and subordinate groups: The positive-negative asymmetry effect and normative processes. British Journal of Social Psychology, 44(Pt. 2), 289-308. doi:10.1348/014466604X17605

Arbuckle

J. L.

(1996). Full information estimation in the presence of incomplete data. In Marcoulides

G. A.

Schumacker

R. E.

(Eds.), Advanced structural equation modeling: Issues and techniques (pp. 243-277). Mahwah, NJ: Lawrence Erlbaum.

Asbrock

Sibley

C. G.

Duckitt

(2010). Right-wing authoritarianism and social dominance orientation and the dimensions of generalized prejudice: A longitudinal test. European Journal of Personality, 24, 324-340. doi:10.1002/per.746

Asparouhov

(2005). Sampling weights in latent variable modeling. Structural Equation Modeling, 12, 411-434. doi:10.1207/s15328007sem1203_4

Banaji

M. R.

Baron

A. S.

Dunham

Olson

(2008). The development of intergroup social cognition: Early emergence, implicit nature, and sensitivity to group status. In Levy

S. R.

Killen

(Eds.), Intergroup attitudes and relations in childhood through adulthood (pp. 197-236). New York, NY: Oxford University Press.

Bergh

Akrami

Sidanius

Sibley

(in press). Is group membership necessary for understanding generalized prejudice? A re-evaluation of why prejudices are interrelated. Journal of Personality and Social Psychology.

Brewer

M. B.

(1999). The psychology of prejudice: Ingroup love or outgroup hate? Journal of Social Issues, 55, 429-444.

Byrne

B. M.

Shavelson

R. J.

Muthén

(1989). Testing for the equivalence of factor covariance and mean structures: The issue of partial measurement invariance. Psychological Bulletin, 105, 456-466. doi:10.1037/0033-2909.105.3.456

10.

Curran

P. J.

Hussong

A. M.

(2003). The use of latent trajectory models in psychopathology research. Journal of Abnormal Psychology, 112, 526-544. doi:10.1037/0021-843X.112.4.526

11.

Dambrun

Kamiejski

Haddadi

Duarte

(2009). Why does social dominance orientation decrease with university exposure to the social sciences? The impact of institutional socialization and the mediating role of “geneticism.” European Journal of Social Psychology, 39, 88-100.

12.

Dhont

Roets

Van Hiel

(2011). Opening closed minds: The combined effects of intergroup contact and need for closure on prejudice. Personality and Social Psychology Bulletin, 37, 514-528. doi:10.1177/0146167211399101

13.

Dhont

Van Hiel

Hewstone

(2013). Changing the ideological roots of prejudice: Longitudinal effects of ethnic intergroup contact on social dominance orientation. Group Processes & Intergroup Relations, 17(1), 27-44.

14.

Duncan

T. E.

Duncan

S. C.

Strycker

L. A.

(2006). An introduction to latent variable growth curve modeling: Concepts, issues, and applications. New York, NY: Lawrence Erlbaum.

15.

Duriez

Soenens

(2006). Personality, identity styles and authoritarianism: An integrative study among late adolescents. European Journal of Personality, 20, 397-417. doi:10.1002/per.589

16.

Dyck

J. J.

Pearson-Merkowitz

(2014). To know you is not necessarily to love you: The Partisan mediators of intergroup contact. Political Behavior, 36, 553-580. doi:10.1007/s11109-013-9240-6

17.

Ekehammar

Akrami

Gylje

Zakrisson

(2004). What matters most to prejudice: Big five personality, social dominance orientation, or right-wing authoritarianism? European Journal of Personality, 18, 463-482. doi:10.1002/per.526

18.

Enders

C. K.

(2010). Applied missing data analysis. New York, NY: Guilford Press.

19.

Fischer

Hanke

Sibley

C. G.

(2012). Cultural and institutional determinants of social dominance orientation: A cross-cultural meta-analysis of 27 societies. Political Psychology, 33, 437-467.

20.

Guimond

Dambrun

Michinov

Duarte

(2003). Does social dominance generate prejudice? Integrating individual and contextual determinants of intergroup cognitions. Journal of Personality and Social Psychology, 84, 697-721. doi:10.1037/0022-3514.84.4.697

21.

A. K.

Sidanius

Cuddy

A. J.

Banaji

M. R.

(2013). Status boundary enforcement and the categorization of black-white biracials. Journal of Experimental Social Psychology, 49, 940-943.

22.

A. K.

Sidanius

Kteily

N. S.

Sheehy-Skeffington

Pratto

Henkel

. . . Stewart

(2015). The nature of social dominance orientation: Theorizing and measuring preferences for intergroup inequality using the new SDO-7 scale. Journal of Personality and Social Psychology, 109, 1003-1028. doi:10.1037/pspi0000033

23.

A. K.

Sidanius

Pratto

Levin

Thomsen

Kteily

N. S.

Sheehy-Skeffington

(2012). Social dominance orientation: Revisiting the structure and function of a variable predicting social and political attitudes. Personality and Social Psychology Bulletin, 38, 583-606. doi:10.1177/0146167211432765

24.

Hurlbert

Lombardi

(2009). Final collapse of the Neyman-Pearson decision theoretic framework and rise of the neoFisherian. Annales Zoologici Fennici, 46, 311-349.

25.

Jost

J. T.

Thompson

E. P.

(2000). Group-based dominance and opposition to equality as independent predictors of self-esteem, ethnocentrism, and social policy attitudes among African Americans and European Americans. Journal of Experimental Social Psychology, 36, 209-232. doi:10.1006/jesp.1999.1403

26.

Kandler

Bleidorn

Riemann

(2012). Left or right? Sources of political orientation: The roles of genetic factors, cultural transmission, assortative mating, and personality. Journal of Personality and Social Psychology, 102, 633-645.

27.

Kreindler

S. A.

(2005). A dual group processes model of individual differences in prejudice. Personality and Social Psychology Review, 9, 90-107. doi:10.1207/s15327957pspr0902_1

28.

Krosch

A. R.

Berntsen

Amodio

D. M.

Jost

J. T.

Van Bavel

J. J.

(2013). On the ideology of hypodescent: Political conservatism predicts categorization of racially ambiguous faces as Black. Journal of Experimental Social Psychology, 49, 1196-1203.

29.

Kteily

N. S.

A. K.

Sidanius

(2012). Hierarchy in the mind: The predictive power of social dominance orientation across social contexts and domains. Journal of Experimental Social Psychology, 48, 543-549.

30.

Kteily

N. S.

Sidanius

Levin

(2011). Social dominance orientation: Cause or “mere effect”? Journal of Experimental Social Psychology, 47, 208-214. doi:10.1016/j.jesp.2010.09.009

31.

Kugler

M. B.

Cooper

Nosek

B. A.

(2010). Group-based dominance and opposition to equality correspond to different psychological motives. Social Justice Research, 23, 117-155. doi:10.1007/s11211-010-0112-5

32.

Lee

Pratto

Johnson

B. T.

(2011). Intergroup consensus/disagreement in support of group-based hierarchy: An examination of socio-structural and psycho-cultural factors. Psychological Bulletin, 137, 1029-1064.

33.

Lee

Ashton

M. C.

Ogunfowora

Bourdage

J. S.

Shin

K.-H.

(2010). The personality bases of socio-political attitudes: The role of honesty–humility and openness to experience. Journal of Research in Personality, 44, 115-119. doi:10.1016/j.jrp.2009.08.007

34.

Lehmiller

J. J.

Schmitt

M. T.

(2007). Group domination and inequality in context: Evidence for the unstable meanings of social dominance and authoritarianism. European Journal of Social Psychology, 37, 704. doi:10.1002/ejsp.383

35.

Levin

(1996). A social psychological approach to understanding intergroup attitudes in the United States and Israel (Unpublished doctoral dissertation). University of California, Los Angeles, USA.

36.

McFarland

(2010). Authoritarianism, social dominance, and other roots of generalized prejudice. Political Psychology, 31, 453-477.

37.

Merelman

R. M.

(1972). The adolescence of political socialization. Sociology of Education, 45, 134-166.

38.

Mueller

R. O.

Hancock

G. R.

(2010). Structural equation modeling. In Hancock

G. R.

Mueller

R. O.

(Eds.), The reviewer’s guide to quantitative methods in the social sciences (pp. 371-384). New York, NY: Taylor & Francis.

39.

Muthén

L. K.

Muthén

B. O.

(2012). Mplus user’s guide (7th ed.). Los Angeles, CA: Author.

40.

Oud

J. H. L.

(2002). Continuous time modeling of the cross-lagged panel design. Kwantitatieve Methoden, 69, 1-26.

41.

Pratto

Sidanius

Levin

(2006). Social dominance theory and the dynamics of intergroup relations: Taking stock and looking forward. European Review of Social Psychology, 17, 271-320. doi:10.1080/10463280601055772

42.

Pratto

Sidanius

Stallworth

L. M.

Malle

B. F.

(1994). Social dominance orientation: A personality variable predicting social and political attitudes. Journal of Personality and Social Psychology, 67, 741-763. doi:10.1037/0022-3514.67.4.741

43.

Preacher

K. J.

Wichman

A. L.

MacCallum

R. C.

Briggs

N. E.

(2008). Latent growth curve modeling (Vol. 157). Thousand Oaks, CA: SAGE.

44.

Raabe

Beelmann

(2011). Development of ethnic, racial, and national prejudice in childhood and adolescence: A multinational meta-analysis of age differences. Child Development, 82, 1715-1737.

45.

Raudenbush

S. W.

Bryk

A. S.

(2002). Hierarchical linear models: Applications and data analysis methods (Vol. 1). Thousand Oaks, CA: SAGE.

46.

Reynolds

K. J.

Turner

J. C.

Haslam

A. S.

Ryan

M. K.

Bizumic

Subasic

(2007). Does personality explain in-group identification and discrimination? Evidence from the minimal group paradigm. British Journal of Social Psychology, 46, 517-539. doi:10.1348/014466606X153080

47.

Rogosa

(1980). A critique of cross-lagged correlation. Psychological Bulletin, 88, 245-258. doi:10.1037/0033-2909.88.2.245

48.

Satorra

Bentler

P. M.

(2001). A scaled difference chi-square test statistic for moment structure analysis. Psychometrika, 66, 507-514. doi:10.1007/BF02296192

49.

Schmitt

M. T.

Branscombe

N. R.

Kappen

D. M.

(2003). Attitudes toward group-based inequality: Social dominance or social identity? British Journal of Social Psychology, 42, 161-186. doi:10.1348/014466603322127166

50.

Sears

D. O.

(1975). Political socialization. Handbook of Political Science, 2, 93-153.

51.

Sibley

C. G.

Duckitt

(2008). Personality and prejudice: A meta-analysis and theoretical review. Personality and Social Psychology Review, 12, 248-279. doi:10.1177/1088868308319226

52.

Sibley

C. G.

Duckitt

(2010). The ideological legitimation of the status quo: Longitudinal tests of a social dominance model. Political Psychology, 31, 109-137. doi:10.1111/j.1467-9221.2009.00747.x

53.

Sibley

C. G.

Harding

J. F.

Perry

Asbrock

Duckitt

(2010). Personality and prejudice: Extension to the HEXACO personality model. European Journal of Personality, 24, 515-534. doi:10.1002/per.750

54.

Sibley

C. G.

Liu

J. H.

(2010). Social dominance orientation: Testing a global individual difference perspective. Political Psychology, 31, 175-207. doi:10.1111/j.1467-9221.2009.00748.x

55.

Sibley

C. G.

Wilson

M. S.

Duckitt

(2007a). Antecedents of men’s hostile and benevolent sexism: The dual roles of social dominance orientation and right-wing authoritarianism. Personality and Social Psychology Bulletin, 33, 160-172. doi:10.1177/0146167206294745

56.

Sibley

C. G.

Wilson

M. S.

Duckitt

(2007b). Effects of dangerous and competitive worldviews on right-wing authoritarianism and social dominance orientation over a five-month period. Political Psychology, 28, 357-371. doi:10.1111/j.1467-9221.2007.00572.x

57.

Sidanius

Cotterill

Sheehy-Skeffington

Kteily

N. S.

Carvacho

(in press). Social dominance theory: Explorations in the psychology of oppression. Cambridge Handbook of the Psychology of Prejudice.

58.

Sidanius

Kteily

N. S.

Sheehy-Skeffington

A. K.

Sibley

C. G.

Duriez

(2013). You’re inferior and not worth our concern: The interface between empathy and social dominance orientation. Journal of Personality, 81, 313-323. doi:10.1111/jopy.12008

59.

Sidanius

Levin

van Laar

Sears

D. O.

(2010). The diversity challenge: Social identity and intergroup relations on the college campus. New York, NY: The Russell Sage Foundation.

60.

Sidanius

Pratto

(1999). Social dominance: An intergroup theory of social hierarchy and oppression. Cambridge, UK: Cambridge University Press.

61.

Sidanius

Pratto

(2001). Social dominance: An intergroup theory of social hierarchy and oppression. Cambridge, UK: Cambridge University Press.

62.

Sidanius

Pratto

Martin

Stallworth

L. M.

(1991). Consensual racism and career track: Some implications of social dominance theory. Political Psychology, 12, 691-721.

63.

Sidanius

Pratto

Mitchell

(1994). In-group identification, social dominance orientation, and differential intergroup social allocation. The Journal of Social Psychology, 134, 151-167. doi:10.1080/00224545.1994.9711378

64.

Sidanius

Sinclair

Pratto

(2006). Social dominance orientation, gender, and increasing educational exposure. Journal of Applied Social Psychology, 36, 1640-1653. doi:10.1111/j.0021-9029.2006.00074.x

65.

Sinclair

Sidanius

Levin

(1998). The interface between ethnic and social system attachment: The differential effects of hierarchy-enhancing and hierarchy-attenuating environments. Journal of Social Issues, 54, 741-757. doi:10.1111/j.1540-4560.1998.tb01246.x

66.

Thomsen

Green

E. G. T.

A. K.

Levin

Van Laar

Sinclair

Sidanius

(2010). Wolves in sheep’s clothing: SDO asymmetrically predicts perceived ethnic victimization among white and Latino students across three years. Personality and Social Psychology Bulletin, 36, 225-238. doi:10.1177/0146167209348617

67.

Thomsen

Green

E. G. T.

Sidanius

(2008). We will hunt them down: How social dominance orientation and right-wing authoritarianism fuel ethnic persecution of immigrants in fundamentally different ways. Journal of Experimental Social Psychology, 44, 1455-1464. doi:10.1016/j.jesp.2008.06.011

68.

Torney-Purta

(2004). Adolescents’ political socialization in changing contexts: an international study in the spirit of Nevitt Sanford. Political Psychology, 25(3), 465-478. doi:10.1111/j.1467-9221.2004.00380.x

69.

Turner

J. C.

Reynolds

K. J.

(2003). Why social dominance theory has been falsified. British Journal of Social Psychology, 42, 199-206. doi:10.1348/014466603322127184

70.

Vandenberg

R. J.

Lance

C. E.

(2000). A review and synthesis of the measurement invariance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods, 3, 4-70. doi:10.1177/109442810031002

71.

Willett

J. B.

Sayer

A. G.

(1994). Using covariance structure analysis to detect correlates and predictors of individual change over time. Psychological Bulletin, 116, 363-381. doi:10.1037/0033-2909.116.2.363

72.

Wilson

M. S.

Liu

J. H.

(2003). Social dominance theory comes of age, and so must change: A reply to Sidanius & Pratto, and Turner & Reynolds. British Journal of Social Psychology, 42, 221-223. doi:10.1348/014466603322127210

73.

Wodtke

G. T.

(2012). The impact of education on intergroup attitudes a multiracial analysis. Social Psychology Quarterly, 75, 80-106. doi:10.1177/0190272511430234