The Association of Cognitive Ability with Right–Wing Ideological Attitudes and Prejudice: A Meta–Analytic Review

Right–wing ideological attitudes and prejudice: The role of cognition

Prejudice largely involves negative evaluations of and beliefs about out–groups, such as other ethnic groups, women, homosexuals and the elderly. Two broad research traditions have historically sought to unveil the psychological basis of prejudice. Whereas some focused on the impact of situational and contextual factors, such as (intergroup) threat (e.g. Riek et al., 2006), others investigated the role of individual differences and personality predispositions (Adorno et al., 1950; Allport, 1954). In their well–known book The Authoritarian Personality, Adorno et al. (1950) argued that authoritarian personality characteristics lead people to adhere to extreme right–wing parties and prejudice. Since then, other relevant attitudinal individual differences have been related to right–wing ideology and prejudice, including tough–mindedness (Eysenck, 1954), dogmatism (Rokeach, 1960) and conservatism (Wilson, 1973). These attitudes can all be divided under the category of ‘right–wing socio–cultural attitudes’, broadly referring to adherence to traditional values and norms and resistance to change (Altemeyer, 1998; Duckitt & Sibley, 2009; Jost et al., 2003). Right–wing authoritarianism (RWA; Altemeyer 1981, 1998) is a typical indicator of this pattern of broad social beliefs (Duckitt & Sibley, 2009). The three RWA facets—uncritically submitting to authorities, adhering to societal norms and traditions and showing aggressiveness towards individuals who deviate from these conventional norms and values—all tap into the social–cultural domain. Most conservatism scales and dogmatism also refer to the social–cultural domain.

We focus on one prominent research question that has elicited interest among scholars for a long time. Specifically, do individuals with stronger endorsement of right–wing and prejudiced attitudes score higher or lower on specific cognitive characteristics, differentiating them from individuals with rather left–wing and less prejudiced attitudes? In the social psychological literature, there is a long and widely accepted tradition linking limited cognitive resources with intergroup biases and ideological attitudes. For instance, the well–known study by Gilbert and Hixon (1991), a staple in most prejudice textbooks, demonstrated that when cognitively busy (e.g. rehearsing a digit sequence or performing a visual search task while performing the central task), participants are less likely to activate stereotypes but are significantly more likely to apply (i.e. ‘use’) stereotypes that are activated (or salient). Others have observed that people under high (versus low) cognitive load are more likely to recall stereotypic traits (e.g. Macrae, Hewstone & Griffiths, 1993; Pratto & Bargh, 1991). With regard to ideological attitudes, Eidelman, Crandall, Goodman and Blanchar (2012) have demonstrated that inducing low–effort thinking (e.g. alcohol consumption, cognitive load and time pressure) results in stronger endorsement of politically conservative ideologies, such that ‘conservatism may be a process consequence of low–effort thought’ (Eidelman et al., 2012, p. 808). Underlying each of these experimental methodologies is the assumption that lower availability of cognitive resources elicits more social conservatism and prejudice–relevant thinking. This basic premise, when framed experimentally within a social psychology context, courts little controversy. Yet, when exported to the realm of individual differences, such that those lower in cognitive abilities are purportedly higher in social conservatism and prejudicial attitudes, this idea has less traction. However, the underlying principles are very closely related if not analogous—to the extent that lesser mental ability is related to these outcomes; this should be true whether the ability limitation is based on an individual difference or is experimentally induced (see E. T. Higgins, 2000, for a discussion of ‘general principles’ across person and situation effects).

Indeed, some scholars have argued that individuals with lower cognitive skills are relatively ill–equipped to process complex and new social information and to understand constantly changing societal contexts. Therefore, they are more likely to stick to what is presently known and considered acceptable, rather than being open–minded and appreciating multidimensional perspectives (Deary et al., 2008; Heaven, Ciarrochi & Leeson, 2011; McCourt, Bouchard, Lykken, Tellegen & Keyes, 1999; Stankov, 2009). By emphasizing societal traditions, the preservation of the status quo and strict group boundaries, ideologies endorsing resistance to social change, that is, right–wing ideologies (Jost et al., 2003), should be particularly appealing to those with lower cognitive abilities (e.g. Heaven et al., 2011; Keiller, 2010; Stankov, 2009). According to this theoretical perspective, therefore, right–wing ideologies provide well–structured and ordered views about society and intergroup relations, thereby psychologically minimizing the complexity of the social world. Theoretically, therefore, those with fewer cognitive resources drift towards right–wing conservative ideologies in an attempt to increase psychological control over their context. Furthermore, studies have shown that being cognitively restricted when engaging in complex mental processing and high–effort thinking (Eidelman et al., 2012) facilitates reliance on simple heuristics in social judgments, which in turn lead to quick and biased views about other groups (Heaven et al., 2011; Hodson & Busseri, 2012; Keiller, 2010). Although initially under–investigated, cognitive ability has gradually gained importance in the field of ideological attitudes and prejudice (e.g., Deary et al., 2008; Dhont & Hodson, 2014; Hodson, 2014; Van Hiel et al., 2010). In the next sections, we provide an overview of the empirical studies investigating associations of cognitive ability with both right–wing ideological attitudes and prejudice.

Cognitive ability and right–wing ideological attitudes

Adorno et al. (1950) already investigated the relationship between cognitive ability and authoritarianism, and many studies followed this first inquiry, reporting negative relations between cognitive ability and authoritarianism (e.g. Adorno et al., 1950; Christie, 1954; Davids & Eriksen, 1957; Jacobson & Rettig, 1959; Siegel, 1956). Similar results were obtained with the Dogmatism Scale (Long & Ziller, 1965; Thompson & Michel, 1972; Uhes & Shaver, 1970) and, more recently, for a range of measures tapping right–wing ideological attitudes (e.g. Bouchard, Segal, Tellegen, McGue, Keyes & Krueger, 2003; Kanazawa, 2010; Keiller, 2010; McCourt et al., 1999). Further evidence for this association was obtained on the cross–national level by Stankov (2009), who found that inhabitants of conservative countries typically show lower average performance across different ability tests. A meta–analysis (Van Hiel et al., 2010), including data from 50 studies, reported an average correlation of r = −.26 between cognitive ability and right–wing ideological attitudes, including conservatism, authoritarianism and dogmatism. Moreover, several longitudinal studies, some using large representative samples, provided evidence for the direction of the relation, demonstrating that lower general cognitive ability in childhood relates to stronger endorsement of conservative and authoritarian attitudes later in life (Block & Block, 2006; Deary et al., 2008; Fraley, et al., 2012; Heaven et al., 2011; Schoon, Cheng, Gale, Batty & Deary, 2010).

Remarkably, despite the use of various cognitive ability measures across studies, scholars paid little attention to the potential differential impact of the various types of cognitive ability. However, a few studies simultaneously contrasted various types of cognitive abilities. For instance, Heaven et al. (2011) reported that among 12– to 13–year–olds, lower verbal abilities predict RWA five years later in time, whereas numerical abilities did not show such longitudinal effect. Kemmelmeier (2008) observed weak significant negative associations of verbal ability with two of three indicators for conservative attitudes and a positive association with a third indicator (i.e. anti–regulation attitudes), yet numerical ability did not yield any significant association. Choma, Hodson, Hoffarth, Charlesford and Hafer (2014) similarly found numerical reasoning to be unrelated to RWA. Deary et al. (2008) reported negative relationships of comparable strength between conservatism (i.e. items tapping into political distrust, social conservatism and anti–working women) on the one hand and two verbal ability tasks (mean r across sexes = −.18 and −.17), a reasoning task (mean r = −.15) and a short–time memory task (mean r = −.10) on the other hand. In sum, the results of these few studies yielded rather mixed findings with respect to potential differences between cognitive ability types.

Cognitive ability and prejudice

In the early studies, ethnocentrism, a form of prejudice towards ethnic out–groups in general, was particularly studied. Adorno et al. (1950) devoted an entire chapter on the relationship between cognitive ability and ethnocentrism, which was found to be negative. This initial finding was replicated in other studies across diverse samples and with diverse indicators of cognitive ability, such as reasoning tests (e.g. Eysenck, 1954; Kutner & Gordon, 1964; O'Connor, 1952), verbal ability tests (e.g. Egan, 1989) and general cognitive ability tests (Rokeach, 1951). More recently, Meeusen, de Vroome and Hooghe (2013) also revealed negative correlations between measures of verbal and mathematical abilities and ethnocentrism.

Other studies recently focused on prejudice expressed towards specific out–groups. Keiller (2010) reported that abstract reasoning is negatively related to prejudice against homosexuals. Costello and Hodson (2014) found that White children who are more able to comprehend hierarchical relations among objects (e.g. that cars and trucks belong to the superordinate category vehicles) express less racial bias towards Black children. Furthermore, longitudinal studies provided evidence for the effect of cognitive ability early in life on subsequent levels of prejudice. Analysing large, representative datasets, Deary et al. (2008), Schoon et al. (2010) and Hodson and Busseri (2012) demonstrated that cognitive ability in childhood leads not only to increased right–wing ideological attitudes but also to increased racism later in adulthood. A recent literature review on this topic by Dhont and Hodson (2014) concluded that ‘the field will benefit from a recognition of, and open discussion about, differences in cognitive abilities between those lower versus higher in prejudice’ (p. 454).

Similar to the literature on cognitive ability and right–wing ideological attitudes, little attention has been devoted to the possible differential impact of distinct types of cognitive ability. Only one study (Deary et al., 2008), using a large and representative dataset, reported associations of various strengths between racism and two verbal ability tasks (mean rs across sexes = −.23 and −.19), a reasoning task (mean r = −.15) and a short–time memory task (mean r = −.12).

The present study

The present study is a meta–analytic integration of empirical research on the relationship of cognitive ability with both right–wing social–cultural ideological attitudes and intergroup prejudice. The present meta–analyses extend the meta–analysis on cognitive ability by Van Hiel et al. (2010) in three important ways. First, since Van Hiel et al. collected samples for their meta–analysis in early 2009, several important studies, some with large samples or longitudinal data, have been published (e.g. Heaven et al., 2011; Hodson & Busseri, 2012; Kanazawa, 2010; Keiller, 2010; Schoon et al., 2010; Stankov, 2009; Xu, Mar & Peterson, 2013), which we included in the present analysis. Based on this literature and the meta–analysis of Van Hiel et al. (2010), we expected that lower cognitive abilities predict more right–wing ideological attitudes and greater prejudice. Second, whereas Van Hiel et al. only examined relationships with ethnocentrism as an indicator of prejudice, we also included studies that administered other indicators of prejudice (e.g. Costello & Hodson, 2014; Keiller, 2010).

Third, the present meta–analysis aimed to investigate the potential differential effects of specific types of cognitive ability, which was not examined by Van Hiel and colleagues. Research to date has typically used very diverse types and measures of cognitive ability, but little attention has been directed towards the question of whether particular types of cognitive ability might be especially relevant in ideology and prejudice. In other words, is the association between cognitive ability and right–wing ideological attitudes and prejudice consistent across all types of cognitive ability, or are these relationships largely driven by particular types of cognitive abilities? To investigate this, we divided all samples included in our meta–analysis into categories according to the measure of cognitive ability employed. The cognitive ability measures were classified according to the CHC theory of cognitive abilities (McGrew, 2005; Schneider & McGrew, 2012), considered one of the most empirically well–supported and widely accepted comprehensive theoretical frameworks of the structure of cognitive abilities (e.g. Flanagan, Genshaft & Harrison, 2012). Specifically, the CHC model proposed three levels: g is the highest level, representing general cognitive ability. Underlying g are nine primary broad domains who each contribute to the higher–order g–factor. These nine primary domains are: fluid reasoning (Gf), the broad ability to reason and solve novel problems; comprehension–knowledge ability (Gc), static abilities based on one's previously acquired knowledge; short–term memory (Gsm), the ability to encode, maintain and manipulate information in the immediate situation; long–term storage and retrieval (Glr), the ability to store and retrieve information in long–term memory; visual–spatial processing (Gv), the ability to perceive, generate, store and retrieve visual and spatial information; auditory processing (Ga), abilities involved in detecting and interpreting sounds; cognitive processing speed (Gs), the ability to quickly and fluently perform relatively simple cognitive tasks; reading and writing (Grw), individual's depth and breadth of reading and writing knowledge and skills; and finally, quantitative knowledge (Gq), the individual's depth and breadth of quantitative or mathematical knowledge and skills.

In sum, the present meta–analysis extends the meta–analysis of Van Hiel et al. (2010) by the following actions: (i) including new and recent studies for the relationship between cognitive ability and right–wing ideological attitudes; (ii) including various measures of prejudice and (iii) investigating whether the specific type of cognitive ability influences the strength of the relationship. Because it is widely acknowledged that right–wing ideological attitudes and prejudice are conceptually distinct (e.g. Altemeyer, 1981; Duckitt, 2001), we perform two separate meta–analyses, one for the relationship between cognitive ability and right–wing ideological attitudes and one for the relationship between cognitive ability and prejudice (including both measures of generalized prejudice and prejudice towards specific groups).

Selection of studies: Search strategies and inclusion criteria

Studies for this meta–analysis were selected by using a variety of search strategies. First, we identified the relevant studies included in Van Hiel et al.'s (2010) meta–analysis, excluding the studies focusing on education rather than pure cognitive abilities. Second, we searched the databases ISI Web of Knowledge and Google Scholar for studies published until November 2014. We used a variety of keywords in various combinations. Keywords for right–wing ideological attitudes were authoritarianism, conservatism and dogmatism. Keywords for intergroup prejudice were racism, sexism, ethnocentrism, prejudice, ethnic prejudice, racial prejudice and intolerance. Keywords for cognitive ability were cognitive ability, mental ability, reasoning, intelligence and IQ. Third, we checked the references list of each relevant article for additional studies of interest. Finally, we contacted researchers in the field to share relevant unpublished data. We distributed our call for unpublished data via the websites or mailing lists of the International Society of Political Psychology, Society for Personality and Social Psychology, European Association of Social Psychology and the Social Psychology Network. Three researchers contacted us and shared their unpublished data.

Studies had to meet several criteria to be included in the meta–analysis. First, studies had to administer at least one measure of right–wing ideological attitudes or intergroup prejudice and at least one measure of cognitive ability. Furthermore, no sample overlap was allowed because samples included in a meta–analysis have to be statistically independent (Card, 2012; Mullen, 1989). When studies included multiple types of ideological or out–group attitudes or multiple types of cognitive ability in the same sample, we selected the type least prevalent across the other samples (see also Onraet et al., 2013; Van Hiel et al., 2010). When multiple indicators of a single type of right–wing ideological or out–group attitudes (e.g. multiple subscales of a conservatism scale) or multiple indicators of a single primary domain of cognitive ability (e.g. different indicators of fluid ability) were administered, the mean correlation was calculated and used for the analyses.

Study characteristics and coding

For the relationship between cognitive ability and right–wing ideological attitudes, we found 67 samples, with a total of 84 017 unique participants, meeting the inclusion criteria for the meta–analysis. For the relationship between cognitive ability and prejudice, we found 23 samples, with a total of 27 011 unique participants, meeting the inclusion criteria for the meta–analysis. All studies are displayed in Table 1, and forest plots with each study's effect size and 95% confidence intervals (CIs) are displayed in Figure 1 (for right–wing ideological attitudes) and Figure 2 (for prejudice). In total, over both meta–analyses combined, we now have 82 unique samples, with 94 398 unique participants. We retrieved 32 additional studies, which comprised 78 840 extra participants compared with the meta–analysis of Van Hiel et al. (2010).

OPEN IN VIEWER

Table 1

An overview of all studies included in the meta–analysis

Study	N	Age group	Gender composition	Location	Type of ideological attitudes/prejudice	General or broad ability	Type of primary broad ability	Effect size (r)
Adams and Vidulich (1962)	36	Young adults	Mix	USA/Canada	Dogmatism	Broad	Short–term memory	−0.44
Adorno et al. (1950)	342	Young adults	Male only	USA/Canada	Authoritarianism	Broad	Writing and reading	−0.20
Adorno et al. (1950)	168	Young adults	Male only	USA/Canada	Ethnocentrism	Broad	Writing and reading	−0.08
Adorno et al. (1950)	104	Adults	Male only	USA/Canada	Ethnocentrism	General	—	−0.32
Adorno et al. (1950)	104	Adults	Male only	USA/Canada	Authoritarianism	General	—	−0.48
Adorno et al. (1950)	77	Adults	Male only	USA/Canada	Ethnocentrism	General	—	−0.28
Berkowitz and Wolkon (1964)	76	Young adults	Mix	USA/Canada	Authoritarianism	General	—	−0.27
Bettinghaus et al. (1970)	120	Mix	Mix	USA/Canada	Dogmatism	Broad	Fluid	−0.19
Block and Block (2006)	46	Children	Male only	USA/Canada	Conservatism	General	—	−0.30
Block and Block (2006)	49	Children	Female only	USA/Canada	Conservatism	General	—	−0.28
Bobo and Licari (1989)	1473	Adults	Mix	USA/Canada	Prejudice	Broad	Comprehension–knowledge	−0.22
Boshier (1973)	100	Mix	Mix	Other	Conservatism	Broad	Short–term memory	0.03
Bouchard et al. (2003)	355	Adults	Mix	Other	Conservatism	General	—	−0.23
Carl (2014)	12589	Adults	Mix	USA/Canada	Authoritarianism	Broad	Comprehension–knowledge	−0.22
Choma et al (2014)	198	Young adults	Mix	USA/Canada	Authoritarianism	Broad	Fluid	−0.09
Christensen (1963)	117	Young adults	Female only	USA/Canada	Dogmatism	General	—	−0.01
Christensen (1963)	49	Young adults	Male only	USA/Canada	Dogmatism	General	—	0.00
Christie (1954)	182	Young adults	Mix	USA/Canada	Authoritarianism	General	—	−0.48
Clark (1968)	35	Young adults	Male only	USA/Canada	Dogmatism	Broad	Comprehension–knowledge	−0.52
Costello and Hodson (2014)	20	Children	Mix	USA/Canada	Prejudice	Broad	Fluid	−0.39
Costello and Hodson (2014)	53	Children	Mix	USA/Canada	Prejudice	General	—	−0.45
Costin (1965)	67	Young adults	Mix	USA/Canada	Dogmatism	Broad	Comprehension–knowledge	−0.11
Crowson et al. (2007)	276	Young adults	Mix	USA/Canada	Conservatism	General	—	−0.02
Davids (1955)	20	Young adults	Male only	USA/Canada	Authoritarianism	General	—	−0.40
Davids and Eriksen (1957)	48	Young adults	Male only	USA/Canada	Authoritarianism	General	—	−0.24
Deary et al. (2008)	3412	Children	Male only	Europe/UK	Prejudice	Broad	Short–term memory	−0.12
Deary et al. (2008)	3412	Children	Male only	Europe/UK	Conservatism	Broad	Short–term memory	−0.13
Deary et al. (2008)	3658	Children	Female only	Europe/UK	Prejudice	Broad	Short–term memory	−0.13
Deary et al. (2008)	3658	Children	Female only	Europe/UK	Conservatism	Broad	Short–term memory	−0.13
Egan (1989)	94	Adolescents	Mix	Europe/UK	Ethnocentrism	Broad	Comprehension–knowledge	−0.43
Egan (1989)	94	Adolescents	Mix	Europe/UK	Conservatism	Broad	Comprehension–knowledge	−0.54
Eidelman (Unpublished data)	55	Young adults	Mix	USA/Canada	Authoritarianism	Broad	Fluid	0.15
Eisenman and Cherry (1970)	263	Young adults	Mix	USA/Canada	Authoritarianism	Broad	Long–term memory	−0.20
Eysenck (1954)	86	Adults	Male only	Europe/UK	Ethnocentrism	Broad	Fluid	−0.25
Eysenck (1954)	86	Adults	Male only	Europe/UK	Authoritarianism	Broad	Fluid	−0.28
Fraley et al. (2012)	635	Children	Mix	USA/Canada	Conservatism	General	—	−0.12
Francis (1997).	711	Adolescents	Mix	Europe/UK	Dogmatism	Broad	Fluid	−0.18
Gough (1951)	271	Adolescents	Mix	USA/Canada	Authoritarianism	General	—	−0.43
Heaven et al. (2011)	375	Adolescents	Mix	Other	Authoritarianism	Broad	Writing and reading	−0.26
Hello et al. (2006)	301	Young adults	Mix	Europe/UK	Prejudice	Broad	Comprehension–knowledge	−0.17
Hello et al. (2006)	301	Young adults	Mix	Europe/UK	Authoritarianism	Broad	Comprehension–knowledge	−0.25
Himmelweit and Swift (1971)	614	Adolescents	Male only	Europe/UK	Authoritarianism	General	—	−0.27
Iyer et al. (2012)	8651	Adults	Mix	USA/Canada	Conservatism	Broad	Fluid	−0.05
Jacobson and Rettig (1959)	354	Young adults	Mix	USA/Canada	Authoritarianism	Broad	Comprehension–knowledge	−0.22
Jones (1957)	220	Young adults	Mix	USA/Canada	Authoritarianism	General	—	−0.22
Kanazawa (2010)	13058	Adolescents	Mix	USA/Canada	Conservatism	Broad	Comprehension–knowledge	−0.24
Kanazawa (2010)	13034	Adults	Mix	USA/Canada	Conservatism	Broad	Comprehension–knowledge	−0.07
Katz (1988)	100	Young adults	Mix	Other	Conservatism	Broad	Fluid	0.21
Katz (1988)	110	Young adults	Mix	Other	Conservatism	Broad	Fluid	0.03
Katz (1990)	100	Young adults	Mix	Other	Conservatism	Broad	Processing speed	0.20
Keiller (2010)	257	Young adults	Mix	USA/Canada	Prejudice	General	—	−0.36
Keiller (2010)	257	Young adults	Mix	USA/Canada	Authoritarianism	General	—	−0.28
Kemmelmeier (2008)	5893	Young adults	Mix	USA/Canada	Conservatism	Broad	Comprehension–knowledge	−0.04
Kranou (2013), unpublished data	425	Young adults	Mix	Europe/UK	Conservatism	Broad	Fluid	−0.12
Kutner and Gordon (1964)	33	Children	Mix	USA/Canada	Ethnocentrism	Broad	Fluid	−0.32
Lapsley and Enright (1979)	94	Young adults	Mix	USA/Canada	Dogmatism	Broad	Fluid	−0.26
Long and Ziller (1965)	72	Young adults	Female only	USA/Canada	Dogmatism	Broad	Visual–spatial	−0.20
Marks and McDougall (1959)	57	Adolescents	Mix	USA/Canada	Authoritarianism	General	—	−0.68
McCourt et al. (1999)	274	Adults	Mix	USA/Canada	Authoritarianism	General	—	−0.37
Meeusen et al. (2013)	1910	Mix	Mix	Europe/UK	Ethnocentrism	Broad	Comprehension–knowledge	−0.29
Messick and Frederiksen (1958)	232	Young adults	Mix	USA/Canada	Authoritarianism	Broad	Visual–spatial	0.01
Moore et al. (1984)	40	Adolescents	Mix	USA/Canada	Prejudice	General	—	0.00
O'Connor (1952)	57	Young adults	Mix	USA/Canada	Ethnocentrism	Broad	Fluid	−0.37
Oskarsson et al. (in press)	1946	Adults	Male only	Europe/UK	Conservatism	General	—	0.18
Rokeach (1951)	144	Young adults	Mix	USA/Canada	Ethnocentrism	General	—	−0.28
Rubinstein (2003)	111	Young adults	Mix	Other	Authoritarianism	Broad	Long–term memory	−0.67
Scarr (1981)	914	Mix	Mix	Usa/Canada	Authoritarianism	General		−0.35
Schoon et al. (2010)	4537	Children	Female only	Europe/UK	Prejudice	Broad	Fluid	−0.15
Schoon et al. (2010)	4537	Children	Female only	Europe/UK	Conservatism	Broad	Fluid	−0.15
Schoon et al. (2010)	4267	Children	Male only	Europe/UK	Prejudice	Broad	Fluid	−0.17
Schoon et al. (2010)	4267	Children	Male only	Europe/UK	Conservatism	Broad	Fluid	−0.18
Shook and Fazio (2009)	58	Young adults	Mix	Usa/Canada	Conservatism	Broad	Short–term memory	0.08
Sidanius and Lau (1989)	5655	Young adults	Mix	USA/Canada	Prejudice	General	—	−0.18
Siegel (1956)	60	Young adults	Male only	USA/Canada	Authoritarianism	General	—	−0.19
Siegel (1956)	60	Adults	Male only	USA/Canada	Authoritarianism	General		−0.21
Stankov (2009)	732	Young adults	Mix	USA/Canada	Conservatism	General	—	−0.35
Stankov (2009)	430	Young adults	Mix	USA/Canada	Conservatism	Broad	Comprehension–knowledge	−0.40
Stankov (2009)	288	Young adults	Mix	Other	Conservatism	Broad	Fluid	−0.23
Taylor and Dunnette (1974)	79	Adults	Male only	USA/Canada	Dogmatism	General	—	0.29
Thompson and Michel (1972)	379	Young adults	Mix	USA/Canada	Dogmatism	General	—	−0.20
Uhes and Shaver (1970)	316	Adolescents	Mix	USA/Canada	Dogmatism	Broad	Long–term memory	−0.26
Vezzali et al. (2013), unpublished data	122	Children	Mix	Europe/UK	Prejudice	Broad	Fluid	−0.13
Vezzali et al. (2013), unpublished data	395	Children	Mix	Europe/UK	Prejudice	Broad	Fluid	0.06
Von Stülpnagel and Steffens (2010)	148	Young adults	Mix	Europe/UK	Prejudice	Broad	Processing speed	−0.13
Wegmann (1992)	50	Young adults	Mix	USA/Canada	Authoritarianism	Broad	Fluid	−0.42
Wegmann (1992)	29	Adults	Mix	USA/Canada	Authoritarianism	Broad	Fluid	−0.49
Xu et al. (2013)	486	Adults	Mix	USA/Canada	Conservatism	Broad	Fluid	−0.17
Xu et al. (2013)	540	Adults	Mix	USA/Canada	Conservatism	Broad	Fluid	−0.10
Xu et al. (2013)	460	Adults	Mix	USA/Canada	Conservatism	Broad	Fluid	−0.12
Zagona and Zurcher (1965)	60	Young adults	Mix	USA/Canada	Dogmatism	Broad	Comprehension–knowledge	−0.18

OPEN IN VIEWER

Figure 1

Forest plot for the meta–analysis for ideological right–wing attitudes, grouped by type of ideological attitude (authoritarianism, conservatism and dogmatism).

OPEN IN VIEWER

Figure 2

Forest plot for the meta–analysis for prejudice, grouped by type of prejudice.

We coded each sample for design, sample and publication characteristics. First, we coded each study according to the specific type of right–wing ideological attitudes or prejudice used: authoritarianism, conservatism ² or dogmatism for right–wing ideological attitudes and ethnocentrism or prejudice towards specific groups as types of prejudice. Next, we also coded the type of cognitive ability according to the CHC model. We decided to create two moderator variables for the distinction between different types of cognitive. First, we distinguished between different levels of cognitive ability. More specifically, we coded whether the measure tapped the following: (i) a higher–order factor of general ability (studies using scores of entire intelligence tests or studies combining several types of broad abilities into one factor) or (ii) one specific type of broad primary ability (without specifying which type of broad primary ability). For the second moderator variable, we coded the latter studies according to the specific type of primary broad domains of the CHC model. For this second moderator analysis on type of cognitive ability, the studies using general ability measures were not included.

Sample characteristics were according to age group in which cognitive ability was administered [children (0–11 years old), adolescents (12–17 years old), young adolescents (18–27 years old) and adults (27+ years old)], sex composition (mixed sex, males only and females only) and geographic location of the sample (United States/Canada, Europe and other). As an indicator of publication characteristic, the year of data acquisition was coded into three categories, 1950–1969, 1970–1989 and 1990–present.

Statistical analyses

Pearson product–moment correlation coefficients (rs) between cognitive ability and right–wing ideological attitudes/prejudice were used as effect size estimates. For studies reporting mean differences in scores on right–wing ideological attitudes or prejudice across groups with low or high cognitive ability, the reported test statistics (F–value, t–value or p–values) were used for calculating the effect sizes (Borenstein, Hedges, Higgins & Rothstein, 2005). Three samples only reported a p–value. In these cases, we used the lower limit effect size estimates from the reported significance level. This meta–analytic decision is commonly used but a conservative strategy that tends to underestimate the true magnitude of effect sizes (Durlak & Lipsey, 1991).

For the statistical meta–analyses, we used the software Comprehensive Meta–analysis version 2.2 (Borenstein et al., 2005) in combination with metafor, a meta–analysis package for R (Viechtbauer, 2010). We applied a random–effects model to compute the overall effects, because we assumed that effect sizes would vary across studies. Random–effects models produce results that can be generalized to future studies with different designs (Hedges & Vevea, 1998). First, Fisher–Z coefficients were calculated based on the Pearson correlations to permit an unbiased comparison of effect sizes. Second, we computed mean weighted effect sizes and 95% CIs around the point estimate of the combined estimates. Next, for interpretation convenience, the effect size estimates were transformed back to correlations. According to Cohen (1988), effect sizes (rs) of .10 are considered small effects, .30 are considered moderate effects and .50 are considered large effects. Based on an analysis of meta–analysis in psychological research, Hemphill (2003) recommended interpreting effect sizes of .10, .20 and .30 as small, moderate and large effects, respectively, to better reflect effect sizes in psychology per se.

Homogeneity analyses were conducted to test whether the sets of effect sizes were homogeneous at the population level and allowed us to examine the role of potential moderator variables. We conducted these moderation analyses using categorical testing procedures (Lipsey & Wilson, 2001). A significant within–groups Q (Q_w) indicates that the effect sizes within each moderator category are heterogeneous, whereas a significant between–groups Q (Q_b) estimate indicates that the effect sizes of the moderator subgroups are significantly different. I² indices (J. P. T. Higgins & Thompson, 2002) indicate the percentage of variability in point estimates due to between–study heterogeneity, rather than sampling error (I² = 0 indicates that all variability in effect estimates is caused by sampling error; I² values on the order of 25, 50 and 75 represent low, moderate and high between–study heterogeneity).

Finally, we addressed the robustness of the meta–analytical estimates by performing sensitivity analyses. These evaluate (and adjust for) the impact of publication bias as well as the impact of outliers and influential studies. The accuracy of a meta–analysis strongly depends upon the representativeness of the sample of studies analysed. Publication bias is one source that potentially jeopardizes the representativeness of a meta–analytical sample set because the likelihood that a study becomes published hinges upon the presence of significant results (Kepes, Banks, McDaniel & Whetzel, 2012; Kepes, McDaniel, Brannick & Banks, 2013; Rothstein, Sutton & Borenstein, 2005). Hence, publication bias might overestimate the meta–analytical effect sizes. Another important source that potentially questions meta–analytic conclusions is the presence of outliers and influential cases (Kepes et al., 2013; Viechtbauer & Cheung, 2010). Even though it is not uncommon to observe extreme effect size values when conducting a meta–analysis, it is mandatory to illuminate to which extent these outliers/influential cases weigh upon the meta–analytic conclusions.

These sensitivity analyses involved the combination of four recommended techniques (Kepes et al., 2012, 2013; Stanley & Doucouliagos; 2014; Viechtbauer & Cheung, 2010). First, we addressed the presence of publication bias and outliers by visually inspecting contour–enhanced funnel plots (Peters, Sutton, Jones, Abrams & Rushton, 2008; Kepes et al., 2013; Sterne et al., 2011). Contour–enhanced funnel plots display the magnitude of each study's effect size as a function of the sample's standard error (Sterne et al., 2011), allowing visual inspection of the symmetry of the meta–analytical study distribution. In the funnel plots, the grey–shaded areas indicate different levels of statistical significance, whereas the white (unshaded) area indicates the non–significance of the study effect sizes. In the presence of publication bias, it is expected that the bottom of a plot (where the smaller studies are located) will show a higher concentration of studies on one side of the mean than the other. The missing studies in the underrepresented area would primarily be located in the white (non–significant) regions of the plot. If these two visual features are present, this would reflect publication bias, resulting from the phenomenon that smaller studies only become published if they have larger–than–average effects, because this makes them more likely to attain statistical significance criteria (Peters et al., 2008; Sterne et al., 2011).

Second, we applied the trim–and–fill method (Duval & Tweedie, 2000) to identify and adjust for publication bias. This method is an iterative, statistical procedure based upon the notion that—in the absence of publication bias—the funnel plot would be symmetrically dispersed around the summary effect. The trim–and–fill procedure estimates the number of missing studies in a dataset by ‘trimming’ the funnel plot until it is symmetrical and then ‘filling’ in both sides of the funnel in a way that maintains symmetry. Based upon the imputation (‘filling’) of missing effect sizes, the procedure then re–estimates an adjusted pooled effect size as sensitivity analysis. We added the trim–and–fill imputations to the contour–enhanced funnel plots because they help to inform the likely location of missing studies (Peters et al., 2008).

Third, because the trim–and–fill method has recently been criticized in the literature (e.g. lower ability to detect publication bias and tendency for under–correction of publication bias; Carter & McCullough, 2014; Peters et al., 2008), we also applied Egger's linear regression procedure (Egger, Smith, Schneider & Minder, 1997; Sterne & Egger, 2001). This test is also based upon the funnel plot but does not implicitly assume that publication bias is the only cause of funnel plot asymmetry. Egger's test can be described as a weighted least squares regression model in which the effect size is predicted by the standard error. The (non–)significance of the coefficient associated with the standard error (i.e. the slope, b₁) is interpreted as a test of funnel plot asymmetry and leads to a conclusion on the absence/presence of publication bias (Sterne & Egger, 2001). Interestingly, the use of this model is recently expanded by the recognition that in Egger's regression equation, the model's intercept (b₀) can be interpreted as an estimate of the underlying effect size, which is theoretically uninfluenced by publication bias (see, for extensive discussions, Carter & McCullough, 2014; Stanley & Doucouliagos, 2014). This expansion of the use of Egger's test is referred to as the ‘precision–effect test’ (PET; Stanley & Doucouliagos, 2014). Simulation studies have now shown that PET estimations are highly accurate when the true meta–analytic effect is zero but tend to overcorrect when the true effect is non–zero. In these cases, the intercept of a regression model in which the effect size is predicted by the variance (i.e. standard error squared) provides more optimal estimates of the adjusted pooled effect size. This is referred to as a ‘precision–effect with standard error’ (or PEESE; Carter & McCullough, 2014; Stanley & Doucouliagos, 2014). We adopted this conditional PET–PEESE procedure (i.e. if b₀ using PET is significant, PEESE–adjusted estimates should be interpreted as unbiased effects) to calculate more accurate pooled effects adjusted for publication bias.

In the final step of the sensitivity analyses, we addressed the impact of potential outliers and influential cases in the two core meta–analyses. In addition to the contour–enhanced funnel plot providing a first visual sense of potential outliers (Kepes et al., 2013), we applied the set of outlier and influence diagnostics for the meta–analyses proposed by Viechtbauer and Cheung (2010). This allows identification of particularly influential studies in both meta–analyses. Subsequently, we re–ran the meta–analyses without these outliers to evaluate to what extent these influential cases distort the conclusions of the meta–analyses. This type of sensitivity analyses may either strengthen the conclusions of the meta–analyses or leave some doubts regarding their robustness (Viechtbauer & Cheung, 2010). All sensitivity analyses were conducted using the metafor package in R (Viechtbauer, 2010).

Cognitive ability and right–wing ideological attitudes

Fifty–seven studies showed negative relations, nine showed positive relations and one showed a correlation of approximately 0. The meta–analysis (for all results, see Table 2) revealed an overall moderate negative relation, r = −.20, p < .001. In other words, lower cognitive ability was associated with the stronger endorsement of right–wing ideological attitudes.

OPEN IN VIEWER

Table 2

Moderators of effect sizes for studies on the relationship of cognitive ability and right–wing ideological attitudes

Moderator	N	k	r	95%CI	Q b	Q w	I 2
Total set	84 017	67	−0.20 **	[−0.23, −0.17]		1058.06 **	93.76
General or primary broad ability					1.64		39.16
General ability	7 820	25	−0.24 **	[−0.34, −0.14]		442.18 **	94.57
Primary broad ability	76 197	42	−0.17 **	[−0.21, −0.14]		614.59 **	93.33
Type of primary broad ability †					27.41 **		78.11
Comprehension–knowledge	45 915	11	−0.23 **	[−0.30, −0.16]		385.15 **	97.40
Fluid	21 207	18	−0.13 **	[−0.18, −0.09]		102.01 **	83.34
Short–term memory	7 264	5	−0.12 **	[−0.17, −0.07]		9.67	58.64
Long–term memory and retrieval	690	3	−0.39 **	[−0.62, −0.11]		30.33 **	93.41
Processing speed	100	1	0.20	[0.00, 0.38]		0.00	0.00
Visual–spatial processing	304	2	−0.07	[−0.27, 0.13]		2.40	58.33
Writing and reading	717	2	−0.23 **	[−0.30, −0.16]		0.71	0.00
Type of ideological attitudes					24.13 **		91.71
Authoritarianism	18 142	27	−0.30 **	[−0.34, −0.24]		147.22 **	82.34
Conservatism	63 740	27	−0.13 **	[−0.17, −0.08]		657.63 **	96.05
Dogmatism	2 135	13	−0.17 **	[−0.25, −0.09]		35.60 **	66.29
Age group					21.36 **		81.27
Children	16 604	7	−0.15 **	[−0.17, −0.13]		9.5	36.85
Adolescents	15 496	8	−0.32 **	[−0.39, −0.25]		42.52 **	84.62
Young adults	12 090	35	−0.19 **	[−0.25, −0.13]		239.18 **	88.40
Adults	38 693	14	−0.15 **	[−0.22, −0.07]		442.50 **	84.62
Mix	1 134	3	−0.19	[−0.40, 0.05]		15.85 **	87.38
Gender composition of sample					10.23 **		80.45
Female only	8 433	5	−0.14 **	[−0.16, −0.12]		4.06	1.42
Male only	11 168	15	−0.18 **	[−0.28, −0.08]		256.26 **	94.54
Mix	64 416	47	−0.21 **	[−0.25, −0.17]		775.46 **	94.07
Location					1.59		0.00
USA/Canada	62 427	48	−0.21 **	[−0.25, −0.18]		710.10 **	93.38
Europe/UK	20 051	11	−0.17 **	[−0.24, −0.10]		244.78 **	95.91
Other	1 539	8	−0.14	[−0.31, 0.04]		86.32 **	91.89
Time of data collection					4.86		58.88
1950–1969	2 462	20	−0.28 **	[−0.36, −0.19]		85.07 **	77.66
1970–1989	3 269	13	−0.17 **	[−0.27, −0.07]		86.82 **	86.18
1990–present	78 286	34	−0.17 **	[−0.21, −0.13]		821.27 **	95.98

Note: All cognitive ability measures are coded so that higher scores reflect higher cognitive ability.

N, number of participants; k, number of studies; CI, confidence interval; Q_b, homogeneity statistic between classes; Q_w, homogeneity statistic within classes. I², homogeneity statistic (percentage of heterogeneity).

*

p < .05;

**

p < .01;

***

p < .007 (.05/7).

†

For this moderator analyses, we excluded the 27 samples measuring general ability, because these studies relate to a combination of different types of cognitive ability. Hence, this moderator analysis is based only on the 42 samples with measures for one specific type of broad ability.

Cognitive ability and prejudice

We found 23 studies reporting correlations between cognitive ability and prejudice, with 21 studies showing negative relations, 1 showing a positive relation and 1 showing a correlation of approximately 0. The overall effect size for this relationship (Table 3) was comparable with the effect size for right–wing ideological attitudes, r = −.19, p < .001, indicating that lower cognitive ability was associated with stronger endorsement of prejudice.

OPEN IN VIEWER

Table 3

Moderators of effect sizes for studies on the relationship of cognitive ability and prejudice

Moderator	N	k	r	95% CI	Q b	Q w	I 2
Total set	27 011	23	−0.19 ***	[−0.23, −0.16]		109.78 ***	79.96
General or primary broad ability					3.6		72.21
General ability	6 330	7	−0.27 ***	[−0.36, −0.18]		18.67 **	67.86
Primary broad ability	20 681	16	−0.17 ***	[−0.21, −0.13]		85.76 ***	82.51
Type of primary broad ability †					12.58 *		68.21
Comprehension–knowledge	3 778	4	−0.26 ***	[−0.32, −0.19]		10.70 *	71.97
Fluid	9 517	8	−0.15 ***	[−0.21, −0.08]		25.41 ***	72.45
Short–term memory	7 070	2	−0.13 ***	[−0.15, −0.10]		0.18	0.00
Processing speed	148	1	−0.13	[−0.29, 0.03]		0.00	0.00
Writing and reading	168	1	−0.08	[−0.23, 0.07]		0.00	0.00
Type of prejudice					11.78 ***		91.51
Ethnocentrism	2 673	9	−0.28 ***	[−0.34, −0.22]		10.87	26.37
Prejudice towards specific groups	24 338	14	−0.16 ***	[−0.19, −0.13]		56.49 ***	76.99
Age group					29.36 ***		86.38
Children	16 497	9	−0.13 ***	[−0.17, −0.09]		30.57 ***	73.83
Adolescents	134	2	−0.24	[−0.60, 0.20]		5.56 *	82.03
Young adults	6 730	7	−0.21 ***	[−0.28, −0.14]		15.25 *	60.66
Adults	1 740	4	−0.23 ***	[−0.27, −0.19]		1.37	.00
Mix	1 910	1	−0.29 ***	[−0.33, −0.25]		0.00	0.00
Gender composition of sample					6.81 *		70.64
Female only	8 195	2	−0.14 ***	[−0.16, −0.12]		0.84	0.00
Male only	8 114	6	−0.16 ***	[−0.21, −0.11]		11.38 *	56.04
Mix	10 702	15	−0.23 ***	[−0.28, −0.17]		72.27 ***	80.63
Location					5.17 *		80.67
USA/Canada	18 930	12	−0.25 ***	[−0.30, −0.19]		25.14 **	56.24
Europe/UK	8 081	11	−0.16 ***	[−0.21, −0.11]		74.14 ***	86.51
Time of data collection					2.75		27.28
1950–1969	583	6	−0.25 ***	[−0.34, −0.16]		6.98	28.37
1970–1989	7 348	5	−0.22 ***	[−0.28, −0.15]		10.25 *	60.97
1990–present	19 080	12	−0.17 ***	[−0.22, −0.12]		83.51 ***	86.83

Note: All cognitive ability measures are coded so that higher scores reflect higher cognitive ability.

N, number of participants; k, number of studies; CI, confidence interval; Q_b, homogeneity statistic between classes; Q_w, homogeneity statistic within classes; I², homogeneity statistic (percentage of heterogeneity).

*

p < .05;

**

p < .01;

***

p < .007 (.05/7).

†

For this moderator analyses, we excluded the seven samples measuring general ability, because these studies relate to a combination of different types of cognitive ability. Hence, this moderator analysis is based only on the 16 samples with measures for one specific type of broad ability.

Sensitivity analyses

The validity of these two meta–analyses was addressed with sensitivity analyses evaluating the impact of publication bias and potential outliers/influential studies. These results are organized as an answer to three questions: (1) ‘Is there evidence of publication bias?’; (2) ‘How do the pooled effect sizes change if we adjust for potential publication bias?’; and (3) ‘Which studies are outliers and how do they affect the meta–analytic estimates?’. Publication bias and outlier analyses were run on the full set of studies for the two meta–analyses, with the publication bias analyses also run on smaller sub–distributions (Kepes et al., 2012). We chose to analyse publication bias for each type of right–wing ideological attitudes and prejudice, as analyses identified these as the strongest moderator variables (explaining 93% and 42% of between–study variance, respectively).

Is there evidence of publication bias?

This question is addressed by the joint interpretation of three procedures: contour–enhanced funnel plots, the trim–and–fill procedure and significance of Egger's regression tests (Kepes et al., 2012; Peters et al., 2008). Figure 3 presents the contour–enhanced funnel plot with trim–and–fill imputations for the (sub–)meta–analyses on cognitive ability and right–wing ideological attitudes. The trim–and–fill method on the overall meta–analysis (Figure 3A) suggests that 11 studies are missing at the right of the mean effect size. As the contour–enhanced funnel plot allows evaluation of the missing studies in the context of statistical significance, it can be noted that 10 of 11 studies are located in the grey–shaded areas of the plot. This suggests that funnel plot asymmetry primarily results from factors other than publication bias. Indeed, the separate publication bias analyses for each type of right–wing ideological attitudes (Figure 3B–D) strengthen this hypothesis as there is strong moderator–induced between–study heterogeneity. More specifically, within the sub–distributions for both authoritarianism (Figure 3B) and conservatism (Figure 3C), trim–and–fill method no longer imputes studies at any side of the mean. Only for dogmatism (Figure 3D) are two studies imputed in the grey area. However, as this sub–distribution only includes 13 studies, these imputations can also be caused by the small set of studies in this meta–analysis. For all (sub–)meta–analyses, Egger's regression tests for funnel plot asymmetry were performed. However, in none of these regression models were significant slope coefficients found. This also strengthens the conclusion that publication bias is unlikely to distort the pooled estimate between cognitive ability and right–wing ideological attitudes.

OPEN IN VIEWER

Figure 3

Contour–enhanced funnel plots with trim–and–fill imputations for cognitive ability and right–wing ideological attitudes. The black points represent the real studies; the white points represent the imputed studies. If missing studies are imputed in areas where non–significant studies would be plotted (no shading), the observed asymmetry may be due to publication bias. If missing studies are imputed in areas of statistical significance (darker shading), publication bias is less likely.

Similar results are found for the (sub–)meta–analyses on the association between cognitive ability and prejudice. Figure 4 presents the contour–enhanced funnel plot with trim–and–fill imputations for the overall meta–analysis on prejudice (Figure 4A) and the specific types of prejudices: ethnocentrism (Figure 4B) and prejudice towards specific groups (Figure 4C). The trim–and–fill method on the overall meta–analysis imputes four studies at the right of the mean, with only two of them located in the white (non–significant) area of the plot. The trim–and–fill method imputed three missing studies in the meta–analysis of ethnocentrism and one in the meta–analysis prejudice towards specific groups. However, funnel plot asymmetry in these subsets might be primarily related to the limited number of included studies. Also, for all (sub–)meta–analyses, Egger's regression tests yield no significant slope coefficients. This strengthens the conclusion that publication bias is unlikely to distort the relation between cognitive ability and prejudice.

OPEN IN VIEWER

Figure 4

Contour–enhanced funnel plots with trim–and–fill imputations for cognitive ability and prejudice. The black points represent the real studies; the white points represent the imputed studies. If missing studies are imputed in areas where non–significant studies would be plotted (no shading), the observed asymmetry may be due to publication bias. If missing studies are imputed in areas of statistical significance (darker shading), publication bias is less likely.

How do the pooled effect sizes change if we adjust for potential publication bias?

Table 4 presents the original and adjusted effect sizes provided by the trim–and–fill procedure and extended Egger's regression procedure (i.e. PET–PEESE estimators). In line with the previous analyses, adjusted effect sizes are strongly similar to the original meta–analytic estimates. All estimates remain significant (p < .001), and hence, the PEESE results can be regarded as the best estimates of the effect sizes corrected for publication bias (Carter & McCullough, 2014; Stanley & Doucouliagos, 2014). Notably, these PEESE estimates provide a somewhat less negative effect between cognitive ability and right–wing ideological attitudes (r = −.15) and between cognitive ability and authoritarianism (r = −.23). All other PEESE estimates resemble the original estimates.

OPEN IN VIEWER

Table 4

Effect sizes based on sensitivity analyses correcting for publication bias and outliers/influential studies

	k	Original estimate [95% CI]	Trim and fill	PET †	PEESE †	k	Without outliers
Right–wing ideological attitudes	67	−0.20 [−0.23, −0.17]	−0.15 [−0.18, −0.12]	−0.13 [−0.17, −0.10]	−0.15 [−0.17, −0.12]	64	−0.20 [−0.23, −0.17]
Authoritarianism	27	−0.30 [−0.34, −0.24]	−0.30 [−0.34, −0.24]	−0.21 [−0.26, −0.17]	−0.23 [−0.27, −0.20]	26	−0.28 [−0.32, −0.23]
Conservatism	27	−0.13 [−0.17, −0.08]	−0.13 [−0.17, −0.08]	−0.12 [−0.19, −0.06]	−0.12 [−0.17, −0.08]	26	−0.14 [−0.18, −0.10]
Dogmatism ‡	13	−0.17 [−0.25, −0.09]	−0.13 [−0.22, −0.04]	−0.20 [−0.16, −0.08]	−0.18 [−0.29, −0.06]	12	−0.20 [−0.26, −0.14]
Prejudice	23	−0.19 [−0.23, −0.16]	−0.18 [−0.22, −0.14]	−0.15 [−0.19, −0.11]	−0.17 [−0.19, −0.14]	22	−0.18 [−0.21, −0.15]
Ethnocentrism ‡	9	−0.28 [−0.34, −0.22]	−0.26 [−0.32, −0.19]	−0.28 [−0.38, −0.20]	−0.28 [−0.34, −0.22]	8	−0.28 [−0.36, −0.19]
Prejudice towards specific groups ‡	14	−0.16 [−0.19, −0.13]	−0.16 [−0.19, −0.12]	−0.15 [−0.20, −0.10]	−0.16 [−0.18, −0.13]	14	−0.16 [−0.19, −0.13]

Note: All estimates are significant at p < .001.

†

PET–PEESE is a conditional estimating procedure. Because PET results are all significant (hence, rejecting the null hypothesis b₀ = 0), the PEESE results should be interpreted as the best estimates of the true effects corrected for publication bias.

‡

Estimates for these sub–distributions should be interpreted with caution as these are based on a limited set of studies (Sterne & Egger, 2001).

Finding 1: General relationship of cognitive ability with ideology and prejudice

The first main finding of our study was that people with greater cognitive resources are more likely to adhere to left–wing beliefs and tend to be less prejudiced, whereas those having lower cognitive abilities are more likely to endorse right–wing beliefs and be more prejudiced (average effect sizes of r = −.20, 95% CI [−0.23, −0.17]; and r = −.19, 95% CI [−0.23, −0.16], respectively). The confidence interval reported provides considerable support for the notion that these findings are both meaningful and replicable. We further assessed the validity of our meta–analytical estimates, with sensitivity analyses addressing the impact of both publication bias and outliers/influential studies. In general, these analyses underscore robustness in our meta–analytical study because the implications of the meta–analytic results and conclusions remain largely unaltered when adjusting for potential publication bias or when omitting outliers and overly influential studies.

A straightforward comparison between the results of the present meta–analysis and the meta–analysis conducted by Van Hiel et al. (2010), who also investigated the relationship between cognitive ability and ideological attitudes, is not possible. First, Van Hiel et al. (2010) included ideological attitudes and ethnocentrism in the same analysis, whereas we performed two separate analyses. Second, and most importantly, Van Hiel et al. (2010) also included years of education as a proxy of cognitive ability, whereas we chose not to include this variable as an indicator of cognitive ability and instead focused only on objective tasks measuring cognitive ability. As can be seen from their results, years of education yields the strongest effect size compared with other indicators of cognitive ability, making their overall effect sizes for the different types of ideology higher than the present effect sizes.

Moderator analyses revealed that the effect size in the present meta–analysis was relatively stable across different sample characteristics, such as gender, location and time frame, attesting to the robustness of the relationship. Only one moderator related to sample characteristics yielded a significant effect. Specifically, the strength of the effect size differed across age groups, with the strongest effect size for the relationship between cognitive ability and right–wing ideological attitudes in the group of adolescents. Although we do not have a clear explanation for this effect, it should be noted that adolescence constitutes the formative years of political ideology (e.g. Altemeyer, 1998; Alwin, Cohen, & Newcomb, 1991) and that cognitive ability might then yield its biggest impact. Future research can further examine this potential.

Our effect sizes varied around r = −.20 and were largest in magnitude for authoritarianism (r = −.30, 95% CI [−0.34, −0.24]) and ethnocentrism (r = −.28, 95% CI [−0.34, −.22]). Clearly, these effects are not inconsequential but rather are of a similar magnitude of other meta–analytic relations such as between contact and prejudice (Pettigrew & Tropp, 2006), stereotypes and prejudice (Dovidio, Brigham, Johnson & Gaertner, 1996), discrimination and prejudice (Talaska, Fiske & Chaiken, 2008; Schütz & Six, 1996), personality factors and prejudice (Sibley & Duckitt, 2008) and even stronger than relations between religiosity and prejudice (Hall, Matz & Wood, 2010). In fact, these effect sizes approximate those of the vast majority of personality and social psychology findings more generally (Richard, Bond & Stokes–Zoota, 2003). With meta–analytic relations of this magnitude reliably observed, a strong case can be made that cognitive abilities are as important, and often more important, in explaining ideology and prejudice than many of the constructs commonly discussed in personality and social psychology textbooks. Hence, we strongly advise future models and theories aiming to uncover the psychological basis of ideology and prejudice to provide a key role for cognitive ability (Dhont & Hodson, 2014; Hodson, 2014; Hodson & Busseri, 2012).

An important reason for this long–standing interest in the cognitive basis of ideology lies in an attempt to uncover the scientific inaccuracy or invalidity of certain ideologies, and the superiority of other ideologies (Durrheim, 1997). Specifically, ideologies might seem inferior when they attract less intelligent people, whereas ideologies that attract intelligent people may appear to be more ‘correct’. To be clear, any attempt to show whether right–wing or left–wing ideologies are accurate or valid on the basis of the level of cognitive ability of their adherents is based on false premises and certainly not the goal of our present synthesis.

Moreover, we appreciate that this issue is a very delicate and controversial one (Dhont & Hodson, 2014; Hodson, 2014) that, for this reason, speaks to the need for a cumulative–science approach (i.e. meta–analysis). Moreover, we would like to stress that, although right–wing ideological attitudes relate to conservative and right–wing political party affiliation (e.g. Altemeyer, 1996; Jost et al., 2003), our findings cannot be generalized to party identification. In other words, the current findings not necessarily imply that adherents of right–wing parties have lower cognitive abilities than adherents of left–wing parties.

Finding 2: Different effect sizes for different types of broad cognitive ability

The second important question of the present study pertained to the potential differences in effect sizes for different types of cognitive abilities. Previous studies (Deary et al., 2008; Heaven et al., 2011; Kemmelmeier, 2008) yielded somewhat conflicting outcomes. In our meta–analysis, we investigated two moderators in order to answer this question. First, we found that the strength of the relationship did not differ between studies focusing on one specific type of cognitive ability and those focusing on general cognitive ability. Second, we compared the effect sizes between different types of broad cognitive abilities as defined by the CHC theory (McGrew, 2005; Schneider & McGrew, 2012). Our analysis for right–wing attitude effect sizes varied significantly across ability types. To interpret this effect, we focus on comprehension–knowledge, fluid abilities and short–term memory, because we have a smaller number of participants (<1000) for the other types of broad abilities. The strongest effect size was obtained for comprehension–knowledge (r = −.23), which refers to abilities based on previously acquired knowledge and skills valued by one's culture. It includes general verbal information, language development, lexical knowledge, listening and communication abilities and grammar sensitivity (Schneider & McGrew, 2012). The effect size was considerably smaller for fluid abilities (r = −.13), referring to abilities to solve unfamiliar problems and abstract reasoning, and for short–term memory (r = −.12), referring to abilities to encode, maintain and manipulate information in the immediate situation. Although the moderator for type of primary broad ability was not significant in our analysis for prejudice, the same pattern of results emerged, with the strongest effect size for comprehension–knowledge (r = −.26) and weaker effect sizes for fluid abilities (r = −.15) and short–term memory (r = −.13).

These findings corroborate the studies of Heaven et al. (2011) and Kemmelmeier (2008) who found that verbal abilities are more strongly related to ideological attitudes compared with numerical and mathematical reasoning. As noted by Heaven et al. (2011), ideologies are relevant to verbal narratives, arguments and point of views, but not directly to numerical abilities. Similarly, other researchers argue that ideology can be considered as a schema or a learned knowledge structure, including norms and values, beliefs and opinions (e.g. Fiske, Lau & Smith, 1990; Hamill, Lodge & Blake, 1985). Hence, this might explain why comprehension–knowledge abilities may be especially relevant in relationship with ideology.

Also of relevance to the present findings is the recognition that cognitive ability is associated with particular personality traits, which also lie at the basis of right–wing ideological attitudes and prejudice. More specifically, high openness (Costa & McCrae, 1992; Digman, 1990), which refers to a preference for novelty, variety and intense experience, lies at the basis of lower authoritarianism and conservatism and lower prejudice (Block & Block, 2006; Sibley & Duckitt, 2008). At the same time, several researchers have shown that openness is also related to cognitive ability, with the strongest associations for comprehension–knowledge abilities, compared with other types of cognitive ability, such as fluid ability (Ackerman & Heggestad, 1997; Ashton, Lee, Vernon & Jang, 2000). This common personality association may also provide insights into the stronger relationships of comprehension–knowledge abilities with right–wing ideological attitudes and prejudice.

Whereas the effect of comprehension–knowledge ability is strongest, it is important to note that the other types of broad abilities (for which we obtained enough studies) also yielded negative and significant effects. This finding further attests to the robustness of this general relationship. However, on the basis of the present state of the literature, we cannot make conclusive statements about all types of cognitive abilities. First, studies on cognitive ability and ideology and prejudice did not include every type of ability (e.g. to our knowledge, no study investigated auditory processing or quantitative knowledge). Second, for other types of cognitive ability, most notably long–term memory, processing speed and visual–spatial processing, we found only a few studies, which elicited potential statistical power issues in our analysis. Hence, we should be cautious not to over–interpret the results for these specific abilities. Future research administering a wide range of cognitive ability measures at the same time can provide a more decisive answer to this intriguing question.

Finding 3: Different effect sizes for different types of ideological attitudes and prejudice

Our analyses revealed that cognitive ability is not related to all types of right–wing ideological attitudes or prejudices to the same extent. First, in our analysis of right–wing ideological attitudes, we found the strongest effect size for authoritarianism as compared with conservatism and dogmatism. One possible explanation lies in particular rigid cognitive styles linked with cognitive ability, and most notably NFC. Research indicated that lower cognitive ability is associated with a higher NFC (Webster & Kruglanski, 1994), which relates to an individual's desire to come to a relatively quick closure in decisions and judgments (Kruglanski, 1989; Kruglanski & Webster, 1996). Individuals with a high NFC have a strong preference for structure, certainty and predictability and are repelled by ambiguity. Because of these needs, individuals with a high NFC are more attracted to right–wing ideological attitudes, because these attitudes stress the importance of traditional values and the resistance towards change (Roets & Van Hiel, 2006). Of interest, NFC is more strongly related to authoritarianism than other types of ideological attitudes, like conservatism (Chirumbolo, 2002; Crowson, Thoma & Hestevold, 2005; Van Hiel, Pandelaere & Duriez, 2004). Authoritarianism may be especially functional in reducing ambiguity and providing a sense of certainty, because ‘RWA attitudes… would provide for individuals the strongest level of closure on social and political issues’ (Crowson et al., 2005, p. 574). In the present context, individuals with lower cognitive abilities may similarly benefit from more certainty and closure, which, in the context on ideas about society, are best met through authoritarian beliefs relative to other right–wing attitudes. Whereas cognitive styles may provide an explanation for the strongest effects found between cognitive ability and authoritarianism, we should note that the sensitivity bias analyses (Table 4) revealed that the actual effect size of authoritarianism might be somewhat lower, bringing them closer to the effect sizes of conservatism and dogmatism. Hence, we should interpret these findings with caution.

A second interesting finding concerned two types of prejudices. More specifically, ethnocentrism, which can be considered an indicator of generalized prejudice, yielded a stronger effect size than prejudices directed towards specific out–groups. In other words, cognitive ability related more strongly to negative attitudes and a dislike of other groups in general, rather than specific groups such as ethnic minorities or the lesbian, gay, bisexual and transgendered community. A similar effect can be found in research on the personality basis of prejudice. More specifically, Sibley and Duckitt (2008) found that (low) agreeableness or openness to experience is more strongly related to measures of generalized than specific prejudices. These authors argued that, because specific types of prejudice are partly rooted in knowledge about and (personal) experiences with these particular groups, these situational and group–specific factors might attenuate the generalized effect of personality (see also Akrami, Ekehammar & Bergh, 2011). Generalized prejudice, on the other hand, is more abstract and less influenced by group–specific attitudes and cognitions, yielding a stronger effect of personality. Similar processes can explain the present findings for cognitive ability as well. Cognitive ability might yield its strongest effects on generalized prejudice because of the reliance on general cognitive heuristics during judgment formation that is similar across all out–groups; in contrast, specific prejudice types are additionally influenced by factors uniquely related to the particular social group in question (including the social context).

Finding 4: Little evidence for publication bias or distortion by outliers

Given the controversial nature of studying the relationship of cognitive ability with right–wing ideological attitudes and prejudice, a crucial question is to what extent the studies included in the current meta–analyses represent the entire sample collection. It might be that studies retrieving null findings or positive findings are more likely to be put away in file drawers. A definite strength of this meta–analysis is that we addressed this issue by comprehensively addressing the impact of publication bias. These analyses, based upon visual inspection, regression and trim–and–fill methods, do not yield evidence that publication bias is likely in the present meta–analytic study collection. Nevertheless, it remains important for future updates of this meta–analysis to further search for additional unpublished studies. Despite our widespread call for unpublished data, only three researchers shared their unpublished data, and more unpublished datasets might be out somewhere.

Taken together, our sensitivity analyses accounting for publication bias and removing outliers/influential studies underscore and validate the robustness of our meta–analytical conclusions. Some other potential biases, however, could not be completely ruled out. First, it remains possible that the current meta–analytical conclusions do not reflect the true effects, owing to measurement error. Even though adjusting for measurement error is recommended by some meta–analysts (particularly in the psychometric meta–analytical tradition of Hunter & Schmidt, 2004; Kepes et al., 2013), we found insufficient information in the original studies to conduct such artefact adjustments. Also, we acknowledge that it is possible that the results of our meta–analysis are somewhat influenced by indirect range restriction (Hunter & Schmidt, 2004; Kepes et al., 2013). Many studies consist of volunteers, and volunteers may tend to have a higher–than–average cognitive ability. However, as information of the population variances on the outcome measures is essentially lacking in the original studies, it was not possible to correct for this artefact as part of the current sensitivity analyses.

Alternative accounts

Some scholars have criticized a pure ‘cognitive’ explanation of the negative relationship of cognitive ability with ideology and prejudice and suggest alternative accounts explaining these relationships. More specifically, social desirability, education and socio–economic status (SES) have often been proposed as explaining these effects. We elaborate on these alternative accounts in the next sections.

Towards new theories on the cognitive basis of right–wing ideology and prejudice

We introduce two new perspectives in the field of cognition and ideology, which hold promise for future research. First, we discuss the need to focus more research attention towards the interplay between cognitive ability and cognitive style in explaining ideology and prejudice. Second, we argue to study affective processes as well and discuss the interplay between cognition and affect in the context of ideology and prejudice.

Limitations of the current state of the literature

As a final note, we would like to address two limitations of the current state of the literature concerning cognitive ability, ideology and prejudice, which we think are important issues to tackle in future research. First, whereas a large focus is placed on social–cultural right–wing attitudes in relationship with cognitive ability, few studies have investigated economic–hierarchical attitudes in this context. We will further reflect on the possible relationship. Second, we discuss the need for cross–cultural validation of the obtained relationships.