Not All Education is Equally Liberal : The Effects of Science Education on Political Attitudes

Education stands as a potent predictor of political attitudes; however, the underlying mechanisms and moderators of this relationship are not well-understood. We hypothesize that the liberalizing effect of education is moderated by discipline, and that the scientific ethos that serves to guide empirical inquiries facilitates the development of more liberal political attitudes via concerns about fairness and equality. As predicted, being educated in a science-related discipline, as opposed to a non-science discipline, was associated with greater political liberalism; importantly, this effect could not be accounted for by self-selection (Study 1). Furthermore, concerns about fairness and equality, as captured by an individual’s social dominance orientation, mediated the relationship between studying science and political liberalism (Study 2). Study 3 replicated these findings and attest to their generalizability. Study 4 directly assessed the underlying mechanism, endorsement of the scientific ethos, and replicated the mediational model; those who endorsed the scientific ethos more strongly reported more liberal political attitudes, and this was mediated by their lower social dominance orientation.

A variety of studies have been conducted on the impact of college education on the values, beliefs, and behaviors across a variety of domains, including sexuality (e.g., at Aristotle University in Thessaloniki, Greece-Papaharitou et al., 2008), politics (e.g., at ten Southern California universities in the USA-McClintock & Turner, 1962; at the University of Missouri in St. Louis, Missouri, USA -Thumin, 1972; for a more recent review, see Hastie, 2007), environmentalism (e.g., at Bloomberg University in Pennsylvania, USA -Lang, 2011), and social attitudes (e.g., at a private prestigious university in the Eastern USA- Lottes & Kuriloff, 1994).The general consensus has been that a university education has a liberalizing effect, such that college seniors report more liberal attitudes than freshman (e.g., at Bennington College in Vermont- Newcomb, 1943Newcomb, /1957; for review, see Feldman & Newcomb, 1969a, 1969b).Cognitive theorists have attempted to explain this effect by positing that education promotes the ability to link new ideas, which in turn allows for a more liberal orientation (Gaasholt & Togeby, 1995).
Beyond the general liberalizing effect of higher education, existing research has also begun to examine how the specific content of one's education predicts attitudes and behavioral outcomes.Studies that have focused on attitudinal correlates of educational discipline have largely focused on political attitudes, and found that relative to those whose fields focus on corporations and profit, those whose fields focus on individuals and social groups tend to be more politically liberal (e.g., among members of professional groups, like APA psychologists -Spaulding & Turner, 1968; among participants of the Family Survey Dutch Population- Van de Werfhorst & de Graaf, 2004).
Additional studies have extended these findings to behavioral outcomes.Relative to other majors, economics students emphasize self-interest in ultimatum games and allocate smaller amounts of money to the responder than did arts and science students (e.g., at the College of the Holy Cross in Massachusetts, USA -Carter & Irons, 1991), display less cooperation in the prisoner's dilemma game (for review, see Frank, Gilovich, & Regan, 2000), donate less frequently to charitable causes (e.g., at the University of Washington, USA- Bauman & Rose, 2011), behave in less environmentally-friendly ways (e.g., at Bloomberg University in Indiana, USA -Lang, 2011), and show less favorable attitudes towards the disenfranchised (e.g., at an Ontario University, Canada -Guimond, Begin, & Palmer, 1989;Guimond & Palmer, 1996).Recent empirical research has attempted to account for these findings by demonstrating that students whose studies focus on the financial sector possess more positive and less immoral associations with greed (for review, see Wang, Malhotra, & Murnighan, 2011).
Along a slightly different vein, studies have also found that the content of one's college education influences more general tendencies towards openness to experience, with those in science fields displaying more openness than those majoring in other fields (e.g., at Saint Louis University in Missouri, USA -Chibnall, Blaskiewicz, & Detrick, 2009).These attitudes have been corroborated with additional studies that have found that both arts and science students tend be more tolerant and accepting of taboo sexual behaviors, including extramarital relationships, homosexuality, and controversial sexual practices (e.g., at Aristotle University in Greece- Papaharitou et al., 2008).
More generally, however, choice of major also influences the norms individuals rely on when making moral judgments (e.g., at a large regional public university in the Southeastern USA-Livingstone, Derryberry, King, & Vendetti, 2006) and attributions of causality (e.g., at an Ontario University, Canada- Guimond et al., 1989).
While there is considerable consensus regarding the reliable relationship between education and attitudes about the social world, an ongoing debate exists regarding the process that accounts for such a relationship.Two primary hypotheses have been put forth.On the one hand, the self-selection hypothesis posits that individuals select disciplines that closely match their preexisting worldviews and ideologies (Astin, 1979;Feldman & Newcomb, 1969a, 1969b).Existing findings concerning personality differences between individuals who choose different academic disciplines support this hypothesis (e.g., at a medium-sized public university in the Eastern USA -Pringle, DuBose, & Yankey, 2010), as does a growing body of research that suggests a broader link between cognitive ability and social attitudes (e.g., Hodson & Busseri, 2012;McClosky, 1958).
On the other hand, the socialization hypothesis posits that educational disciplines vary in their underlying ideological assumptions, and that participation in higher education alters the individuals' attitudes to match those in their chosen discipline.Paz-y-Miño and Espinosa (2009) examined the difference in views on evolution between biology and non-biology majors at Roger Williams University in Rhode Island, USA.As would be expected, biology majors scored higher on beliefs in evolution than non-biology majors.However, they also found an increase of acceptance for evolution within the biology major that increased with class affiliation.This suggests that as students progressed in this major, they became more oriented towards one of its central ideas-evolution.Lang (2011) found a similar pattern when comparing the political values between incoming business majors and non-business majors at Bloomberg University in Indiana, USA.His findings suggest that business and economics majors score lower on environmentalism and awareness compared to other majors, with a strengthening of beliefs from freshmen to seniors.
In sum, the existing literature has compellingly demonstrated that educational attainment plays an important role in arbitrating attitudes towards a variety of social objects and, most notably, influences political orientation.Apart from this general consensus, however, the aforementioned research is largely equivocal on the conditions under which education exerts a liberalizing effect and the underlying process that accounts for this relationship.While various studies have examined the effect of studying discipline-specific constructs on attitudes towards that construct (e.g., biology major's beliefs concerning the evolutions of species-Paz-y-Miño & Espinosa, 2009; business majors' attitudes towards profit and economic gain- Wang et al., 2011; social science major's attitudes towards social change and status hierarchies- Guimond & Palmer, 1990), few have extended beyond discipline-specific relationships in order to investigate higher-order processes that account for the education-politics link.Is there a broader ethos that characterizes the sciences and that distinguishes it from the humanities?To what extent do these differences moderate the relationship between educational attainment and political attitudes?
Psychologists have long noted that personal values influence both the questions that are asked across academic disciplines and the methods used in arriving at the answers.Many scholars have consistently been concerned with the moral and social ramifications of their empirical endeavors (Howard, 1985;Kurtines, Alvarez, & Azmitia, 1990) and concede that intellectual inquiries are value-laden (Becker, 1968;Bhaskar, 1975;Fiske & Shweder, 1986;Howard, 1985;Toulmin, 1953).Given that the scientific method of acquiring information about the world uses empirical models based on impartiality, rationality and fairness, we predict that the study of scientific disciplines in particular facilitates lower social dominance-that is, less endorsement of the belief that social inequality between groups is acceptable (Pratto, Sidanius, Stallworth, & Malle, 1994)-and greater political liberalism.
We hypothesized that the same scientific ethos that serves to guide empirical inquiries also facilitates the development of more liberal political attitudes via concerns about social dominance and hierarchy.Given science's emphasis on rationality, impartiality, fairness, progress, and the idea that we are to use these rational tools for the mutual benefit of all people in society (Taylor, 2004), we predicted that studying science should lead to greater rejection of legitimizing myths that promote existing hierarchies and dominant social orientations.Given that social dominance hierarchies are maintained by the belief that existing social inequalities between groups are natural and acceptable (i.e., hierarchy-enhancing (HE) legitimizing myths), and undermined by the belief that such social inequalities are unacceptable (i.e., hierarchy-attenuating (HA) legitimizing myths; Sidanius, Van Laar, Levin, & Sinclair, 2003), we predicted that the impartiality of science would render it more hierarchy-attenuating than hierarchy-enhancing.
Importantly, our prediction that social dominance orientation would emerge as a mediator in accounting for the predicted link between science education and political liberalism is consistent with current conceptualizations of SDO-namely, the Group Socialization Model, which posits that people's social dominance orientation changes as a function of their social context (Guimond, Dambrun, Michinov, & Duarte, 2003).Empirical studies have supported this notion that SDO changes as a function of a variety of social factors, including power/status (e.g., belonging to subordinate vs. dominant groups- Sidanius & Pratto, 1999;Sinclair, Sidanius, & Levin, 1998), prestige (e.g., studying business and law vs.other fields- Sidanius, Pratto, Martin, & Stallworth, 1991;Guimond et al., 2003), and information about the role of nature vs. nurture in shaping behavior (e.g., believing in genetic determinism as a function of exposure to psychology- Dambrun, Kamiejski, Haddadi, & Duarte, 2009).
The present research builds upon the Group Socialization Model of SDO (Guimond et al., 2003) by examining the role of studying "hard" science versus social science or non-science fields on individuals' social dominance and, subsequently, their political attitudes.The present studies diverge from previous findings related to socialization processes involved in the development of SDO, however, in two important ways.First, our focus is on comparing a different array of disciplines-i.e., science vs. non-science-rather than focusing on the potential prestige or opportunity for financial gain between different disciplines (i.e., business and law versus other fields).
Second, we are interested in the repercussions of disciplines and SDO on a wide range of political attitudes, which previous studies have not tested.Third, we aim to revisit the issue of whether socialization or self-selection can account for the relationship between academic discipline and SDO.Given the mixed evidence on whether selfselection processes (e.g., Norman & Redlo, 1952;Pringle et al., 2010;Sidanius et al., 2003) or socialization forces (e.g., Lang, 2011;Paz-y-Miño & Espinosa, 2009) account for discipline-specific effects on social attitudes, further investigation is needed to understand the conditions under which these differential processes are at work.We empirically examined the link between educational discipline and political attitudes across four studies.We predicted that being educated in a hard science discipline (i.e., one that relies on the systematic use of the scientific method as the primary way of acquiring information about the world), as opposed to a non-science or social science discipline, would predict greater political liberalism, and that this effect would not be accounted for by solely self-selection (Studies 1, 2).Furthermore, we predicted that an individual's social dominance orientation (SDO-Pratto et al., 1994), would mediate the relationship between studying science and political liberalism (Study 2, 3).In Study 4, we directly assessed the predicted underlying mechanism presumed to drive these effects, the scientific ethos.We tested the extent to which the group difference findings from Studies 1-3 would replicate using individual differences in the endorsement of the scientific ethos to predict political liberalism.

Study 1
Study 1 examined the relationship between individuals' choice of academic discipline and political attitudes.Undergraduate students across disciplines completed a questionnaire concerning their choice of college concentration and their attitudes towards a variety of political issues, including healthcare (e.g., equal healthcare accessibility), the economy (e.g., government regulations of the free market), foreign policy (e.g., U.S. support of allies), and social issues (e.g., abortion).We predicted that individuals in the hard sciences, or those who rigorously use the scientific method as the primary way of acquiring information, would exhibit more politically liberal attitudes across domains relative to individuals in the non-sciences and more social sciences.Specifically, we predicted that this effect would emerge among those who had substantive exposure to discipline-specific education (i.e., upper years), but not among those who had not (i.e., lower years).

Method Participants and Procedure
One hundred and ninety-six students from a university in New England participated voluntarily in an online study.
The average age was 20.60 years.For a complete breakdown of participants' major by gender and year, please see Table 1.Participants were recruited from the listservs of campus dormitories and received entry into a drawing for six $25 Amazon.comgift cards in return for participation.Study 3 relied on national sample of both current college enrollees and graduates, and thus class status in college was not computed given that a portion of the sample had graduated.

Measures
Participants completed a political attitudes questionnaire that included 18 statements reflecting perspectives on current political issues.
i The statements were varied to express both liberal and conservative beliefs (e.g., "Welfare programs make people dependent on the government for provisions" and "Welfare is necessary to bring fairness to American economic life"; see Appendix).Participants rated their level of agreement with each statement on a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree).The scale was created by the authors to assess salient political issues that divided U.S. political liberals and conservatives.Composite scores for each of the four subcategories of political attitudes were computed.Responses to statements reflecting conservative attitudes were reversed coded so that for all composites, higher scores reflect more liberal attitudes.Given that we did not have any a priori hypotheses concerning the particular type of political attitude, they were averaged to yield an overall measure of political attitudes (α = .58;M = 3.89, SD = 0.43). ii In addition, participants entered their intended or declared major (i.e., "what is your intended or declared major?") and year in college, along with basic demographic information (i.e., age, gender).Participants' choice of discipline was coded as science, social science, or non-science.Disciplines that relied on the systematic application of empirical methods to construct consistent, predictive models of the physical and natural world (e.g., Gilbert, 1991) were coded as "science"; science disciplines included majors that belonged to the following six scientific fields: all forms of biology (50%), psychology (16%), chemistry (10%), all forms of physics (10%), all forms of engineering sciences (10%), and environmental science (4%).All other disciplines that did not meet this criteria were coded as social science or non-science; disciplines that focused on social phenomena but that nevertheless relied on systematic methods were coded as social science, and those that did not were coded as non-sciences.Social science disciplines included five majors: government/political science (38%), economics (25%), social studies (15%), sociology (19%), linguistics (2%).Majors that did not qualify as either a science or a social science were coded as non-sciences, which included a wide array of different majors, the most prevalent of which included history (38%), various languages and literatures (23%), and various forms of comparative studies (21%).
iii Finally, exposure (to discipline) was operationalized using year in college; first and second years were categorized as low exposure/lower years whereas third and fourth years were categorized as high exposure, or upper years. iv

Results
There was no effect of gender, and no interactions involving gender on political attitudes (both Fs < 1); thus, gender was not included in the analyses presented below.

Discussion
Study 1 offered preliminary evidence for a divergence in political attitudes between individuals in hard sciences vs. social science and non-science disciplines.Across domains, those who are in scientific fields exhibited greater political liberalism compared to those in non-hard-scientific fields.Furthermore, we cast strong doubt on the self-selection explanation that posits the role of pre-existing individual differences in accounting for these effects.
By demonstrating that the link between scientific study and political liberalism only holds for those who had relatively high levels of exposure to discipline-specific education (i.e., third and fourth years), but not those who had yet been extensively exposed (i.e., first and second years), these findings suggest that discipline-specific socialization processes-and not merely self-selection-account for the relationship between academic discipline and political attitudes.However, the question remains as to the underlying process that accounts for such a relationship.

Study 2
As a psychological construct, political liberalism has been linked to a number of different variables, including measures of resistance to change (e.g., openness to experience, uncertainty tolerance, integrative complexity, need for order/structure/closure, fear of threat and loss, dogmatism, intolerance of ambiguity-for review, see Jost, Glaser, Kruglanski, & Sulloway, 2003), attitudes towards inequality and social hierarchy (e.g., social dominance orientation- Pratto et al., 1994), and general cognitive abilities (e.g., intelligence or g- Schoon et al., 2010), to name a few.We posit that among these different correlates of political liberalism, social dominance orientation (SDO) should be particularly relevant to disciplinary differences between the sciences and non-sciences.Past work by Sidanius et al. (1991) has found that certain majors (i.e., business, law) were higher in SDO and racism than those in other majors, and that such an effect was moderated by college exposure; furthermore, Van Laar, Sidanius, and Levin (2008) found SDO to decrease as a function of exposure to specific courses.We contend that SDO should be particularly relevant here because science's emphasis on rationality, impartiality, fairness, progress, and the idea that we are to use these rational tools for the mutual benefit of all people in society (Taylor, 2004) should lead to greater rejection of hierarchy legitimizing myths and dominant social orientations.Existing research has shown that social dominance orientation (SDO) is a predictor of political attitudes (Cohrs, Moschner, Maes, & Kielmann, 2005;Pratto et al., 1994;Pratto, Stallworth, & Conway-Lanz, 1998;Pratto, Stallworth, & Sidanius, 1997;Mirisola, Sibley, Boca, & Duckitt, 2007;Van Hiel & Mervielde, 2002;Whitley & AEgisdóttir, 2000).
Given that education has been shown to reduce the expression of SDO (Sidanius, Pratto, & Bobo, 1996;Sidanius, Sinclair, & Pratto, 2006), we predicted that the effects of a science education on political liberalism would be mediated by social dominance, such that those who study science would be lower in SDO, and lower SDO would predict more liberal political attitudes.
To account for the possibility that other mechanisms may explain the link between science and political liberalism, we also tested a potential alternative mediator, individual differences in resistance to ambiguity and change.Like SDO, resistance to change has been consistently linked to political attitudes, and numerous studies using various operationalizations of this construct (e.g., openness to experience, uncertainty tolerance, integrative complexity, need for order/structure/closure, fear of threat and loss, dogmatism, intolerance of ambiguity) have found convergent evidence that those more resistant to change tend to be more politically conservative (for review, see Jost et al., 2003).Given that different disciplines rely on distinct methods for drawing conclusions and building models, an alternative explanation could be that the scientific method-which encourages the use of empiricism to continually test the validity of our existing models and their boundary conditions-promotes less aversion to ambiguity and change.Furthermore, this relationship between how individuals react to ambiguous or conflicting information and

Method Participants and Procedure
One hundred undergraduate students from the same university in New England as in Study 1 (78% female) participated voluntarily in the study (see Table 1).The average age was 20.36 years.Participants were recruited from the listservs of campus dormitories and entered into a raffle for six $15 gift cards for local food-based establishments in return for participation.
Participants subsequently rated their agreement with three current controversial political issues: same-sex marriage, affirmative action, and the Affordable Care Act.These specific issues were chosen because of their relevance to the current political sphere and bipartisan nature (i.e., they are contested issues that tend to divide American liberals and conservatives).Participants indicated their support for each issue on a scale from 1 (strongly do not support) to 7 (strongly support).Responses on the three political issues showed acceptable levels of internal consistency (α = .68)and thus were averaged to form an overall political composite, with higher values reflecting greater liberalism (M = 5.27, SD = 1.29).
Finally, participants indicated their intended or declared college major from a list of all majors offered and their year in college, along with demographic information (age, gender, race, and subjective socioeconomic status).
Participants' majors were coded as hard science, social science, or non-science using the same criteria from the previous study; representation of majors were similar to that of Study 1: among the sciences, biology (51%) and psychology (33%) were by far the most represented; among the social sciences, political science/government (33%), economics (28%), and social studies (28%) were the primary fields represented; among the non-sciences, history (39%), comparative studies (19%), and various languages/literatures (13%) were the most commonly represented.Participants' exposure was also coded as high or low using the same criteria. v

Results
There was no effect of gender and no gender by major interaction on political attitudes (ps > .28).However, gender differences emerged on the mediator, SDO; women displayed lower levels of SDO than men, F(1, 87) = 5.41, p = .022.However, because we did not make any a priori predictions about gender, gender was not related to the outcome variable of interest, and gender did not interact with the predictor variable, gender was not included in the analyses.
A univariate ANOVA revealed a marginal effect of major on political attitudes, F( 2 To test the possibility of mediated moderation, discipline (hard science vs. non-hard science), year (upper vs. lower classmen) and SDO were entered into a bootstrap analysis predicting political attitudes, as recommended for mediation analyses with small samples (see Hayes, 2013).In these analyses, mediation is significant if the 95% bias corrected and accelerated confidence interval (CI) for the indirect effect does not include zero (Preacher & Hayes, 2004;Preacher, Rucker, & Hayes, 2007).Results based on 1,000 bootstrapped samples indicated that the conditional direct effects of discipline on political attitudes for upper and lower classmen were non-significant (B = . .Taken together, these findings suggest that at least for those with significant exposure to their discipline (i.e., upperclassmen), studying science is associated with more liberal political attitudes because studying science leads to lower levels of social dominance orientation.
We also tested the alternative mediator, individual differences in resistance to ambiguity and change (i.e., dialecticism).While this individual difference measure significantly predicted political attitudes, as expected based on the past literature (r = -.21,p = .036),it did not predict choice of major between science vs. non-science (r = .057,p = .57).
In addition, we examined demographic variables like race and socioeconomic status as potential confounds given past findings that have suggested factors like race and class to be linked to both participation in specific academic fields (e.g., Lubienski, 2001;Tyson, Lee, Borman, & Hanson, 2007), and political spheres (e.g., Gilroy, 1991).
Neither race nor class predicted choice of major or political attitudes (all ps > .21).

Discussion
Together, Studies 1 and 2 provided convergent evidence that studying science, relative to studying the non-sciences or social sciences, promoted more politically liberal attitudes across a variety of domains.However, both studies were conducted using samples from a singular New England University.Thus, to test the generalizability and replicability of the effects, Study 3 was conducted using a more diverse sample of participants and a more general measure of political liberalism.

Measures
As in the previous study, participants completed a measure of SDO (Pratto et al., 1994; M = 2.66, SD = 1.11; α = .72),along with information about their college and major.In addition, participants rated the degree to which they were political liberal on a scale from 1 to 7, with higher values reflecting greater liberalism (M = 4.87, SD = 1.51).Self-assessed political liberalism has been reliably used in many existing studies as a valid proxy for political orientation (e.g., Graham, Haidt, & Nosek, 2009;Inbar, Pizarro, & Bloom, 2009;Kanai, Feilden, Firth, & Rees, 2011;Lee, Sohn, & Fowler, 2013) that converges highly with other measures of political liberalism, including implicit measures of political liberalism (e.g., Graham et al., 2009) and attitudes on a wide range of political issues (e.g., Inbar et al., 2009).

Results
There was no effect of gender and no gender by major interaction on the dependent variable of interest, political liberalism (both Fs < 1).However, there was a marginal effect of gender on the mediator, SDO; women displayed somewhat lower levels of SDO than men, t(55) = 1.88, p = .065.However, given that we did not make any a priori predictions about gender, gender was not related to the outcome variable of interest, and gender did not interact with the predictor variable, gender was not included in the analyses.
A univariate ANOVA revealed a significant effect of major on political attitudes, F( 2 To test whether the effects of major on political liberalism were mediated by SDO, we conducted the following regressions comparing hard science and non-hard-science majors.In the first regression equation, political liberalism was regressed on major (0 = hard science, 1 = non-hard science).This regression equation is similar to the ANOVA results reported above; as expected, major significantly predicted political liberalism, β = -.34,p = .009.
In the second regression equation, SDO was regressed on major.Those who studied science showed lower SDO relative to those who studies non-science majors, β = .29,p = .043.Finally, in the third regression equation, political liberalism was regressed on major and SDO simultaneously.The results indicate that the effect of SDO remained significant, β = .46,p = .001,whereas the effect of major did not, β = -.19,p = .14.

Education and Political Attitudes 152
As recommended for small samples, we used nonparametric bootstrapping analyses (see Preacher & Hayes, 2004;Preacher et al., 2007) to test the significance of SDO as a mediator of the relationship between discipline and political liberalism.In these analyses, mediation is significant if the 95% bias corrected and accelerated confidence interval for the indirect effect does not include zero (Preacher & Hayes, 2004;Preacher et al., 2007).
Results based on 1,000 bootstrapped samples indicated that the total effect of major on political liberalism was significant (B = -1.05,SE = 0.44, p = .021),as was the direct effect (B = -.64,SE = 0.17; p =.0008).SDO mediated the relationship between discipline and political liberalism (CI [-.98, -.041]), such that participants who majored in the sciences showed lower levels of SDO, and through less SDO, they were more politically liberal.Because zero was not in the 95% confidence interval, the indirect effect was significantly different from zero at p < .05(twotailed).Taken together, these findings are consistent with the hypothesis that majoring in a science is associated with increased political liberalism because doing so makes you exhibit less of SDO.

Discussion
Studies 1-3 relied on choice of major to predict political attitudes, and found convergent evidence that majoring in a science-related field led to greater political liberalism.In explaining this effect, it was theorized that majoring in a science served as a proxy for endorsement of the broader scientific ethos, and endorsement of this ethos drove the effects on political attitudes.In order to directly assess the assumed underlying factor, endorsement of the scientific ethos, and assess the extent to which individual differences in this ethos would lead to the same results observed in the previous studies, Study 4 was conducted.

Study 4
In line with past research on other dimensions (e.g., interdependence- Markus & Kitayama, 1991;Singelis, 1994; dialecticism-Chen, English, & Peng, 2006) that have highlighted the utility of employing both the individual-level and group-level approach for studying the same construct, Study 4 examined whether individual differences in endorsement of the scientific ethos would parallel the group-level effects between science and non-science majors.
That is, Study 4 did not measure major and instead, measured belief in the scientific ethos as a predictor of political liberalism.An additional aim of Study 4 was to more stringently test the generalizability of our findings by relying on a different sample than those used in the previous three studies.Given that Studies 1 and 2 were both conducted at the same university, and Study 3 relied on an online sample, these offered limited opportunity to test the extent to which the present findings would generalize to other populations.Thus, Study 4 was tested on two new samples: undergraduates from a Canadian university and a community sample from the larger Boston area.

Measures
Participants completed a series of four questions about their endorsement of the scientific ethos that were specifically devised for the present study by the authors: "How much do you believe in science?"; "To what extent do you think that science is used for the benefit of society?"; "To what extent do you believe in rationality (as opposed to subjectivity)?" and "To what extent do you believe in impartiality and fairness as the best ways to acquire knowledge?"Participants responded to all questions on a 1 (not at all) to 5 (very much) scale.Responses to the four questions were averaged (α = .63)to yield an overall measure of scientific ethos endorsement.Higher scores reflected greater endorsement of science (M = 4.26, SD = 0.56).
In addition, participants completed the same SDO scale (Pratto et al., 1994

Results and Discussion
Gender did not predict self-reported political liberalism, or endorsement of the scientific ethos (ps >.19); however, gender was marginally associated with SDO, r = -.23,p = .059.Women, on average, displayed lower levels of SDO than men.However, given that gender was not linked to our predictor or outcome variables, it was not included in the analyses.
As predicted, endorsement of the scientific ethos predicted greater self-reported political liberalism, β = .27,t(104) = 2.88, p = .005.vii Those who endorsed the scientific ethos self-reported to be more politically liberal.To test the replicability of the previous finding that the effects of science on political liberalism were mediated by SDO, we conducted the following regressions using self-reported political liberalism.
In the initial regression equation, political liberalism was regressed on endorsement of the scientific ethos, reported above.In the second regression equation, SDO was regressed on scientific ethos endorsement.Those who endorsed science more showed lower SDO relative to those who endorsed science less, β = -.21,t(116) = 2.30, p = .023.Finally, in the third regression equation, political liberalism was regressed on scientific ethos endorsement and SDO simultaneously.The results indicate that the effect of SDO remained significant, β = -.36,t(104) = 3.86, p <.001, whereas the effect of scientific ethos endorsement was not, β = .15,t(104) = 1.69, p = .094.
Once again, we used nonparametric bootstrapping analyses (see Preacher & Hayes, 2004;Preacher et al., 2007) to test the significance of SDO as a mediator of the relationship between science endorsement and political liberalism.In these analyses, mediation is significant if the 95% bias corrected and accelerated confidence interval for the indirect effect does not include zero (Preacher & Hayes, 2004;Preacher et al., 2007).Results based on 1,000 bootstrapped samples indicated that the total effect of scientific ethos endorsement on political liberalism was significant (B = .84,SE = 0.29; p = .0048),but the direct effect was not (B = .48,SE = 0.29; p =.094).SDO mediated the relationship between scientific ethos endorsement and political liberalism (CI [0.11, 0.67]), such that participants who endorsed science more showed lower levels of SDO, and through less SDO, they were more politically liberal.Because zero was not in the 95% confidence interval, the indirect effect was significantly different from zero at p < .05(two-tailed).Taken together, these findings are consistent with the hypothesis that endorsing science is associated with increased political liberalism because the scientific ethos makes you exhibit less of SDO.

General Discussion
Across four studies, we demonstrated the discipline-specific liberalizing effects of higher education.Relative to those studying non-sciences, students in the sciences exhibited greater political liberalism across a variety of domains (including foreign policy, healthcare, the economy) and a variety of social issues (e.g., gay marriage, affirmative action), as well as in general self-reported liberalism.These effects could not be accounted for merely by self-selection, but rather, emerged as the product of discipline-contingent socialization pressures.Furthermore, we elucidated one underlying mechanism that accounts for the relationship between science education and political liberalism, and demonstrated that those who study scientific disciplines are more politically liberal because they exhibit lower levels of social dominance orientation.Importantly, we replicated this mediational model using individual differences in endorsement of the scientific ethos.
These findings have several important implications.First, they offer a more nuanced look into the relationship between educational attainment and political attitudes.Although the existing literature has long noted that greater time spent in higher education is a powerful predictor of liberal political orientations (Feldman & Newcomb, 1969a, 1969b;Newcomb, 1943Newcomb, /1957)), existing studies on the precise nature of the attitudinal effects of being educated have been limited in two major ways.First, they have been equivocal on the question of whether such effects are a product of self-selection or socialization (e.g., Hastie, 2007;Lang, 2011;Norman & Redlo, 1952;Paz-y-Miño & Espinosa, 2009;Pringle et al., 2010).Second, they have largely been limited to examining attitudes towards discipline-specific constructs (e.g., evolution of species, business profit, etc.).The present research addresses these questions left by the current literature and offers both more definitive support for the discipline-specific socializing effects of education and keener insight into the broader sociopolitical ramifications of such educational socialization.
Second, the present results suggest that the study of science itself holds normative implications and leads to political outcomes.Previous research has noted that science can be value-laden insofar as the extent to which personal values influence both the questions that are asked and the methods used in arriving at the answers (Becker, 1968;Bhaskar, 1975;Fiske & Shweder, 1986;Howard, 1985;Kurtines et al., 1990;Toulmin, 1953).
These findings suggest that beyond what values scientists may introduce to the process of scientific investigation, the act of studying science itself produces important psychological and political consequences.These findings suggest that the same scientific ethos that serves to guide empirical inquiries also facilitates attitudes towards social dominance and hierarchy-attenuating myths more broadly.Consequently, these results can contribute to a better understanding socio-political attitudes, including how institutions influence the maintenance of existing social hierarchies.
Third, our findings offer further evidence for the Group Socialization Model of Social Dominance, or the conceptualization of social dominance orientation as a mediator that changes as a function of social context (Guimond et al., 2003).The present findings are consistent with previous empirical studies that have similarly demonstrated that SDO shifts as a function of the social environment (Dambrun et al., 2009;Guimond et al., 2003;Sidanius et al., 1991;Sidanius & Pratto, 1999;Sinclair et al., 1998) and has a mediating effect on the relationship between discipline and social attitudes (Dambrun et al., 2009;Guimond et al., 2003).
Education and Political Attitudes 156 assess the degree of political orientation change throughout college.The cross-sectional nature of our studies may be particularly limiting insofar as they do not capture what happens in cases where students change their discipline, and the present studies did not assess this possibility.Similarly, the correlational nature of our findings does not rule out the alternative explanation that intelligence may be driving the observed effects (i.e., the possibility that more intelligent students opted for the hard sciences, and greater intelligence accounted for their increased political liberalism).The present studies did not measure intelligence and thus could not rule out this possibility; however, the findings that scientific ethos endorsement predicts political liberalism via SDO is strongly suggestive that there is something specific to subscribing to science that impacts political orientation.More longitudinal research methods are needed to assess how political orientation changes as a function of time spent in specific disciplines.
One final limitation of the present research is the reliance on explicitly reported political attitudes.Given the relative liberal environment of college campuses in general, one possibility is that disciplinary differences in social desirability concerns exist, and those in science-related disciplines explicitly reported greater political liberalism and lower social dominance because of such concerns regarding what is socially desirable.Future studies can more closely tackle these limitations by examining implicit political attitudes and their link to the explicit political attitudes reported her in order to investigate whether the disciplinary divergence in political orientation also extends to implicit measures of political orientation.

Conclusion
In conclusion, the present work elucidates an important feature of how and when higher education emerges as a potent predictor of political attitudes.Four studies have offered convergent evidence that the liberalizing effect of education is moderated by discipline; those majoring in the hard sciences tend to display more liberal political attitudes than those in other disciplines.The scientific ethos that serves to guide empirical inquiries in the hard sciences facilitated the development of more liberal political attitudes via concerns about fairness and equality.

Notes
i) Participants also completed additional questions on their attitudes towards the environment, but these were not included in the political liberalism composite because they did not theoretically reflect liberal versus conservative political attitudes.
ii) Given the suboptimal alpha, we also ran the analysis including the nature of the political topic as a factor to test if the effects are specific to certain political domains but not others.We conducted a mixed model ANOVA with major and class status (upper vs. lower years) as between-subject participant factors, along with subtopic as a within-subject participant factor, in predicting political attitudes.The results were consistent with the primary findings described in the main text (i.e., a main effect of major, F(2, 177) = 3.60, p = .029,as well as an interaction between major and exposure, F(2, 177) = 4.03, p = .019).In addition, a main effect of subtopic emerged, F(3, 531) = 41.47,p < .001;an examination of the means shows that participants were more liberal overall on topics related to healthcare and social issues than the economy or foreign policy (Ms = 3.83,3.92,3.99,4.14 and SDs = 0.37,0.38,0.39,0.23,respectively).However, no interactions involving subtopic emerged (all Fs < 1).
iii) There were five cases where a participant indicated a double major; in all such cases, both majors belonged to the same discipline: history and East-Asian studies (N = 1), history and literature (N = 2), chemistry and physics (N = 1), and physics and astrophysics (N = 1).The breakdown of majors represented within fields reflects the first major indicated in these cases where two majors were included.
v) We collapsed across first/second years and third/fourth years for the same reasons outlined in Study 1.In addition, we re-ran our analysis including year in college as a factor; the same pattern of results emerged as those reported in the main text: a marginal effect of discipline emerged, F(2, 88) = 2.73, p = .071,and a significant interaction between discipline and year, F(6, 88) = 2.45, p = .031.Examination of the means indicate that for science majors, liberalism generally increased over the years (Ms = 5.41,5.12,6.58,6.05 and SDs = 0.88,1.13,0.88,0.69,respectively),whereas for the other disciplines, this pattern did not emerge (Ms = 4.77,5.83,5.95,5.12 and SDs = 0.50, 0.43, 1.57, 1.05 for the non-sciences; Ms = 4.66,4.00,6.05,4.61 and SDs = 2.50,2.01,0.70,1.43 for the social sciences).
vi) There were four cases of double majors, two that involved majors within the same discipline (anthropology/political science, english/history) and two that did not (psychology/english; psychology/history).In the two cases where one of the majors was a hard science and the other was not, they were coded as hard science, given that our hypothesis centers on exposure to science and double majors concentrating on a hard science meets that criteria.As with Study 1, representation of the different majors within a discipline reflect the first major indicated in all such cases of double majors.
vii) also ran these analyses with sample (i.e., York University undergraduates; Decision Science Lab community participants) as a covariate.While the two samples did not differ in political liberalism, t(104) = 1.62, p = .11,or SDO, t(116) = 1.63, p = .11,they marginally differed in scientific ethos endorsement.Those from the Decision Science Laboratory exhibited greater endorsement of science (M = 4.31, SD = 0.58) than those from York University (M = 4.11, SD = 0.49), t(115) = 1.72, p = .089.However, including sample as a covariate in the analyses did not change the pattern of findings; belief in science remained a significant predictor of political liberalism, β = .26,t(104) = 2.74, p = .007.
political attitudes has been confirmed with neuroimaging data(Amodio, Jost, Master, & Yee, 2007).Thus, one possibility is that individual differences in how people respond to ambiguity, contradiction, and change could play an important role in accounting for the link between studying science and political attitudes.To test this alternative explanation, we used an established individual difference measure that captures general attitudes towards ambiguity, contradiction and change (i.e., the Dialecticism Scale-Choi & Choi, 2002;Spencer-Rodgers, Peng, Wang, & Hou, 2004); we subsequently examined its relationship to individuals' selection of major and political attitudes.
eighteen participants (35% men) from the subject participation pools at York University in York, Canada (N = 33) and the Decision Science Lab in Cambridge, Massachusetts (N = 85) participated as part of larger studies.The former consisted of undergraduates enrolled in introductory psychology courses participating for course credit.The latter consisted of a community sample of paid participants recruited from the larger Boston area; to allow us to test the generalizability of the previous findings, participants who were not undergraduates at the New England university sampled in Studies 1 and 2 were specifically recruited from the community in this latter sample.

Table 1
Breakdown of participants for studies by field, gender, and class.