Generalized Prejudice: Lessons about social power, ideological conflict, and levels of abstraction

ABSTRACT Prejudices tend to come as a package deal; people who are more racist tend to also be more sexist, and so on. Documentations of such attitudinal patterns – generalised prejudice – have a rich history, but the theoretical understanding thereof has been lagging. In recent years, we have seen clear theoretical advancement in the understanding of this phenomenon. We discuss classic premises in this research along with newer research that challenges some of the most ingrained ideas about generalised prejudice. For instance, we discuss research challenging the notions that generalised prejudice is an “us” versus “them” phenomenon, as well as a product of conservative ideology. We further argue that prejudice generalisations need to be studied at different levels of abstraction. Finally, we propose integrative perspectives on generalised and target-specific prejudice, with greater emphasis on processes of generalisation, rather than static generalised prejudice factors.


Introduction
People who are more prejudiced against ethnic minorities tend to also be more prejudiced against religious and sexual minorities, and they tend to denigrate women and people with disabilities, and so on (e.g., Ekehammar & Akrami, 2003;Zick et al., 2008). Different labels have been used for such observations, but the term generalised prejudice is perhaps the most intuitive and fitting for the observations at hand (see, Allport, 1954). As we use the term here, it is meant to be descriptive rather than endowed with explanatory assumptions in itself. It describes the empirical observation that prejudices come as a package deal.
Over 75 years, generalised prejudice has been found to be replicable in different cultures and with various measures (e.g., Asbrock et al., 2010;Bergh et al., 2012;Bierly, 1985;Cantal et al., 2015;Cohrs et al., 2012;Cunningham et al., 2004;Duckitt, 2001;Ekehammar & Akrami, 2003;Hartley, 1946;Kogan, 1961;McFarland & Crouch, 2002). For the most part, however, these demonstrations left major theoretical questions unanswered. For instance, what groups fit versus do not fit within the same attitude cluster? What is the common denominator among traditionally studied target groups? Are there multiple kinds of generalised prejudice? How is generalised prejudice related to political intolerance?
In this review, we address these types of questions. After setting the stage with a brief history of generalised prejudice research, we describe different levels of attitude generalisations across groups. This allows us to situate the concept of generalised prejudice in discussions about even broader attitude orientations (e.g., misanthropy), as well as narrower prejudice clusters (e.g., prejudice against different ethnic groups) and more particular associations between them (e.g., secondary transfer effects of heightened/reduced prejudice). Finally, we provide some reflections on emerging and future research (e.g., the mechanism behind connections between generalised prejudice and animal exploitation and network perspectives on generalised prejudice).

Defining generalised prejudice
For something to be generalised it must extend across a range of conditions. Therefore, the notion of generalised prejudice calls for a broad definition of prejudice that is not limited to just a handful of social groups. At the same time, for a concept to have discriminant validity, it cannot involve boundless generalisation and should not apply to any possible target (e.g., groups, individuals, and products). Scholars have incorporated both of these features into definitions of generalised prejudice. For example, generalised prejudice has been defined as a negative attitude about all or most outgroups (for a review, see, Bergh & Akrami, 2017). This involves a broad definition of prejudice (a negative group attitude) and a boundary of generalisation (the targets are outgroups). This definition has limitations, however, and so we adopt a different approach.
We define prejudice as a devaluing sentiment about a group (Bergh & Brandt) This definition encompasses both absolute negativity, as well as "less positive" evaluations of some group compared to another (e.g., sexism which involves evaluative ambivalence and positivity, but also devaluation of one group compared to another one; Glick & Fiske, 2001; see also, Dovidio & Gaertner, 2010). Generalised prejudice is then a devaluing sentiment that is directed towards multiple groups. Technically, "multiple groups" imply that we only need two separate targets to talk about a generalised (not fully targetspecific) sentiment. This would, however, be very narrow, and research has historically been interested in a greater extent of generalisation (e.g., Allport, 1954). In practice, we consider generalised prejudice to fall somewhere between "more than two groups" and "all groups". How to best delineate boundary conditions of generalised prejudice is a key focus of this review.

Historical ebbs and flows of generalised prejudice research
One of the first and most important demonstrations of generalised prejudice came from Hartley (1946). His participants rated their desired social distance from numerous ethnic, national, and religious groups, along with three fictitious groups (e.g., Pireneans). The group evaluations were strongly correlated, including the ones concerning fictitious targets. This suggested that there is an aspect of prejudice that had less do with the specific characteristics of the target and more to do with the characteristics of person expressing it. Based on such findings, Allport (1954) noted that "one of the facts of which we are most certain is that people who reject one out-group will tend to reject other outgroups. If a person is anti-Jewish, he is likely to be anti-Catholic, anti-Negro, anti any out-group" (p. 68). Adorno et al. (1950) similarly argued that individuals with an authoritarian personality have positive attitudes about their own groups and negative attitudes about all other groups.
The pioneering work on such a broad-spanning prejudice, ethnocentric ideology, and authoritarian personality came to face extensive methodological and theoretical criticism. For instance, poorly constructed measures of authoritarianism cast doubt on most of the correlational findings (e.g., Bass, 1955;Duckitt, 1992;Hyman & Sheatsley, 1954), including those showing commonalities between different prejudices (e.g., Fink, 1971). An unfortunate result of this criticism was that the generalised prejudice baby was thrown out with the authoritarian personality bathwater. That baby was long forgotten (see, e.g., Reis, 2010), but it turned out to be vital (Akrami et al., 2011;Hodson & Dhont, 2015). Using better measures, including peer reports and implicit measures, many studies later corroborated the notion that conceptually distinct prejudices (e.g., ageism and racism) are substantially correlated (for a review, see, Bergh & Akrami, 2017).
Work on generalised prejudice resurrected around the turn of the century, with work suggesting that there were multiple kinds of values and dispositions associated with prejudice (Altemeyer, 1998;Duckitt, 2001;S. McFarland, 2001). This updated work on right-wing authoritarianism (RWA) and social dominance orientation (SDO; Duckitt & Sibley, 2017;Sibley & Duckitt, 2008) made two key contributions to the literature on generalised prejudice. First, this work highlighted the role of personality, beyond ideological value orientations (e.g., authoritarianism) for understanding generalised prejudice. The idea is that personality traits associated with authoritarianism and social dominance ought to be also be related to generalised prejudice. Consistent with this idea, generalised prejudice has been linked to several personality variables, including low agreeableness, low openness to experience, low honesty-humility, and low empathic concern (e.g., Bäckström & Björklund, 2007;S. McFarland, 2010;Sibley & Duckitt, 2008;Sibley et al., 2010).
The second important lesson was that generalised prejudice is not a homogeneous attitude towards any kind of outgroup. Based on distinct motives associated with authoritarianism and social dominance, Duckitt and Sibley (2007) proposed that there are three sorts of generalised prejudice: against socially derogated, dangerous, and dissident groups, respectively (we describe these in more detail later). This work showed that prejudice generalises across many groups, but groups with similar target characteristics are more strongly related.
While these developments were generative, the wedding of research on generalised prejudice, authoritarianism, and social dominance has also been restrictive. In particular, it meant that generalised prejudice was operationalised and tested to specifically fit the mould of existing theories about people's personality and ideology. The research focuses on particular types of prejudice that RWA and SDO (and related variables) should explain. This is useful for testing perspectives related to these constructs, but it will not result in a full accounting of prejudices' generalisation. In the upcoming sections, we discuss newer research on generalised prejudice, divorced from much of its traditional theoretical cloak. The premise is to start with a better representation of the phenomenon to explain, that is, commonalities in prejudices against different groups. With better representations in hand, we can then compare different explanations.

Degrees of generalisations across targets
What are the constraints on the generalisability of prejudice across target groups? The original idea was that generalised prejudice represents a bias against most any outgroup (Adorno et al., 1950;Allport, 1954). This has been challenged from several angles.  demonstrated that generalised prejudice has little basis in ingroup-outgroup dynamics. Duckitt and Sibley (2007) argued that the classic notion is too broad and that we should study narrower factors instead, such as the derogated, dangerous, and dissident group factors. Moving in the opposite direction, Crawford and Brandt (2019) proposed that the classic notion is too narrow and that a broader negativity orientation can be extracted when more targets are included.
These perspectives seem at odds with each other. Yet, they can be reconciled if one considers that the generalisation of group attitudes can be studied at different levels of abstraction, with higher levels of abstraction being associated with increasingly broad generalisability across target groups. Schematically, we distinguish between four such levels (see, Figure 1): Attitudes concerning most any group (Level 1), devaluing attitudes about broad clusters of groups (Level 2), attitudes about smaller subsets of groups (Level 3), and attitudes towards a specific target group (Level 4). Level 2 has been the principal focus of generalised prejudice research. The upcoming sections situate research on generalised prejudice in relation to work at each level, starting from the top.

Level 1: Generalised prejudice and misanthropy
The highest level of abstraction focuses on generalised evaluations of all groups, or at least a very broad range thereof. The idea is that some individuals are negative towards any and all sorts of people. Crawford and Brandt (2019) referred to such a broader form of negativity as broad generalised prejudice. Here, we highlight that such negativity is conceptually different from generalised prejudice. Generalised prejudice research assumes some form of demarcation line between favoured and disfavoured groups (Allport, 1954), but one could also imagine that some individuals are negative towards any and all sorts of people -which we call misanthropy.
To examine attitudes that generalise to a wide variety of target groups, Crawford and Brandt (2019) analysed data from four samples (N = 7,543) where they assessed negativity towards a wide variety of target groups, Figure 1. Illustration of levels of abstraction and degree of attitude generalisations across groups. The emphasis on the second level indicates focus of (most) generalised prejudice research. t the third level, we provide examples of narrower facets of generalised prejudice, but they should not be interpreted as primary compared to other clusters of biases that are not included.
including groups stereotyped as high and low status (e.g., wealthy and poor people), progressive and conservative (e.g., feminists and pro-life proponents), and in between (e.g., middle-class people). From these ratings, indices were created for both a typical set of generalised prejudice targets (e.g., gay men, Muslims, and Black people) as well as misanthropy (negativity towards all groups). Individuals high on openness to experience (metaanalytic r = −.17) and agreeableness (meta-analytic r = −.25) both expressed less generalised prejudice using the typical selection of targets. However, they found that the misanthropic index was correlated with personality traits in different ways. In particular, openness to experience was unrelated to misanthropy (meta-analytic r = −.03). At the same time, high agreeableness was still associated with less misanthropy (meta-analytic r = −.23). Extraversion was also weakly associated with less misanthropy (meta-analytic r = −.09).
The personality correlations found for the narrower and more typical selection of target groups closely resemble those previously found for generalised prejudice (see, Sibley & Duckitt, 2008), whereas the correlations for misanthropy do not. The latter findings are instead in line with personality research in interpersonal contexts, where low agreeableness and low extraversion are associated with distrust and antisocial behaviours, for instance (e.g., Evans & Revelle, 2008;Mehl et al., 2006). Indeed, misanthropic individuals are expected to be negative towards most people, perhaps regardless if the evaluated target is a group or an individual.
The measure of misanthropy adopted by Crawford and Brandt (2019) was an average of negativity ratings towards all of the targets. This simple approach does not take into account the fact that some group attitudes are negatively correlated. When some targets are negatively correlated, the average rating does not represent negativity towards all groups. Instead, it represents a person's negativity towards the majority of targets minus his/her negativity towards inversely related targets. An alternative is to use bi-factor modelling to extract general negativity (misanthropy) net of more specific factors (e.g., DeMars, 2013; Rodriguez et al., 2016).
Using this approach, Bergh & Brandt modelled a general negativity factor (misanthropy) for a wide array of groups, independent of devaluing (prejudiced) sentiments. Groups that could be described as stereotypically neutral in terms of progressiveness or status (e.g., middle-class people and parents) loaded the highest on the general negativity factor (see , Table 1). In contrast, groups that are more extreme in terms of progressive-conservative values, or status, loaded quite modestly. Examining the correlates of this factor, across two samples (studies 2a and 2b in Bergh & Brandt), we found that misanthropy was more likely among people with lower levels of agreeableness (rs = −.28 and −.15), honesty-humility (rs = −.21 and −.12), extraversion (rs = −.30 and −.16), conscientiousness (rs = −.19 and −.11), openness to experience (rs = −.19 and −.21), and altruism (rs = −.32 and −.30). These results replicate the findings from Crawford and Brandt (2019) for extraversion and agreeableness. Yet, they also suggest that misanthropic attitudes have subpar discriminant validity (i.e. similarly correlated with most personality traits we examined), when operationalised as independent of prejudice. The generalised prejudice factors discussed at Level 2 have more distinct criteria correlations.

Level 2: Principal forms of generalised prejudice
At the second level of generalisation (see, Figure 1), the guiding principle is that people favour some groups over many other groups. The question then becomes, what groups are similarly disfavoured and how do these attitudes cluster? We propose that there are three major forms of generalised prejudice at this level, at least in the United States (see, Figure 1 and the heading "Three Principal Forms of Generalized Prejudice"). Before we get there, we discuss research laying the groundwork for such a model. It starts with the premise that stereotypic target attributes help explain commonalities in evaluative biases (e.g., Cuddy et al., 2007), more so than acknowledged in the classic generalised prejudice literature.

Generalised prejudice against dissident and derogated groups
Why are some groups disfavoured? The first answer comes from Duckitt and Sibley (2007). Using factor analysis, they distinguished three forms of prejudice in New Zealand, based on whether the target groups were socially derogated (e.g., people being unattractive or having a handicap), dangerous (e.g., terrorists and violent criminals), or dissident (e.g., protestors and feminists). The same structure has been found in other cultures (e.g., Brazil and Germany; Asbrock et al., 2010;Cantal et al., 2015). Prejudices towards dangerous, dissident, and derogated groups were also associated with different predictors. Individuals scoring high on RWA were prejudiced against groups that were perceived to violate the traditional order in society (dangerous and dissident groups) and those with high scores on SDO care more keeping certain groups in a subordinate position (derogated groups; Duckitt & Sibley, 2007). This initial answer to the question of why are some groups disfavoured is not complete, however. For example, Duckitt and Sibley's (2007) prejudice factors are not orthogonal. Prejudices against "dissident" and "derogated" groups are strongly correlated (Bergh, 2013;Cantal et al., 2015). Moreover, Hodson et al. (2017) showed that commonalities in evaluations of dissident and derogated groups are strongly associated with the common variance in RWA and SDO, suggesting that a common factor may explain why the prejudices are correlated. What might this common factor be?

Classic generalised prejudice = devaluation of marginalised groups (not outgroups)
Generalised prejudice, including prejudice towards derogated, dangerous, and dissident, are all associated with RWA and SDO. One way to identify what binds prejudices together is to consider commonalities between RWA and SDO (see also, Hodson et al., 2017). 1 One of the things that RWA and SDO have in common is a desire to leave existing group hierarchies intact (Bergh & Akrami, 2017). This desire could be associated with somewhat different motives: People who score high on RWA seem to defend hierarchies because they think that any disruption to the existing order is bad (e.g., Duckitt & Sibley, 2017), whereas those high on SDO simply value when some people have control over others (e.g., Ho et al., 2015;Sidanius & Pratto, 1999). Yet, regardless of the more distal reasons, in unequal social systems both RWA and SDO proximally motivate hierarchy preservation.
The vast majority of target groups in the generalised prejudice literature are groups that are socially marginalised and stigmatised. Based on such observations,  hypothesised that social marginalisation explains why prejudice generalises as it does, at least among commonly studied targets. They compared this marginalisation hypothesis with the theoretically orthogonal idea that people are "anti-outgroup," as proposed by Adorno et al. (1950) and Allport (1954). The anti-outgroup perspective has been endorsed, in slightly different forms, in practically every paper published on generalised prejudice over the last 20 years (see also, Bergh & Akrami, 2017).
Consider the case of prejudice against overweight people. It has no relation with people's own weight; thin and overweight people are just as likely to express anti-overweight biases (Crandall, 1994). At the same time, anti-overweight biases are strongly related to prejudice against ethnic, religious, and sexual minorities (Bergh et al., 2012Crandall, 1994). This raises the question: Do we even need to consider "us" versus "them" to understand generalised prejudice? A more nuanced answer is given in the section Three Principal Forms of Generalised Prejudice, but the short answer is that only a limited range of prejudices provide clear evidence of an ingroup-outgroup dynamic. Across a variety of targets, there is instead good evidence against a broad-spanning bias towards outgroups. Anecdotal evidence comes from observations that women, older adults, and overweight people all derogate their ingroups if they score high on other prejudices, such anti-gay biases and racism (Bergh et al., 2012;Crandall, 1994;Glick & Fiske, 1996).  provided a more systematic inquiry, repeatedly comparing the role of ingroup-outgroup dynamics with a general tendency to devalue marginalised groups. First, minimal group experiments were used to compare generalised prejudice to biases between arbitrary and randomly created groups. These "experimentally purified" ingroup and outgroup attitudes, stripped of social stereotypes about the targets, were largely unrelated to generalised prejudice (using similar targets as in previous research, such as racism and sexual prejudice). In fact, 13 of 17 correlations were negative (rs ranging from −.23 to +.11) and suggesting that those who score high on generalised prejudice are less inclined to ingroup favouritism, if anything. Furthermore, Big Five agreeableness and openness to experience accounted for 29% of variance in generalised prejudice (in line with previous research; e.g., Ekehammar & Akrami, 2003) but no more than 5% of the variance in pure outgroup negativity, or any indicator of ingroup or outgroup evaluations. Also noteworthy: there was one personality variable that predicted pure ingroup biases -empathic concern. Yet, empathically concerned people showed more ingroup biases (r = .28), which is the exact opposite of the finding for generalised prejudice (r = -. 32; see also, S. McFarland, 2010). From these studies, it would seem that ingroup biases stripped of confounds (e.g., cultural stereotypes about the targets) are quite different from traditionally studied generalised prejudice, despite theoretical arguments that these concepts are essentially interchangeable (e.g., Cunningham et al., 2004;Kinder & Kam, 2010).
Subsequent studies by  focused on substantiating the alternative hypothesis that some people are generally prejudiced against marginalised groups, which can also involve ingroup devaluation. Data from Sweden (N = 861 and 10,600) showed that sexism was associated with generalised prejudice against ethnic minorities and gay people, for instance, in similar ways among male and female respondents. In the larger study, based a representative sample of Swedish teenagers, multi-group confirmatory factor analysis revealed that the standardised loading of sexism on generalised prejudice was 0.55 among male respondents and 0.48 among female respondents. In other words, for understanding commonalties among the studied prejudices it does not seem to matter much whether biases against women represent an evaluation of an ingroup or outgroup. However, prejudice against women fit conceptually and empirically with the idea that prejudice generalises because it is directed at marginalised groups, irrespective of ingroup-outgroup dynamics.
A possible alternative explanation is that prejudice towards marginalised groups generalises because these groups are perceived to hold similar values and share a similar worldview (e.g., Brandt et al., 2015). Another study by , conducted among American Mechanical Turk workers (N = 419), aimed to rule this out. The study included high and low status groups that were also associated with progressive and conservative values (e.g., Greenpeace canvassers and corporate lawyers), but more importantly, another set of high and low status groups that are neutral in terms of political values (e.g., tall and short people). There is no reason to expect that shared worldviews would explain commonalities in evaluations of Greenpeace canvassers and short people, but there is from a status perspective. Results showed that all low status groups loaded significantly on a single factor (the mean standardised loading was 0.43 [ranging from 0.24 to 0.74]). Evaluations of high status groups were disjointed (the mean standardised loading was 0.15 [ranging from 0.08 to 0.71]).
In retrospect, however, it was premature to suggest that there is no coherent generalised prejudice against high status groups. Based on the work by , the targets were not varied or numerous enough for such an inference. Also, the findings do not rule out that there are other generalised prejudice factors to be found with a broader array of targets (e.g., groups characterised by political values). The studies were not designed to test that possibility; they merely aimed to show that low status can explain generalised prejudice independent of group membership or value connotations of the targets.

Worldview conflicts and new kinds of generalised prejudice
A core hypothesis in the work by Adorno et al. (1950) was that political conservatism would be related to generalised prejudice. People holding relatively conservative and right-wing values systems express more anti-Black, anti-gay, anti-atheist, and anti-immigrant prejudices (among others) compared to people holding relatively liberal and left-wing values systems (e.g., Crawford et al., 2016;Federico & Sidanius, 2002;Rowatt, 2019;Sidanius & Pratto, 1993). These differences emerge on both implicit (e.g., Greenwald et al., 2009) and explicit measures (e.g., Crawford et al., 2016), including extreme self-reports where people place social groups on scale ranging from animal to human (Kteily et al., 2015). Indices of generalised prejudice against such groups are also strongly associated with the right-wing ideological values, captured by RWA and SDO (RWA meta-analytic r = .49, SDO metaanalytic r = .55, Sibley & Duckitt, 2008; see also, Duckitt, 2001;Ekehammar et al., 2004;S. McFarland, 2010). The totality of the work in this area indicates that political conservatives express more prejudice, including more generalised prejudice, than political liberals.
Recent work has challenged the generality of this conclusion and suggests that there may be alternative dimensions of generalised prejudice. This work starts with the observation that the targets used in generalised prejudice research tend to focus on groups that are marginalised or groups that are emblematic of liberal causes. This focus makes sense given the societal and moral stakes of prejudice towards these groups; however, this focus can lead to premature conclusions about prejudice in general . To address this gap, a growing body of work has assessed how political ideology and values are associated with prejudice towards a larger array of groups ; see also, Koch et al., 2016). This includes marginalised groups, groups that are emblematic of liberal causes, high status and dominant groups, groups emblematic of conservative causes, and groups that are relatively "neutral" in terms of status or political associations (e.g., middle-class people and parents).
The work that takes this "many groups" approach typically finds that people express prejudice towards groups that are perceived to have different political values than themselves (for a review see, Brandt & Crawford, 2020). Said another way, conservatives tend to express prejudice against groups that are perceived as liberal and liberals express prejudice against groups that are perceived as conservative. Given well known processes of similarity-liking (Byrne, 1969), ingroup favouritism (e.g., Brewer, 1979;Tajfel & Turner, 1979), and the rejection of value-conflicting groups (Brandt & Crawford, 2020), this is not surprising.
The consistent demonstration that liberals also express prejudice, but towards different groups than those that conservatives target with their ire, has two major implications for work on generalised prejudice. First, it shows that prejudice is also directed towards conservative targets, suggesting that perceived ideology is a relevant dimension for understanding how prejudice is expressed towards different social groups. This dovetails with other work showing the importance of perceived ideology in social cognition (Koch et al., 2016). In turn, it also suggests that there may be other forms of generalised prejudice that target groups sharing values, beyond marginalised, socially derogated and dissident generalised prejudices (e.g., Duckitt & Sibley, 2007), The second implication has to do with the role of ideological values compared to status and power in explanations of generalised prejudice. Social status can explain patterns of generalised prejudice independent of conservative or progressive values (see previous heading; . However, progressive versus conservative values can also explain broadspanning group biases independent of status (Brandt, 2017). These pieces of evidence are not necessarily in contradiction. Finding a generalised prejudice factor based on status does not rule out other kinds of generalised prejudice. Similarly, finding that target status did not explain the strength and direction of biases of more or less conservative and progressive respondents (Brandt, 2017), does not rule out other effects of status or power for understanding prejudice. 2 In essence, status and ideological values could be complementary for explaining generalised prejudice.
Research on value or ideological conflicts have not formally examined how different target evaluations are associated (e.g., using factor analysis), but there are indications of clear commonalities among anti-conservative prejudices. One indication is that studies with alternative sets of targets generate similar findings (e.g., Brandt, 2017;Crawford et al., 2015). Another indication comes from the precision of ideology-based predictive models. Brandt (2017) showed that an ideology-based model of prejudice estimated in one dataset (N = 4,940) provides good precision for predicting the direction and magnitude of the association between ideology and prejudices in several new datasets (total N = 2,093), including new target groups the model had not encountered before. It would be hard to cross-validate the model if anti-conservative biases were not coherent, especially with a wide range of targets.

Three principal forms of prejudice
We have so far discussed separate perspectives on how and why different group biases go together and reveal patterns of generalised prejudice: A generic ingroup-outgroup perspective (e.g., Adorno et al., 1950;Cunningham et al., 2004), a tripartite perspective on prejudice against dangerous, dissident, and derogated groups (Duckitt & Sibley, 2007), a status and power perspective  and a worldview conflict perspective (e.g., Brandt, 2017;Brandt & Crawford, 2020). How do they all fit together?
We initiated a project to reconcile and integrate ideas from all of them (Bergh & Brandt), using a broader and more theory-agnostic set of targets for outlining principal forms of generalised prejudice. We were inspired by research on principal stereotype dimensions by Koch et al. (2016). They generated lists of groups using minimal prompts, like "name 40 social groups." Using the groups generated with this method, they ran seven studies (total N = 4,451) where participants sort the groups and rate them on a range of stereotype dimensions. Multi-dimensional scaling was subsequently used to uncover the spontaneous dimensions people use to categorise groups. They found that status (or "agency"/ "socio-economic success") and conservativeprogressive values are the primary and most spontaneously used stereotype dimensions (Koch et al., 2020, acknowledged that warmth is important too, but there is less consensus on what groups fit this stereotype; it appears to be more idiosyncratic). Based on Koch et al. (2016), we thought that prejudices might similarly cluster as a function of status and conservative-progressive political values. To start, we used exploratory methods to uncover specific delineations of different kinds of generalised prejudice (e.g., we were uncertain how prejudice against high status groups might relate to other forms of prejudice).
We (Bergh & Brandt) first re-analysed a dataset from Brandt (2017, Study 4) with feeling thermometer ratings 3 for 42 target groups (N = 432). These targets that should be reasonably representative of salient groups in the United States, based on a minimal listing instruction (adapted from Koch et al., 2016, Study 5a). This choice is important because one criticism of generalised prejudice research (e.g.,  is that target groups have been picked to fit certain theoretical assumptions (e.g., what authoritarianism should predict; e.g., Adorno et al., 1950;Duckitt & Sibley, 2007), rather than groups that participants' notice in their worlds.
Analytically, we initially conducted exploratory factor analysis and extracted four factors based on a parallel test. The first three factors were well defined and captured negativity towards (1) privileged and conservative groups (e.g., rich people, henceforth privileged/conservative), (2) negativity towards underrepresented and historically disadvantaged groups (e.g., poor people, henceforth marginalised 4 ), and (3) negativity towards groups defying traditional values (e.g., gay people, henceforth unconventional), respectively. The third factor also contained negative loadings for conservative groups (e.g., Republicans). This suggests that it has a bipolar meaning, such that individuals either like conservative groups and dislike unconventional groups, or vice versa. The fourth factor was completely based on double-loadings, and it was most strongly associated with stereotypically neutral and normative groups (e.g., middle-class people and White people). We suspected that the last factor would be more indicative of misanthropic attitudes (i.e. indiscriminate negativity) than prejudice (i.e. devaluing group attitudes).
Subsequent analyses used bi-factor modelling to estimate factors of "excess" negativity net of generalised negativity towards all groups. This operationalises our definition of prejudice as devaluing sentiments about groups (see also, Graziano et al., 2007). The reference point here is baseline, target-agnostic negativity towards most anyone. We pre-registered and provided confirmatory tests of such a structure across three datasets from the United States (total N = 1,296, for an illustration, see, Figure 2). We also conducted multigroup analyses to formally examine the robustness of the results across studies (see ,  Table 1), as well as across ethnic groups and liberals/conservative respondents. Overall, the analyses showed that our three principal types of generalised prejudice were robust across studies and groups of respondents.
One take-away is that status and progressive-traditional values are complementary for explaining the connections between specific prejudices. Some groups seem to face prejudice primarily on the basis of their social status (e.g., poor people), other groups on the basis of their values (e.g., atheists), and for some groups it can be a mix of the two (e.g., gay people; see, Table 1). 4 We use the term marginalised for groups that have lower status or power, or have been historically mistreated compared to other (typically majority) groups. This would include certain groups that are stereotypically considered "competent," or high in "socio-economic success" (e.g., Jews and Asians; Fiske et al., 2002;Koch et al., 2016), but nevertheless treated as subpar compared to the majority population, just like other disadvantaged groups.
We would also imagine that these evaluations could change over time or in particular contexts (e.g., the perceived progressiveness of Black people after Black Lives Matter protests). There are also other caveats to mention about our three kinds of generalised prejudice. First, we only tested this model in convenience samples (mostly on Amazon Mechanical Turk) in the United States. Thus, the generalisability to other contexts is unclear. Other research has documented similar patterns of prejudice and personality associations in different types of samples (including representative ones) and in different countries (e.g., Brandt et al., 2015;Cantal et al., 2015;Cohrs et al., 2012), so we think there will be some degree of generalisability. A second caveat is that these models leave considerable variability to explain. Most of our confirmatory models had reasonable fit according to heuristic criteria (e.g., CFI ≥ .94, SRMR ≤ .07), but they were not excellent. We also know that cross-loadings had relatively little impact on the model fit and we did not see evidence that the overall structure is wrong (e.g., involving the wrong Figure 2. Results from confirmatory factor analysis for three principal forms of generalised prejudice across Two Pre-registered studies. Note. Loadings are standardised. Groups are labelled here as they were shown to participants, following Koch et al. (2016). Within each factor, groups were randomly assigned to parcels. number of factors or unreliable characteristics). This suggests that much of misfit may be traced to many residual associations between specific prejudices (for details, see, Bergh & Brandt,Bergh and BrandtBergh & Brandt, Bergh and BrandtBergh & Brandt,Bergh & Brandt,Bergh and BrandtBergh & Brandt,Bergh & Brandt,Bergh & Brandt,Bergh & Brandt,). In sum, we found consistent evidence for our three broad generalised prejudice factors, but clearly they do not provide a full picture of prejudice. We discuss more specific, unaccounted variability in upcoming sections, related to Level 3 and 4 in Figure 1.
We also identified criteria associations of the three types of generalised prejudice (Bergh & Brandt,Bergh and BrandtBergh & Brandt,Bergh and BrandtBergh & Brandt,Bergh & Brandt,Bergh and BrandtBergh & Brandt, Bergh & Brandt,Bergh & Brandt,Bergh & Brandt,). In terms of personality, we found that openness to experience was negatively related to prejudice against unconventional and marginalised groups (rs ranging from −.24 to −.32), in line with previous generalised prejudice studies (e.g., Sibley & Duckitt, 2008). In line with a worldview conflict perspective (Brandt, 2017;Brandt et al., 2015), we also found that openness was associated with more prejudice against privileged/conservative groups (rs ranging from .19 to .29). For the privileged/conservative targets, we also found negative associations with extraversion (rs ranging from −.10 to −.29). Extraversion is typically not implicated in personality-prejudice research, but the associations are consistent with the notion that extroverted people are more well connected with high status people (e.g., Anderson et al., 2001).
Speaking to the role of ingroup-outgroup dynamics in generalised prejudice, we found no consistent associations between people's own socioeconomic identification and prejudice against marginalised groups (rs ranging from .00 to −.05). This is in line with findings from , suggesting that generalised prejudice against marginalised groups is not primarily about one's own group identification. The other prejudice factors were not predicted by socio-economic identification either (│rs │< .08, all non-significant), but they were predicted by political (progressiveconservative) identification (│rs │ranging from .38 to .72 [signs being negative for privileged/conservative groups]). Unsurprisingly, we found especially strong associations between such identifications and prejudice against unconventional groups (rs > .60).
In a final study, we also examined the relation between perceived threats and the three prejudice factors. These analyses did not concern particular outgroups threatening an ingroup (cf., Stephan et al., 1998), but rather perceived threats to important values in society and the economic welfare of most Americans. We asked about such threat perceptions for each target group in the study. Although the threat perceptions did not reduce to factors perfectly matching the prejudice factors, we could document strong associations (│rs│ranging from .43 to .78) between broad threat perceptions and all three forms of generalised prejudice. In other words, to understand why prejudices cluster as they do, perceptions of their threats to social values and economic welfare are important.
Taken together, this work provides a comprehensive model for integrating work from multiple models of generalised prejudice Brandt, 2017;Duckitt & Sibley, 2007). The model suggests that groups perceived to have conflicting political values are the ones that fit best with an "us" versus "them" conceptualisation of generalised prejudice (cf., Allport, 1954;see also, Cikara et al., 2017). At the same time, this is but one form of generalised prejudice and it is not the form that has been the focus of most research. Within the factor for marginalised targets, we find the cases where people are known to express prejudice towards their own group (e.g., women and Black people), 5 if they derogate the other groups in this cluster. Finally, the factor for prejudice against privileged/conservative groups is the least explored in previous research. Within this factor, we find groups that benefit the most from maintained status hierarchies between groups, but not necessarily groups stereotyped as agentic or competent (e.g., Jews and Asian people do not belong to this factor despite agentic stereotypes for these groups; cf., Cuddy et al., 2007;Koch et al., 2016). This pattern is consistent with previous studies on generalised prejudice (e.g., Zick et al., 2008), suggesting that a concern that more rights, power, or resources to any of these groups are viewed as a threat to the majority and/or participants themselves (Bergh & Akrami, 2017;Bergh & Brandt). Our theory-agnostic methods of group selection and factor estimation reveal findings that would have been unlikely based on studying generalised prejudice guided by theories and models about other concepts (e.g., cultural stereotypes, authoritarianism, or social dominance).

Level 3: Facets of generalised prejudice
Between the level of broadly generalised prejudice and target-specific biases, it is possible to identify an intermediate level of narrower generalised prejudice (Level 3 in Figure 1). Some biases are more closely related, above and beyond the principal types of generalised prejudice discussed at Level 2. These could be identified as sub-factors or residual correlations net of higher-order prejudice factors. For example, evaluations of sexual minorities seem to be related above and beyond their commonalities with more distinct prejudices, like racism and anti-overweight biases (e.g., Bergh et al., 2012). Technically, one could imagine several levels of abstraction in between 5 Figure 2 and Table 1 suggests that sexism is not particularly indicative of prejudice against marginalised groups. However, this is likely due to our use of feeling thermometers. These are unlikely to pick up on devaluing sentiments about women, net of overall warmth. Research with other measures clearly indicate that sexism belongs in this cluster (e.g., .
broadly generalised and target-specific prejudice. For instance, prejudice against bisexuals, gay men, and lesbians could be viewed as a cluster of heteronormative biases that are nested within a prejudice orientation concerning gender roles, which would also include sexism and prejudice against transgender people. For the remainder of this section, we discuss generalised ethnic prejudice and gender-role prejudice, as two illustrative examples of narrower Level 3 facets of generalised prejudice.

Generalised ethnic prejudice: ethnocentrism revisited
Overlapping evaluations of various ethnic groups is perhaps the most prominent illustration of generalised prejudice, and it guided much of the early theorising about the nature of generalised prejudice (Adorno et al., 1950;Allport, 1954;Hartley, 1946). Indeed, generalised prejudice has often been equated with ethnocentrism, by definition (e.g., Bratt, 2005;Cantal et al., 2015;Cunningham et al., 2004;Kinder & Kam, 2010;Krauss, 2002;Ray & Lovejoy, 1986). However, the terms should not be treated as synonymous. First, generalised prejudice is not confined to ethnic or cultural groups, as typically emphasised in work on ethnocentrism (Bizumic et al., 2009;Kinder & Kam, 2010). Ethnocentrism is a narrower concept that deals with a subset of groups studied in generalised prejudice research at Level 2 (e.g., only a handful of the targets in Table 1 are ethnic groups). Second of all, ethnocentric biases are by definition due to ingroup-outgroup categorisations, whereby ingroups are evaluated more favourable than all outgroups. In contrast, when we use the term generalised prejudice we are agnostic about what produces biases against many groups and why they are correlated. If some Black Americans are prejudiced against White Americans as well as both Mexican and Russian immigrants, then one could say that they are ethnocentric (and generally prejudiced). However, if they express prejudice against other Black people rather than White people (and there is reason to expect such a result; see, Ho et al., 2015), then they would still be generally prejudiced -but not ethnocentric. In sum, the term ethnocentrism is more specific than generalised prejudice, both in terms of the range of targets as well as propositions for how and why biases are correlated. Just as we argued that an "us" versus "them" mechanism could be misleading for explaining generalised prejudice at Level 2, there are similar challenges at this level. Based on observations that (some) ethno-cultural prejudices are correlated (generalised ethnic prejudice), it is not a warranted to conclude that people are ethnocentric (i.e. favouring their ingroup[s] over all outgroups).
The problem is that most ethnocentrism research focuses entirely on biases against ethnic minorities, or foreigners, among participants who belong to the ethnic majority, or the most powerful ethnic group in the society (for reviews of such work; see, Bizumic, 2019;Kinder & Kam, 2010). This leads to a confound between ingroup favouritism over all outgroups (the ethnocentrism hypothesis) and system justifying or hierarchy enhancing preferences (e.g., Jost & Banaji, 1994;Sidanius & Pratto, 1999). If outgroup negativity is most central, then ethnocentric people should derogate any ethnic outgroup compared to their own, irrespective of whether those outgroups are more or less socially valued (e.g., Bizumic et al., 2009;Kinder & Kam, 2010;LeVine & Campbell, 1972;Sumner, 1906). A good test then of the ethnocentrism hypothesis is to examine if those who derogate lower status ethnic groups similarly derogate groups that have higher status than their own (see also, . Brewer and Campbell (1976) conducted a classic study that allows a separation of outgroup negativity from negativity towards marginalised ethnic groups. They examined ingroup and outgroup evaluations of 30 ethnic groups in Kenya, Tanzania, and Uganda (total N = 1,500). The average respondent evaluated the ingroup more favourably than any outgroup. At the same time, groups with greater status and power were rated more favourably than other groups. These findings suggest that ingroup favouritism is widespread in this context (in line with ethnocentrism hypotheses), but outgroups are not devalued in a uniform manner (contrary to ethnocentrism hypotheses). More recently, Axt et al. (2014) showed a similar pattern of results in the United States. Members of different ethnic and religious groups rated their own group more favourable than any outgroup, on average, but the majority groups were also consistently rated more favourable than minority groups. From a generalised prejudice perspective, however, the most important question is not how ethnicities are rated on average, but rather whether it is the same individuals who favour their own group over both high and low status groups.  argued that people with supposedly ethnocentric personalities (e.g., scoring low on openness to experience and honestyhumility) are systematically biased against lower status ethnic groups only -and not against those above them in the hierarchy. Results from a large, representative sample in New Zealand (N = 4,037) support that idea. Among Māori participants, biases against other ethnic minorities were predicted by personality in a similar fashion as anti-minority biases among European New Zealanders. In both groups, people with similar personalities express prejudice towards people with a lower standing in the social hierarchy. In contrast, Māori biases against European New Zealanders (i.e. based on an upward social comparison) were not associated with such personality traits (see, Figure 3).
In sum, work on ethnocentrism has been revitalised in important ways (e.g., Bizumic, 2019;Kinder & Kam, 2010), but it provides an oversimplified and uncorroborated view on the generality of outgroup rejection (see also, Brewer, 1999). To understand ethnic biases under a variety of social conditions, new theory and data are needed. The research discussed in section 2 could provide a stepping stone in that regard. For example, ingroup favouritism at an individual level might only be systematic and generalised towards ethnic outgroups that are lower, but not higher, in status and power. It may also be more prevalent when ethnic groups are seen as holding conflicting socio-political values. These ideas call for further attention. Eagly and Diekman (2005) argued that prejudice can be understood as lowering or declining evaluations of members of group who are seen as occupying (or potentially occupying) roles that are incongruent with stereotypic role prescriptions. For example, women in leadership roles are devalued on the basis of violating the prescriptive stereotypes that leaders are male and that a women's role is to support them. Such stereotypic role violations help explain some patterns of Level 3 generalised prejudice, specifically towards groups that are perceived to occupy stereotypeinconsistent roles in similar ways.

Generalised prejudice based on gender roles
A good example is generalised prejudice against people who are not following stereotypic gender roles, such as feminists, gay men and lesbians, and transgender people. The perception that these groups do not adhere to stereotypic gender roles may be the "glue" that binds this generalisation together. This is consistent with the correlation between the endorsement of traditional gender roles and prejudice against gay men, lesbians, and transgender people (e.g., Hill & Willoughby, 2005;MacDonald et al., 1973;Tebbe & Moradi, 2012). This perspective also fits with the observation that negativity towards feminists (who also are seen to violate stereotypic gender roles) cluster with prejudice against gay people and other "dissident" groups, whereas negativity against housewives cluster with "derogated" groups (Duckitt & Sibley, 2007).
When we move from gender roles to broader concerns about traditional or progressive group behaviour, however, the focus naturally shifts towards the broader types of generalised prejudice at Level 2 in Figure 1. Notably, women face devaluation regardless of whether they endorse traditional roles (in the case of housewives) or not (feminists; Duckitt & Sibley, 2007). Thus, although role violations bind prejudices towards some groups together at Level 3, we do not think it represents a defining feature of the broadest forms of generalised prejudice at Level 2.

Level 4: Specific prejudices and implications for studying generalised prejudice
Prejudices towards specific groups are the most specific and least abstract level. These are the many specific prejudices that make up the broader forms at Levels 1, 2, and 3. Although prejudices cluster and share variance, there is unshared variance that represents prejudice emerging from the distinct social position, stereotypes, and historical experience of the specific group in the society. For example, although prejudice towards gay men and lesbian women are strongly correlated and part of a shared sexual prejudice facet of generalised prejudice, these groups face distinct prejudices due different intersecting gender identities (e.g., Herek & Capitanio, 1999). Next, we turn to discussing how theories about specific prejudice (Level 4) relate to generalised prejudice (Level 2).

Generalised prejudice and target-specific explanations for behaviour
The unique experiences and social position of specific groups have been used to elaborate theories of specific group prejudices (e.g., Cottrell & Neuberg, 2005;Mackie et al., 2000). One way to think about the relationship between specific prejudice and generalised prejudice starts from the premise that our perceptions of specific groups are not only based on that particular group, but how that group is perceived and embedded in the broader social system. These additional perceptions will include factors (e.g., status, value conflict) that are shared across multiple groups (i.e. multiple groups can have low status or have perceived value conflicts with the perceiver). Generalised prejudice captures these perceptions that are shared across groups, whereas target-specific prejudices capture perceptions that are idiosyncratic to the group. Both perceptions are relevant.
This way of thinking about the link between target specific and generalised prejudices can be thought of in terms of shared variance (generalised prejudice) and unshared variance (specific prejudices). In this way, target-specific and generalised prejudices can be viewed as statistically independent and thereby inviting potentially different explanations (Akrami et al., 2011;Bergh & Akrami, 2017;Meeusen et al., 2017;Meeusen & Kern, 2016). In particular, generalised prejudice could predict broad behavioural patterns at the same time as specific group prejudices predict specific group behaviours. This is consistent with findings about attitudes generally. General attitudes can predict broad behavioural patterns at the same time as specific evaluations predict specific behaviours (e.g., Ajzen & Fishbein, 2005).
Our expectation that generalised prejudice is informative about behaviours across group-domains is also shared by potential victims of unfair group treatment. Sanchez et al. (2017) showed that female participants expected to be treated poorly on the basis of their gender from someone who expresses racism. Likewise, members of ethnic minorities expected racist treatments from someone expressing sexism. In other words, people expect behavioural patterns from others that correspond perfectly to individuals acting out generalised prejudiced against marginalised groups.
In sum, target-specific and generalised forms of prejudice are both relevant for understanding group-based attitudes and behaviours. Importantly, there are also points of connection that can strengthen and integrate these disjointed approaches to prejudice. We discuss one example, secondary transfer effects (e.g., Pettigrew, 2009), which is also important for reevaluating assumptions about common causes for generalised prejudice (see Future Directions).

Secondary transfer effects and generalised prejudice
A commonly studied method to reduce outgroup prejudice is to increase positive contact between groups (Paluck et al., 2019;Pettigrew & Tropp, 2006). In the prototypical study, the positive contact focuses on a specific outgroup and shows that positive contact reduces prejudice with that specific outgroup (e.g., Mousa, 2020). A number of studies suggest that contact is not only associated with reduced prejudice against the primary target, but also other groups (Pettigrew & Tropp, 2006; for comprehensive reviews, see also, Lolliot et al., 2013;Vezzali et al., 2021). For example, in a large representative sample of Germans, Pettigrew (2009) showed that contact with immigrants was not only associated with more positive attitudes about immigrants, but indirectly, also with less bias against gay and homeless people. In other words, changes in prejudice generalise across groups, or transfer to secondary targets. The process of change, though, is different from what is typically assumed for generalised prejudice. Generalised prejudice is often believed to stem from personality and ideological values, so when it changes it is expected to be slow and happen in a top-down, or common cause fashion: A whole cluster of prejudices move together, like a slow-marching band that follow their director (the directing force here is personality/ideological orientation; e.g., Asbrock et al., 2010;Bratt et al., 2016;Osborne et al., 2020). In contrast, secondary-transfer effects reflect a bottom-up process: It starts with changes in prejudice towards one specific and focal group (e.g., immigrants) that spill-over to changes in other prejudices.
As noted by Pettigrew (2009), it can be difficult to distinguish which perspective better accounts for changes in prejudice, and they are not necessarily mutually exclusive (Vezzali et al., 2021). Yet, some longitudinal and experimental data document specific secondary changes that are consistent with transfer effects, but difficult to account for from a traditional generalised prejudice perspective. For example, Van Laar et al. (2005) found that college students assigned to live with Black or Hispanic roommates became less prejudiced towards both groups over time, but students with an Asian roommate became more prejudiced towards the other two ethnic minorities. Studies like these suggest that prejudice changes are not homogeneous across targets (see also, Harwood et al., 2011), as assumed from a traditional generalised prejudice perspective (e.g., Adorno et al., 1950). In fact, many, repeated spill-over effects could be sufficient to explain the emergence of generalised in the first place. We discuss this issue next.

Common cause and network perspectives on generalised prejudice
The concept of generalised prejudice resembles the notion of general intelligence. For a long time, substantial positive correlations observed among cognitive tests were argued to reflect a common cause within the individuala psychological entity of generalised intelligence or "g" (e.g., Gottfredson, 1998;Jensen, 1998). Likewise, it has long been argued that prejudices correlate because they have a common cause, and Adorno et al. (1950) explicitly used the analogy of positively correlated intelligence tests in their writing. Today, researchers recognise multiple forms of generalised prejudice (e.g., Bergh & Brandt Duckitt & Sibley, 2017). Yet, the basic logic of a common cause remains the same: Clusters of prejudices "hang together," and change together, because they have the same cause. Some researchers have made these causal assumptions explicit in longitudinal studies Bratt et al., 2016;Osborne et al., 2020). They are not alone, however, in promoting a common cause argument. Correlated prejudices have been argued to "reflect", "derive from", or be "rooted in" an "underlying tendency" or "disposition" (see, e.g., Bergh et al., 2012;Cunningham et al., 2004;Ekehammar & Akrami, 2003;Hodson et al., 2017;S. McFarland, 2010;Zick et al., 2008).
The challenge with these inferences is that the correlational patterns between prejudices do not have to reflect a common cause; they can also arise from other processes. One possibility is that changes in one prejudice spill over to changes in other, similar prejudices, in line with the notion of secondary transfer effects in the contact literature (e.g., Pettigrew, 2009). This is not to say that outgroup contact is ultimately responsible for all kinds of generalised prejudice (see previous sections for limitations of an ingroupoutgroup psychology singularly explaining generalised prejudice). The point is rather that irrespective of the trigger (e.g., contact or hierarchy changes among some groups), transfer effects could explain why group attitudes towards many groups are correlated. The existing longitudinal studies focused on aggregate changes in generalised prejudice (e.g., Osborne et al., 2020), do not rule out a more precise causal dynamic whereby prejudices change one after the other. The latter possibility is consistent with a network perspective on psychological constructs.
Network perspectives have a long history in other fields, for instance, to map relationships ("edges" or connections) between people ("nodes") in sociology (e.g., Cook & Whitmeyer, 1992). More recently, network perspectives have been applied to psychological concepts, such as psychopathology, intelligence, personality, attitudes, and ideological beliefs (Borsboom & Cramer, 2013;Brandt & Sleegers, 2021;Cramer et al., 2012;Dalege et al., 2017;Van Der Maas et al., 2006). A key prediction in network models is that changes in one node can have domino effects that over time produce the joint characteristic of the psychological construct (e.g., insomnia → fatigue →concentration problems → depressed mood →feelings of self-reproach = depression; Borsboom & Cramer, 2013;Cramer et al., 2016). Compared to a common cause model, this also represents an alternative account for the correlational pattern of generalised prejudice.
From a network perspective on generalised prejudice, each node represents prejudice towards a specific group. Some prejudices are directly connected, others are indirectly connected. Connections can vary in strength and sign. For instance, prejudices against Muslims and Jews are positively associated, beyond their commonalities with other prejudices . In contrast, prejudice against Christians and atheists is likely to have a direct inverse connection, as indicated by such residual correlations in our factor analytic work (Bergh & Brandt) Strongly associated prejudices make up clusters in the network, and these are equivalent to distinct factors in factor analysis (see, Golino & Epskamp, 2017). Our three principal prejudice factors, as discussed earlier, suggest that privileged, marginalised, and unconventional groups should still be evident in a network structure.
The way that specific prejudices might influence each other and form clusters provides another way to think about how Level 4 specific prejudices and Level 2 generalised prejudices (see, Figure 1) are related. A network model suggests that Level 4 specific prejudices are more strongly correlated with each other, which form clusters in the network. Sufficiently large clusters are what we think of as Level 2 generalised prejudices, whereas smaller clusters of more closely aligned groups might represent the more specific Level 3 generalised prejudices.
Connections between specific prejudices provide a framework for bringing together secondary transfer effects and generalised prejudice into a unifying framework for understanding cross-target prejudice changes. More specifically, secondary-transfer effects are consistent with spill-over changes between directly and strongly connected prejudices. For instance, Black and Hispanic people are proximate in multidimensional mapping of stereotypes (Koch et al., 2016), so it is reasonable to also expect prejudices against these groups to be strongly connected and thereby more likely to change after one and other (see, Van Laar et al., 2005).
Beyond such anecdotic compatibility with a network perspective, we consider it an important topic for future research. As theoretical guidance, a good starting point here would be the framework by Brandt and Sleegers (2021), who discussed socio-political beliefs and values from a network perspective. They simulated an exogenous treatment on one node (akin to intergroup contact) and documented spill-over effects (akin to secondary transfer) within such a system. The results showed that a prolonged treatment led to spill-over effects, analogous to having a college roommate of a different ethnicity (Van Laar et al., 2005).
Finally, it should be noted that common cause and network perspectives are not mutually exclusive, and they can be integrated . It is possible that a cluster of associated prejudices may change both jointly, for instance, as a function of broader personality changes, and one by one following intergroup contact and other specific changes. This would suggest that changes might go both from a higher to lower level of abstraction ("topdown": Level 2 → Level 3 → Level 4 in Figure 1) as well as from a lower to a higher level ("bottom-up": Level 4 →Level 3 →Level 2). Both alternatives should be explored empirically.

Generalised prejudice and animal exploitation
Another emerging extension of generalised prejudice research comes from work on human exploitation of other animals. Several studies indicate that prejudice against ethnic minorities is associated with exploitation of animals K. Costello & Hodson, 2010;Dhont & Hodson, 2014;Dhont et al., 2016;Everett et al., 2019;Kimberly Costello & Hodson, 2014). Dhont et al. (2016) further argued that "human outgroup prejudices (such as racial and ethnic prejudice) and speciesism share common ideological motives, including the desire for group-based dominance and inequality, indicated by SDO" (p. 508). Indeed, they found that SDO accounts for the common variance in speciesism and ethnic prejudice. However, the precise mechanism is unclear because desires for group-based dominance and inequality are not the same.
Early definitions of SDO focused more on group-based dominance, defined as a person's "desires that one's in-group dominate and be superior to outgroups" (Pratto et al., 1994, p. 742). Later, Pratto et al., 2006 underscored both desires for "group-based dominance and inequality" (p. 281; emphasis added). Most recently, definitions of SDO have shifted to a principled support for group inequality, irrespective of one's own group membership (i.e. group-based dominance is de-emphasised; see, Ho et al., 2015Ho et al., , 2012. The latter definition aligns with observations that SDO is associated with ingroup favouritism in high power groups, while being associated with outgroup favouritism and ingroup derogation in disadvantaged groups (Ho et al., 2015;Levin et al., 2012). This raises an important question about the connection between human prejudice and speciesism: Do people exploit animals because animals are the "others" in relation to "us" humans? Or is it because people who exploit animals tend to like all sorts of hierarchies and do not care about the groups and animals that become exploited in those arrangements?
These explanations are typically confounded. Humans occupy the top of the animal hierarchy and most participants belong to groups in the same position (e.g., most participants are White; Dhont et al., 2016). In future work, it is relevant to disentangle these explanations because they make different predictions about other attitudes and behaviours. For instance, the ingroup-outgroup explanation suggests that Black Americans who exploit animals will also be prejudiced against White Americans. In contrast, principal hierarchy preservation suggests that animal exploitation should be related to ingroup derogation among Black Americans (for analogous tests for social groups, see, . Several levels of abstraction could be necessary to understand attitudes about animals, similar to the levels we outline for generalised prejudice against humans (Figure 1). For instance, we could imagine that some attitudes pertain to all or most animals, whereas other concern exploitation of certain types of animals (e.g., farm animals) but not others (pets), as well as more specific attitudes for narrower clusters of species or breeds (see, Podberscek, 2009;Serpell, 2009). From this perspective, we would expect the strongest connections between human prejudice and animal exploitation when the levels of abstraction are matched, in line with correspondence principle in attitude research (e.g., Ajzen & Fishbein, 2005). The clustering of prejudices at Level 2 might also translate to different reasons for exploiting animals. For instance, many animals might be exploited because people do not consider them "worthy" enough (perhaps especially plausible for exploitation linked to SDO; Dhont et al., 2016), but other exploitation might be better explained by traditional norms in different societies (e.g., views on Whaling in Japan and the United States, independent of dominance orientations).
These reflections aside, the broader point in this literature is intriguing, namely that there are important parallels between (some) prejudice and exploitation of (some) animals Dhont et al., 2016;Everett et al., 2019). The relations between group derogation and animal exploitation open for new discussion about the boundaries and nature of generalised prejudice. It suggests that some prejudices may generalise across species.

Concluding Remarks
The notion of generalised prejudice was foundational in two of the most influential books ever written about prejudice (see, Adorno et al., 1950;Allport, 1954). After those publications, it took fifty years before we started seeing novel approaches to better understand these attitudinal patterns, especially why different individuals are generally prejudiced against different clusters of targets (Altemeyer, 1998;Duckitt, 2001;Sidanius & Pratto, 1999). At present, we believe that time has come to provide another fundamental revision of what we think we know about generalised prejudice. This includes a recognition that generalised prejudice is not necessarily about "us" versus "them" , that prejudice exists in different forms on the political left and right and focuses on different clusters of targets (Brandt & Crawford, 2020), that generalisations should be considered at different levels of abstraction (Bergh & Akrami, 2017;, and finally, that more attention should be dedicated to the relation between target-specific and generalised forms of prejudice (Akrami et al., 2011;Meeusen et al., 2017;Meeusen & Kern, 2016).

Disclosure statement
No potential conflict of interest was reported by the author(s).