Can the relationship between higher education and

cosmopolitanism be explained by field of study?

Rahsaan Maxwell

Department of Political Science

University of North Carolina at Chapel Hill

[email protected]

October 12, 2020

Acknowledgements

Cameron Ballard-Rosa and Marc Hetherington provided helpful comments on previous ver- sions.

1 Abstract

One of the most divisive debates of our time is whether should be more or less cosmopolitan. One for the divisiveness is the overlap between cosmopoli- tanism and societal cleavages. For example, post-secondary qualifications are strongly associated with greater cosmopolitanism. This relationship is remarkably ro- bust but it is unclear whether educational attainment directly affects cosmopolitanism or whether pro and anti-cosmopolitan people sort into different educational trajec- tories. In this paper, I explore whether contradictory findings can be explained by previous studies mis-specifying the aspect of education that affects cosmopolitanism.

I use the Dutch LISS Panel to focus on field of study; an important source of vari- ation within post-secondary education. I find that respondents with humanities or social sciences degrees are more cosmopolitan than respondents with physical science or technical degrees. However, controlling for field of study cannot account for the cosmopolitanism gap across educational attainment. Moreover, there is no evidence that a post-secondary degree in humanities or social science subjects causes more cos- mopolitanism. It is more likely that cosmopolitan people sort into humanities and social science subjects. This has implications for our understanding of the relationship between cosmopolitanism and education as well as the formulation of worldviews more broadly.

2 1 Introduction

One of the most divisive debates across Western Europe and North America is whether societies should be more or less cosmopolitan (Kriesi et al. 2012). The central principle of cosmopolitanism is that are part of one (Appiah 2006). Cosmopolitans believe humans should not be limited by -state boundaries and societies should be open to immigration, , and robust international institutions (Beck and Grande 2007; Maxwell 2019). Opponents of cosmopolitanism believe societies should limit ethnic diversity, privilege national sovereignty, and maintain historical . The debate be- tween these perspectives is so intense, it may restructure social cleavages across Europe and North America (de Wilde et al. 2019; Hooghe and Marks 2018). One reason for the divisiveness of cosmopolitanism is its overlap with other societal cleav- ages. Most notably, studies show that higher educational qualifications are associated with more support for cosmopolitanism (Hainmueller and Hiscox 2007; Hakhverdian et al. 2013; Igarashi and Saito 2014). This is a remarkably robust relationship, and is consistent over time and across country contexts. However, there is an ongoing debate about how to inter- pret the relationship between educational attainment and cosmopolitanism. Some scholars find evidence that educational attainment has a direct causal effect on cosmopolitanism (Cavaillé and Marshall 2019; d’Hombres and Nunziata 2016; Margaryan, Paul and Siedler 2019). Other studies suggest the divide is due to pro and anti-cosmopolitan people sorting into different educational trajectories (Finseraas, Skorge and Strøm 2018; Kunst, Kuhn and van de Werfhorst 2020; Lancee and Sarrasin 2015). In this paper, I explore whether these contradictory findings can be explained by previ- ous studies mis-specifying the aspect of education that affects cosmopolitanism. The logic of the causal relationship is that higher education provides information, cognitive skills and exposure to a broad range of people, all of which sensitize students to the value of cosmopoli- tan diversity (Nie, Junn and Stehlik-Barry 1996; Stubager 2008; Surridge 2016). However, scholars may have overlooked important variation within post-secondary education in how

3 information, skills and operate. The key variation within post-secondary education that I examine is field of study. Post- secondary education includes a wide range of subjects, from art history to mechanical engi- neering. Within that range, humanities and the social sciences are the most likely to promote contextual understanding of social issues, which may be directly compatible with an openness to diversity and cosmopolitanism. In comparison, physical sciences and technical subjects are less likely to engage social issues and are more likely to rely on formal codified rules (Guimond, Begin and Palmer 1989; Ladd and Lipset 1975; Persson 2012). This does not im- ply that physical sciences and technical subjects are associated with anti-cosmopolitanism. Instead, the key insight is that humanities and the social sciences are more strongly related to cosmopolitanism. Therefore, the relationship between post-secondary education and cos- mopolitanism might only be present (or might be stronger) for people who study humanities or social sciences. I use the Dutch Longitudinal Studies for the Social Sciences (LISS) panel. The longitudinal data structure is particularly useful for my inquiry because it allows me to observe cosmopolitanism before and after respondents receive post-secondary degrees in dif- ferent fields of study. This provides leverage on whether education affects cosmopolitanism in ways that would not be possible with cross-sectional data. The LISS panel is also useful because it provides information on respondents’ field of study; a level of detail not often found in large panel studies. The LISS panel also allows me to analyze multiple indicators of cosmopolitanism: immigration attitudes, support for the European Union (EU), support and social contacts with non-Dutch people. I find that field of study is associated with cosmopolitanism. Respondents with hu- manities or social sciences degrees are generally more cosmopolitan than respondents with physical science or technical degrees. In addition, the cosmopolitanism gap across fields of study is roughly similar to the cosmopolitanism gap between respondents with and without post-secondary education. However, controlling for field of study cannot account for the

4 cosmopolitanism gap across educational attainment. Moreover, there is no evidence that getting a post-secondary degree in humanities or social science subjects causes more cos- mopolitanism. It is more likely that cosmopolitan people sort into humanities and social science subjects. The results in this paper have several implications. First, by examining field of study, I offer a new perspective for the established literature on education and cosmopolitanism. The main focus of previous research has been educational attainment, but I show that respondents who do not have post-secondary degrees but study the humanities or social sciences are just as cosmopolitan as respondents with post-secondary degrees in physical sciences or technical subjects. If we want to understand how educational stratification relates to cosmopolitanism, a new line of inquiry on field of study may be just as important as standard analyses of educational attainment. This paper also engages debates about whether educational attainment has a causal effect on cosmopolitanism. Respondents with post-secondary degrees in humanities and social sciences subjects are the most cosmopolitan subset, but there is no evidence that their education caused an increase in cosmopolitanism. Instead, my results suggest they were more cosmopolitan prior to their education, which is consistent with the logic of sorting. Sorting suggests that cosmopolitanism may be a sticky disposition that is formed early in life (Hetherington and Weiler 2018; Sears and Valentino 1997). This connects to research which finds educational divides in political tolerance (Dra˘zanová 2017; Sniderman et al. 1996) and (Stubager 2008) already exist before education occurs. In addition, the evidence for sorting challenges conventional wisdom that cosmopolitanism is most prevalent among the young and fades as people age (Favell 2008; Recchi 2015). My findings engage debates about when and how worldviews are incubated (Gross 2013; Gross and Simmons 2014; Lerch, Russell and Ramirez 2017), and suggest that early-life socialization is a key stage for future analysis. Finally, this paper deepens our understanding of how cosmopolitanism overlaps with

5 other societal divides. If educational attainment and field of study are equally important for understanding the connection between education and cosmopolitanism, cross-cutting cleavages may open possibilities for compromise. Future studies should explore this in more detail as cosmopolitanism will likely remain a salient social and political issue across Europe and North America for the foreseeable future.

2 Fields of study and cosmopolitanism

The analysis in this paper departs from the insight that fields of study involve different subject matters, forms of analysis, and cognitive skills, all of which may be related to social and political attitudes and behavior (Ladd and Lipset 1975; Lopez, Gurin and Nagda 1998). For the purposes of this paper, I distinguish between humanities and social science degrees on one side and physical science and technical degrees on the other.1 Research suggests that people who study physical sciences or technical subjects are more likely than humanities or social science students to hold authoritarian values (Carnevale et al. 2020; Stubager 2009; Surridge 2016) and are less likely to be civically engaged (Hillygus 2005; Niemi and Hanmer 2010). I extend that line of inquiry to cosmopolitanism. Humanities and the social sciences are concerned with the social and cultural world. Those topics lend themselves to contextual thinking because social and cultural phenomena are generally not amenable to universal laws (Sil 2000). For example, evaluating art is a notoriously subjective endeavor as people bring different psychological and cultural associa- tions to their understanding of art. There is also no one set of laws or public policies that is ideal for all times and places. Studying the social and cultural world through the lens of conditional dynamics that vary according to context is consistent with cosmopolitanism. The central principle of cos-

1One could examine more fine-grained distinctions across specific degree subjects (Sur- ridge 2016). However, this would require large sample sizes unavailable in the LISS.

6 mopolitanism is that humans are part of one community. Therefore, humans should treat each other with mutual respect and no national, ethnic, or religious is necessarily more legitimate than another (Appiah 2006; Beck and Grande 2007). From the cosmopoli- tan perspective, reality and truth are conditional on specific (social and cultural) contexts, which is the central principle of humanities and social science endeavors. Admittedly, there are important distinctions between humanities and the social sciences. Humanities subjects are more likely to embrace postmodern approaches that are skeptical of rationality and view social and cultural phenomena through the lens of contingent particu- larities (Carnevale et al. 2020; van Dijk 2008). The social sciences are more likely to embrace positivist approaches that generalize from empirical observations and may be less amenable to extreme cosmopolitanism (Brady and Collier 2004; King, Keohane and Verba 1994).2 Yet despite that variation, a common thread that unifies humanities and social sciences is the contextual diversity of social and cultural phenomena (Camic, Gross and Lamont 2011). Physical sciences and technical subjects engage the natural and physical world where there are concrete and universal laws. For example, Newton’s law of universal gravitation applies everywhere on earth and the formula for water is always one oxygen and two hydrogen atoms. As a result, these areas of study are less likely to embrace the concepts of relativism and subjectivity (Bunge 1967; Gibbons et al. 2005). The universality of these subjects does not preclude them from incorporating contextual variation. The strength of gravitational pull varies according to mass, which is why the force is stronger on earth than on our moon. Nonetheless, physical sciences and technical subjects tend to be focused on general laws and the practicalities of how to make things work (Guimond, Begin and Palmer 1989; Ladd and Lipset 1975; Persson 2012). It is not clear that these subjects in and of themselves should

2LISS data confirm these expectations as humanities degree recipients are more cosmopoli- tan than social science recipients. However, sample sizes are not large enough for reliable detailed analysis of these finer-grained distinctions.

7 be consistent with a strong anti-cosmopolitan perspective. Yet, in comparison to humanities and social sciences, the physical sciences and technical subjects may be less consistent with cosmopolitanism.

3 Hypotheses

The analysis in this paper proceeds in two steps. The first is cross-sectional and exam- ines whether field of study confounds the relationship between educational attainment and cosmopolitanism. The relationship between educational attainment and cosmopolitanism could be spurious if people who study humanities and the social sciences are more likely to get post-secondary degrees and field of study is the more important underlying relationship between education and cosmopolitanism. Therefore, the first hypothesis is that the rela- tionship between educational attainment and cosmopolitanism is no longer (statistically or substantively) significant after controlling for field of study.

H1: The relationship between educational attainment and cosmopolitanism is no longer significant after controlling for field of study.

The second hypothesis leverages the panel structure of the data to ask whether the re- lationship between field of study and cosmopolitanism is causal. Education teaches new facts and skills and exposes people to new ways of understanding the world (Nie, Junn and Stehlik-Barry 1996). To the extent that humanities and social sciences teach people to view the world through a contextual and relativistic lens, they may increase cosmopoli- tanism among students. In addition, research suggests that interactions with educators and other students are central to the education experience (Stubager 2008; van de Werfhorst and de Graaf 2004). Socializing with other people who are engaged in a humanities or so- cial science analysis that privileges contextual thinking might also contribute to increased

8 cosmopolitanism. Previous research on the effect of post-secondary education on cosmopoli- tanism may been inconclusive because it failed to account for the subset of post-secondary humanities and social science degrees that cause the increase in cosmopolitanism.

H2: Receiving a post-secondary humanities or social science education causes people to become more cosmopolitan.

4 The Dutch case

The Netherlands is a useful case for two main . First, as in many European countries, cosmopolitanism is a central political conflict in the Netherlands (De Vries 2018a). Cities like Amsterdam and Rotterdam have long been cosmopolitan centers open to global business and diverse (Favell 2008). However, there is also a long of populist reaction to cosmopolitanism, with politicians like Pim Fortuyn, Rita Verdonk and Geert Wilders (leader and founder of the Party for Freedom) leading movements to strengthen the national focus of Dutch (Vossen 2017). Second, the Netherlands’ education system requires early specialization. In all coun- tries, students in post-secondary education choose subjects that can be classified as humani- ties/social sciences or technical/physical sciences. In the Netherlands, these distinctions also exist in secondary education. Students in non-vocational (HAVO/VWO) secondary schools choose one of four profiles: ‘Culture and society’, ‘Economy and society’, ‘Nature and health’ or ‘Nature and technology’, with more specific specializations within each profile. Vocational secondary education (VMBO) is divided into four subsections: economy, technology, care and welfare or agriculture (Jenkins, Skinner and Trines 2018). The early specialization of Dutch education is useful for my inquiry because it allows me to analyze separately the relationships between educational attainment and cosmopolitanism

9 and field of study and cosmopolitanism, thereby isolating the potential effects of educational attainment and field of study. This would not be possible in countries where specialization only exists for students who advance to post-secondary education. The early specialization of the Netherlands also suggests that the patterns I find are even stronger elsewhere in the world. Early specialization makes the Netherlands a most-likely case for finding evidence that field of study affects cosmopolitanism. Research finds that early specialization transmits stronger educational effects on students and produces more highly-trained graduates (Malamud 2011). It should be easier for me to find evidence of causal effects in the Netherlands, but I do not. Instead, I find evidence of selection, for which the Netherlands is a least-likely case. Early specialization forces students to choose an educational path before they have matured and fully explored their options. This leads to worse matches between students and their subjects and a greater chance of costly ca- reer switches (Malamud 2010). Student sorting into fields of study consistent with their pre-existing values should be stronger in late specialization systems. This suggests that cosmopolitan sorting into humanities and social sciences subjects in the early specialization Netherlands is a conservative estimate of that dynamic elsewhere in the world.

5 Data

I analyze the Longitudinal Internet Studies for the Social sciences (LISS). The panel is conducted by the CentERdata research institute at Tilbourg University. The panel uses a true probability sample of households drawn from the population register by Statistics Netherlands. LISS is conducted online and households that do not have sufficient computing resources are given a computer and internet connection. The LISS panel began in 2007 and is conducted annually (Scherpenzeel 2009).3

3Panel members complete monthly surveys on specific modules of the survey. All modules are administered once a year.

10 The longitudinal design of the LISS is valuable for evaluating H2. Without exogenous allocation to specific fields of study (a dynamic which does not exist to my knowledge), cross- sectional analyses do not provide leverage on whether receiving a specific type of degree changes cosmopolitan attitudes. Panel data allow me to observe cosmopolitan attitudes before and after receiving degrees in specific fields of study. One concern with panel data is attrition. To compensate for attrition, LISS has periodically recruited new panel members. Panel attrition is concentrated among respondents aged 75 years or older (de Vos 2009), which should not pose a threat to the core findings in the paper because it represents less than five percent of the sample. A unique feature of LISS that makes it well-suited to my inquiry is a question that asks about field of study for respondents’ highest degree. Respondents are given a list of sub- jects, which I code into Humanities/Social Sciences or Physical sciences/Technical. I code the following six as Humanities/Social Sciences: Teacher training or education; Art; - ities (modern or classical languages, history, theology, etc.); Social and behavioral studies (including organization studies, media, culture, sports and leisure studies, etc.); Economics, management, business administration, accountancy; Law, public administration. I code the following nine as Physical sciences/Technical: Mathematics, physics, Information Technol- ogy; Technology (including architecture, industry, crafts, construction, etc.); Agriculture, forestry, environment; Medical, health services, nursing, etc.; Personal care services; Cater- ing, recreation; Transport, logistics; Telecommunication; Public order and safety (police, army, fire brigade, etc.). The sixteenth subject option was ‘General or no specific field’, which is not included in my analyses. This category accounted for 15.8 percent of respondents but is impossible to classify and cannot be analyzed.4 A seventeenth response option was ‘Do not know’,

4Respondents who select ‘General or no specific field’ have low levels of cosmopolitanism and educational attainment. Among these respondents, 20 percent have no higher certifica- tion than primary school and 92 percent have no higher certification than secondary school.

11 which accounted for 1.4 percent of respondents and was not included in my analyses. A final response option ‘Other’ accounted for 5.5 percent of respondents. These respondents were given an open-ended question that asked the field of study for their highest degree. I was able to manually code 82 percent of the open-ended responses into either ‘Humanities/Social Sciences’ or ‘Physical sciences/Technical subjects’ and include them in my analyses.5 LISS includes ten items that tap into multiple dimensions of cosmopolitanism. Six items are about immigration attitudes and one item is about support for European unification. These are two of the central issues in European debates about whether societies should be more or less cosmopolitan (Clarke, Goodwin and Whiteley 2017; Maxwell 2020). The immigration items are: ‘Where would you place yourself on a scale of 1 to 5, where 1 means that immigrants can retain their own culture and 5 means that they should adapt entirely?’ and a series of items that ask the extent to which respondents agree or disagree with the following statements: ‘It is good if society consists of people from different cultures’, ‘It should be made easier to attain asylum in the Netherlands’, ‘Legally residing foreigners should be entitled to the same social security as Dutch citizens’, ‘There are too many people of foreign origin or descent in the Netherlands’, ‘It does not help a neighborhood if many people of foreign origin or descent move in’. I use these five items to create an additive index for immigration attitudes, coded from 0 (anti-immigration) to 1 (pro-immigration). The European integration question asks: ‘Where would you place yourself on a scale from 1 to 5, where 1 means that European unification should go further and 5 means that it has already gone too far?’ I recode this item from 0 (anti-EU) to 1 (pro-EU). Another two items are about support for political parties. One is support for the Green

Among respondents who identify a field of study, 5 percent have no higher certification than primary school and 33 percent have no higher certification than secondary school.

5The remaining 18 percent of answers were either blank or did not specify a field of study. All results are substantively the same with or without including the re-coded respondents.

12 Party (which is organized around pro-cosmopolitan policies) and the second is support for the Party for Freedom (PVV) (which is organized around nationalist anti-cosmopolitan policies) (De Vries 2018b). The final measure is whether respondents have someone with non-Dutch ethnicity among their five closest acquaintances. This variable is coded ‘0’ if all acquaintances are Dutch and ‘1’ if at least one acquaintance is Turkish, Moroccan, Antillean, Surinamese, Indonesian, another non-Western country, or another Western country. These were the cate- gories given to respondents by LISS, and while not perfectly exhaustive of all possible ethnic categories, they tap personal engagement with cultures outside the Netherlands (Recchi et al. 2019). The appendix includes full details on the question wording and coding. To measure educational attainment, I use an item that codes respondents’ highest degree into the official Statistics Netherlands categories. Primary school, intermediate secondary education (VMBO), and secondary education (HAVO/VWO) are coded as ‘0’ - no post- secondary education. Intermediate vocational education (MBO), higher vocational education (HBO) and university (WO) are coded as ‘1’ - post-secondary education. I limit my analysis to respondents born in the Netherlands. Cosmopolitanism is openness to international influences, including immigrants. As a result, immigrants are generally more cosmopolitan than natives. Among foreign-born respondents, there is not much difference in cosmopolitanism between those who study humanities/social sciences or technical/physical science subjects. There is also not much difference in cosmopolitanism between foreign-born respondents with and without post-secondary degrees. This suggests that the relationship between education and cosmopolitanism is not the same for immigrants and natives; a dis- tinction that could be explored in more detail elsewhere. At the time of analysis, up to 12 waves (2007-2020) were available for most of the relevant LISS modules.6 Limiting my analysis to respondents born in the Netherlands provides a sample of 50,912 person-year observations and 4,897 respondents.

6Twelve waves were available for Religion and Ethnicity, Work and Schooling, Personality, Politics and Values and Economic Situation: Income. Eleven waves were available for Social

13 6 An overview of education and cosmopolitanism

Figure 1 provides an overview of cosmopolitanism across degree level and field of study. The top of each panel suggest respondents with post-secondary qualifications are generally more cosmopolitan than respondents without post-secondary qualifications, which is con- sistent with expectations. An exception is the measure for non-Dutch friends, where there is no difference between respondents with and without post-secondary qualifications. The bottom of each panel suggests that respondents with humanities or social science degrees are more cosmopolitan than respondents with physical science or technical degrees, which is also consistent with expectations. For the first four indicators, the size of the cosmopoli- tanism gap across educational attainment is roughly similar to the gap across field of study. For non-Dutch friends, the field of study gap is larger than the educational attainment gap. These results suggest that field of study may be at least as important as standard analyses of educational attainment for understanding educational divides over cosmopolitanism.

7 Can field of study account for cosmopolitanism dif-

ferences across degree levels?

The first stage of analysis (H1) explores whether field of study can account for the cos- mopolitanism gap between respondents with and without post-secondary degrees. Respon- dents who study humanities or the social sciences are more likely than respondents who study physical sciences or technical subjects to have post-secondary degrees, which means that field of study could be a confound for the relationship between educational attainment and cosmopolitanism.7 The results in table 1 explore this with two models for each measure

Integration and Leisure.

7Among respondents who study humanities or the social sciences, 75 percent have a post- secondary degree. Among respondents who study physical science or technical subjects, 62

14 Figure 1: Cosmopolitanism across degree level and field of study

Immigration attitudes European unification

Post−sec ● Post−sec ●

No Post−sec ● No Post−sec ●

Hum/Soc ● Hum/Soc ●

Phys/Tech ● Phys/Tech ●

0.35 0.40 0.45 0.50 0.30 0.35 0.40 0.45

PVV support Green support

Post−sec ● Post−sec ●

No Post−sec ● No Post−sec ●

Hum/Soc ● Hum/Soc ●

Phys/Tech ● Phys/Tech ●

0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55

Non−Dutch friends

Post−sec ●

No Post−sec ●

Hum/Soc ●

Phys/Tech ●

0.00 0.05 0.10 0.15

LISS Panel 2007-2020 Mean scores surrounded by 95 percent confidence intervals. Respondents born in the Netherlands. Top two rows compare respondents with and without post-secondary degrees. Third and fourth rows compare respondents with humanities/social science and physical science/technical degrees.

of cosmopolitanism. For each measure of cosmopolitanism, the first model includes a co- variate for degree level and the second includes covariates for degree level and field of study. If field of study confounds the relationship between degree level and cosmopolitanism, the coefficients for degree level should be reduced (and no longer statistically significant) after

percent have a post-secondary degree.

15 the inclusion of the field of study variables.

Table 1: Cosmopolitanism and education

Immigration EU PVV (1) (2) (3) (4) (5) (6) Degree Level 0.061*** 0.054*** 0.084*** 0.077*** -0.093*** -0.084*** (0.005) (0.005) (0.007) (0.007) (0.007) (0.007)

Field of Study 0.043*** 0.049*** -0.058*** (0.004) (0.007) (0.007) Observations 38,243 38,243 36,513 36,513 36,521 36,521 Respondents 7,955 7,955 7,748 7,748 7,709 7,709 R2 0.042 0.056 0.036 0.043 0.053 0.064

Green Friends (7) (8) (9) (10) Degree Level 0.048*** 0.046*** 0.038 -0.003 (0.006) (0.007) (0.069) (0.070)

Field of Study 0.035*** 0.307*** (0.009) (0.065) Observations 35,628 35,628 39,985 39,985 Respondents 7,594 7,594 8,720 8,720 R2 0.048 0.052 Pseudo R2 0.013 0.016

∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 LISS Panel 2007-2020. Models 1-8 are linear regressions and models 9-10 are logistic regressions. All models include standard errors clustered by respondent and controls for survey year, age, sex, employment status, and personal income. Respondents born in the Netherlands. ‘Degree level’ is 0-No post-secondary, 1-Post-secondary. ‘Field of study’ is 0-Physical Science/Technical, 1-Humanities/Social Science.

Results in table 1 indicate that degree level is associated with greater cosmopolitanism on the first four measures (as expected from the results in figure 1, it is not associated with the measure of non-Dutch friends). For each of these measures, the coefficients are statistically significant at p < 0.001. However, contrary to H1, there is no meaningful reduction in the degree level coefficients after including field of study. These results do not support H1,

16 because they suggest field of study is unlikely to be a confound for degree level. Nonetheless, results in table 1 suggest that degree level and field of study are impor- tant predictors of cosmopolitanism. To gauge the relative importance of each predictor, I estimate regressions based on the models in table 1, with additional covariates for the inter- action between degree level and field of study. Figure 2 presents predicted scores from those interaction terms (full model results are in appendix table 2). The key comparison in figure 2 is between the combination of humanities and social science degrees without post-secondary qualifications and physical science and technical de- grees with post-secondary education. This provides insight on whether field of study or degree level is more strongly related to cosmopolitanism. Figure 2 suggests that these two combinations mostly have similar levels of cosmopolitanism, across the five measures.8 This suggests degree level and field of study have relationships with cosmopolitanism that are similar in strength. Educational attainment alone may be insufficient for understanding the connection between education and cosmopolitanism.

8 Does field of study affect cosmopolitanism?

The next analyses examine whether post-secondary humanities or social science education is the key combination that causes more cosmopolitanism (H2). I use the LISS panel to examine cosmopolitanism before and after respondents receive post-secondary humanities or social science degrees. I compare the cosmopolitanism time trend of people who receive post- secondary humanities or social sciences degrees (the ‘treatment group’) with the time trend

8For immigration attitudes and Green support the two groups are statistically indistin- guishable. Respondents with physical science or technical degrees and no post-secondary qualifications are slightly more cosmopolitan on the European unification (0.03 points) and PVV support (0.02 points) measures. Respondents with post-secondary humanities or social science degrees are slightly more cosmopolitan on the friendship measure (0.04 points).

17 Figure 2: The interaction between degree level and field of study

Immigration attitudes European unification

HS PS ● HS PS ●

HS no PS ● HS no PS ●

PT PS ● PT PS ●

PT no PS ● PT no PS ●

0.30 0.35 0.40 0.45 0.50 0.30 0.35 0.40 0.45 0.50

PVV support Green support

HS PS ● HS PS ●

HS no PS ● HS no PS ●

PT PS ● PT PS ●

PT no PS ● PT no PS ●

0.20 0.25 0.30 0.35 0.40 0.40 0.45 0.50 0.55 0.60

Non−Dutch friends

HS PS ●

HS no PS ●

PT PS ●

PT no PS ●

0.00 0.05 0.10 0.15 0.20

LISS Panel 2007-2020. Calculated from models in Appendix Table 2. Results for Immigration attitudes, European unification, PVV support and Green support are predicted scores. Results for Non-Dutch friends are predicted probabilities. ‘HS PS’ is humanities/social science with post-secondary, ‘HS NoPS’ is humanities/social science without post-secondary. ‘PT PS’ is physical science/technical with post-secondary. ‘PT NoPS’ is physical science/technical without post-secondary.

of people who do not (the ‘control group’). This difference-in-difference strategy accounts for the fact that cosmopolitanism may change over time for reasons unrelated to education. I estimate linear regressions with person fixed effects and control variables for survey year, age, employment status, and personal income. There are three key covariates. One is a bivariate measure of whether respondents received a post-secondary humanities or social

18 science degree since the previous survey wave.9 The second key covariate is a categorical variable for years prior to receiving a post-secondary humanities or social science degree. This is necessary for evaluating the parallel trends assumption of difference-in-difference analysis: there can be no difference in time trends between control and treatment groups during the years before receiving a degree. The final covariate of interest is a categorical variable for years after receiving a post-secondary humanities or social science degree. This would be evidence of the causal effect, because it indicates whether the change over several time intervals is different for people who receive a post-secondary humanities or social science degree as opposed to those who do not. Figure 3 presents results for the most general comparison: people who receive a post- secondary humanities or social science degree versus those who do not. In most panels of figure 3, confidence intervals overlap with zero before and after receiving a degree. This sug- gests the cosmopolitanism changes over time are similar for both groups. The one exception is PVV support, where coefficients suggest that people who receive post-secondary human- ities or social science degrees may become slightly more likely to support the PVV after receiving the degree. This runs counter to the prediction (H2) of greater cosmopolitanism after receiving the degree. However, the trend is present before receiving the degree, which violates the parallel trends assumption. Any cosmopolitan time trend differences between the two groups cannot be due to the effect of receiving a degree. The comparison in figure 3 may be too imprecise for identifying the effect of humanities and social science education. Therefore, I estimate additional models that focus only on people who receive post-secondary degrees. Results in figure 4 compare cosmopolitanism time trends for people who receive a post-secondary humanities or social science degree

9There are 328 respondents who receive a post-secondary humanities or social science degree during the survey. In all analyses, years are presented only if they include at least 50 person-year degree recipients in the data.

19 Figure 3: Cosmopolitanism time trends for getting a post-secondary degree in the humanities or social sciences Immigration attitudes European unification

0.25 0.25

0.15 0.15 ● ● ● ● ● ● ● 0.05 0.05 ● ● ● ●

● ● ● ● ● ● ● ● ● ● −0.05 ● −0.05

−0.15 −0.15

−4 −3 −2 −1 0 1 2 3 4 5 6 −4 −3 −2 −1 0 1 2 3 4 5 6

PVV support Green support

0.25 0.25

● 0.15 ● 0.15

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.05 0.05 ● ● ● ●

● −0.05 −0.05

−0.15 −0.15

−4 −3 −2 −1 0 1 2 3 4 5 6 −4 −3 −2 −1 0 1 2 3 4 5 6

Non−Dutch friends

0.25

0.15

● 0.05 ● ● ● ● ● ● ● ● ● ●

−0.05

−0.15

−4 −3 −2 −1 0 1 2 3 4 5 6

LISS Panel 2007-2020. Linear regression models with person fixed effects and controls for survey year, age, employment status, and personal income. Respondents born in the Netherlands. Coefficients and 95 percent standard error estimate whether change over different time intervals is different for people who earn post-secondary humanities/social science degrees during the panel as opposed to those who do not. Results presented only for years with at least 50 degree recipients in the model. versus those who receive a post-secondary physical science or technical degree. However, the results in figure 4 are consistent with figure 3. Among respondents who receive post- secondary degrees, there is no evidence that receiving a post-secondary humanities or social science degree leads to greater cosmopolitanism.

20 Figure 4: Cosmopolitanism time trends among post-secondary degree recipients

Immigration attitudes European unification

0.25 0.25

0.15 0.15

● ● ● ● ● ● ● 0.05 0.05 ● ● ● ●

● ● ● ● ● ● ● ● ● ● ● −0.05 −0.05

−0.15 −0.15

−4 −3 −2 −1 0 1 2 3 4 5 6 −4 −3 −2 −1 0 1 2 3 4 5 6

PVV support Green support

0.25 0.25

0.15 0.15 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.05 0.05 ● ● ● ●

● −0.05 −0.05

−0.15 −0.15

−4 −3 −2 −1 0 1 2 3 4 5 6 −4 −3 −2 −1 0 1 2 3 4 5 6

Non−Dutch friends

0.25

0.15

● ● 0.05 ● ● ● ● ● ● ● ● ●

−0.05

−0.15

−4 −3 −2 −1 0 1 2 3 4 5 6

LISS Panel 2007-2020. Linear regression models with person fixed effects and controls for survey year, age, employment status, and personal income. Respondents born in the Netherlands. Coefficients and 95 percent standard error estimate whether change over different time intervals is different for people who earn post-secondary humanities/social science degrees during the panel as opposed to those who earn post-secondary physical science/technical degrees during the panel. Results presented only for years with at least 50 degree recipients in the model.

Another possibility is that the combination of degree level and degree type is important, so the effect of post-secondary humanities or social science education is only observable relative to those who never earn post-secondary degrees. Figure 5 presents cosmopolitan time trend analyses comparing these groups. The results are consistent with those from figures

21 3 and 4, as there is no compelling evidence of the effect of education. There is suggestive evidence that respondents who receive post-secondary humanities or social science degrees may become slightly more likely to support over time, in comparison to those who never receive post-secondary degrees. However, confidence intervals for post-degree coefficients mostly overlap with zero and there is no strong evidence of a trend. As a robustness check, I explore whether educational attainment and field of study affect cosmopolitanism as distinct and separate educational processes. Appendix table 3 presents results from models analyzing cosmopolitanism time trends for respondents who receive post-secondary degrees versus those who do not receive post-secondary degrees. Appendix table 4 presents results from models analyzing cosmopolitanism time trends for respondents who receive humanities or social science degrees versus those who receive physical science or technical degrees. In neither table is there any evidence that educational attainment or field of study directly affects cosmopolitanism time trends. It is possible that cosmopolitanism evolves over a longer period than the 6 year post- education window analyzed with LISS data. If so, my analyses might be too short to capture those trends. However, if cosmopolitanism develops over a longer time period, that would suggest dynamics that occur after receiving the degree are more important than a direct relationship between post-secondary humanities social science degrees and cosmopolitanism. It is beyond the scope of the LISS to conduct a detailed analysis of how education and post-education experiences may interact to affect cosmopolitanism. However, to check the plausibility of that relationship, I subset the data according to whether respondents’ educa- tion was suitable for their current work. I focus on respondents who are above the midpoint (5) on the scale of education-work suitability (from 0 to 10). These respondents pursued careers in fields that are related to their education. They may also be most affected by their education and the best subgroup for observing the effect of post-secondary humanities or social science education on cosmopolitanism.10

10It is also possible that effects are strongest among respondents with the least education-

22 Figure 5: Cosmopolitanism time trends for post-secondary humanities/social science degrees versus those who never get post-secondary degrees

Immigration attitudes European unification

0.25 0.25

0.15 0.15 ● ● ● ● ● ● ● ● 0.05 0.05 ● ● ●

● ● ● ● ● ● ● ● ● ● −0.05 ● −0.05

−0.15 −0.15

−4 −3 −2 −1 0 1 2 3 4 5 6 −4 −3 −2 −1 0 1 2 3 4 5 6

PVV support Green support

0.25 0.25

● 0.15 ● 0.15 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.05 0.05 ● ● ● ●

● −0.05 −0.05

−0.15 −0.15

−4 −3 −2 −1 0 1 2 3 4 5 6 −4 −3 −2 −1 0 1 2 3 4 5 6

Non−Dutch friends

0.25

0.15

● ● 0.05 ● ● ● ● ● ● ● ● ●

−0.05

−0.15

−4 −3 −2 −1 0 1 2 3 4 5 6

LISS Panel 2007-2020. Linear regression models with person fixed effects and controls for survey year, age, employment status, and personal income. Respondents born in the Netherlands. Coefficients and 95 percent standard error estimate whether change over different time intervals is different for people who earn post-secondary humanities/social science degrees during the panel as opposed to those who never earn post-secondary degrees. Results presented only for years with at least 50 degree recipients in the model.

Appendix table 5 presents results from time trend analyses that compare people who work suitability because they have no additional effects of work that is related to their education. However, 76 percent of respondents are above the midpoint on the suitability scale and only 18 percent are below the midpoint. There are not enough respondents with

23 received post-secondary humanities and social science degrees and have high education/work suitability versus all others. Appendix table 6 presents results from time trend analyses that compare people who received post-secondary humanities and social science degrees and have high education/work suitability versus people who earn post-secondary physical science or technical degrees. Finally, appendix table 7 presents results from time trend analyses that compare people who received post-secondary humanities and social science degrees and have high education/work suitability versus people who never receive post-secondary degrees. None of these subset analyses provide any evidence to suggest that post-secondary humanities or social science degrees cause greater cosmopolitanism.

9 Is selection a more promising explanation?

Results thus far indicate important cosmopolitan divides across field of study. However, there is no evidence that field of study can account for cosmopolitanism divides in educational attainment and there is no evidence that field of study directly causes people to become more cosmopolitan. Therefore, selection may be a more promising explanation of why humanities and social science graduates are more cosmopolitan than physical science and technical graduates. Fully modeling selection into different fields of study is beyond the scope of the LISS data. Nonetheless, one way to get insight on the likelihood of selection is to examine mean levels of cosmopolitanism across field of study, before and after post-secondary degrees. Results in figure 6 suggest that people who study the humanities and social sciences are consistently more cosmopolitan than people who study physical sciences or technical subjects. This gap exists before and after receiving the post-secondary degree and across each cosmopolitanism indicator. Admittedly, the confidence intervals often overlap, in part because of small sample sizes within each year. However, the overall pattern is clear. The largest gaps are for party support measures, and in particular for the . The field low education-work suitability for reliable analysis.

24 Figure 6: Cosmopolitanism before and after post-secondary degrees

Immigration attitudes European unification

0.55 0.55

0.50 0.50 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.45 ● ● ● 0.45 ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.40 0.40 ● ● ● ● ● 0.35 0.35

0.30 0.30

−4 −3 −2 −1 0 1 2 3 4 5 6 −4 −3 −2 −1 0 1 2 3 4 5 6

PVV support Green support

0.40 0.60 ● ● ● ● ● ● ● 0.35 0.55 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.30 ● ● 0.50 ● ● ● ● ● ● ● ● ● ● 0.25 ● 0.45 ● ● ●

0.20 0.40

0.15 0.35

−4 −3 −2 −1 0 1 2 3 4 5 6 −4 −3 −2 −1 0 1 2 3 4 5 6

Non−Dutch friends

0.25

0.20 ● ● ●● ● 0.15 ● ● ● ● ● ● ● ● ● ● ● ● ● 0.10 ● ● ● ● 0.05

0.00

−4 −3 −2 −1 0 1 2 3 4 5 6

LISS Panel 2007-2020. Mean cosmopolitanism scores surrounded by 95 percent confidence intervals. The top two panels limited to respondents who earn degrees during the LISS panel (for years with at least 50 degree recipients). For these panels, x-axis is the amount of time before and after receiving the degree. ‘0’ is the period the degree was earned. Black points are respondents who receive humanities/social science post-secondary degrees. Grey points are respondents who receive physical science/technical post-secondary degrees.

of study gap in Green party support appears to get slightly larger after receiving degrees, but not because humanities and social science post-secondary graduates increase their support. Instead, there is a slight decrease in support from physical science and technical graduates. Figure 6 suggests that selection may account for cosmopolitan divides across field of study, because those divides are already present at least four years before people receive

25 post-secondary degrees. This implies that cosmopolitanism may be a sticky disposition that is stable over time. A full test of cosmopolitan stability across the lifespan requires a panel longer than 12 years. However, for a suggestive indication, I plot cosmopolitanism scores across field of study and the full age range of respondents in the LISS panel (15-100 years old). Full results are in appendix figure 1 and have two main findings. First, across the five measures, cosmopolitanism levels are relatively stable across the age range.11 Second, humanities and social science degree recipients are always more cosmopolitan than physical science and technical degree recipients, across all ages. This is further evidence that the cosmopolitan difference between humanities/social science and physical science/technical types has its roots in early life socialization and not in the education process.

10 Discussion

This paper explored the relationship between cosmopolitanism and field of study. I examined whether variation across field of study could explain inconsistent results in research on the relationship between cosmopolitanism and educational attainment. I find evidence that humanities and social science graduates are more cosmopolitan than physical science and technical graduates, but those differences cannot account for cosmopolitanism gaps across degree level. Nonetheless, the cosmopolitanism gap across field of study is at least as large as the gap between those with and without post-secondary qualifications. This suggests field of study is an important dimension that should be considered in future studies of the relationship between education and cosmopolitanism. The second stage of my analysis explores whether post-secondary humanities and social science degrees cause people to become more cosmopolitan, but I find no evidence of any

11Older respondents are slightly less likely than younger respondents to be in favor of the EU and to have non-Dutch friends. However, older respondents are also less likely than younger respondents to support the PVV.

26 direct effects. Instead, the evidence is consistent with selection, because post-secondary humanities and social science graduates are already more cosmopolitan than physical science and technical graduates before receiving their degrees. My results question conventional wisdom that cosmopolitanism is strongest among the young (Favell 2008; Recchi 2015). My results suggest that cosmopolitanism is formed during early life socialization, which builds on research about the importance of childhood years for shaping worldviews (Lerch, Russell and Ramirez 2017). Admittedly, it is beyond the scope of my data to fully disentangle age, period and cohort effects, but future research should explore this in more detail with data that track cosmopolitanism among individuals across their entire life-course. I find no evidence that field of study affects cosmopolitanism, but education may have causal effects on other attitudes (Elchardus and Spruyt 2009). In addition, data limitations restrict my analyses to broad field of study and educational attainment categories, but one could examine more fine-grained differences across specific subjects or educational tracks (e.g. university versus vocational) (Paredes, Paserman and Pino 2020; Persson 2012). An- other possibility is that selection matters for who chooses different educational trajectories, but smaller-scaled interventions might be possible at the classroom level. Recent research highlights how very specific interventions can nudge tolerance and cosmopolitan attitudes (Adida, Lo and Platas 2018; Kokkonen et al. 2015). It remains to be seen whether these micro-level interventions can impact broader societal divides, but future research should explore all of these nuances in more detail. Yet another possibility is that causal effects and selection processes are both important for the education-cosmopolitanism link. Each dynamic might be more or less important depending on the geographic or temporal context, or the dimension of cosmopolitanism being studied. To properly evaluate this would require rich cross-national longitudinal data, which is beyond the scope of this paper. The data in this paper are from the Netherlands, but the findings are relevant well beyond. Early educational specialization in the Netherlands should weaken the effects of selection.

27 This suggests that the dynamic of cosmopolitan people sorting into humanities and social science subjects should be even more pronounced in other countries. Cosmopolitan and educational cleavages are likely to resonate in European and North American societies for years to come.

28 References

Adida, Claire, Adeline Lo and Melina Platas. 2018. “Perspective taking can promote short- term inclusionary behavior toward Syrian refugees.” Proceedings of the National Academy of Sciences of the United States of America 115(38):9521–9526.

Appiah, Kwame Anthony. 2006. Cosmopolitanism: Ethics in a World of Strangers. : W.W. Norton and Company.

Beck, Ulrich and Edgar Grande. 2007. Cosmopolitan Europe. Malden, MA: Polity Press.

Brady, Henry and David Collier, eds. 2004. Rethinking Social Inquiry: Diverse Tools, Shared Standards. Lanham, MD: Rowman and Littlefield.

Bunge, Mario. 1967. Scientific Research II: The Search for Truth. Berlin: Springer-Verlag.

Camic, Charles, Neil Gross and Michèle Lamont, eds. 2011. Social Knowledge in the Making. Chicago, IL: University of Chicago Press.

Carnevale, Anthony, Nicole Smith, Lenka Dra˘zanová, Artem Gulish and Kathryn Campbell. 2020. “The Role of Education in Taming Authoritarian Attitudes.” Georgetown Center on Education and the Workforce Report.

Cavaillé, Charlotte and John Marshall. 2019. “Education and Anti-Immigration Attitudes: Evidence from Compulsory Schooling Reforms Across Western Europe.” American Political Science Review 113(1):254–263.

Clarke, Harold, Matthew Goodwin and Paul Whiteley. 2017. Brexit: Why Britain Voted to Leave the European Union. Cambridge, UK: Cambridge University Press. de Vos, Klaas. 2009. “Panel attrition in LISS.” Tilburg: CentERdata.

29 De Vries, Catherine. 2018a. “The cosmopolitan-parochial divide: changing patterns of party and electoral competition in the Netherlands and beyond.” Journal of European Public Policy 25(11):1541–1565.

De Vries, Catherine. 2018b. Euroscepticism and the Future of European Integration. Oxford: Oxford University Press. de Wilde, Pieter, Ruud Koopmans, Wolfgang Merkel, Oliver Strijbis and Michael Zürn, eds. 2019. The Struggle Over Borders: Cosmopolitanism and . Cambridge: Cambridge University Press. d’Hombres, Béatrice and Luca Nunziata. 2016. “Wish you were here? Quasi-experimental evidence on the effect of education on self-reported attitude toward immigrants.” European Economic Review 90:201–224.

Dra˘zanová, Lenka. 2017. Education and Tolerance: A Comparative Quantitative Analysis of the Educational Effect on Tolerance. Frankfurt am Main: Peter Lang.

Elchardus, Mark and Bram Spruyt. 2009. “The Culture of Academic Disciplines and the Sociopolitical Attitudes of Students: A Test of Selection and Socialization Effects.” Social Science Quarterly 90(2):446–460.

Favell, Adrian. 2008. Eurostars and eurocities: Free movement and mobility in an integrating Europe. Malden, MA: Blackwell Publishing.

Finseraas, Henning, Øyvind Skorge and Marte Strøm. 2018. “Does education affect immi- gration attitudes? Evidence from an education reform.” Electoral Studies 55:131–135.

Gibbons, Michael, Camille Limoges, Helga Nowotny, Simon Schwartzman, Peter Scott and Martin Trow. 2005. The New Production of Knowledge: The Dynamics of Science and Research in Contemporary Societies. London: Sage.

30 Gross, Neil. 2013. Why Are Professors Liberal and Why Do Conservatives Care? Cambridge, MA: Harvard University Press.

Gross, Neil and Solon Simmons, eds. 2014. Professors and Their Politics. Baltimore, MD: Johns Hopkins University Press.

Guimond, Serge, Guy Begin and Douglas Palmer. 1989. “Education and Causal Attributions: The Development of “Person-Blame” and “System-Blame” .” Social Psychology Quarterly 52(2):126–140.

Hainmueller, Jens and Michael Hiscox. 2007. “Educated Preferences: Explaining Attitudes Toward Immigration in Europe.” International Organization 61(2):399–442.

Hakhverdian, Armen, Erika van Elsas, Wouter van der Bruge and Theresa Kuhn. 2013. “Euroscepticism and education: A longitudinal study of 12 EU member states, 1973-2010.” European Union Politics 14(4):522–541.

Hetherington, Marc and Jonathan Weiler. 2018. Prius or Pickup?: How the Answers to Four Simple Questions Explain America’s Great Divide. New York: Houghton Mifflin Harcourt.

Hillygus, D. Sunshine. 2005. “The Missing Link: Exploring the Relationship between Higher Education and Political Engagement.” Political Behavior 27(1):25–47.

Hooghe, Liesbet and Gary Marks. 2018. “Cleavage theory meets Europe’s crises: Lipset, Rokkan, and the transnational cleavage.” Journal of European Public Policy 25(1):109– 135.

Igarashi, Hiroki and Hiro Saito. 2014. “Cosmopolitanism as Cultural Capital: Exploring the Intersection of , Education and Stratification.” Cultural Sociology 8(3):222– 239.

Jenkins, Viktoria, Makala Skinner and Stefan Trines. 2018. “Education in the Netherlands.” World Education News and Reviews. December 11.

31 King, Gary, Robert Keohane and Sidney Verba. 1994. Designing Social Inquiry: Scientific Inference in Qualitative Research. Princeton, NJ: Princeton University Press.

Kokkonen, Andrej, Stefan Dahlberg, Eelco Harteveld and Wouter van der Brug. 2015. “Group Gender Composition and Tolerance of Immigrants’ .” European Sociolog- ical Review 31(1):65–76.

Kriesi, Hanspeter, Edgar Grande, Martin Dolezal, Marc Helbling, Dominic Höglinger, Swen Hutter and Bruno Wüest. 2012. Political Conflict in Western Europe. Cambridge: Cam- bridge University Press.

Kunst, Sander, Theresa Kuhn and Herman van de Werfhorst. 2020. “Does education decrease Euroscepticism? A regression discontinuity design using compulsory schooling reforms in four European countries.” European Union Politics 21(1):24–42.

Ladd, Everett and Seymour Lipset. 1975. The Divided Academy: Professors and Politics. New York: McGraw-Hill.

Lancee, Bram and Oriane Sarrasin. 2015. “Educated Preferences or Selection Effects? A Longitudinal Analysis of the Impact of Educational Attainment on Attitudes Towards Immigrants.” European Sociological Review 31(4):490–501.

Lerch, Julia, S. Garnett Russell and Francisco Ramirez. 2017. “Wither the Nation-State? A Comparative Analysis of in Textbooks.” Social Forces 96(1):153–180.

Lopez, Gretchen, Patricia Gurin and Biren Nagda. 1998. “Education and Understanding Structural Causes for Group Inequalities.” Political Psychology 19(2):305–329.

Malamud, Ofer. 2010. “Breadth versus Depth: The Timing of Specialization in Higher Education.” LABOUR 24(4):359–390.

Malamud, Ofer. 2011. “Discovering One’s Talent: Learning from Academic Specialization.” ILR Review 64(2):375–405.

32 Margaryan, Shushanik, Annemarie Paul and Thomas Siedler. 2019. “Does Education Affect Attitudes towards Immigration? Evidence from Germany.” Journal of Human Resources .

Maxwell, Rahsaan. 2019. “Cosmopolitan immigration attitudes in large European cities: Contextual or compositional effects?” American Political Science Review 113(2):456–474.

Maxwell, Rahsaan. 2020. “Geographic divides and cosmopolitanism: Evidence from Switzer- land.” Comparative Political Studies 53(13):2061–2090.

Nie, Norman, Jane Junn and Kenneth Stehlik-Barry. 1996. Education and Democratic Citi- zenship in America. Chicago: University of Chicago Press.

Niemi, Richard and Michael Hanmer. 2010. “Voter Turnout Among College Students: New Data and a Rethinking of Traditional Theories.” Social Science Quarterly 91(2):301–323.

Paredes, Valentina, Daniele Paserman and Francisco Pino. 2020. “Does Economics Make You Sexist?” NBER Working Paper No. 27070.

Persson, Mikael. 2012. “Does Type of Education Affect Political Participation? Results from a Panel Survey of Swedish Adolescents.” Scandinavian Political Studies 35(3):198–221.

Recchi, Ettore. 2015. Mobile Europe: The Theory and Practices of Free Movement in the EU. Basingstoke: Palgrave Macmillan.

Recchi, Ettore, Adrian Favell, Fulya Apaydin, Roxana Barbulescu, Michael Braun, Irina Ciornei, Niall Cunningham, Juan Diez Madrano, Deniz Duru, Laurie Hanquinet, Janne Jensen, Steffen Pötzschke, David Reimer, Justyna Salamonska, Mike Savage and Albert Varela. 2019. Everyday Europe: Social in an Unsettled Continent. Bris- tol: Polity Press.

Scherpenzeel, Annette. 2009. “Start of the LISS panel: Sample and recruitment of a probability-based Internet panel.” Tilburg: CentERdata.

33 Sears, David and Nicholas Valentino. 1997. “Politics Matters: Political Events as Catalysts for Preadult Socialization.” American Political Science Review 91(1):45–65.

Sil, Rudra. 2000. “The Division of Labor in Social Science Research: Unified Methodology or “Organic Solidarity”?” Polity 32(4):499–531.

Sniderman, Paul, Joseph Fletcher, Peter Russell and Philip Tetlock. 1996. The Clash of Rights: Liberty, Equality, and Legitimacy in Pluralist . New Haven: Yale University Press.

Stubager, Rune. 2008. “Education effects on authoritarian-libertarian values: a question of socialisation.” British Journal of Sociology 59(2):327–350.

Stubager, Rune. 2009. “Education-based group identity and consciousness in the authoritarian-libertarian value conflict.” European Journal of Political Research 48(2):204– 233.

Surridge, Paula. 2016. “Education and : Pursuing the Link.” Oxford Review of Education 42(2):146–164. van de Werfhorst, Herman and Nan de Graaf. 2004. “The sources of political orientations in post-industrial society: social class and education revisited.” British Journal of Sociology 55(2):211–235. van Dijk, Teun. 2008. Discourse and Context: A Sociocognitive Approach. Cambridge: Cambridge University Press.

Vossen, Koen. 2017. The Power of Populism: Geert Wilders and the Party for Freedom in the Netherlands. Oxon and New York: Routledge.

34 Appendix

Cosmopolitanism items

Immigration attitudes: ‘Where would you place yourself on a scale of 1 to 5, where 1 means that immigrants can retain their own culture and 5 means that they should adapt entirely?’ 1- immigrants can retain their own culture... 5- immigrants should adapt entirely to Dutch culture ‘It is good if society consists of people from different cultures’, ‘It should be made easier to attain asylum in the Netherlands’, ‘Legally residing foreigners should be entitled to the same social security as Dutch citizens’, ‘There are too many people of foreign origin or descent in the Netherlands’, ‘It does not help a neighborhood if many people of foreign origin or descent move in’ 1- fully disagree, 2- disagree, 3- neither agree nor disagree, 4- agree, 5- fully agree European Union attitudes: ‘Where would you place yourself on a scale from 1 to 5, where 1 means that European unification should go further and 5 means that it has already gone too far?’ 1- European unification should go further... 5- European unification has already gone too far Party evaluations: ‘What do you think of GroenLinks (green party)’?, ‘What do you think of the PVV (Wilders freedom party)’? 0- very unsympathetic... 10- very sympathetic Social contacts: Ethnicity of five closest acquaintances. 0- all acquaintances are Dutch, 1- at least one acquaintance is Turkish, Moroccan, Antillean, Surinamese, Indonesian, another non-Western country, or another Western country.

35 Appendix Table 1: Summary statistics (respondents born in the Netherlands)

Mean SD Min Max Immigration attitudes 0.45 0.17 0 1 EU attitudes 0.38 0.29 0 1 PVV evaluations 0.28 0.27 0 1 Green Party evaluations 0.50 0.22 0 1 Non-Dutch friends 0.09 0.29 0 1 Post-secondary education 0.67 0.47 0 1 Humanities/Social science 0.40 0.49 0 1 Age 49.78 16.88 14 100 Male 0.50 0.50 0 1

36 Appendix Table 2: Degree level and Field of study with Interactive models

Imm EU PVV Green Friends (1) (2) (3) (4) (5) Degree Level 0.044*** 0.077*** -0.084*** 0.046*** 0.033 (0.008) (0.009) (0.009) (0.007) (0.087)

Field of Study 0.046*** 0.049*** -0.057*** 0.035*** 0.367*** (0.008) (0.011) (0.012) (0.009) (0.111)

Level × Field -0.003 0.000 -0.002 -0.005 -0.086 (0.009) (0.013) (0.014) (0.011) (0.133) Observations 38,243 36,513 36,521 35,628 39,985 Respondents 7,955 7,748 7,709 7,594 8,720 R2 0.056 0.043 0.064 0.052 Pseudo R2 0.016

∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 LISS Panel 2007-2020. Models 1-4 are linear regressions and model 5 is logistic regression. All models include standard errors clustered by respondent and controls for survey year, age, sex, employment status, and personal income. Respondents born in the Netherlands. ‘Degree level’ is 0-No post-secondary, 1-Post-secondary. ‘Field of study’ is 0-Physical Science/Technical, 1-Humanities/Social Science.

37 Appendix Table 3: Cosmopolitanism time trends and postsecondary degrees

Imm EU PVV Green Friends (1) (2) (3) (4) (5) 8 years before degree 0.006 0.035 0.122*** -0.030 0.103* (0.016) (0.040) (0.031) (0.029) (0.044) 7 years before degree -0.020 0.055 0.111*** 0.015 0.090* (0.015) (0.038) (0.029) (0.028) (0.042) 6 years before degree -0.018 0.031 0.084** -0.000 0.077 (0.015) (0.036) (0.027) (0.026) (0.039) 5 years before degree -0.014 0.040 0.078** -0.016 0.064 (0.014) (0.034) (0.027) (0.026) (0.038) 4 years before degree -0.008 0.052 0.091*** 0.002 0.072 (0.014) (0.033) (0.026) (0.025) (0.037) 3 years before degree -0.001 0.066* 0.078** 0.026 0.088* (0.013) (0.032) (0.025) (0.024) (0.036) 2 years before degree -0.002 0.061 0.056* 0.010 0.065 (0.013) (0.031) (0.024) (0.024) (0.035) 1 year before degree 0.006 0.064* 0.057* 0.013 0.060 (0.013) (0.031) (0.024) (0.023) (0.035) Year of degree 0.009 0.070* 0.054* 0.021 0.050 (0.013) (0.030) (0.024) (0.023) (0.034) 1 year after degree 0.010 0.066* 0.049* 0.036 0.046 (0.013) (0.031) (0.024) (0.023) (0.035) 2 years after degree 0.022 0.065* 0.034 0.042 0.057 (0.013) (0.032) (0.024) (0.024) (0.035) 3 years after degree 0.026* 0.091** 0.048 0.052* 0.073* (0.013) (0.032) (0.025) (0.024) (0.036) 4 years after degree 0.020 0.093** 0.038 0.043 0.026 (0.014) (0.033) (0.026) (0.025) (0.037) 5 years after degree 0.033* 0.099** 0.026 0.046 0.023 (0.014) (0.034) (0.026) (0.025) (0.038) 6 years after degree 0.035* 0.126*** 0.025 0.054* 0.055 (0.015) (0.035) (0.027) (0.026) (0.040) 7 years after degree 0.036* 0.110** 0.026 0.039 0.030 (0.015) (0.037) (0.029) (0.027) (0.043) 8 years after degree 0.020 0.082* 0.058 0.031 0.037 (0.017) (0.040) (0.031) (0.030) (0.049) Observations 44,106 41,947 41,986 40,908 45,755 Respondents 8,749 8,525 8,470 8,347 9,486 Overall R2 0.003 0.014 0.012 0.018 0.008

∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 LISS Panel 2007-2020. Linear regression models with person fixed effects and controls for year, age, employment status, and personal income. Respondents born in the Netherlands. Coefficients (with 95 percent standard errors in parentheses) estimate whether change over different time intervals is different for people who earn post-secondary degrees during the panel as opposed to those who do not. Results presented for years with at least 50 degree recipients in the model. 38 Appendix Table 4: Cosmopolitanism time trends and field of study

Imm EU PVV Green Friends (1) (2) (3) (4) (5) 5 years before degree -0.050** 0.019 0.108*** -0.045 -0.078 (0.018) (0.042) (0.033) (0.031) (0.054) 4 years before degree -0.024 0.022 0.118*** -0.057* -0.056 (0.016) (0.038) (0.030) (0.028) (0.048) 3 years before degree -0.035* 0.037 0.100*** -0.003 -0.064 (0.015) (0.036) (0.028) (0.027) (0.045) 2 years before degree -0.043** 0.009 0.053* -0.008 -0.056 (0.014) (0.034) (0.026) (0.025) (0.042) 1 year before degree -0.040** -0.005 0.073** -0.002 -0.075 (0.014) (0.033) (0.026) (0.024) (0.041) Year of degree -0.030* 0.010 0.065** 0.005 -0.070 (0.012) (0.030) (0.024) (0.022) (0.038) 1 year after degree -0.026 0.013 0.084*** 0.019 -0.073 (0.013) (0.033) (0.025) (0.024) (0.041) 2 years after degree -0.016 0.011 0.058* 0.025 -0.067 (0.014) (0.033) (0.026) (0.025) (0.042) 3 years after degree -0.019 0.034 0.085** 0.034 -0.017 (0.015) (0.035) (0.027) (0.026) (0.044) 4 years after degree -0.040** 0.018 0.088** 0.024 -0.093* (0.015) (0.037) (0.029) (0.027) (0.046) 5 years after degree -0.023 0.041 0.070* 0.043 -0.081 (0.017) (0.040) (0.031) (0.029) (0.051) 6 years after degree -0.027 0.019 0.088** 0.046 -0.026 (0.018) (0.042) (0.033) (0.031) (0.054) 7 years after degree -0.032 0.040 0.059 0.042 -0.030 (0.019) (0.046) (0.036) (0.034) (0.059) Observations 4,752 4,388 4,369 4,162 5,264 Respondents 1,111 1,071 1,058 1,032 1,216 Overall R2 0.003 0.042 0.003 0.010 0.004

∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 LISS Panel 2007-2020. Linear regression models with person fixed effects and controls for year, age, post-secondary degree, employment status, and personal income. Respondents born in the Netherlands. Coefficients (with 95 percent standard errors in parentheses) estimate whether change over different time intervals is different for people who earn humanities or social science degrees during the panel as opposed to those who earn physical science or technical degree during the panel. Results presented for years with at least 50 degree recipients in the model.

39 Appendix Table 5: Cosmopolitanism time trends for post-secondary humanities/social science degrees with high education/work suitability versus all others

Imm EU PVV Green Friends (1) (2) (3) (4) (5) 4 years before degree -0.031 0.052 0.152*** -0.016 0.038 (0.025) (0.062) (0.045) (0.045) (0.068) 3 years before degree -0.037 0.076 0.128** 0.030 0.019 (0.025) (0.060) (0.044) (0.044) (0.066) 2 years before degree -0.044 0.063 0.084 0.043 0.022 (0.024) (0.059) (0.043) (0.044) (0.065) 1 year before degree -0.035 0.057 0.106* 0.045 0.020 (0.024) (0.058) (0.042) (0.043) (0.064) Year of degree -0.018 0.076 0.089* 0.049 0.015 (0.023) (0.057) (0.042) (0.042) (0.063) 1 year after degree -0.017 0.075 0.092* 0.065 -0.001 (0.024) (0.058) (0.042) (0.043) (0.064) 2 years after degree -0.005 0.062 0.067 0.072 0.016 (0.024) (0.059) (0.043) (0.043) (0.065) 3 years after degree -0.002 0.100 0.089* 0.078 0.076 (0.024) (0.059) (0.043) (0.044) (0.066) 4 years after degree -0.020 0.103 0.075 0.069 0.007 (0.025) (0.062) (0.045) (0.046) (0.069) 5 years after degree 0.003 0.126* 0.054 0.081 0.017 (0.026) (0.063) (0.046) (0.046) (0.071) 6 years after degree -0.011 0.083 0.074 0.103* 0.047 (0.026) (0.064) (0.048) (0.048) (0.073) Constant 0.488*** 0.603* 0.221 0.503** 0.301 (0.101) (0.243) (0.183) (0.174) (0.282) Observations 36,127 34,509 34,507 33,665 37,242 Respondents 7,460 7,278 7,228 7,123 8,092 Overall R2 0.001 0.011 0.001 0.012 0.007

∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 LISS Panel 2007-2020. Linear regression models with person fixed effects and controls for year, age, employment status, and personal income. Respondents born in the Netherlands. Coefficients (with 95 percent standard errors in parentheses) estimate whether change over different time intervals is different for people who earn post-secondary humanities/social science degrees during the panel and have high education/work suitability in their post-degree work as opposed to all others. Results presented for years with at least 50 degree recipients in the model.

40 Appendix Table 6: Cosmopolitanism time trends for post-secondary humanities/social science degrees with high education/work suitability versus post-secondary physical science/technical degrees

Imm EU PVV Green Friends (1) (2) (3) (4) (5) 4 years before degree -0.022 0.063 0.134** -0.007 0.026 (0.028) (0.063) (0.050) (0.048) (0.080) 3 years before degree -0.029 0.086 0.112* 0.037 0.014 (0.027) (0.061) (0.048) (0.047) (0.077) 2 years before degree -0.038 0.071 0.074 0.049 0.022 (0.027) (0.061) (0.048) (0.046) (0.076) 1 year before degree -0.031 0.064 0.100* 0.048 0.024 (0.026) (0.059) (0.046) (0.046) (0.075) Year after degree -0.018 0.076 0.090 0.052 0.021 (0.026) (0.058) (0.046) (0.045) (0.074) 1 year after degree -0.019 0.071 0.095* 0.067 0.013 (0.026) (0.059) (0.047) (0.045) (0.075) 2 years after degree -0.010 0.056 0.077 0.073 0.033 (0.027) (0.060) (0.047) (0.046) (0.076) 3 years after degree -0.009 0.091 0.100* 0.075 0.101 (0.027) (0.061) (0.048) (0.047) (0.077) 4 years after degree -0.030 0.089 0.092 0.064 0.036 (0.028) (0.063) (0.050) (0.049) (0.081) 5 years after degree -0.011 0.112 0.078 0.079 0.054 (0.028) (0.065) (0.051) (0.049) (0.083) 6 years after degree -0.030 0.060 0.101 0.100 0.088 (0.029) (0.066) (0.053) (0.051) (0.086) Observations 3,805 3,514 3,478 3,298 4,175 Respondents 837 815 799 775 896 Overall R2 0.000 0.029 0.002 0.007 0.002

∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 LISS Panel 2007-2020. Linear regression models with person fixed effects and controls for year, age, employment status, and personal income. Respondents born in the Netherlands. Coefficients (with 95 percent standard errors in parentheses) estimate whether change over different time intervals is different for people who earn post-secondary humanities/social science degrees during the panel and have high education/work suitability in their post-degree work as opposed to people who earn post-secondary physical science/technical degrees. Results presented for years with at least 50 degree recipients in the model.

41 Appendix Table 7: Cosmopolitanism time trends for post-secondary humanities/social science degrees with high education/work suitability versus people who never get post-secondary degrees

Imm EU PVV Green Friends (1) (2) (3) (4) (5) 4 years before degree -0.033 0.041 0.156** -0.019 0.042 (0.026) (0.065) (0.049) (0.049) (0.070) 3 years before degree -0.036 0.063 0.134** 0.025 0.011 (0.026) (0.063) (0.048) (0.047) (0.068) 2 years before degree -0.047 0.051 0.088 0.040 0.027 (0.025) (0.062) (0.047) (0.047) (0.067) 1 year before degree -0.031 0.057 0.101* 0.040 0.024 (0.025) (0.061) (0.046) (0.046) (0.066) Year after degree -0.017 0.070 0.078 0.055 0.022 (0.024) (0.059) (0.045) (0.045) (0.064) 1 year after degree -0.018 0.073 0.081 0.067 0.000 (0.025) (0.061) (0.046) (0.046) (0.066) 2 years after degree -0.003 0.063 0.055 0.076 0.014 (0.025) (0.061) (0.047) (0.047) (0.067) 3 years after degree -0.003 0.110 0.080 0.080 0.076 (0.025) (0.062) (0.047) (0.047) (0.067) 4 years after degree -0.021 0.110 0.062 0.079 0.007 (0.026) (0.065) (0.049) (0.049) (0.071) 5 years after degree 0.002 0.136* 0.038 0.089 0.017 (0.027) (0.066) (0.050) (0.049) (0.072) 6 years after degree -0.010 0.092 0.060 0.117* 0.049 (0.028) (0.067) (0.052) (0.051) (0.075) Observations 11,365 10,694 10,778 10,418 11,804 Respondents 2,444 2,356 2,340 2,285 2,633 Overall R2 0.004 0.024 0.005 0.012 0.016

∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001 LISS Panel 2007-2020. Linear regression models with person fixed effects and controls for year, age, employment status, and personal income. Respondents born in the Netherlands. Coefficients (with 95 percent standard errors in parentheses) estimate whether change over different time intervals is different for people who earn post-secondary humanities/social science degrees during the panel and have high education/work suitability in their post-degree work as opposed to people who never earn post-secondary degrees. Results presented for years with at least 50 degree recipients in the model.

42 Appendix Figure 1: Cosmopolitanism across the lifespan

Immigration attitudes European unification

0.55 0.55

● 0.50 ● ● 0.50 ● ● ● ● ● ● 0.45 ● 0.45 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.40 0.40 ● ● ● ● ● 0.35 0.35 ● ● ● ● ● ● 0.30 0.30

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

PVV support Green support

0.40 0.60 ●

● ● 0.35 ● ● 0.55 ● ● ● ● ● ● ● ● ● ● ● ● 0.30 ● 0.50 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.25 0.45 ● ● ● ● ● 0.20 ● 0.40

0.15 0.35

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Non−Dutch friends

0.25

0.20

● ● 0.15 ● ● ● ● ● ● ● 0.10 ● ● ● ● ● ● ● ● ● ● 0.05 ●

0.00

1 2 3 4 5 6 7 8 9 10

LISS Panel 2007-2020. Mean cosmopolitanism scores (surrounded by 95 percent confidence intervals) across age deciles (on the x-axis): 1: 15-24 years old, 2: 25-33 years old, 3: 34-39 years old, 4: 40-45 years old, 5: 46-51 years old, 6: 52-56 years old, 7: 57-61 years old, 8: 62-66 years old, 9: 67-72 years old, 10: 73-100 years old. The vertical line at 2.5 on the x-axis indicates the mean age for receiving post-secondary degrees (32.1 years old). Black points are respondents who receive humanities/social science degrees. Grey points are respondents who receive physical science/technical degrees.

43