The Co-Evolution of Education and Tolerance: Evidence from England

The Co-Evolution of Education and Tolerance: Evidence from England Abstract Using data from several periods of English history and building on the literature on culture and institutions, we analyze the co-evolution of education and attitudes toward women’s and minority rights. First, we establish a strong association between the size of twenty-first-century educational institutions in a given location and the attitudes prevalent there: the proximity of a large college or university is associated with individual support for women’s and minority rights, regardless of whether the individual has been to university. Second, we present evidence for high inter-temporal persistence in the geographical variation in the density of educational institutions over several centuries. Third, we show that the geographical distribution of later educational institutions depends not only on the distribution of medieval institutions, but also on correlates of medieval exposure to ethnic and religious diversity that are likely to have influenced attitudes. Institutions and culture co-evolve, and the inter-temporal persistence of the density of educational institutions is one mechanism (though probably not the only one) that explains the association between medieval exposure to diversity and twenty-first-century attitudes. Introduction There is now a sizeable literature in economics and political science on the connections between culture and formal institutions. However, Alesina and Giuliano (2015) note some of this literature’s limitations: culture and institutions are likely to co-evolve, but most papers focus on one direction of causality, and a chronology of the changes in culture and institutions is often lacking. In this paper, we attempt to address these concerns in a study of the association between a particular set of cultural characteristics (attitudes toward minority groups and the role of women) and a particular set of institutions (educational ones). Research in sociology and social psychology has provided detailed evidence on the positive effect of education on the attitudes of students toward minority groups and women’s rights. This effect could be through higher levels of cognitive sophistication (Bobo and Licari 1989), through socialization involving liberal social norms (Coenders and Scheepers 2003; Hello et al. 2004), or through broadening the social network of which the individual is a part (Hello, Scheepers, and Sleegers 2006). There is evidence that education is associated with positive attitudes toward immigrants (e.g., Mayda 2006; Semyonov and Raijman 2006) and people of other ethnicities (e.g., Hooghe, Meeusen, and Quintelier 2013) and sexual orientations (e.g., McVeigh and Diaz 2009; Ohlander, Batalova, and Treas 2005). There is also evidence for an association with positive attitudes toward women’s rights; this association holds for both women (Pierotti 2013) and men (Banaszak and Plutzer 1993). However, the literature overlooks a potential effect of educational institutions on attitudes at the aggregate level: it is possible that the presence of a college or university in town affects attitudes across the whole town, including the attitudes of those who have never been to university. (From now on, we use the word “university” to refer to any tertiary education institution.) University students and faculty could promote positive attitudes toward outgroups or women’s rights through interactions with the wider community, so that the socialization and social network effects noted by the literature extend beyond the campus. Alternatively, the presence of a university might attract a certain type of resident. In this way, the university could be part of a wider social network that promotes certain social norms. Moreover, the effect of educational institutions on attitudes is only half of the story, because the location of universities could itself be a function of geographical variation in attitudes: a university might be more welcome in a town where the local community is open to institutions that promote engagement with that which is foreign. Schofer and Meyer (2005) find that the cross-country variation in the rate of university expansion is explained partly by variation in the strength of a country’s international linkages, an effect consistent with the interpretation of university education (and the university as an institution) as an expression of cosmopolitan values (Meyer et al. 2008). It is possible that the within-country variation exhibits a similar pattern, reflecting regional variation in attachment to these values. In order to explore the association between attitudes and the location of educational institutions, we compile cross-sectional data from several periods of English history.1 First of all, we measure the size of the association between university size and local attitudes in twenty-first-century opinion surveys. Second, we measure the magnitude of inter-temporal persistence in the importance of educational institutions, using geographical variation in (i) twenty-first-century university size and (ii) the density of educational institutions prior to the middle of the nineteenth century (which we term the “premodern” period). The premodern institution that we focus on is the library, libraries having become increasingly widespread following the invention of the printing press in the fifteenth century. Third, we measure the extent to which geographical variation in the density of premodern libraries, and of the bookstores that supplied them, is correlated with local communities’ exposure to ethnic and religious diversity in the twelfth and thirteenth centuries. Such a correlation suggests that exposure to diversity can change preferences in a town, making it more open to institutions that embody new ideas. Our results complement the existing evidence for long-run persistence in attitudes toward outgroups (Fielding 2018; Jha 2013; Voigtländer and Voth 2013) and toward the role of women (Alesina, Giuliano, and Nunn 2013). We extend this literature by showing how institutions have been a vehicle for the inter-temporal persistence of cultural variation. Explaining the Geographical Variation in Twenty-First-Century Attitudes The first part of our analysis is designed to estimate the effect of twenty-first-century universities on attitudes in the surrounding area. We measure these attitudes using data from the 2010 and 2015 rounds of the British Election Study (BES): see www.britishelectionstudy.com and www.britishelectionstudy.com/data-objects/panel-study-data/. One key advantage of the BES is the breadth of its geographical coverage: respondents are drawn randomly from the electoral roll, with observations for every parliamentary constituency. Three key 2015 BES questions relating to attitudes toward people of other ethnicities and sexual orientations, and toward women’s rights, are as follows: “Do you agree with the statement that equal opportunities for ethnic minorities have gone too far?” “Do you agree with the statement that equal opportunities for gays and lesbians have gone too far?” “Do you agree with the statement that equal opportunities for women have gone too far?” There is also a variety of questions about immigrants, though the responses to these questions are highly correlated. The immigrant question that we will use is as follows: “Do you agree with the statement that immigrants are a burden on the welfare state?” The possible responses to these questions are: “strongly disagree” (to which we allocate a value of one), “disagree” (two), “don’t know” (three), “agree” (four), and “strongly agree” (five). We use the responses to construct four opinion variables for each respondent; each variable has a maximum possible value of five (indicating maximal opposition to equal opportunities and immigration) and a minimum possible value of one. Our sample comprises all white respondents in the 460 English constituencies outside London.2 When we restrict the sample to those respondents also reporting personal characteristics (which permits their inclusion in the data analysis that follows), we are left with just under 20,000 observations. London is excluded from our sample because, as the nation’s capital, it contains an atypical set of public institutions that could affect attitudes in ways that are difficult to measure. Another highly salient political question that may relate to attitudes toward outgroups concerns membership of the European Union. Previous studies (e.g., Bruter 2005, appendix 3) have found education to be associated with more favorable attitudes toward the EU, and one interpretation of this effect is that opposition to the EU is motivated partly by antipathy toward outgroups defined by nationality. BES respondents were asked about their voting intention in the upcoming referendum on Britain’s EU membership, and we construct a further variable that equals one if the respondent indicated an intention to vote Remain, three if the respondent indicated an intention to vote Leave, and two if the respondent did not know. A higher value of this variable indicates greater opposition to European integration, just as higher values of the other variables indicate greater opposition to equal opportunities. For each of the five 2015 attitude variables, table 1 shows the proportion of people giving each response. Opposition to immigration and ethnic minority rights is somewhat higher than opposition to gay rights, which is somewhat higher than opposition to women’s rights. Nevertheless, for each statement, there is a substantial proportion of people agreeing and a substantial proportion disagreeing. For each statement and for each response, table 2 provides a measure of local university size for the relevant group of respondents. This measure is the average across respondents of the ratio of the university student population in the respondent’s constituency to the rest of the constituency population (denoted student ratio);3 data are taken from the 2011 census. It can be seen that the ratio is consistently lower among those respondents who oppose equal opportunities, immigration, and European integration. This correlation is consistent with the hypothesis outlined in the introduction: universities promote liberal attitudes in the area around them. However, table 2 does not in itself constitute evidence in favor of the hypothesis. First, there might be systematic variation in the personal characteristics of respondents that is correlated both with attitudes and with the size of the local student population. For example, inhabitants of university towns might be younger, on average, or more highly educated, or have higher incomes. Second, attitudes might affect local student numbers. For example, not all students live on campus, and their choice of neighborhood might depend on the attitudes of their potential neighbors. We must address both of these problems in order to establish that the size of the student population affects local attitudes. Table 1. Distribution of Responses to Attitude Questions in the 2015 BES   (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]    Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Strongly disagree  1,091  5.5%  591  3.0%  1,542  7.8%  1,138  5.7%      Disagree [remain]  3,115  15.7%  2,272  11.5%  5,653  28.5%  3,082  15.5%  7,434  37.5%  Don’t know  9,501  47.9%  8,515  42.9%  10,114  51.0%  4,674  23.6%  3,553  17.9%  Agree [leave]  3,266  16.5%  4,878  24.6%  1,889  9.5%  5,898  29.7%  8,856  44.6%  Strongly agree  2,865  14.4%  3,582  18.1%  640  3.2%  5,050  25.5%      Total  19,838    19,838    19,838    19,842    19,843      (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]    Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Strongly disagree  1,091  5.5%  591  3.0%  1,542  7.8%  1,138  5.7%      Disagree [remain]  3,115  15.7%  2,272  11.5%  5,653  28.5%  3,082  15.5%  7,434  37.5%  Don’t know  9,501  47.9%  8,515  42.9%  10,114  51.0%  4,674  23.6%  3,553  17.9%  Agree [leave]  3,266  16.5%  4,878  24.6%  1,889  9.5%  5,898  29.7%  8,856  44.6%  Strongly agree  2,865  14.4%  3,582  18.1%  640  3.2%  5,050  25.5%      Total  19,838    19,838    19,838    19,842    19,843    Note: The sample comprises white respondents in all English parliamentary constituencies outside London. Table 1. Distribution of Responses to Attitude Questions in the 2015 BES   (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]    Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Strongly disagree  1,091  5.5%  591  3.0%  1,542  7.8%  1,138  5.7%      Disagree [remain]  3,115  15.7%  2,272  11.5%  5,653  28.5%  3,082  15.5%  7,434  37.5%  Don’t know  9,501  47.9%  8,515  42.9%  10,114  51.0%  4,674  23.6%  3,553  17.9%  Agree [leave]  3,266  16.5%  4,878  24.6%  1,889  9.5%  5,898  29.7%  8,856  44.6%  Strongly agree  2,865  14.4%  3,582  18.1%  640  3.2%  5,050  25.5%      Total  19,838    19,838    19,838    19,842    19,843      (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]    Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Strongly disagree  1,091  5.5%  591  3.0%  1,542  7.8%  1,138  5.7%      Disagree [remain]  3,115  15.7%  2,272  11.5%  5,653  28.5%  3,082  15.5%  7,434  37.5%  Don’t know  9,501  47.9%  8,515  42.9%  10,114  51.0%  4,674  23.6%  3,553  17.9%  Agree [leave]  3,266  16.5%  4,878  24.6%  1,889  9.5%  5,898  29.7%  8,856  44.6%  Strongly agree  2,865  14.4%  3,582  18.1%  640  3.2%  5,050  25.5%      Total  19,838    19,838    19,838    19,842    19,843    Note: The sample comprises white respondents in all English parliamentary constituencies outside London. Table 2. Average Values of Student ratio in the Constituencies of Respondents in the 2015 BES, Disaggregated by Attitudes   (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]  Strongly disagree  4.95%  5.12%  4.55%  5.35%    Disagree [remain]  4.66%  4.87%  4.08%  4.66%  4.52%  Don’t know  3.95%  4.05%  3.89%  3.97%  3.90%  Agree [leave]  3.46%  3.53%  3.63%  3.67%  3.53%  Strongly agree  3.42%  3.56%  3.58%  3.55%      (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]  Strongly disagree  4.95%  5.12%  4.55%  5.35%    Disagree [remain]  4.66%  4.87%  4.08%  4.66%  4.52%  Don’t know  3.95%  4.05%  3.89%  3.97%  3.90%  Agree [leave]  3.46%  3.53%  3.63%  3.67%  3.53%  Strongly agree  3.42%  3.56%  3.58%  3.55%    Note: In order to compute the figures in each of columns (i–iv) of the table, respondents in the 2015 BES are grouped according to their response to the statement heading that column. For each statement and for each response, the figure is the average across respondents of the size of the student population in the respondent’s parliamentary constituency as a percentage of the rest of the population. In column (v), respondents are grouped according to their voting intention in the EU membership referendum. The sample is the same as in table 1. Table 2. Average Values of Student ratio in the Constituencies of Respondents in the 2015 BES, Disaggregated by Attitudes   (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]  Strongly disagree  4.95%  5.12%  4.55%  5.35%    Disagree [remain]  4.66%  4.87%  4.08%  4.66%  4.52%  Don’t know  3.95%  4.05%  3.89%  3.97%  3.90%  Agree [leave]  3.46%  3.53%  3.63%  3.67%  3.53%  Strongly agree  3.42%  3.56%  3.58%  3.55%      (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]  Strongly disagree  4.95%  5.12%  4.55%  5.35%    Disagree [remain]  4.66%  4.87%  4.08%  4.66%  4.52%  Don’t know  3.95%  4.05%  3.89%  3.97%  3.90%  Agree [leave]  3.46%  3.53%  3.63%  3.67%  3.53%  Strongly agree  3.42%  3.56%  3.58%  3.55%    Note: In order to compute the figures in each of columns (i–iv) of the table, respondents in the 2015 BES are grouped according to their response to the statement heading that column. For each statement and for each response, the figure is the average across respondents of the size of the student population in the respondent’s parliamentary constituency as a percentage of the rest of the population. In column (v), respondents are grouped according to their voting intention in the EU membership referendum. The sample is the same as in table 1. In order to address the first problem, we fit a set of ordered Probit models of the responses in table 1, each model corresponding to a different question. In each model, each observation of the dependent variable is the response of one individual to one particular question; this variable takes a value between one and five, with higher values indicating greater opposition to equal rights, immigration, or European integration. Our key explanatory variable is the logarithm of student ratio,4 but our set of explanatory variables also incorporates a range of individual and constituency characteristics that might be correlated with student ratio and also affect attitudes. These variables are discussed in more detail in appendix A; they include fixed effects for the county in which the individual’s constituency is located, the constituency’s historical population size and current population density, the ratio of its non-white population to its total population,5 the individual’s age, gender, education level, marital status, employment status, and religious affiliation, whether the individual is a student, the per capita income level of the individual’s household, whether the household contains children or elderly dependents, and several different measures of the individual’s psychological characteristics. Table 3 includes the estimates of the coefficients on the logarithm of student ratio along with the corresponding t-ratios; these t-ratios are based on standard errors that allow for clustering at the constituency level. Estimates of the coefficients on the other explanatory variables are omitted from table 3 but appear in appendix A. Because an ordered Probit model is non-linear, the coefficients in table 3 do not in themselves indicate the average effect of the logarithm of student ratio on the individual responses. For this reason, table 3 also reports a marginal effect alongside each coefficient. If we assume a causal interpretation of the coefficients in table 3, the marginal effect is the effect of a unit increase in the logarithm of student ratio on the probability that the dependent variable will take a value greater than the central one, that is, that the individual will agree or strongly agree with the statement, or will choose Leave in the EU referendum. Note that this marginal effect measures the effect of student ratio while holding constant all of the other personal and constituency characteristics in the model. Table 3. Estimated Coefficients on the Logarithm of Student Ratio in Ordered Probit Models of 2015 and 2010 BES Attitudes   Attitudes in the 2015 BES  Attitudes in the 2010 BES    (i)  (ii)  (iii)  (iv)  (v)  (vi)  (vii)    Equal ops for gays/lesbians have gone too far.  Equal ops for eth. minorities have gone too far.  Equal ops for women have gone too far.  Immigrants are a burden on the welfare state.  Voting intention in the EU referendum  Ranking of immigration as an election issue  Disapproval of the EU  Coefficient  −0.028  −0.058  −0.056  −0.070  −0.070  −0.063  −0.091  t-ratio  −1.65*  −3.29***  −3.69***  −3.45***  −3.87***  −1.93*  −2.97***  Marginal effect  −0.010  −0.021  −0.023  −0.024  −0.026  −0.024  −0.031  Sample size  19,838  19,838  19,838  19,842  19,843  8,364  8,048    Attitudes in the 2015 BES  Attitudes in the 2010 BES    (i)  (ii)  (iii)  (iv)  (v)  (vi)  (vii)    Equal ops for gays/lesbians have gone too far.  Equal ops for eth. minorities have gone too far.  Equal ops for women have gone too far.  Immigrants are a burden on the welfare state.  Voting intention in the EU referendum  Ranking of immigration as an election issue  Disapproval of the EU  Coefficient  −0.028  −0.058  −0.056  −0.070  −0.070  −0.063  −0.091  t-ratio  −1.65*  −3.29***  −3.69***  −3.45***  −3.87***  −1.93*  −2.97***  Marginal effect  −0.010  −0.021  −0.023  −0.024  −0.026  −0.024  −0.031  Sample size  19,838  19,838  19,838  19,842  19,843  8,364  8,048  Note: The table reports the coefficients on the logarithm of student ratio in ordered Probit models of different attitudes of respondents in the BES. Student ratio is the ratio of the student population to the rest of the population in the respondent’s parliamentary constituency in 2011. The attitudes are measured on a five-point scale, except for column (v) with a three-point scale and column (vi) with a four-point scale. For columns (i–iv), higher values on the scale indicate greater agreement with the statement heading that column. For column (v), higher values indicate a preference for Leave versus Remain in the EU referendum. For column (vi), the scale corresponds to the ranking of immigration among the most important election issues, higher values indicating a higher rank. In column (vii), higher values on the scale indicate less approval of the UK’s membership of the EU. See the main text for a discussion of the other regressors. Marginal effects measure the impact of a unit increase in the logarithm of student ratio on the probability of an outcome higher than the central value, or in column (vi) on the probability of a rank of third or higher. One asterisk (*) indicates a coefficient significantly different from zero at the 10 percent level; three asterisks (***) indicate significance at the 1 percent level. The 2015 sample is the same as in table 1. Table 3. Estimated Coefficients on the Logarithm of Student Ratio in Ordered Probit Models of 2015 and 2010 BES Attitudes   Attitudes in the 2015 BES  Attitudes in the 2010 BES    (i)  (ii)  (iii)  (iv)  (v)  (vi)  (vii)    Equal ops for gays/lesbians have gone too far.  Equal ops for eth. minorities have gone too far.  Equal ops for women have gone too far.  Immigrants are a burden on the welfare state.  Voting intention in the EU referendum  Ranking of immigration as an election issue  Disapproval of the EU  Coefficient  −0.028  −0.058  −0.056  −0.070  −0.070  −0.063  −0.091  t-ratio  −1.65*  −3.29***  −3.69***  −3.45***  −3.87***  −1.93*  −2.97***  Marginal effect  −0.010  −0.021  −0.023  −0.024  −0.026  −0.024  −0.031  Sample size  19,838  19,838  19,838  19,842  19,843  8,364  8,048    Attitudes in the 2015 BES  Attitudes in the 2010 BES    (i)  (ii)  (iii)  (iv)  (v)  (vi)  (vii)    Equal ops for gays/lesbians have gone too far.  Equal ops for eth. minorities have gone too far.  Equal ops for women have gone too far.  Immigrants are a burden on the welfare state.  Voting intention in the EU referendum  Ranking of immigration as an election issue  Disapproval of the EU  Coefficient  −0.028  −0.058  −0.056  −0.070  −0.070  −0.063  −0.091  t-ratio  −1.65*  −3.29***  −3.69***  −3.45***  −3.87***  −1.93*  −2.97***  Marginal effect  −0.010  −0.021  −0.023  −0.024  −0.026  −0.024  −0.031  Sample size  19,838  19,838  19,838  19,842  19,843  8,364  8,048  Note: The table reports the coefficients on the logarithm of student ratio in ordered Probit models of different attitudes of respondents in the BES. Student ratio is the ratio of the student population to the rest of the population in the respondent’s parliamentary constituency in 2011. The attitudes are measured on a five-point scale, except for column (v) with a three-point scale and column (vi) with a four-point scale. For columns (i–iv), higher values on the scale indicate greater agreement with the statement heading that column. For column (v), higher values indicate a preference for Leave versus Remain in the EU referendum. For column (vi), the scale corresponds to the ranking of immigration among the most important election issues, higher values indicating a higher rank. In column (vii), higher values on the scale indicate less approval of the UK’s membership of the EU. See the main text for a discussion of the other regressors. Marginal effects measure the impact of a unit increase in the logarithm of student ratio on the probability of an outcome higher than the central value, or in column (vi) on the probability of a rank of third or higher. One asterisk (*) indicates a coefficient significantly different from zero at the 10 percent level; three asterisks (***) indicate significance at the 1 percent level. The 2015 sample is the same as in table 1. In order to check the robustness of our results using data from the 2015 round of the BES, we also fit ordered Probit models of two attitude variables from the 2010 round. The range of questions in the 2010 round is narrower, but there are questions relating to immigrants and EU membership. First, respondents were asked to rank a list of issues in terms of their importance for the election, one issue being immigration. We construct a variable that equals four if immigration is ranked first, three if it is ranked second, two if it is ranked third, and one if it is unranked.6 (In the context of the election, it is very likely that a high rank reflected an unfavorable view of immigrants.) Second, respondents were asked about their level of approval of Britain’s membership in the EU. We construct a variable that equals one for responses of “strongly approve,” two for “approve,” three for “don’t know,” four for “disapprove,” and five for “strongly disapprove.” Results for these two additional variables also appear in table 3. All of the coefficients in table 3 are negative and significantly different from zero at the 10 percent level, and five out of seven are significant at the 1 percent level. In column (i), the reported marginal effect is –0.01, indicating that a unit increase in the logarithm of student ratio is associated with a one-percentage-point fall in the probability that the individual will agree (or strongly agree) that gay rights have gone too far. In order to interpret the magnitude of this effect, we note that the standard deviation of the logarithm of student ratio is 0.67, so, for example, a three-standard-deviation increase in this variable is associated with a two-percentage-point fall in the probability. The marginal effects for the other attitude variables are between two and three times as large as this. For example, the marginal effect in column (iv) is –0.024, indicating that a three-standard-deviation increase in the logarithm of student ratio reduces the probability of agreeing (or strongly agreeing) that immigrants are a burden on the welfare state by about five percentage points. Recall from table 1 (column iv) that the overall proportion of respondents agreeing or strongly agreeing is 55 percent. The table 3 results imply that in the average constituency, a large increase in student numbers (i.e., an increase in the logarithm of student ratio of over three standard deviations) will turn this anti-immigration majority into a minority. This suggests that the presence of a large university can make a substantial difference to those public opinions, such as opinions about immigration, which currently hold the attention of politicians. One caveat to a causal interpretation of these results is that we have still not dealt with our second problem: the potential endogeneity of student ratio to attitudes. For this reason, appendix A includes an alternative set of results that have been produced using an Instrumental Variables (IV) estimator. This estimator exploits exogenous variation in the size of the university-age population in order to identify the effect of student ratio on attitudes. The IV estimates are somewhat less precise, but five out of the seven coefficients are still significantly different from zero at the 5 percent level. The corresponding marginal effects are a little larger than in table 3, but this difference is not statistically significant. This gives us some confidence in asserting that in the twenty-first century, holding constant other determinants of individual attitudes (such as the individual’s own age, education, and income level), these attitudes are significantly more liberal in constituencies with large numbers of university students. Locations with large universities are indeed more liberal. Our next task is to establish whether there is any substantial inter-temporal persistence in the geographical variation in the density of educational institutions. Inter-Temporal Persistence and the Geographical Distribution of Twenty-First-Century Universities Until the middle of the nineteenth century, England had only two universities, so we cannot focus exclusively on universities if we wish to measure inter-temporal persistence in the density of educational institutions over very long time horizons. Instead, we will explore the association between our twenty-first-century variable (the value of student ratio in each constituency) and the density of another type of educational institution, the privately funded library. The proliferation of libraries in the premodern period was a consequence of the fall in book prices and rise in literacy rates that followed the invention of moveable type in the late fifteenth century (Dittmar 2011; Raven 2006). Growth in book production and in the number of libraries was especially high in the eighteenth century; the most distinctive institution of this period was the private subscription library, which served the needs of a wide range of readers, including those from the growing middle class and skilled working class, and supplemented older types of library (for example the parish church library) that had arisen after the invention of moveable type. Libraries were places where like-minded people could meet and discuss the books they were reading: they were not teaching institutions, but they were institutions of learning. There was no public provision of library services until the Public Libraries Act of 1850, so regional variation in the location of libraries before this date is likely to have reflected regional variation in demand. Our information about the location of English libraries comes from the list compiled by Alston (2011). Using this list, it is possible to construct a dataset of the total number of libraries in each English town before 1850, and then to construct a constituency-level dataset in which each observation is the number of libraries in the largest town in the twenty-first-century parliamentary constituency.7 Table 4 illustrates the correlation between these library numbers and student ratio by grouping constituencies according to the number of premodern libraries in their largest town and showing the average value of student ratio for each group. It can be seen that on average, those constituencies with towns that were rich in libraries are also constituencies with large numbers of university students, and this correlation is consistent with inter-temporal persistence in the density of educational institutions. There are two possible explanations for such a correlation. First, there might be inter-temporal persistence in some of the local characteristics that have always been favorable to the establishment of educational institutions; such characteristics include the level of urbanization and prosperity of the local area. Second and more interestingly, there might be inter-temporal persistence in the density of educational institutions conditional on these characteristics. In other words, among constituencies that are otherwise similar, having had a large number of libraries is associated with a higher value of student ratio. Such an association would arise if there were persistent geographical variation in local preferences, or if the establishment of libraries created a local educational infrastructure that later reduced the cost of establishing universities. Table 4. Average Values of Student ratio Disaggregated by the Number of Premodern Libraries Number of premodern libraries in the largest town in the constituency  Number of constituencies  Average value of student ratio across constituencies  0–9  242  2.58%  10–19  85  2.83%  20–29  38  4.13%  30–39  11  7.03%  40–49  14  7.43%  50-plus  70  8.65%  Number of premodern libraries in the largest town in the constituency  Number of constituencies  Average value of student ratio across constituencies  0–9  242  2.58%  10–19  85  2.83%  20–29  38  4.13%  30–39  11  7.03%  40–49  14  7.43%  50-plus  70  8.65%  Note: Each row of the table corresponds to a different group of constituencies, constituencies being grouped according to the number of premodern libraries in their largest town. The first column indicates the number of libraries, the second column the group size, and the third column the average value of student ratio for that group. Student ratio is the ratio of the student population to the rest of the constituency population in 2011. The sample consists of all 460 English parliamentary constituencies outside London. Table 4. Average Values of Student ratio Disaggregated by the Number of Premodern Libraries Number of premodern libraries in the largest town in the constituency  Number of constituencies  Average value of student ratio across constituencies  0–9  242  2.58%  10–19  85  2.83%  20–29  38  4.13%  30–39  11  7.03%  40–49  14  7.43%  50-plus  70  8.65%  Number of premodern libraries in the largest town in the constituency  Number of constituencies  Average value of student ratio across constituencies  0–9  242  2.58%  10–19  85  2.83%  20–29  38  4.13%  30–39  11  7.03%  40–49  14  7.43%  50-plus  70  8.65%  Note: Each row of the table corresponds to a different group of constituencies, constituencies being grouped according to the number of premodern libraries in their largest town. The first column indicates the number of libraries, the second column the group size, and the third column the average value of student ratio for that group. Student ratio is the ratio of the student population to the rest of the constituency population in 2011. The sample consists of all 460 English parliamentary constituencies outside London. In order to investigate whether there is any conditional persistence in the density of educational institutions, we fit an Ordinary Least Squares (OLS) regression equation in which the dependent variable is the logarithm of student ratio and one of the explanatory variables is the number of premodern libraries in the largest town in the constituency. The distribution of the number of libraries is quite highly skewed—it has a mean of 26 and a maximum value of 237—so the explanatory variable is trimmed so that its maximum value is 50; this variable is denoted libraries in table 5.8 The other explanatory variables include county fixed effects and two measures of urbanization: the constituency population per hectare in the 2011 census (denoted population density) and the population of the largest town in the constituency in the 1841 census, that is, at the end of the premodern period (denoted town population in 1841). The 1841 population numbers are measured in tens of thousands and trimmed at 75,000; the figures are taken from Bennett (2011, appendix 3). Also included are an indicator variable for constituencies with a port (denoted port town) and a measure of the level of social and economic development in the constituency. There are a number of alternative measures of social and economic development that are all highly correlated with one another. The results in table 5 are based on a model that includes the ratio of the number of workers in a constituency who are in a professional occupation (i.e., in occupational categories 1–2) to the number who are not, as reported in the 2011 census; this variable is denoted professional ratio. Results using alternative measures such as unemployment rates or ACORN classifications are very similar, and are available on request. Descriptive statistics for all of the variables appear in appendix B.9 Table 5. Determinants of the Logarithm of Student Ratio: Ordinary Least Squares Estimates   Coefficient  t-ratio  Logarithm of professional ratio  0.298  3.15***  Population density  0.014  6.83***  Port town  0.201  3.21***  Town population in 1841  −0.024  −1.11  Libraries  0.016  5.44***  R2  0.72    Coefficient  t-ratio  Logarithm of professional ratio  0.298  3.15***  Population density  0.014  6.83***  Port town  0.201  3.21***  Town population in 1841  −0.024  −1.11  Libraries  0.016  5.44***  R2  0.72  Note:Student ratio is the ratio of the student population to the rest of the constituency population in 2011; the sample is the same as in table 4. Professional ratio is the ratio of workers in social classes 1–2 to other classes in the constituency in 2011; population density is the constituency population per hectare in 2011; port town is a binary variable equal to one in constituencies with a port and zero otherwise; town population in 1841 is the population of the largest town in the constituency (in tens of thousands) in 1841, trimmed at 75,000; libraries is the number of premodern libraries established in the largest town in the constituency, trimmed at 50. The set of explanatory variables also includes county fixed effects and the variables discussed in the second part of appendix A: estimates of the relevant coefficients are available on request. Three asterisks (***) indicate a coefficient significantly different from zero at the 1 percent level. Table 5. Determinants of the Logarithm of Student Ratio: Ordinary Least Squares Estimates   Coefficient  t-ratio  Logarithm of professional ratio  0.298  3.15***  Population density  0.014  6.83***  Port town  0.201  3.21***  Town population in 1841  −0.024  −1.11  Libraries  0.016  5.44***  R2  0.72    Coefficient  t-ratio  Logarithm of professional ratio  0.298  3.15***  Population density  0.014  6.83***  Port town  0.201  3.21***  Town population in 1841  −0.024  −1.11  Libraries  0.016  5.44***  R2  0.72  Note:Student ratio is the ratio of the student population to the rest of the constituency population in 2011; the sample is the same as in table 4. Professional ratio is the ratio of workers in social classes 1–2 to other classes in the constituency in 2011; population density is the constituency population per hectare in 2011; port town is a binary variable equal to one in constituencies with a port and zero otherwise; town population in 1841 is the population of the largest town in the constituency (in tens of thousands) in 1841, trimmed at 75,000; libraries is the number of premodern libraries established in the largest town in the constituency, trimmed at 50. The set of explanatory variables also includes county fixed effects and the variables discussed in the second part of appendix A: estimates of the relevant coefficients are available on request. Three asterisks (***) indicate a coefficient significantly different from zero at the 1 percent level. One potential concern with the OLS estimates is that the level of social and economic development in a constituency could be endogenous to the presence of universities there. For this reason, appendix B includes an alternative set of results from a model fitted using an IV estimator, with instruments that are based on the relative intensity of coal mining across the country. These results are very similar to those reported in table 5. Table 5 shows that constituencies with higher levels of professional ratio and population density have significantly higher levels of student ratio, as do constituencies with a port town. However, population levels in the 1841 census have no significant association with the level of student ratio. Conditional on all of these effects, there is a significant association between student ratio and libraries. The estimated value of the coefficient is 0.016, implying that each additional library is associated with a value of student ratio that is about 1.6 percent higher. To put it another way, the difference between the maximum and minimum values of the library variable (i.e., 50 libraries) corresponds to an increase in the logarithm of student ratio of 0.016 × 50 = 0.8, which is slightly greater than one standard deviation of the dependent variable. This effect is significant at the 1 percent level and corresponds to a high level of conditional persistence in the density of educational institutions. Our next task is to explain the geographical distribution of premodern libraries, and to measure the extent to which the location of libraries was influenced by medieval exposure to ethnic and religious diversity. Explaining the Geographical Variation in Premodern Libraries In this section, we will present some estimates of the factors that explain the variation across towns in the number of premodern libraries, and of the bookstores that supplied them. We begin with a discussion of two town-level characteristics that are likely to be correlated with medieval exposure to ethnic and religious diversity, explaining how these characteristics might be associated with attitudes toward outgroups, and therefore with openness to educational institutions (such as libraries) that embody new ideas. This discussion is informed by the literature on intergroup contact theory originating with Allport (1954). Allport contends that the context in which two groups meet is important in determining whether the contact leads to positive or negative views of the other group, and the subsequent literature has identified several characteristics that are associated with a positive effect on attitudes toward outgroups. These characteristics include an absence of competition between the two groups and equal social status: see for example the surveys of Dovidio, Glick, and Rudman (2005) and Pettigrew and Tropp (2006, 2012). Exposure to Diversity through Jewish Immigration The largest immigrant group in early medieval England was the Jews, the first Jewish communities arriving from France soon after the Norman Conquest. Hillaby (2003) discusses tax records that suggest that by the late twelfth century there were Jewish communities in over 20 English towns, while Mundill (2010) explains the two main economic functions of these communities: to provide financial services (lending money at interest being forbidden between Christians) and to provide a tax base for the king that was independent from the normal feudal system and therefore from the influence of the barons. In the thirteenth century, the state began to regulate the settlement of Jewish communities more directly, and Jews were officially permitted to live only in designated towns that contained a chest (or archa) where contracts between Jews and Christians had to be deposited (Brand 2003; Brown and McCartney 2005). By the mid-thirteenth century, there were archae in 30 towns; appendix C includes a list of these locations taken from Hillaby and Hillaby (2013). Growing anti-Semitism finally led to the expulsion of all Jews from England in 1290, after which there was no substantial foreign immigration into England until the arrival of the Huguenots in the late seventeenth century.10 The prevalence of anti-Semitic attitudes in medieval England was probably very high, and racially motivated attacks on individual Jews seem to have been common. However, as noted by Mundill (2010), genocidal attacks on whole communities appear to have been restricted to two periods. The first of these was 1189–1190, following clashes between Jews and Christians at the coronation of Richard I, and the second was the civil war of 1263–1265. The late twelfth-century attacks often seem to have been perpetrated by people from outside town: for example, foreign merchants instigated the attack at Lynn.11 Attacks during the civil war were perpetrated by baronial armies seeking to terrorize communities who supported the king, including the Jews (Hillaby 2003; Stacey 2003). Compared with fourteenth-century Germany during the Black Death (Voigtländer and Voth 2013), there is relatively little evidence of genocidal attacks on English Jews perpetrated by their near neighbors. Fielding (2018) discusses whether direct contact with a Jewish community would be likely to improve or worsen attitudes toward the Jews, on average. Noting that the majority of ordinary people would interact with the Jews on the basis of equal social standing (the Jews as vassals of the king, the common people as vassals of the local lord) and during peaceful trade (Jews buying goods from local merchants: Jews lent money only to wealthy individuals and organizations with collateral, and were barred from economic activities that would put them in competition with Christians), intergroup contact theory suggests that contact would lead to an improvement in attitudes among the bulk of the population. Langmuir (1963) reviews the evidence on this point, concluding that “the majority of the little evidence that there is suggests that it was primarily those who lived in close contact with Jews who were friendly with them.” If these arguments are correct, then we should expect the presence of an archa in the thirteenth century to indicate a town with a history of intergroup contact promoting greater openness to other groups and a more favorable environment for the establishment of educational institutions. Exposure to Diversity through the Crusades Although foreign travel was generally very uncommon in medieval England, one specific group of Englishmen who did travel were the crusaders. What was the likely effect of crusading on their opinions about other races and religions? In order to address this question, it is important to note that the majority of crusaders probably had very little choice about whether to fight. With the possible exception of the First Crusade (1095–1099), crusades were the initiative of senior churchmen or nobles. Although in theory their vassals were not required to accompany them, there was a great deal of social pressure to follow one’s lord to the Holy Land, sometimes accompanied by financial inducements such as the cancellation of debts (Benjamin 2015). Low-ranking knights and their retainers went on crusade if their lord did. For the vassals who made up the bulk of a crusading army, participation in the crusade probably represented a treatment effect, not a selection effect. This treatment might include exposure to more intense xenophobic propaganda and face-to-face fighting that induced negative views of the enemy population. However, having established states in Antioch, Edessa, Tripoli, and Acre/Jerusalem, the crusaders were required to engage in trade and diplomacy with neighboring Arab states, and to manage local economies that were based on the output of Arab farmers. These activities involved men of many different ranks, and historical research suggests that this led to a level of understanding of Arab culture, religion, and language that stood in marked contrast to the ignorance prevalent in Western Europe (Attiya 1999; Hamilton 1997). Moreover, the division of labor within a crusader state between the Arab population (primary production) and the European population (administration and services) meant that day-to-day contact between them did not involve competition, and this absence of competition could have fostered positive attitudes toward the outgroup. Historical accounts by crusaders such as Jean de Joinville include sympathetic depictions of individual Arabs, both rulers and those of lower rank (Khanmohamadi 2010; Rouleau 2005). This eventually influenced popular Western European fiction, in which Arabs were depicted not only as villains but also as heroes (Calkin 2012; Hamilton 1997). One vehicle for this influence was the return of crusading forces following their various defeats, for example the fall of Edessa (1144), Jerusalem (1187), Antioch (1268), Tripoli (1289), and Acre (1291). The salience of participation in the crusades to many returning English nobles was reflected in the modification of their coats of arms, most commonly by the addition of a Saracen’s (i.e., Arab’s) head. In interpreting the Saracen’s head, it is important to note that heads can be displayed on coats of arms in a variety of different ways, each with its own heraldic terminology. Violence and disrespect to an enemy are indicated by the appearance of a head that is erased (i.e., decapitated), on the point of a pheon (i.e., spear), or distilling blood. Such depictions have been popular among noble families in Wales and Scotland, but they are relatively rare in England. Of the 84 English coats of arms featuring Saracen’s heads that are listed in Burke (1884), only eight depict violence of this kind. English arms typically display a Saracen’s head that is couped (i.e., depicted in the style of a king’s head on a coin), a neutral representation of contact with a foreign culture that does not imply disrespect. One popular equivalent of a noble’s coat of arms was the inn sign, many signs being modeled on the arms of a local lord. Therefore, one indication of the salience of the crusades to a local community is an inn named the Saracen’s Head. One limitation that we face is that there is no medieval documentary source that lists inns of this name; however, it is possible to use the address list in the 1851 census to locate Saracen’s Head inns that survived to the nineteenth century, and a list of these 88 locations is provided in appendix C. This list can be used to create an indicator variable for the presence or absence of a Saracen’s Head inn in each location.12 Both the original distribution of Saracen’s Head inns in the twelfth and thirteenth centuries and the rate of attrition up to the nineteenth century are likely to be influenced by a number of confounding factors: for example, the original distribution is likely to depend on medieval population densities and the rate of attrition on subsequent population growth rates. However, if we control for these factors, then we can interpret our indicator variable as a correlate of the salience to a local community of participation in the crusades, the phenomenon that brought medieval Englishmen most directly into contact with large numbers of foreigners. If the arguments above are correct, then we should expect the presence of a Saracen’s Head inn to reflect a history of intergroup contact promoting greater openness to other groups and a more favorable environment for the establishment of educational institutions. Other Relevant Medieval Institutions Aside from exposure to ethnic and religious diversity, there are some other characteristics of medieval towns that might later have affected the demand for educational institutions. If these other characteristics are also correlated with the location of archae or of Saracen’s Head inns, then it will be important to control for them in our empirical model of premodern library location. Institution #1: The School Libraries might have been easier to establish in towns where there was already some form of formal education. Until the mid-nineteenth century, there were only two universities in England, and most formal education took place in schools. Our information about medieval schools is drawn from Orme (2006), who provides a comprehensive discussion of medieval English education. The evidence that Orme presents shows that medieval schools were sophisticated, diverse, and independent of control by any central authority. They provided a context for encounters with foreign languages and cultures. Despite the differences between modern and medieval worldviews, medieval schools fulfilled the same functions as do modern ones: medieval schools were “the same thing in different circumstances.” Before the Norman Conquest, schools were few in number and largely restricted to monastic institutions, but the twelfth century saw the establishment of other types of school, especially grammar schools with secular teachers and students. Although monastic schools were solely for the purpose of educating those in religious orders, cathedral schools were open to a wider range of students from different walks of life. Our data on medieval school location are taken from Orme (2006, 190–94). Appendix C includes a list of these 83 locations; the list includes all known schools established in the twelfth or thirteenth century,13 except monastic establishments from which the local population would have been excluded. Institution #2: The Manuscript Library Even before the invention of moveable type, which was introduced into England toward the end of the fifteenth century, large medieval schools and monasteries were supported by libraries of handwritten books. The typical library contained a relatively small number of books, but the subject matter of these books was very diverse (Kibre 1946). The existence of such a manuscript library in a town may have facilitated the later development of libraries with printed books. Appendix C includes a list of 97 locations where a library is known to have been established before the year 1500; this information comes from Alston (2011). The Distribution of Medieval Schools, Libraries, Archae, and Saracen’s Head Inns Appendix C includes an analysis of the geographical distribution of medieval libraries, schools, archae, and Saracen’s Head inns. Although our four variables are positively correlated, the correlations are explained by simple factors such as medieval town size and distance from London. Conditional on these factors, there are no significant correlations, so we do have four distinct characteristics of medieval towns. In the data analysis that follows, we estimate the extent to which the location of premodern libraries and bookstores depends on our four medieval variables. We interpret the effect of medieval libraries and schools as persistence in the importance of educational institutions, which could be explained either by local variation in preferences for education (a cultural characteristic) or by the medieval institutions generating sector-specific human and physical capital in certain areas. We interpret the effect of archae and Saracen’s Head inns as the influence of attitudes toward outgroups on later institutional location. It is the latter that will provide unambiguous evidence for the effect of culture on institutions. It is important to make one further point regarding the interpretation of the archa and Saracen’s Head effects. The preceding discussion highlights the possibility that the presence of Jews and returning crusaders had a treatment effect on the local community, and this treatment effect could be interpreted in terms of intergroup contact theory. An alternative possibility is that the presence of Jews and returning crusaders reflected a selection effect: that is, the medieval migrants were more likely to settle in places where the existing community was already relatively cosmopolitan. In the absence of further medieval data, it is difficult to estimate the relative importance of treatment and selection effects. The absence of a significant conditional correlation between the location of archae and the location of Saracen’s Head inns suggests that a treatment effect is more likely: if certain places were already relatively cosmopolitan, why didn’t Jews and returning crusaders both settle in these places? However, the sample of medieval towns used in appendix C is quite small, which limits the power of the test of the significance of the conditional correlation. Nevertheless, regardless of the relative importance of the treatment and selection effects, our archa and Saracen’s Head inn variables identify those places that were relatively cosmopolitan in the Middle Ages. Analyzing the Libraries Data The figures in table 6 illustrate the association between the number of premodern libraries in a town (excluding the medieval ones) and the location of medieval schools, libraries, archae, and Saracen’s Head inns. The libraries data are again based on Alston (2011), using a sample of 838 English towns for which adequate population data are available from Bennett (2011, appendix 3); London is again excluded from our analysis. Among these 838 towns, there are 690 with at least one premodern library and 6,057 libraries in total, compared with 122 medieval libraries in 95 towns. There are 19 towns with over 50 libraries, including the two university towns of Oxford and Cambridge; the largest number—237—is for Liverpool. None of the results in this section are very different if we exclude these 19 towns (or just Oxford and Cambridge) from the sample. Table 6. Average Number of Premodern Libraries, Disaggregated by Medieval Characteristics Medieval characteristic of the town  Average number of premodern libraries when the medieval characteristic is absent  Average number of premodern libraries when the medieval characteristic is present  Medieval school  5.51  21.35  Medieval library  5.64  18.32  Archa  6.06  35.45  Saracen’s Head inn  4.68  28.07  Medieval characteristic of the town  Average number of premodern libraries when the medieval characteristic is absent  Average number of premodern libraries when the medieval characteristic is present  Medieval school  5.51  21.35  Medieval library  5.64  18.32  Archa  6.06  35.45  Saracen’s Head inn  4.68  28.07  Note: The sample comprises 838 English towns outside London. Here, unlike in tables 4 and 5, the premodern library numbers exclude medieval libraries (i.e., ones founded before 1500). Table 6. Average Number of Premodern Libraries, Disaggregated by Medieval Characteristics Medieval characteristic of the town  Average number of premodern libraries when the medieval characteristic is absent  Average number of premodern libraries when the medieval characteristic is present  Medieval school  5.51  21.35  Medieval library  5.64  18.32  Archa  6.06  35.45  Saracen’s Head inn  4.68  28.07  Medieval characteristic of the town  Average number of premodern libraries when the medieval characteristic is absent  Average number of premodern libraries when the medieval characteristic is present  Medieval school  5.51  21.35  Medieval library  5.64  18.32  Archa  6.06  35.45  Saracen’s Head inn  4.68  28.07  Note: The sample comprises 838 English towns outside London. Here, unlike in tables 4 and 5, the premodern library numbers exclude medieval libraries (i.e., ones founded before 1500). Each row of table 6 corresponds to a different medieval characteristic; one column of the table shows the average number of premodern libraries in towns where the characteristic is present, while another column shown the average number of libraries in towns where the characteristic is absent. It can be seen that the average number of libraries is always much higher in towns where the characteristic is present. These associations do not necessarily entail a direct link between the medieval characteristics and library numbers: they could simply reflect the fact that larger medieval towns were more likely to have the characteristics than smaller ones, larger premodern towns had more libraries, and there is some inter-temporal persistence in the relative sizes of towns. For this reason, we also fit a regression equation in which the dependent variable is the number of premodern libraries in a town. The set of explanatory variables includes binary variables indicating the presence of a medieval school, library, archa, or Saracen’s Head inn (denoted medieval school, medieval library, archa, saracen), plus county fixed effects (capturing regional characteristics such as distance to London), the logarithm of the town’s population in the 1841 census (denoted population in 1841), and the logarithm of the town’s population in the mid-seventeenth century (denoted population in c17). The population measures are taken from Bennett (2011, appendix 3). Also included in the regression equation is a medieval population measure: this variable (denoted population in c14) is equal to the logarithm of the town’s adult male population in the 1377 Poll Tax records, if the town is large enough to appear in the list of Dyer (2000).14 Finally, there is a binary variable indicating the presence of the town in Dyer’s list (denoted town in c14). Descriptive statistics for these variables appear in appendix B. The distribution of libraries across towns is highly skewed and approximately equal to a Poisson distribution (see figure 1), so we fit a Poisson model to the data. This means that each of the estimated coefficients can be interpreted as the percentage change in the number of libraries associated with a unit change in the explanatory variable. The first set of results in table 7 (column (i)) shows the values of these coefficients and the corresponding t-ratios. The results indicate that the number of libraries is strongly associated with the size of the town in the seventeenth and nineteenth centuries, but for a given population level, a town with a medieval school can be expected to have 27 percent more libraries and a town with a Saracen’s Head inn can be expected to have 30 percent more libraries; both of these effects are significant at the 1 percent level. The medieval library and archa effects are statistically insignificant. Figure 1. View largeDownload slide Histogram of Libraries (excluding towns in which Libraries >50) Figure 1. View largeDownload slide Histogram of Libraries (excluding towns in which Libraries >50) It is possible that these results do not adequately capture all of the town characteristics that might be associated both with the location of medieval institutions and with premodern library density. For this reason, the second set of results in table 7 (column (ii)) shows coefficients and t-ratios from a model that includes an additional set of explanatory variables. These are: an indicator variable for whether the town is a port (denoted port town), the number of unskilled laborers in town as a fraction of the total working population (denoted laborers), the total number of inns in the town (denoted inns),15 laborers,2 and inns.2 These figures are based on information about the number of laborers and innkeepers in each parish over the years 1813–1820: see Shaw-Taylor, Wrigley, and Kitson (2006). The fraction of unskilled laborers in the town is included as a measure of relative poverty. The number of inns is included in order to capture variation in the number of visitors to the town, which may influence the number of other facilities there, including libraries. It is also possible that port towns are better connected to London and so have more extensive facilities. Note that the Shaw-Taylor et al. dataset covers only half of the counties in England, and this reduces the column (ii) sample size from 838 towns to 348. Column (ii) shows that there is a statistically significant association between the number of libraries and the number of inns; conditional on this effect, there is no significant association between the number of libraries and number of medieval schools, medieval libraries, or archae. However, the size of the association with the number of Saracen’s Head inns is only slightly smaller than in column (i), and this effect is still significant at the 1 percent level. Table 7. Determinants of the Number of Premodern Libraries in a Town: Poisson Model Estimates   (i)  (ii)    Full sample (838 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  Coeff.  t-ratio  Town in c14  0.201  2.00**  0.235  1.72*  Logarithm of population in c14  −0.132  −1.61  −0.087  −0.64  Logarithm of population in c17  0.180  2.39**  0.216  2.01**  Logarithm of population in 1841  0.762  16.26***  0.746  11.87***  Medieval school  0.220  2.56***  0.183  1.03  Medieval library  0.107  1.14  0.054  0.31  Archa  0.031  0.32  0.066  0.28  Saracen  0.336  3.81***  0.270  2.44***  Laborers      0.486  0.51  Laborers2      −0.136  −0.09  Port town      0.204  1.51  Inns      0.291  2.45***  Inns2      −0.050  −2.40***    (i)  (ii)    Full sample (838 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  Coeff.  t-ratio  Town in c14  0.201  2.00**  0.235  1.72*  Logarithm of population in c14  −0.132  −1.61  −0.087  −0.64  Logarithm of population in c17  0.180  2.39**  0.216  2.01**  Logarithm of population in 1841  0.762  16.26***  0.746  11.87***  Medieval school  0.220  2.56***  0.183  1.03  Medieval library  0.107  1.14  0.054  0.31  Archa  0.031  0.32  0.066  0.28  Saracen  0.336  3.81***  0.270  2.44***  Laborers      0.486  0.51  Laborers2      −0.136  −0.09  Port town      0.204  1.51  Inns      0.291  2.45***  Inns2      −0.050  −2.40***  Note: The sample in column (i) is the same as in table 6; the sample in column (ii) omits towns lacking data for the variables laborers and inns. The number of premodern libraries in each town excludes medieval libraries (i.e., ones founded before 1500). Town in c14 is a binary variable equal to one if the town appears in the list of fourteenth-century towns in Dyer (2000), and zero otherwise; population in c14 is the population level recorded in Dyer’s list (or set equal to one if the town is not in the list); population in c17 and population in 1841 are population estimates for the mid-seventeenth century and for 1841. Laborers denotes the share of unskilled laborers in the total working population of the town; port town is a binary variable equal to one if the town is a port and zero otherwise; inns denotes the number of inns in the town. Medieval school (medieval library, archa, saracen) is a binary variable equal to one if there was a medieval library (medieval school, Jewish settlement, Saracen’s Head inn) in the town, and zero otherwise. The set of explanatory variables also includes county fixed effects. One asterisk (*) indicates a coefficient significantly different from zero at the 10 percent level; two asterisks (**) indicate significance at the 5 percent level, and three asterisks (***) indicate significance at the 1 percent level. Table 7. Determinants of the Number of Premodern Libraries in a Town: Poisson Model Estimates   (i)  (ii)    Full sample (838 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  Coeff.  t-ratio  Town in c14  0.201  2.00**  0.235  1.72*  Logarithm of population in c14  −0.132  −1.61  −0.087  −0.64  Logarithm of population in c17  0.180  2.39**  0.216  2.01**  Logarithm of population in 1841  0.762  16.26***  0.746  11.87***  Medieval school  0.220  2.56***  0.183  1.03  Medieval library  0.107  1.14  0.054  0.31  Archa  0.031  0.32  0.066  0.28  Saracen  0.336  3.81***  0.270  2.44***  Laborers      0.486  0.51  Laborers2      −0.136  −0.09  Port town      0.204  1.51  Inns      0.291  2.45***  Inns2      −0.050  −2.40***    (i)  (ii)    Full sample (838 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  Coeff.  t-ratio  Town in c14  0.201  2.00**  0.235  1.72*  Logarithm of population in c14  −0.132  −1.61  −0.087  −0.64  Logarithm of population in c17  0.180  2.39**  0.216  2.01**  Logarithm of population in 1841  0.762  16.26***  0.746  11.87***  Medieval school  0.220  2.56***  0.183  1.03  Medieval library  0.107  1.14  0.054  0.31  Archa  0.031  0.32  0.066  0.28  Saracen  0.336  3.81***  0.270  2.44***  Laborers      0.486  0.51  Laborers2      −0.136  −0.09  Port town      0.204  1.51  Inns      0.291  2.45***  Inns2      −0.050  −2.40***  Note: The sample in column (i) is the same as in table 6; the sample in column (ii) omits towns lacking data for the variables laborers and inns. The number of premodern libraries in each town excludes medieval libraries (i.e., ones founded before 1500). Town in c14 is a binary variable equal to one if the town appears in the list of fourteenth-century towns in Dyer (2000), and zero otherwise; population in c14 is the population level recorded in Dyer’s list (or set equal to one if the town is not in the list); population in c17 and population in 1841 are population estimates for the mid-seventeenth century and for 1841. Laborers denotes the share of unskilled laborers in the total working population of the town; port town is a binary variable equal to one if the town is a port and zero otherwise; inns denotes the number of inns in the town. Medieval school (medieval library, archa, saracen) is a binary variable equal to one if there was a medieval library (medieval school, Jewish settlement, Saracen’s Head inn) in the town, and zero otherwise. The set of explanatory variables also includes county fixed effects. One asterisk (*) indicates a coefficient significantly different from zero at the 10 percent level; two asterisks (**) indicate significance at the 5 percent level, and three asterisks (***) indicate significance at the 1 percent level. Libraries and Bookstores The expansion of libraries in the seventeenth and eighteenth centuries was facilitated by a growing provincial book trade (Feather 2006; Suarez 2009), and another indicator of local openness to education is the existence of a bookstore. Shaw-Taylor, Wrigley, and Kitson (2006) also provide data on the number of booksellers in each parish: this number was rarely greater than one, so it does not make sense to try to model the number of booksellers, but we can still explore whether any of the medieval characteristics is significantly associated with the presence of at least one bookseller in town. Table 8 shows some results with regard to booksellers: the structure of this table resembles that of table 7, but here the dependent variable takes a value of one if there was at least one bookseller in town during 1813–1820, and a value of zero if there was no bookseller. The table 8 results are based on a Probit model, so the table also includes a set of marginal effects: these numbers indicate the average change in the probability of there being a bookseller in town when there is a unit increase in the explanatory variable. Table 8. Determinants of the Probability of a Bookseller in Town in a Town: Probit Model Estimates   (i)  (ii)    Full sample (348 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  m.e.  Coeff.  t-ratio  m.e.  Town in c14  0.285  0.62  0.07  0.225  0.59  0.05  Logarithm of population in c14  −0.123  −0.24  −0.03  −0.145  −0.29  −0.03  Logarithm of population in c17  −0.059  −0.31  −0.01  −0.162  −0.82  −0.03  Logarithm of population in 1841  0.709  4.74***  0.16  0.968  5.29***  0.18  Medieval school  0.958  2.46***  0.31  1.055  2.53***  0.31  Medieval library  −0.297  −0.79  −0.06  −0.596  −1.47  −0.08  Archa  0.993  1.65*  0.33  1.076  1.83*  0.32  Saracen  0.273  0.93  0.07  0.410  1.41  0.09  Laborers        6.227  2.29**    Laborers2        −15.480  −3.44***    Port town        0.140  0.46  0.03  Inns        −0.020  −0.07    Inns2        −0.017  −0.28      (i)  (ii)    Full sample (348 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  m.e.  Coeff.  t-ratio  m.e.  Town in c14  0.285  0.62  0.07  0.225  0.59  0.05  Logarithm of population in c14  −0.123  −0.24  −0.03  −0.145  −0.29  −0.03  Logarithm of population in c17  −0.059  −0.31  −0.01  −0.162  −0.82  −0.03  Logarithm of population in 1841  0.709  4.74***  0.16  0.968  5.29***  0.18  Medieval school  0.958  2.46***  0.31  1.055  2.53***  0.31  Medieval library  −0.297  −0.79  −0.06  −0.596  −1.47  −0.08  Archa  0.993  1.65*  0.33  1.076  1.83*  0.32  Saracen  0.273  0.93  0.07  0.410  1.41  0.09  Laborers        6.227  2.29**    Laborers2        −15.480  −3.44***    Port town        0.140  0.46  0.03  Inns        −0.020  −0.07    Inns2        −0.017  −0.28    Note: The sample is the same as in table 7 (column (ii)). M.e. stands for the marginal effect of the variable; marginal effects for laborers and inns are not shown because these values are not constant: more information is available on request. See table 7 for other notes. Table 8. Determinants of the Probability of a Bookseller in Town in a Town: Probit Model Estimates   (i)  (ii)    Full sample (348 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  m.e.  Coeff.  t-ratio  m.e.  Town in c14  0.285  0.62  0.07  0.225  0.59  0.05  Logarithm of population in c14  −0.123  −0.24  −0.03  −0.145  −0.29  −0.03  Logarithm of population in c17  −0.059  −0.31  −0.01  −0.162  −0.82  −0.03  Logarithm of population in 1841  0.709  4.74***  0.16  0.968  5.29***  0.18  Medieval school  0.958  2.46***  0.31  1.055  2.53***  0.31  Medieval library  −0.297  −0.79  −0.06  −0.596  −1.47  −0.08  Archa  0.993  1.65*  0.33  1.076  1.83*  0.32  Saracen  0.273  0.93  0.07  0.410  1.41  0.09  Laborers        6.227  2.29**    Laborers2        −15.480  −3.44***    Port town        0.140  0.46  0.03  Inns        −0.020  −0.07    Inns2        −0.017  −0.28      (i)  (ii)    Full sample (348 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  m.e.  Coeff.  t-ratio  m.e.  Town in c14  0.285  0.62  0.07  0.225  0.59  0.05  Logarithm of population in c14  −0.123  −0.24  −0.03  −0.145  −0.29  −0.03  Logarithm of population in c17  −0.059  −0.31  −0.01  −0.162  −0.82  −0.03  Logarithm of population in 1841  0.709  4.74***  0.16  0.968  5.29***  0.18  Medieval school  0.958  2.46***  0.31  1.055  2.53***  0.31  Medieval library  −0.297  −0.79  −0.06  −0.596  −1.47  −0.08  Archa  0.993  1.65*  0.33  1.076  1.83*  0.32  Saracen  0.273  0.93  0.07  0.410  1.41  0.09  Laborers        6.227  2.29**    Laborers2        −15.480  −3.44***    Port town        0.140  0.46  0.03  Inns        −0.020  −0.07    Inns2        −0.017  −0.28    Note: The sample is the same as in table 7 (column (ii)). M.e. stands for the marginal effect of the variable; marginal effects for laborers and inns are not shown because these values are not constant: more information is available on request. See table 7 for other notes. Table 8 shows that the probability of there being a bookseller in town is significantly associated with its nineteenth-century population and with the fraction of the population who are unskilled laborers. Conditional on these effects, there is a significant association with both the presence of a medieval school, which is estimated to raise the bookseller probability by 31 percentage points in column (ii), and with the presence of an archa, which is estimated to raise the bookseller probability by 32 percentage points. However, neither the presence of a medieval library nor the presence of a Saracen’s Head inn has any significant effect. Taken together, tables 7 and 8 show evidence that the exposure of a town to medieval ethnic and religious diversity is significantly associated with a greater density of educational institutions at a later date. Towns with a Saracen’s Head inn, which suggests the salience of the crusades to the local community, had more premodern libraries, while towns with a Jewish community in the twelfth and thirteenth century were more likely to have a bookseller in the nineteenth century.16 The effects are large: the mean probability of a town having a bookstore is 0.17, and holding all other factors constant, the presence of a Jewish community is estimated to increase this probability to 0.49. The mean number of pre-1851 libraries is seven, and holding all other factors constant, the presence of a Saracen’s Head inn is estimated to increase this number by 30 percent, giving the town two more libraries. As we have seen, a larger number of pre-1851 libraries implies a larger number of students in the twenty-first century, and this implies a local population with more positive attitudes toward immigration and toward equal rights for gays and lesbians, ethnic minorities, and women. Conclusion In this paper, we have traced a chain of connections between (i) English attitudes toward minority rights, immigration, and European integration in twenty-first-century opinion polls and (ii) exposure to ethnic and religious diversity in the twelfth and thirteenth centuries. Attitudes are significantly more positive in locations that have larger universities (measured by student numbers), university size depends on the local density of earlier educational institutions, and this density is higher in places showing signs of medieval exposure to diversity. The persistence of geographical variation in the density of educational institutions seems to have been a vehicle for the persistence of geographical variation in attitudes. Is it the only vehicle? In appendix D, we present evidence suggesting that the answer to this question is “probably not.” When our indicators of medieval exposure to diversity (archa and saracen) are added to the model of attitudes summarized in table 3, their effect is found to be jointly statistically significant in five out of seven cases, the two exceptions being attitudes toward the European Union in 2010 and 2015. That is, even when we control for modern university size, there is still a significant association between twenty-first-century attitudes and medieval exposure to diversity. This conditional association seems to be largely through the archa effect (reflecting a local Jewish heritage) rather than through the saracen effect (reflecting local salience of the crusades). The other mechanisms responsible for the inter-generational persistence of regional variation in attitudes have yet to be identified, but they could include, for example, informal social interaction that reinforces conformity to local norms in a cultural evolutionary process of the kind proposed by Cavalli-Sforza and Feldman (1973). Further research into these mechanisms will be required for a full understanding of the reasons why social attitudes display such persistent geographical variation. While our results regarding the historical drivers of variation in the density of English educational institutions are consistent with existing evidence regarding the effect of international linkages on university expansion (Schofer and Meyer 2005), this effect is only half of the story. The presence (or absence) of a university has itself contributed to a high degree of local persistence in attitudes toward foreigners and minority groups. This second effect suggests that modern social and economic forces that cause universities to expand or contract could also lead to substantial changes in attitudes, with long-lasting political consequences. Notes 1 Throughout this paper, the term “English” refers specifically to England, excluding Wales and Scotland. Some of the data described in subsequent sections of the paper are not available for Wales or Scotland. 2 Results disaggregating men and women are available on request. Unfortunately, it is not possible to disaggregate the sample by sexual orientation. 3 In terms of census terminology, a university student is any person in full-time tertiary education. 4 We use a logarithmic transformation because the distribution of student ratio is highly skewed, with a few constituencies that have very high student numbers. 5 The immigrant population density is potentially endogenous to attitudes. Following Dustmann and Preston (2001), it is possible to use county-level population density as an instrument for constituency-level population density. Results using such an instrument are very similar to the ones reported in table 3. 6 This immigration variable has four possible values. In this case, the marginal effect measures the impact of a unit increase in the logarithm of student ratio on the probability that the value will be greater than one, that is, that immigration will be ranked as one of the top three issues. 7 Some constituencies do contain more than one town, and in these cases we select the town with the largest population in the 1841 census. The within-constituency ranking of towns by size has changed very little since this time, so our results are very similar if we use population numbers for a later date. Other constituencies comprise just part of one large city, and in these cases the library number is ascribed to all of the constituencies in the city. 8 An alternative approach to dealing with the skewness is to use a set of indicator variables for constituencies with over 30 libraries, constituencies with over 20 libraries, and constituencies with over 10 libraries. This produces results that are very similar to those in table 5. 9 The regression equation also includes the instrumental variables in the IV model of attitudes discussed in the second part of appendix A: these variables are uncorrelated with libraries (r < 0.01), but they do explain some of the variation in the logarithm of student ratio, and their inclusion does increase the precision of the estimates of the effect of other explanatory variables. 10 Huguenot communities assimilated very quickly to the local English culture, and the location of Huguenot settlements turns out not to be a significant factor in the models discussed below. 11 The one attack in which locals were clearly directly implicated was at York; this attack seems to have been organized by members of the petty nobility who had run up large debts to local Jews. 12 In most cases, there is no direct archaeological evidence with which to establish a precise date for the foundation of these inns. However, Saracen’s Head inn sites that have been excavated have produced evidence for a foundation date contemporary with or not much later than the crusading period (Andrews et al. 2003; Bowsher et al. 2007), and in some cases there is documentary evidence for a medieval foundation date (English Heritage 2010). 13 It is also possible to include in our dataset the location of late medieval schools. However, the rapid expansion of schools after 1300 meant that by the end of the Middle Ages almost all towns of moderate size had a school, and the location of early medieval schools is more useful in explaining the location of premodern libraries and bookstores. That is, the popularity of the later types of institution (libraries and bookstores) is predicted by early adoption of the original type of institution (schools). 14 The list includes all towns of at least 480 men. For towns not on the list, population in c14 is set equal to 480. The inclusion of the binary variable town in c14 in the model means that this arbitrary choice does not affect the results. 15 These figures incorporate all inns, public houses, and taverns. 16 Why are Saracen’s Head inns more important for library location and archae more important for bookseller location? Both libraries and bookstores reflect a demand for new ideas, but libraries are the more communal institution. Crusading was a group activity that exposed the individual to foreign culture at the same time as promoting an esprit de corps, and perhaps it created a taste for communal activity as well as openness to new ideas. Another possibility is that the presence of a bookstore indicates a town with an entrepreneurial spirit and a large financial sector that is favorable to small business. The presence of a medieval Jewish community could be associated with this type of business environment, in which case the archa effect is explained by an economic mechanism rather than a cultural one. We can test this conjecture by using the Shaw-Taylor, Wrigley, and Kitson (2006) dataset to identify those towns that were home to businesses selling luxury items other than books (for example, towns that were home to silversmiths), and then fitting a model of the presence of these businesses. If there is an economic mechanism at work, then the location of archae should predict the location of silversmiths as well as the location of booksellers. When we do fit a silversmith model, the estimated archa coefficient is insignificantly different from zero (p > 0.25 in all versions of the model). This suggests that the mechanism is not an economic one. About the Author David Fielding is Donald Reid Chair of Economics at the University of Otago, New Zealand. His current research examines the social and economic consequences of ethnic diversity, and drivers of the salience of ethnic difference, both in Europe and in Africa. He also researches attitudes toward foreign aid and aid allocation. Recent publications include papers in the British Journal of Political Science (2018), Journal of Development Studies (2017), and Oxford Bulletin of Economics and Statistics (2017). Supplementary Material Supplementary material is available at Social Forces online. References Alesina, Alberto, and Paola Giuliano. 2015. “ Culture and Institutions.” Journal of Economic Literature  53( 4): 898– 944. Google Scholar CrossRef Search ADS   Alesina, Alberto, Paola Giuliano, and Nathan Nunn. 2013. “ On the Origins of Gender Roles: Women and the Plough.” Quarterly Journal of Economics  128( 2): 469– 530. Google Scholar CrossRef Search ADS   Allport, Gordon W. 1954. The Nature of Prejudice . Reading, MA: Addison-Wesley. Alston, Robin. 2011. Libraries in England to 1850. www.scribd.com/doc/63097781/Robin-Alston-Library-History-England. Andrews, Philip, J. Lowe, K. Nichols, C. Phillpotts, and A. Powell. 2003. Charter Quay—The Spirit of Change: The Archaeology of Kingston’s Riverside . Salisbury, UK: Wessex Archaeology. Attiya, Hussein M. 1999. “ Knowledge of Arabic in the Crusader States in the Twelfth and Thirteenth Centuries.” Journal of Medieval History  25( 3): 203– 13. Google Scholar CrossRef Search ADS   Banaszak, Lee Ann, and Eric Plutzer. 1993. “ Contextual Determinants of Feminist Attitudes: National and Subnational Influences in Western Europe.” American Political Science Review  87( 1): 145– 57. Google Scholar CrossRef Search ADS   Benjamin, David J. 2015. Support Structures in Crusading Armies, 1095–1241. Doctoral dissertation, Leicester University. Bennett, Robert. J. 2011. Local Business Voice: The History of Chambers of Commerce in Britain, Ireland and Revolutionary America, 1760–2011 . Oxford: Oxford University Press. Google Scholar CrossRef Search ADS   Bobo, Lawrence, and Frederick C. Licari. 1989. “ Education and Political Tolerance: Testing the Effects of Cognitive Sophistication and Target Group Affect.” Public Opinion Quarterly  53( 3): 285– 308. Google Scholar CrossRef Search ADS   Bowsher, David, Tony Dyson, Nick Holder, and Isca Howell. 2007. The London Guildhall: An Archaeological History of a Neighbourhood from Early Medieval to Modern Times . London: MoLAS. Brand, Paul. 2003. “The Jewish Community of England in the Records of English Royal Government.” In The Jews in Medieval Britain: Historical, Literary and Archaeological Perspectives , edited by Patricia Skinner, 73– 85. Woodbridge, UK: Boydell Press. Brown, Reva B., and Sean McCartney. 2005. “ The Exchequer of the Jews Revisited: The Operation and Effect of the Scaccarium Judeorum.” Medieval History Journal  8( 2): 303– 22. Google Scholar CrossRef Search ADS   Bruter, Michael. 2005. Citizens of Europe? The Emergence of a Mass European Identity . Basingstoke, UK: Palgrave Macmillan. Google Scholar CrossRef Search ADS   Burke, Bernard. 1884. The General Armory of England, Scotland, Ireland and Wales . London: Harrison. Calkin, Siobhain B. 2012. “Saracens.” In Heroes and Anti-Heroes in Medieval Romance , edited by Neil Cartlidge, 185– 200. Cambridge: Brewer. Cavalli-Sforza, Luca, and Marcus W. Feldman. 1973. “ Models for Cultural Inheritance I: Group Mean and Within-Group Variation.” Theoretical Population Biology  4( 1): 42– 55. Google Scholar CrossRef Search ADS PubMed  Coenders, Marcel, and Peer Scheepers. 2003. “ The Effect of Education on Nationalism and Ethnic Exclusionism: An International Comparison.” Political Psychology  24( 2): 313– 43. Google Scholar CrossRef Search ADS   Dittmar, Jeremiah E. 2011. “ Information Technology and Economic Change: The Impact of the Printing Press.” Quarterly Journal of Economics  126( 3): 1133– 72. Google Scholar CrossRef Search ADS   Dovidio, John F., Peter E. Glick, and Laurie A. Rudman. 2005. On the Nature of Prejudice: Fifty Years after Allport . Oxford: Blackwell Publishing. Dustmann, Christian, and Ian Preston. 2001. “ Attitudes to Ethnic Minorities, Ethnic Context and Location Decisions.” Economic Journal  111( 470): 353– 73. Google Scholar CrossRef Search ADS   Dyer, Alan. 2000. “Ranking Lists of English Medieval Towns.” In The Cambridge Urban History of Britain (Volume I: 600–1540) , edited by David M. Palliser, Peter Clark, and Martin J. Daunton, 747– 70. Cambridge: Cambridge University Press. Google Scholar CrossRef Search ADS   English Heritage. 2010. Newport Pagnell: Historic Town Assessment Report. Milton Keynes, UK: Milton Keynes Council/Buckinghamshire County Council. Feather, John. 2006. “The Book Trade and Libraries.” In The Cambridge History of Libraries in Britain and Ireland (Volume 2: 1640–1850) , edited by Giles Mandelbrote and K. A. Manley, 301– 12. Cambridge: Cambridge University Press. Fielding, David. 2018. “ Traditions of Tolerance: The Long-Run Persistence of Regional Variation in Attitudes towards English Immigrants.” British Journal of Political Science  48( 1): 167– 88. Google Scholar CrossRef Search ADS   Hamilton, Bernard. 1997. “ Knowing the Enemy: Western Understanding of Islam at the Time of the Crusades.” Journal of the Royal Asiatic Society (Third Series)  7( 3): 373– 87. Google Scholar CrossRef Search ADS   Hello, Evelyn, Peer Scheepers, and Peter Sleegers. 2006. “ Why the More Educated Are Less Inclined to Keep Ethnic Distance: An Empirical Test of Four Explanations.” Ethnic and Racial Studies  29( 5): 959– 85. Google Scholar CrossRef Search ADS   Hello, Evelyn, Peer Scheepers, Ad Vermulst, and Jan R M Gerris. 2004. “ Association between Educational Attainment and Ethnic Distance in Young Adults: Socialization by Schools or Parents?” Acta Sociologica  47( 3): 253– 75. Google Scholar CrossRef Search ADS   Hillaby, Joe. 2003. “Jewish Colonisation in the Twelfth Century.” In The Jews in Medieval Britain: Historical, Literary and Archaeological Perspectives , edited by Patricia Skinner, 15– 41. Woodbridge, UK: Boydell Press. Hillaby, Joe, and Caroline Hillaby. 2013. The Palgrave Dictionary of Medieval Anglo-Jewish History . Basingstoke, UK: Palgrave Macmillan. Google Scholar CrossRef Search ADS   Hooghe, Marc, Cecil Meeusen, and Ellen Quintelier. 2013. “ The Impact of Education and Intergroup Friendship on the Development of Ethnocentrism.” European Sociological Review  29( 6): 1109– 21. Google Scholar CrossRef Search ADS   Jha, Saumitra. 2013. “ Trade, Institutions and Ethnic Tolerance: Evidence from South Asia.” American Political Science Review  107: 806– 32. Google Scholar CrossRef Search ADS   Khanmohamadi, Shirin. 2010. “ Casting a ‘Sideways Glance’ at the Crusades: The Voice of the Other in Joinville’s Vie de Saint Louis.” Exemplaria  22( 3): 177– 99. Google Scholar CrossRef Search ADS   Kibre, Pearl. 1946. “ The Intellectual Interests Reflected in Libraries of the Fourteenth and Fifteenth Centuries.” Journal of the History of Ideas  7( 3): 257– 97. Google Scholar CrossRef Search ADS   Langmuir, Gavin I. 1963. “ The Jews and the Archives of Angevin England: Reflections on Medieval Anti-Semitism.” Traditio  19: 183– 244. Google Scholar CrossRef Search ADS   Mayda, Anna-Maria. 2006. “ Who Is Against Immigration? A Cross-Country Investigation of Individual Attitudes toward Immigrants.” Review of Economics and Statistics  88( 3): 510– 30. Google Scholar CrossRef Search ADS   McVeigh, Rory, and Maria-Elena D. Diaz. 2009. “ Voting to Ban Same-Sex Marriage: Interests, Values, and Communities.” American Sociological Review  74( 6): 891– 915. Google Scholar CrossRef Search ADS   Meyer, John W., Francisco O. Ramirez, David J. Frank, and Evan Schofer. 2008. “Higher Education as an Institution.” In Sociology of Higher Education: Contributions and Their Contexts , edited by Patricia J. Gumport, 187– 221. Baltimore, MD: Johns Hopkins University Press. Mundill, Robert R. 2010. The King’s Jews: Money, Massacre and Exodus in Medieval England . London: Continuum Books. Ohlander, Julianne, Jeanne Batalova, and Judith Treas. 2005. “ Explaining Educational Influences on Attitudes toward Homosexual Relations.” Social Science Research  34( 4): 781– 99. Google Scholar CrossRef Search ADS   Orme, Nicholas. 2006. Medieval Schools: From Roman Britain to Renaissance England . New Haven, CT: Yale University Press. Pettigrew, Thomas F., and Linda R. Tropp. 2006. “ A Meta-Analytic Test of Intergroup Contact Theory.” Journal of Personality and Social Psychology  90: 751– 83. Google Scholar CrossRef Search ADS PubMed  Pettigrew, Thomas F., and Linda R. Tropp. 2012. When Groups Meet: The Dynamics of Intergroup Contact . New York: Psychology Press. Pierotti, Rachael S. 2013. “ Increasing Rejection of Intimate Partner Violence: Evidence of Global Cultural Diffusion.” American Sociological Review  78( 2): 240– 65. Google Scholar CrossRef Search ADS   Raven, James. 2006. “Libraries for Sociability: The Advance of the Subscription Library.” In The Cambridge History of Libraries in Britain and Ireland (Volume 2: 1640–1850) , edited by Giles Mandelbrote and K. A. Manley, 239– 63. Cambridge: Cambridge University Press. Rouleau, Benoit. 2005. L’Islam et les Musulmans dans l’Oeuvre de Jean de Joinville (1224–1318) [Islam and Muslims in the Work of Jean de Joinville (1224–1318)]. Master’s dissertation, Sherbrooke University. Schofer, Evan, and John W. Meyer. 2005. “ The Worldwide Expansion of Higher Education in the Twentieth Century.” American Sociological Review  70( 6): 898– 920. Google Scholar CrossRef Search ADS   Semyonov, Moshe, and Rebeca Raijman. 2006. ‘ The Rise of Anti-Foreigner Sentiment in European Societies, 1988–2000.” American Sociological Review  71( 3): 426– 49. Google Scholar CrossRef Search ADS   Shaw-Taylor, Leigh, Edward A. Wrigley, and Peter M. Kitson. 2006. Male Occupational Data from English Parish Registers (Processed Summary Data), c. 1700–1820. http://dx.doi.org/10.5255/UKDA-SN-5398-1. Stacey, Robert C. 2003. “The English Jews under Henry III.” In The Jews in Medieval Britain: Historical, Literary and Archaeological Perspectives , edited by Patricia Skinner, 41– 55. Woodbridge, UK: Boydell Press. Suarez, Michael F., S.J. 2009. “Towards a Bibliometric Analysis of the Surviving Record, 1701–1800.” In The Cambridge History of the Book in Britain , edited by Michael F. Suarez, S.J., and Michael L. Turner, 37– 65. Cambridge: Cambridge University Press. Google Scholar CrossRef Search ADS   Voigtländer, Nico, and Hans-Joachim Voth. 2013. “ Persecution Perpetuated: The Medieval Origins of Anti-Semitic Violence in Nazi Germany.” Quarterly Journal of Economics  127: 1339– 92. Google Scholar CrossRef Search ADS   © The Author(s) 2018. Published by Oxford University Press on behalf of the University of North Carolina at Chapel Hill. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Social Forces Oxford University Press

The Co-Evolution of Education and Tolerance: Evidence from England

Social Forces , Volume Advance Article (4) – Feb 21, 2018

Loading next page...
 
/lp/ou_press/the-co-evolution-of-education-and-tolerance-evidence-from-england-96LUTOFV0u
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of the University of North Carolina at Chapel Hill. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.
ISSN
0037-7732
eISSN
1534-7605
D.O.I.
10.1093/sf/soy008
Publisher site
See Article on Publisher Site

Abstract

Abstract Using data from several periods of English history and building on the literature on culture and institutions, we analyze the co-evolution of education and attitudes toward women’s and minority rights. First, we establish a strong association between the size of twenty-first-century educational institutions in a given location and the attitudes prevalent there: the proximity of a large college or university is associated with individual support for women’s and minority rights, regardless of whether the individual has been to university. Second, we present evidence for high inter-temporal persistence in the geographical variation in the density of educational institutions over several centuries. Third, we show that the geographical distribution of later educational institutions depends not only on the distribution of medieval institutions, but also on correlates of medieval exposure to ethnic and religious diversity that are likely to have influenced attitudes. Institutions and culture co-evolve, and the inter-temporal persistence of the density of educational institutions is one mechanism (though probably not the only one) that explains the association between medieval exposure to diversity and twenty-first-century attitudes. Introduction There is now a sizeable literature in economics and political science on the connections between culture and formal institutions. However, Alesina and Giuliano (2015) note some of this literature’s limitations: culture and institutions are likely to co-evolve, but most papers focus on one direction of causality, and a chronology of the changes in culture and institutions is often lacking. In this paper, we attempt to address these concerns in a study of the association between a particular set of cultural characteristics (attitudes toward minority groups and the role of women) and a particular set of institutions (educational ones). Research in sociology and social psychology has provided detailed evidence on the positive effect of education on the attitudes of students toward minority groups and women’s rights. This effect could be through higher levels of cognitive sophistication (Bobo and Licari 1989), through socialization involving liberal social norms (Coenders and Scheepers 2003; Hello et al. 2004), or through broadening the social network of which the individual is a part (Hello, Scheepers, and Sleegers 2006). There is evidence that education is associated with positive attitudes toward immigrants (e.g., Mayda 2006; Semyonov and Raijman 2006) and people of other ethnicities (e.g., Hooghe, Meeusen, and Quintelier 2013) and sexual orientations (e.g., McVeigh and Diaz 2009; Ohlander, Batalova, and Treas 2005). There is also evidence for an association with positive attitudes toward women’s rights; this association holds for both women (Pierotti 2013) and men (Banaszak and Plutzer 1993). However, the literature overlooks a potential effect of educational institutions on attitudes at the aggregate level: it is possible that the presence of a college or university in town affects attitudes across the whole town, including the attitudes of those who have never been to university. (From now on, we use the word “university” to refer to any tertiary education institution.) University students and faculty could promote positive attitudes toward outgroups or women’s rights through interactions with the wider community, so that the socialization and social network effects noted by the literature extend beyond the campus. Alternatively, the presence of a university might attract a certain type of resident. In this way, the university could be part of a wider social network that promotes certain social norms. Moreover, the effect of educational institutions on attitudes is only half of the story, because the location of universities could itself be a function of geographical variation in attitudes: a university might be more welcome in a town where the local community is open to institutions that promote engagement with that which is foreign. Schofer and Meyer (2005) find that the cross-country variation in the rate of university expansion is explained partly by variation in the strength of a country’s international linkages, an effect consistent with the interpretation of university education (and the university as an institution) as an expression of cosmopolitan values (Meyer et al. 2008). It is possible that the within-country variation exhibits a similar pattern, reflecting regional variation in attachment to these values. In order to explore the association between attitudes and the location of educational institutions, we compile cross-sectional data from several periods of English history.1 First of all, we measure the size of the association between university size and local attitudes in twenty-first-century opinion surveys. Second, we measure the magnitude of inter-temporal persistence in the importance of educational institutions, using geographical variation in (i) twenty-first-century university size and (ii) the density of educational institutions prior to the middle of the nineteenth century (which we term the “premodern” period). The premodern institution that we focus on is the library, libraries having become increasingly widespread following the invention of the printing press in the fifteenth century. Third, we measure the extent to which geographical variation in the density of premodern libraries, and of the bookstores that supplied them, is correlated with local communities’ exposure to ethnic and religious diversity in the twelfth and thirteenth centuries. Such a correlation suggests that exposure to diversity can change preferences in a town, making it more open to institutions that embody new ideas. Our results complement the existing evidence for long-run persistence in attitudes toward outgroups (Fielding 2018; Jha 2013; Voigtländer and Voth 2013) and toward the role of women (Alesina, Giuliano, and Nunn 2013). We extend this literature by showing how institutions have been a vehicle for the inter-temporal persistence of cultural variation. Explaining the Geographical Variation in Twenty-First-Century Attitudes The first part of our analysis is designed to estimate the effect of twenty-first-century universities on attitudes in the surrounding area. We measure these attitudes using data from the 2010 and 2015 rounds of the British Election Study (BES): see www.britishelectionstudy.com and www.britishelectionstudy.com/data-objects/panel-study-data/. One key advantage of the BES is the breadth of its geographical coverage: respondents are drawn randomly from the electoral roll, with observations for every parliamentary constituency. Three key 2015 BES questions relating to attitudes toward people of other ethnicities and sexual orientations, and toward women’s rights, are as follows: “Do you agree with the statement that equal opportunities for ethnic minorities have gone too far?” “Do you agree with the statement that equal opportunities for gays and lesbians have gone too far?” “Do you agree with the statement that equal opportunities for women have gone too far?” There is also a variety of questions about immigrants, though the responses to these questions are highly correlated. The immigrant question that we will use is as follows: “Do you agree with the statement that immigrants are a burden on the welfare state?” The possible responses to these questions are: “strongly disagree” (to which we allocate a value of one), “disagree” (two), “don’t know” (three), “agree” (four), and “strongly agree” (five). We use the responses to construct four opinion variables for each respondent; each variable has a maximum possible value of five (indicating maximal opposition to equal opportunities and immigration) and a minimum possible value of one. Our sample comprises all white respondents in the 460 English constituencies outside London.2 When we restrict the sample to those respondents also reporting personal characteristics (which permits their inclusion in the data analysis that follows), we are left with just under 20,000 observations. London is excluded from our sample because, as the nation’s capital, it contains an atypical set of public institutions that could affect attitudes in ways that are difficult to measure. Another highly salient political question that may relate to attitudes toward outgroups concerns membership of the European Union. Previous studies (e.g., Bruter 2005, appendix 3) have found education to be associated with more favorable attitudes toward the EU, and one interpretation of this effect is that opposition to the EU is motivated partly by antipathy toward outgroups defined by nationality. BES respondents were asked about their voting intention in the upcoming referendum on Britain’s EU membership, and we construct a further variable that equals one if the respondent indicated an intention to vote Remain, three if the respondent indicated an intention to vote Leave, and two if the respondent did not know. A higher value of this variable indicates greater opposition to European integration, just as higher values of the other variables indicate greater opposition to equal opportunities. For each of the five 2015 attitude variables, table 1 shows the proportion of people giving each response. Opposition to immigration and ethnic minority rights is somewhat higher than opposition to gay rights, which is somewhat higher than opposition to women’s rights. Nevertheless, for each statement, there is a substantial proportion of people agreeing and a substantial proportion disagreeing. For each statement and for each response, table 2 provides a measure of local university size for the relevant group of respondents. This measure is the average across respondents of the ratio of the university student population in the respondent’s constituency to the rest of the constituency population (denoted student ratio);3 data are taken from the 2011 census. It can be seen that the ratio is consistently lower among those respondents who oppose equal opportunities, immigration, and European integration. This correlation is consistent with the hypothesis outlined in the introduction: universities promote liberal attitudes in the area around them. However, table 2 does not in itself constitute evidence in favor of the hypothesis. First, there might be systematic variation in the personal characteristics of respondents that is correlated both with attitudes and with the size of the local student population. For example, inhabitants of university towns might be younger, on average, or more highly educated, or have higher incomes. Second, attitudes might affect local student numbers. For example, not all students live on campus, and their choice of neighborhood might depend on the attitudes of their potential neighbors. We must address both of these problems in order to establish that the size of the student population affects local attitudes. Table 1. Distribution of Responses to Attitude Questions in the 2015 BES   (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]    Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Strongly disagree  1,091  5.5%  591  3.0%  1,542  7.8%  1,138  5.7%      Disagree [remain]  3,115  15.7%  2,272  11.5%  5,653  28.5%  3,082  15.5%  7,434  37.5%  Don’t know  9,501  47.9%  8,515  42.9%  10,114  51.0%  4,674  23.6%  3,553  17.9%  Agree [leave]  3,266  16.5%  4,878  24.6%  1,889  9.5%  5,898  29.7%  8,856  44.6%  Strongly agree  2,865  14.4%  3,582  18.1%  640  3.2%  5,050  25.5%      Total  19,838    19,838    19,838    19,842    19,843      (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]    Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Strongly disagree  1,091  5.5%  591  3.0%  1,542  7.8%  1,138  5.7%      Disagree [remain]  3,115  15.7%  2,272  11.5%  5,653  28.5%  3,082  15.5%  7,434  37.5%  Don’t know  9,501  47.9%  8,515  42.9%  10,114  51.0%  4,674  23.6%  3,553  17.9%  Agree [leave]  3,266  16.5%  4,878  24.6%  1,889  9.5%  5,898  29.7%  8,856  44.6%  Strongly agree  2,865  14.4%  3,582  18.1%  640  3.2%  5,050  25.5%      Total  19,838    19,838    19,838    19,842    19,843    Note: The sample comprises white respondents in all English parliamentary constituencies outside London. Table 1. Distribution of Responses to Attitude Questions in the 2015 BES   (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]    Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Strongly disagree  1,091  5.5%  591  3.0%  1,542  7.8%  1,138  5.7%      Disagree [remain]  3,115  15.7%  2,272  11.5%  5,653  28.5%  3,082  15.5%  7,434  37.5%  Don’t know  9,501  47.9%  8,515  42.9%  10,114  51.0%  4,674  23.6%  3,553  17.9%  Agree [leave]  3,266  16.5%  4,878  24.6%  1,889  9.5%  5,898  29.7%  8,856  44.6%  Strongly agree  2,865  14.4%  3,582  18.1%  640  3.2%  5,050  25.5%      Total  19,838    19,838    19,838    19,842    19,843      (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]    Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Frequency  Pct.  Strongly disagree  1,091  5.5%  591  3.0%  1,542  7.8%  1,138  5.7%      Disagree [remain]  3,115  15.7%  2,272  11.5%  5,653  28.5%  3,082  15.5%  7,434  37.5%  Don’t know  9,501  47.9%  8,515  42.9%  10,114  51.0%  4,674  23.6%  3,553  17.9%  Agree [leave]  3,266  16.5%  4,878  24.6%  1,889  9.5%  5,898  29.7%  8,856  44.6%  Strongly agree  2,865  14.4%  3,582  18.1%  640  3.2%  5,050  25.5%      Total  19,838    19,838    19,838    19,842    19,843    Note: The sample comprises white respondents in all English parliamentary constituencies outside London. Table 2. Average Values of Student ratio in the Constituencies of Respondents in the 2015 BES, Disaggregated by Attitudes   (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]  Strongly disagree  4.95%  5.12%  4.55%  5.35%    Disagree [remain]  4.66%  4.87%  4.08%  4.66%  4.52%  Don’t know  3.95%  4.05%  3.89%  3.97%  3.90%  Agree [leave]  3.46%  3.53%  3.63%  3.67%  3.53%  Strongly agree  3.42%  3.56%  3.58%  3.55%      (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]  Strongly disagree  4.95%  5.12%  4.55%  5.35%    Disagree [remain]  4.66%  4.87%  4.08%  4.66%  4.52%  Don’t know  3.95%  4.05%  3.89%  3.97%  3.90%  Agree [leave]  3.46%  3.53%  3.63%  3.67%  3.53%  Strongly agree  3.42%  3.56%  3.58%  3.55%    Note: In order to compute the figures in each of columns (i–iv) of the table, respondents in the 2015 BES are grouped according to their response to the statement heading that column. For each statement and for each response, the figure is the average across respondents of the size of the student population in the respondent’s parliamentary constituency as a percentage of the rest of the population. In column (v), respondents are grouped according to their voting intention in the EU membership referendum. The sample is the same as in table 1. Table 2. Average Values of Student ratio in the Constituencies of Respondents in the 2015 BES, Disaggregated by Attitudes   (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]  Strongly disagree  4.95%  5.12%  4.55%  5.35%    Disagree [remain]  4.66%  4.87%  4.08%  4.66%  4.52%  Don’t know  3.95%  4.05%  3.89%  3.97%  3.90%  Agree [leave]  3.46%  3.53%  3.63%  3.67%  3.53%  Strongly agree  3.42%  3.56%  3.58%  3.55%      (i)  (ii)  (iii)  (iv)  (v)  Response to statement [or voting intention]  Equal opportunities for gays/lesbians have gone too far.  Equal opportunities for ethnic minorities have gone too far.  Equal opportunities for women have gone too far.  Immigrants are a burden on the welfare state.  [Voting intention in the EU referendum]  Strongly disagree  4.95%  5.12%  4.55%  5.35%    Disagree [remain]  4.66%  4.87%  4.08%  4.66%  4.52%  Don’t know  3.95%  4.05%  3.89%  3.97%  3.90%  Agree [leave]  3.46%  3.53%  3.63%  3.67%  3.53%  Strongly agree  3.42%  3.56%  3.58%  3.55%    Note: In order to compute the figures in each of columns (i–iv) of the table, respondents in the 2015 BES are grouped according to their response to the statement heading that column. For each statement and for each response, the figure is the average across respondents of the size of the student population in the respondent’s parliamentary constituency as a percentage of the rest of the population. In column (v), respondents are grouped according to their voting intention in the EU membership referendum. The sample is the same as in table 1. In order to address the first problem, we fit a set of ordered Probit models of the responses in table 1, each model corresponding to a different question. In each model, each observation of the dependent variable is the response of one individual to one particular question; this variable takes a value between one and five, with higher values indicating greater opposition to equal rights, immigration, or European integration. Our key explanatory variable is the logarithm of student ratio,4 but our set of explanatory variables also incorporates a range of individual and constituency characteristics that might be correlated with student ratio and also affect attitudes. These variables are discussed in more detail in appendix A; they include fixed effects for the county in which the individual’s constituency is located, the constituency’s historical population size and current population density, the ratio of its non-white population to its total population,5 the individual’s age, gender, education level, marital status, employment status, and religious affiliation, whether the individual is a student, the per capita income level of the individual’s household, whether the household contains children or elderly dependents, and several different measures of the individual’s psychological characteristics. Table 3 includes the estimates of the coefficients on the logarithm of student ratio along with the corresponding t-ratios; these t-ratios are based on standard errors that allow for clustering at the constituency level. Estimates of the coefficients on the other explanatory variables are omitted from table 3 but appear in appendix A. Because an ordered Probit model is non-linear, the coefficients in table 3 do not in themselves indicate the average effect of the logarithm of student ratio on the individual responses. For this reason, table 3 also reports a marginal effect alongside each coefficient. If we assume a causal interpretation of the coefficients in table 3, the marginal effect is the effect of a unit increase in the logarithm of student ratio on the probability that the dependent variable will take a value greater than the central one, that is, that the individual will agree or strongly agree with the statement, or will choose Leave in the EU referendum. Note that this marginal effect measures the effect of student ratio while holding constant all of the other personal and constituency characteristics in the model. Table 3. Estimated Coefficients on the Logarithm of Student Ratio in Ordered Probit Models of 2015 and 2010 BES Attitudes   Attitudes in the 2015 BES  Attitudes in the 2010 BES    (i)  (ii)  (iii)  (iv)  (v)  (vi)  (vii)    Equal ops for gays/lesbians have gone too far.  Equal ops for eth. minorities have gone too far.  Equal ops for women have gone too far.  Immigrants are a burden on the welfare state.  Voting intention in the EU referendum  Ranking of immigration as an election issue  Disapproval of the EU  Coefficient  −0.028  −0.058  −0.056  −0.070  −0.070  −0.063  −0.091  t-ratio  −1.65*  −3.29***  −3.69***  −3.45***  −3.87***  −1.93*  −2.97***  Marginal effect  −0.010  −0.021  −0.023  −0.024  −0.026  −0.024  −0.031  Sample size  19,838  19,838  19,838  19,842  19,843  8,364  8,048    Attitudes in the 2015 BES  Attitudes in the 2010 BES    (i)  (ii)  (iii)  (iv)  (v)  (vi)  (vii)    Equal ops for gays/lesbians have gone too far.  Equal ops for eth. minorities have gone too far.  Equal ops for women have gone too far.  Immigrants are a burden on the welfare state.  Voting intention in the EU referendum  Ranking of immigration as an election issue  Disapproval of the EU  Coefficient  −0.028  −0.058  −0.056  −0.070  −0.070  −0.063  −0.091  t-ratio  −1.65*  −3.29***  −3.69***  −3.45***  −3.87***  −1.93*  −2.97***  Marginal effect  −0.010  −0.021  −0.023  −0.024  −0.026  −0.024  −0.031  Sample size  19,838  19,838  19,838  19,842  19,843  8,364  8,048  Note: The table reports the coefficients on the logarithm of student ratio in ordered Probit models of different attitudes of respondents in the BES. Student ratio is the ratio of the student population to the rest of the population in the respondent’s parliamentary constituency in 2011. The attitudes are measured on a five-point scale, except for column (v) with a three-point scale and column (vi) with a four-point scale. For columns (i–iv), higher values on the scale indicate greater agreement with the statement heading that column. For column (v), higher values indicate a preference for Leave versus Remain in the EU referendum. For column (vi), the scale corresponds to the ranking of immigration among the most important election issues, higher values indicating a higher rank. In column (vii), higher values on the scale indicate less approval of the UK’s membership of the EU. See the main text for a discussion of the other regressors. Marginal effects measure the impact of a unit increase in the logarithm of student ratio on the probability of an outcome higher than the central value, or in column (vi) on the probability of a rank of third or higher. One asterisk (*) indicates a coefficient significantly different from zero at the 10 percent level; three asterisks (***) indicate significance at the 1 percent level. The 2015 sample is the same as in table 1. Table 3. Estimated Coefficients on the Logarithm of Student Ratio in Ordered Probit Models of 2015 and 2010 BES Attitudes   Attitudes in the 2015 BES  Attitudes in the 2010 BES    (i)  (ii)  (iii)  (iv)  (v)  (vi)  (vii)    Equal ops for gays/lesbians have gone too far.  Equal ops for eth. minorities have gone too far.  Equal ops for women have gone too far.  Immigrants are a burden on the welfare state.  Voting intention in the EU referendum  Ranking of immigration as an election issue  Disapproval of the EU  Coefficient  −0.028  −0.058  −0.056  −0.070  −0.070  −0.063  −0.091  t-ratio  −1.65*  −3.29***  −3.69***  −3.45***  −3.87***  −1.93*  −2.97***  Marginal effect  −0.010  −0.021  −0.023  −0.024  −0.026  −0.024  −0.031  Sample size  19,838  19,838  19,838  19,842  19,843  8,364  8,048    Attitudes in the 2015 BES  Attitudes in the 2010 BES    (i)  (ii)  (iii)  (iv)  (v)  (vi)  (vii)    Equal ops for gays/lesbians have gone too far.  Equal ops for eth. minorities have gone too far.  Equal ops for women have gone too far.  Immigrants are a burden on the welfare state.  Voting intention in the EU referendum  Ranking of immigration as an election issue  Disapproval of the EU  Coefficient  −0.028  −0.058  −0.056  −0.070  −0.070  −0.063  −0.091  t-ratio  −1.65*  −3.29***  −3.69***  −3.45***  −3.87***  −1.93*  −2.97***  Marginal effect  −0.010  −0.021  −0.023  −0.024  −0.026  −0.024  −0.031  Sample size  19,838  19,838  19,838  19,842  19,843  8,364  8,048  Note: The table reports the coefficients on the logarithm of student ratio in ordered Probit models of different attitudes of respondents in the BES. Student ratio is the ratio of the student population to the rest of the population in the respondent’s parliamentary constituency in 2011. The attitudes are measured on a five-point scale, except for column (v) with a three-point scale and column (vi) with a four-point scale. For columns (i–iv), higher values on the scale indicate greater agreement with the statement heading that column. For column (v), higher values indicate a preference for Leave versus Remain in the EU referendum. For column (vi), the scale corresponds to the ranking of immigration among the most important election issues, higher values indicating a higher rank. In column (vii), higher values on the scale indicate less approval of the UK’s membership of the EU. See the main text for a discussion of the other regressors. Marginal effects measure the impact of a unit increase in the logarithm of student ratio on the probability of an outcome higher than the central value, or in column (vi) on the probability of a rank of third or higher. One asterisk (*) indicates a coefficient significantly different from zero at the 10 percent level; three asterisks (***) indicate significance at the 1 percent level. The 2015 sample is the same as in table 1. In order to check the robustness of our results using data from the 2015 round of the BES, we also fit ordered Probit models of two attitude variables from the 2010 round. The range of questions in the 2010 round is narrower, but there are questions relating to immigrants and EU membership. First, respondents were asked to rank a list of issues in terms of their importance for the election, one issue being immigration. We construct a variable that equals four if immigration is ranked first, three if it is ranked second, two if it is ranked third, and one if it is unranked.6 (In the context of the election, it is very likely that a high rank reflected an unfavorable view of immigrants.) Second, respondents were asked about their level of approval of Britain’s membership in the EU. We construct a variable that equals one for responses of “strongly approve,” two for “approve,” three for “don’t know,” four for “disapprove,” and five for “strongly disapprove.” Results for these two additional variables also appear in table 3. All of the coefficients in table 3 are negative and significantly different from zero at the 10 percent level, and five out of seven are significant at the 1 percent level. In column (i), the reported marginal effect is –0.01, indicating that a unit increase in the logarithm of student ratio is associated with a one-percentage-point fall in the probability that the individual will agree (or strongly agree) that gay rights have gone too far. In order to interpret the magnitude of this effect, we note that the standard deviation of the logarithm of student ratio is 0.67, so, for example, a three-standard-deviation increase in this variable is associated with a two-percentage-point fall in the probability. The marginal effects for the other attitude variables are between two and three times as large as this. For example, the marginal effect in column (iv) is –0.024, indicating that a three-standard-deviation increase in the logarithm of student ratio reduces the probability of agreeing (or strongly agreeing) that immigrants are a burden on the welfare state by about five percentage points. Recall from table 1 (column iv) that the overall proportion of respondents agreeing or strongly agreeing is 55 percent. The table 3 results imply that in the average constituency, a large increase in student numbers (i.e., an increase in the logarithm of student ratio of over three standard deviations) will turn this anti-immigration majority into a minority. This suggests that the presence of a large university can make a substantial difference to those public opinions, such as opinions about immigration, which currently hold the attention of politicians. One caveat to a causal interpretation of these results is that we have still not dealt with our second problem: the potential endogeneity of student ratio to attitudes. For this reason, appendix A includes an alternative set of results that have been produced using an Instrumental Variables (IV) estimator. This estimator exploits exogenous variation in the size of the university-age population in order to identify the effect of student ratio on attitudes. The IV estimates are somewhat less precise, but five out of the seven coefficients are still significantly different from zero at the 5 percent level. The corresponding marginal effects are a little larger than in table 3, but this difference is not statistically significant. This gives us some confidence in asserting that in the twenty-first century, holding constant other determinants of individual attitudes (such as the individual’s own age, education, and income level), these attitudes are significantly more liberal in constituencies with large numbers of university students. Locations with large universities are indeed more liberal. Our next task is to establish whether there is any substantial inter-temporal persistence in the geographical variation in the density of educational institutions. Inter-Temporal Persistence and the Geographical Distribution of Twenty-First-Century Universities Until the middle of the nineteenth century, England had only two universities, so we cannot focus exclusively on universities if we wish to measure inter-temporal persistence in the density of educational institutions over very long time horizons. Instead, we will explore the association between our twenty-first-century variable (the value of student ratio in each constituency) and the density of another type of educational institution, the privately funded library. The proliferation of libraries in the premodern period was a consequence of the fall in book prices and rise in literacy rates that followed the invention of moveable type in the late fifteenth century (Dittmar 2011; Raven 2006). Growth in book production and in the number of libraries was especially high in the eighteenth century; the most distinctive institution of this period was the private subscription library, which served the needs of a wide range of readers, including those from the growing middle class and skilled working class, and supplemented older types of library (for example the parish church library) that had arisen after the invention of moveable type. Libraries were places where like-minded people could meet and discuss the books they were reading: they were not teaching institutions, but they were institutions of learning. There was no public provision of library services until the Public Libraries Act of 1850, so regional variation in the location of libraries before this date is likely to have reflected regional variation in demand. Our information about the location of English libraries comes from the list compiled by Alston (2011). Using this list, it is possible to construct a dataset of the total number of libraries in each English town before 1850, and then to construct a constituency-level dataset in which each observation is the number of libraries in the largest town in the twenty-first-century parliamentary constituency.7 Table 4 illustrates the correlation between these library numbers and student ratio by grouping constituencies according to the number of premodern libraries in their largest town and showing the average value of student ratio for each group. It can be seen that on average, those constituencies with towns that were rich in libraries are also constituencies with large numbers of university students, and this correlation is consistent with inter-temporal persistence in the density of educational institutions. There are two possible explanations for such a correlation. First, there might be inter-temporal persistence in some of the local characteristics that have always been favorable to the establishment of educational institutions; such characteristics include the level of urbanization and prosperity of the local area. Second and more interestingly, there might be inter-temporal persistence in the density of educational institutions conditional on these characteristics. In other words, among constituencies that are otherwise similar, having had a large number of libraries is associated with a higher value of student ratio. Such an association would arise if there were persistent geographical variation in local preferences, or if the establishment of libraries created a local educational infrastructure that later reduced the cost of establishing universities. Table 4. Average Values of Student ratio Disaggregated by the Number of Premodern Libraries Number of premodern libraries in the largest town in the constituency  Number of constituencies  Average value of student ratio across constituencies  0–9  242  2.58%  10–19  85  2.83%  20–29  38  4.13%  30–39  11  7.03%  40–49  14  7.43%  50-plus  70  8.65%  Number of premodern libraries in the largest town in the constituency  Number of constituencies  Average value of student ratio across constituencies  0–9  242  2.58%  10–19  85  2.83%  20–29  38  4.13%  30–39  11  7.03%  40–49  14  7.43%  50-plus  70  8.65%  Note: Each row of the table corresponds to a different group of constituencies, constituencies being grouped according to the number of premodern libraries in their largest town. The first column indicates the number of libraries, the second column the group size, and the third column the average value of student ratio for that group. Student ratio is the ratio of the student population to the rest of the constituency population in 2011. The sample consists of all 460 English parliamentary constituencies outside London. Table 4. Average Values of Student ratio Disaggregated by the Number of Premodern Libraries Number of premodern libraries in the largest town in the constituency  Number of constituencies  Average value of student ratio across constituencies  0–9  242  2.58%  10–19  85  2.83%  20–29  38  4.13%  30–39  11  7.03%  40–49  14  7.43%  50-plus  70  8.65%  Number of premodern libraries in the largest town in the constituency  Number of constituencies  Average value of student ratio across constituencies  0–9  242  2.58%  10–19  85  2.83%  20–29  38  4.13%  30–39  11  7.03%  40–49  14  7.43%  50-plus  70  8.65%  Note: Each row of the table corresponds to a different group of constituencies, constituencies being grouped according to the number of premodern libraries in their largest town. The first column indicates the number of libraries, the second column the group size, and the third column the average value of student ratio for that group. Student ratio is the ratio of the student population to the rest of the constituency population in 2011. The sample consists of all 460 English parliamentary constituencies outside London. In order to investigate whether there is any conditional persistence in the density of educational institutions, we fit an Ordinary Least Squares (OLS) regression equation in which the dependent variable is the logarithm of student ratio and one of the explanatory variables is the number of premodern libraries in the largest town in the constituency. The distribution of the number of libraries is quite highly skewed—it has a mean of 26 and a maximum value of 237—so the explanatory variable is trimmed so that its maximum value is 50; this variable is denoted libraries in table 5.8 The other explanatory variables include county fixed effects and two measures of urbanization: the constituency population per hectare in the 2011 census (denoted population density) and the population of the largest town in the constituency in the 1841 census, that is, at the end of the premodern period (denoted town population in 1841). The 1841 population numbers are measured in tens of thousands and trimmed at 75,000; the figures are taken from Bennett (2011, appendix 3). Also included are an indicator variable for constituencies with a port (denoted port town) and a measure of the level of social and economic development in the constituency. There are a number of alternative measures of social and economic development that are all highly correlated with one another. The results in table 5 are based on a model that includes the ratio of the number of workers in a constituency who are in a professional occupation (i.e., in occupational categories 1–2) to the number who are not, as reported in the 2011 census; this variable is denoted professional ratio. Results using alternative measures such as unemployment rates or ACORN classifications are very similar, and are available on request. Descriptive statistics for all of the variables appear in appendix B.9 Table 5. Determinants of the Logarithm of Student Ratio: Ordinary Least Squares Estimates   Coefficient  t-ratio  Logarithm of professional ratio  0.298  3.15***  Population density  0.014  6.83***  Port town  0.201  3.21***  Town population in 1841  −0.024  −1.11  Libraries  0.016  5.44***  R2  0.72    Coefficient  t-ratio  Logarithm of professional ratio  0.298  3.15***  Population density  0.014  6.83***  Port town  0.201  3.21***  Town population in 1841  −0.024  −1.11  Libraries  0.016  5.44***  R2  0.72  Note:Student ratio is the ratio of the student population to the rest of the constituency population in 2011; the sample is the same as in table 4. Professional ratio is the ratio of workers in social classes 1–2 to other classes in the constituency in 2011; population density is the constituency population per hectare in 2011; port town is a binary variable equal to one in constituencies with a port and zero otherwise; town population in 1841 is the population of the largest town in the constituency (in tens of thousands) in 1841, trimmed at 75,000; libraries is the number of premodern libraries established in the largest town in the constituency, trimmed at 50. The set of explanatory variables also includes county fixed effects and the variables discussed in the second part of appendix A: estimates of the relevant coefficients are available on request. Three asterisks (***) indicate a coefficient significantly different from zero at the 1 percent level. Table 5. Determinants of the Logarithm of Student Ratio: Ordinary Least Squares Estimates   Coefficient  t-ratio  Logarithm of professional ratio  0.298  3.15***  Population density  0.014  6.83***  Port town  0.201  3.21***  Town population in 1841  −0.024  −1.11  Libraries  0.016  5.44***  R2  0.72    Coefficient  t-ratio  Logarithm of professional ratio  0.298  3.15***  Population density  0.014  6.83***  Port town  0.201  3.21***  Town population in 1841  −0.024  −1.11  Libraries  0.016  5.44***  R2  0.72  Note:Student ratio is the ratio of the student population to the rest of the constituency population in 2011; the sample is the same as in table 4. Professional ratio is the ratio of workers in social classes 1–2 to other classes in the constituency in 2011; population density is the constituency population per hectare in 2011; port town is a binary variable equal to one in constituencies with a port and zero otherwise; town population in 1841 is the population of the largest town in the constituency (in tens of thousands) in 1841, trimmed at 75,000; libraries is the number of premodern libraries established in the largest town in the constituency, trimmed at 50. The set of explanatory variables also includes county fixed effects and the variables discussed in the second part of appendix A: estimates of the relevant coefficients are available on request. Three asterisks (***) indicate a coefficient significantly different from zero at the 1 percent level. One potential concern with the OLS estimates is that the level of social and economic development in a constituency could be endogenous to the presence of universities there. For this reason, appendix B includes an alternative set of results from a model fitted using an IV estimator, with instruments that are based on the relative intensity of coal mining across the country. These results are very similar to those reported in table 5. Table 5 shows that constituencies with higher levels of professional ratio and population density have significantly higher levels of student ratio, as do constituencies with a port town. However, population levels in the 1841 census have no significant association with the level of student ratio. Conditional on all of these effects, there is a significant association between student ratio and libraries. The estimated value of the coefficient is 0.016, implying that each additional library is associated with a value of student ratio that is about 1.6 percent higher. To put it another way, the difference between the maximum and minimum values of the library variable (i.e., 50 libraries) corresponds to an increase in the logarithm of student ratio of 0.016 × 50 = 0.8, which is slightly greater than one standard deviation of the dependent variable. This effect is significant at the 1 percent level and corresponds to a high level of conditional persistence in the density of educational institutions. Our next task is to explain the geographical distribution of premodern libraries, and to measure the extent to which the location of libraries was influenced by medieval exposure to ethnic and religious diversity. Explaining the Geographical Variation in Premodern Libraries In this section, we will present some estimates of the factors that explain the variation across towns in the number of premodern libraries, and of the bookstores that supplied them. We begin with a discussion of two town-level characteristics that are likely to be correlated with medieval exposure to ethnic and religious diversity, explaining how these characteristics might be associated with attitudes toward outgroups, and therefore with openness to educational institutions (such as libraries) that embody new ideas. This discussion is informed by the literature on intergroup contact theory originating with Allport (1954). Allport contends that the context in which two groups meet is important in determining whether the contact leads to positive or negative views of the other group, and the subsequent literature has identified several characteristics that are associated with a positive effect on attitudes toward outgroups. These characteristics include an absence of competition between the two groups and equal social status: see for example the surveys of Dovidio, Glick, and Rudman (2005) and Pettigrew and Tropp (2006, 2012). Exposure to Diversity through Jewish Immigration The largest immigrant group in early medieval England was the Jews, the first Jewish communities arriving from France soon after the Norman Conquest. Hillaby (2003) discusses tax records that suggest that by the late twelfth century there were Jewish communities in over 20 English towns, while Mundill (2010) explains the two main economic functions of these communities: to provide financial services (lending money at interest being forbidden between Christians) and to provide a tax base for the king that was independent from the normal feudal system and therefore from the influence of the barons. In the thirteenth century, the state began to regulate the settlement of Jewish communities more directly, and Jews were officially permitted to live only in designated towns that contained a chest (or archa) where contracts between Jews and Christians had to be deposited (Brand 2003; Brown and McCartney 2005). By the mid-thirteenth century, there were archae in 30 towns; appendix C includes a list of these locations taken from Hillaby and Hillaby (2013). Growing anti-Semitism finally led to the expulsion of all Jews from England in 1290, after which there was no substantial foreign immigration into England until the arrival of the Huguenots in the late seventeenth century.10 The prevalence of anti-Semitic attitudes in medieval England was probably very high, and racially motivated attacks on individual Jews seem to have been common. However, as noted by Mundill (2010), genocidal attacks on whole communities appear to have been restricted to two periods. The first of these was 1189–1190, following clashes between Jews and Christians at the coronation of Richard I, and the second was the civil war of 1263–1265. The late twelfth-century attacks often seem to have been perpetrated by people from outside town: for example, foreign merchants instigated the attack at Lynn.11 Attacks during the civil war were perpetrated by baronial armies seeking to terrorize communities who supported the king, including the Jews (Hillaby 2003; Stacey 2003). Compared with fourteenth-century Germany during the Black Death (Voigtländer and Voth 2013), there is relatively little evidence of genocidal attacks on English Jews perpetrated by their near neighbors. Fielding (2018) discusses whether direct contact with a Jewish community would be likely to improve or worsen attitudes toward the Jews, on average. Noting that the majority of ordinary people would interact with the Jews on the basis of equal social standing (the Jews as vassals of the king, the common people as vassals of the local lord) and during peaceful trade (Jews buying goods from local merchants: Jews lent money only to wealthy individuals and organizations with collateral, and were barred from economic activities that would put them in competition with Christians), intergroup contact theory suggests that contact would lead to an improvement in attitudes among the bulk of the population. Langmuir (1963) reviews the evidence on this point, concluding that “the majority of the little evidence that there is suggests that it was primarily those who lived in close contact with Jews who were friendly with them.” If these arguments are correct, then we should expect the presence of an archa in the thirteenth century to indicate a town with a history of intergroup contact promoting greater openness to other groups and a more favorable environment for the establishment of educational institutions. Exposure to Diversity through the Crusades Although foreign travel was generally very uncommon in medieval England, one specific group of Englishmen who did travel were the crusaders. What was the likely effect of crusading on their opinions about other races and religions? In order to address this question, it is important to note that the majority of crusaders probably had very little choice about whether to fight. With the possible exception of the First Crusade (1095–1099), crusades were the initiative of senior churchmen or nobles. Although in theory their vassals were not required to accompany them, there was a great deal of social pressure to follow one’s lord to the Holy Land, sometimes accompanied by financial inducements such as the cancellation of debts (Benjamin 2015). Low-ranking knights and their retainers went on crusade if their lord did. For the vassals who made up the bulk of a crusading army, participation in the crusade probably represented a treatment effect, not a selection effect. This treatment might include exposure to more intense xenophobic propaganda and face-to-face fighting that induced negative views of the enemy population. However, having established states in Antioch, Edessa, Tripoli, and Acre/Jerusalem, the crusaders were required to engage in trade and diplomacy with neighboring Arab states, and to manage local economies that were based on the output of Arab farmers. These activities involved men of many different ranks, and historical research suggests that this led to a level of understanding of Arab culture, religion, and language that stood in marked contrast to the ignorance prevalent in Western Europe (Attiya 1999; Hamilton 1997). Moreover, the division of labor within a crusader state between the Arab population (primary production) and the European population (administration and services) meant that day-to-day contact between them did not involve competition, and this absence of competition could have fostered positive attitudes toward the outgroup. Historical accounts by crusaders such as Jean de Joinville include sympathetic depictions of individual Arabs, both rulers and those of lower rank (Khanmohamadi 2010; Rouleau 2005). This eventually influenced popular Western European fiction, in which Arabs were depicted not only as villains but also as heroes (Calkin 2012; Hamilton 1997). One vehicle for this influence was the return of crusading forces following their various defeats, for example the fall of Edessa (1144), Jerusalem (1187), Antioch (1268), Tripoli (1289), and Acre (1291). The salience of participation in the crusades to many returning English nobles was reflected in the modification of their coats of arms, most commonly by the addition of a Saracen’s (i.e., Arab’s) head. In interpreting the Saracen’s head, it is important to note that heads can be displayed on coats of arms in a variety of different ways, each with its own heraldic terminology. Violence and disrespect to an enemy are indicated by the appearance of a head that is erased (i.e., decapitated), on the point of a pheon (i.e., spear), or distilling blood. Such depictions have been popular among noble families in Wales and Scotland, but they are relatively rare in England. Of the 84 English coats of arms featuring Saracen’s heads that are listed in Burke (1884), only eight depict violence of this kind. English arms typically display a Saracen’s head that is couped (i.e., depicted in the style of a king’s head on a coin), a neutral representation of contact with a foreign culture that does not imply disrespect. One popular equivalent of a noble’s coat of arms was the inn sign, many signs being modeled on the arms of a local lord. Therefore, one indication of the salience of the crusades to a local community is an inn named the Saracen’s Head. One limitation that we face is that there is no medieval documentary source that lists inns of this name; however, it is possible to use the address list in the 1851 census to locate Saracen’s Head inns that survived to the nineteenth century, and a list of these 88 locations is provided in appendix C. This list can be used to create an indicator variable for the presence or absence of a Saracen’s Head inn in each location.12 Both the original distribution of Saracen’s Head inns in the twelfth and thirteenth centuries and the rate of attrition up to the nineteenth century are likely to be influenced by a number of confounding factors: for example, the original distribution is likely to depend on medieval population densities and the rate of attrition on subsequent population growth rates. However, if we control for these factors, then we can interpret our indicator variable as a correlate of the salience to a local community of participation in the crusades, the phenomenon that brought medieval Englishmen most directly into contact with large numbers of foreigners. If the arguments above are correct, then we should expect the presence of a Saracen’s Head inn to reflect a history of intergroup contact promoting greater openness to other groups and a more favorable environment for the establishment of educational institutions. Other Relevant Medieval Institutions Aside from exposure to ethnic and religious diversity, there are some other characteristics of medieval towns that might later have affected the demand for educational institutions. If these other characteristics are also correlated with the location of archae or of Saracen’s Head inns, then it will be important to control for them in our empirical model of premodern library location. Institution #1: The School Libraries might have been easier to establish in towns where there was already some form of formal education. Until the mid-nineteenth century, there were only two universities in England, and most formal education took place in schools. Our information about medieval schools is drawn from Orme (2006), who provides a comprehensive discussion of medieval English education. The evidence that Orme presents shows that medieval schools were sophisticated, diverse, and independent of control by any central authority. They provided a context for encounters with foreign languages and cultures. Despite the differences between modern and medieval worldviews, medieval schools fulfilled the same functions as do modern ones: medieval schools were “the same thing in different circumstances.” Before the Norman Conquest, schools were few in number and largely restricted to monastic institutions, but the twelfth century saw the establishment of other types of school, especially grammar schools with secular teachers and students. Although monastic schools were solely for the purpose of educating those in religious orders, cathedral schools were open to a wider range of students from different walks of life. Our data on medieval school location are taken from Orme (2006, 190–94). Appendix C includes a list of these 83 locations; the list includes all known schools established in the twelfth or thirteenth century,13 except monastic establishments from which the local population would have been excluded. Institution #2: The Manuscript Library Even before the invention of moveable type, which was introduced into England toward the end of the fifteenth century, large medieval schools and monasteries were supported by libraries of handwritten books. The typical library contained a relatively small number of books, but the subject matter of these books was very diverse (Kibre 1946). The existence of such a manuscript library in a town may have facilitated the later development of libraries with printed books. Appendix C includes a list of 97 locations where a library is known to have been established before the year 1500; this information comes from Alston (2011). The Distribution of Medieval Schools, Libraries, Archae, and Saracen’s Head Inns Appendix C includes an analysis of the geographical distribution of medieval libraries, schools, archae, and Saracen’s Head inns. Although our four variables are positively correlated, the correlations are explained by simple factors such as medieval town size and distance from London. Conditional on these factors, there are no significant correlations, so we do have four distinct characteristics of medieval towns. In the data analysis that follows, we estimate the extent to which the location of premodern libraries and bookstores depends on our four medieval variables. We interpret the effect of medieval libraries and schools as persistence in the importance of educational institutions, which could be explained either by local variation in preferences for education (a cultural characteristic) or by the medieval institutions generating sector-specific human and physical capital in certain areas. We interpret the effect of archae and Saracen’s Head inns as the influence of attitudes toward outgroups on later institutional location. It is the latter that will provide unambiguous evidence for the effect of culture on institutions. It is important to make one further point regarding the interpretation of the archa and Saracen’s Head effects. The preceding discussion highlights the possibility that the presence of Jews and returning crusaders had a treatment effect on the local community, and this treatment effect could be interpreted in terms of intergroup contact theory. An alternative possibility is that the presence of Jews and returning crusaders reflected a selection effect: that is, the medieval migrants were more likely to settle in places where the existing community was already relatively cosmopolitan. In the absence of further medieval data, it is difficult to estimate the relative importance of treatment and selection effects. The absence of a significant conditional correlation between the location of archae and the location of Saracen’s Head inns suggests that a treatment effect is more likely: if certain places were already relatively cosmopolitan, why didn’t Jews and returning crusaders both settle in these places? However, the sample of medieval towns used in appendix C is quite small, which limits the power of the test of the significance of the conditional correlation. Nevertheless, regardless of the relative importance of the treatment and selection effects, our archa and Saracen’s Head inn variables identify those places that were relatively cosmopolitan in the Middle Ages. Analyzing the Libraries Data The figures in table 6 illustrate the association between the number of premodern libraries in a town (excluding the medieval ones) and the location of medieval schools, libraries, archae, and Saracen’s Head inns. The libraries data are again based on Alston (2011), using a sample of 838 English towns for which adequate population data are available from Bennett (2011, appendix 3); London is again excluded from our analysis. Among these 838 towns, there are 690 with at least one premodern library and 6,057 libraries in total, compared with 122 medieval libraries in 95 towns. There are 19 towns with over 50 libraries, including the two university towns of Oxford and Cambridge; the largest number—237—is for Liverpool. None of the results in this section are very different if we exclude these 19 towns (or just Oxford and Cambridge) from the sample. Table 6. Average Number of Premodern Libraries, Disaggregated by Medieval Characteristics Medieval characteristic of the town  Average number of premodern libraries when the medieval characteristic is absent  Average number of premodern libraries when the medieval characteristic is present  Medieval school  5.51  21.35  Medieval library  5.64  18.32  Archa  6.06  35.45  Saracen’s Head inn  4.68  28.07  Medieval characteristic of the town  Average number of premodern libraries when the medieval characteristic is absent  Average number of premodern libraries when the medieval characteristic is present  Medieval school  5.51  21.35  Medieval library  5.64  18.32  Archa  6.06  35.45  Saracen’s Head inn  4.68  28.07  Note: The sample comprises 838 English towns outside London. Here, unlike in tables 4 and 5, the premodern library numbers exclude medieval libraries (i.e., ones founded before 1500). Table 6. Average Number of Premodern Libraries, Disaggregated by Medieval Characteristics Medieval characteristic of the town  Average number of premodern libraries when the medieval characteristic is absent  Average number of premodern libraries when the medieval characteristic is present  Medieval school  5.51  21.35  Medieval library  5.64  18.32  Archa  6.06  35.45  Saracen’s Head inn  4.68  28.07  Medieval characteristic of the town  Average number of premodern libraries when the medieval characteristic is absent  Average number of premodern libraries when the medieval characteristic is present  Medieval school  5.51  21.35  Medieval library  5.64  18.32  Archa  6.06  35.45  Saracen’s Head inn  4.68  28.07  Note: The sample comprises 838 English towns outside London. Here, unlike in tables 4 and 5, the premodern library numbers exclude medieval libraries (i.e., ones founded before 1500). Each row of table 6 corresponds to a different medieval characteristic; one column of the table shows the average number of premodern libraries in towns where the characteristic is present, while another column shown the average number of libraries in towns where the characteristic is absent. It can be seen that the average number of libraries is always much higher in towns where the characteristic is present. These associations do not necessarily entail a direct link between the medieval characteristics and library numbers: they could simply reflect the fact that larger medieval towns were more likely to have the characteristics than smaller ones, larger premodern towns had more libraries, and there is some inter-temporal persistence in the relative sizes of towns. For this reason, we also fit a regression equation in which the dependent variable is the number of premodern libraries in a town. The set of explanatory variables includes binary variables indicating the presence of a medieval school, library, archa, or Saracen’s Head inn (denoted medieval school, medieval library, archa, saracen), plus county fixed effects (capturing regional characteristics such as distance to London), the logarithm of the town’s population in the 1841 census (denoted population in 1841), and the logarithm of the town’s population in the mid-seventeenth century (denoted population in c17). The population measures are taken from Bennett (2011, appendix 3). Also included in the regression equation is a medieval population measure: this variable (denoted population in c14) is equal to the logarithm of the town’s adult male population in the 1377 Poll Tax records, if the town is large enough to appear in the list of Dyer (2000).14 Finally, there is a binary variable indicating the presence of the town in Dyer’s list (denoted town in c14). Descriptive statistics for these variables appear in appendix B. The distribution of libraries across towns is highly skewed and approximately equal to a Poisson distribution (see figure 1), so we fit a Poisson model to the data. This means that each of the estimated coefficients can be interpreted as the percentage change in the number of libraries associated with a unit change in the explanatory variable. The first set of results in table 7 (column (i)) shows the values of these coefficients and the corresponding t-ratios. The results indicate that the number of libraries is strongly associated with the size of the town in the seventeenth and nineteenth centuries, but for a given population level, a town with a medieval school can be expected to have 27 percent more libraries and a town with a Saracen’s Head inn can be expected to have 30 percent more libraries; both of these effects are significant at the 1 percent level. The medieval library and archa effects are statistically insignificant. Figure 1. View largeDownload slide Histogram of Libraries (excluding towns in which Libraries >50) Figure 1. View largeDownload slide Histogram of Libraries (excluding towns in which Libraries >50) It is possible that these results do not adequately capture all of the town characteristics that might be associated both with the location of medieval institutions and with premodern library density. For this reason, the second set of results in table 7 (column (ii)) shows coefficients and t-ratios from a model that includes an additional set of explanatory variables. These are: an indicator variable for whether the town is a port (denoted port town), the number of unskilled laborers in town as a fraction of the total working population (denoted laborers), the total number of inns in the town (denoted inns),15 laborers,2 and inns.2 These figures are based on information about the number of laborers and innkeepers in each parish over the years 1813–1820: see Shaw-Taylor, Wrigley, and Kitson (2006). The fraction of unskilled laborers in the town is included as a measure of relative poverty. The number of inns is included in order to capture variation in the number of visitors to the town, which may influence the number of other facilities there, including libraries. It is also possible that port towns are better connected to London and so have more extensive facilities. Note that the Shaw-Taylor et al. dataset covers only half of the counties in England, and this reduces the column (ii) sample size from 838 towns to 348. Column (ii) shows that there is a statistically significant association between the number of libraries and the number of inns; conditional on this effect, there is no significant association between the number of libraries and number of medieval schools, medieval libraries, or archae. However, the size of the association with the number of Saracen’s Head inns is only slightly smaller than in column (i), and this effect is still significant at the 1 percent level. Table 7. Determinants of the Number of Premodern Libraries in a Town: Poisson Model Estimates   (i)  (ii)    Full sample (838 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  Coeff.  t-ratio  Town in c14  0.201  2.00**  0.235  1.72*  Logarithm of population in c14  −0.132  −1.61  −0.087  −0.64  Logarithm of population in c17  0.180  2.39**  0.216  2.01**  Logarithm of population in 1841  0.762  16.26***  0.746  11.87***  Medieval school  0.220  2.56***  0.183  1.03  Medieval library  0.107  1.14  0.054  0.31  Archa  0.031  0.32  0.066  0.28  Saracen  0.336  3.81***  0.270  2.44***  Laborers      0.486  0.51  Laborers2      −0.136  −0.09  Port town      0.204  1.51  Inns      0.291  2.45***  Inns2      −0.050  −2.40***    (i)  (ii)    Full sample (838 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  Coeff.  t-ratio  Town in c14  0.201  2.00**  0.235  1.72*  Logarithm of population in c14  −0.132  −1.61  −0.087  −0.64  Logarithm of population in c17  0.180  2.39**  0.216  2.01**  Logarithm of population in 1841  0.762  16.26***  0.746  11.87***  Medieval school  0.220  2.56***  0.183  1.03  Medieval library  0.107  1.14  0.054  0.31  Archa  0.031  0.32  0.066  0.28  Saracen  0.336  3.81***  0.270  2.44***  Laborers      0.486  0.51  Laborers2      −0.136  −0.09  Port town      0.204  1.51  Inns      0.291  2.45***  Inns2      −0.050  −2.40***  Note: The sample in column (i) is the same as in table 6; the sample in column (ii) omits towns lacking data for the variables laborers and inns. The number of premodern libraries in each town excludes medieval libraries (i.e., ones founded before 1500). Town in c14 is a binary variable equal to one if the town appears in the list of fourteenth-century towns in Dyer (2000), and zero otherwise; population in c14 is the population level recorded in Dyer’s list (or set equal to one if the town is not in the list); population in c17 and population in 1841 are population estimates for the mid-seventeenth century and for 1841. Laborers denotes the share of unskilled laborers in the total working population of the town; port town is a binary variable equal to one if the town is a port and zero otherwise; inns denotes the number of inns in the town. Medieval school (medieval library, archa, saracen) is a binary variable equal to one if there was a medieval library (medieval school, Jewish settlement, Saracen’s Head inn) in the town, and zero otherwise. The set of explanatory variables also includes county fixed effects. One asterisk (*) indicates a coefficient significantly different from zero at the 10 percent level; two asterisks (**) indicate significance at the 5 percent level, and three asterisks (***) indicate significance at the 1 percent level. Table 7. Determinants of the Number of Premodern Libraries in a Town: Poisson Model Estimates   (i)  (ii)    Full sample (838 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  Coeff.  t-ratio  Town in c14  0.201  2.00**  0.235  1.72*  Logarithm of population in c14  −0.132  −1.61  −0.087  −0.64  Logarithm of population in c17  0.180  2.39**  0.216  2.01**  Logarithm of population in 1841  0.762  16.26***  0.746  11.87***  Medieval school  0.220  2.56***  0.183  1.03  Medieval library  0.107  1.14  0.054  0.31  Archa  0.031  0.32  0.066  0.28  Saracen  0.336  3.81***  0.270  2.44***  Laborers      0.486  0.51  Laborers2      −0.136  −0.09  Port town      0.204  1.51  Inns      0.291  2.45***  Inns2      −0.050  −2.40***    (i)  (ii)    Full sample (838 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  Coeff.  t-ratio  Town in c14  0.201  2.00**  0.235  1.72*  Logarithm of population in c14  −0.132  −1.61  −0.087  −0.64  Logarithm of population in c17  0.180  2.39**  0.216  2.01**  Logarithm of population in 1841  0.762  16.26***  0.746  11.87***  Medieval school  0.220  2.56***  0.183  1.03  Medieval library  0.107  1.14  0.054  0.31  Archa  0.031  0.32  0.066  0.28  Saracen  0.336  3.81***  0.270  2.44***  Laborers      0.486  0.51  Laborers2      −0.136  −0.09  Port town      0.204  1.51  Inns      0.291  2.45***  Inns2      −0.050  −2.40***  Note: The sample in column (i) is the same as in table 6; the sample in column (ii) omits towns lacking data for the variables laborers and inns. The number of premodern libraries in each town excludes medieval libraries (i.e., ones founded before 1500). Town in c14 is a binary variable equal to one if the town appears in the list of fourteenth-century towns in Dyer (2000), and zero otherwise; population in c14 is the population level recorded in Dyer’s list (or set equal to one if the town is not in the list); population in c17 and population in 1841 are population estimates for the mid-seventeenth century and for 1841. Laborers denotes the share of unskilled laborers in the total working population of the town; port town is a binary variable equal to one if the town is a port and zero otherwise; inns denotes the number of inns in the town. Medieval school (medieval library, archa, saracen) is a binary variable equal to one if there was a medieval library (medieval school, Jewish settlement, Saracen’s Head inn) in the town, and zero otherwise. The set of explanatory variables also includes county fixed effects. One asterisk (*) indicates a coefficient significantly different from zero at the 10 percent level; two asterisks (**) indicate significance at the 5 percent level, and three asterisks (***) indicate significance at the 1 percent level. Libraries and Bookstores The expansion of libraries in the seventeenth and eighteenth centuries was facilitated by a growing provincial book trade (Feather 2006; Suarez 2009), and another indicator of local openness to education is the existence of a bookstore. Shaw-Taylor, Wrigley, and Kitson (2006) also provide data on the number of booksellers in each parish: this number was rarely greater than one, so it does not make sense to try to model the number of booksellers, but we can still explore whether any of the medieval characteristics is significantly associated with the presence of at least one bookseller in town. Table 8 shows some results with regard to booksellers: the structure of this table resembles that of table 7, but here the dependent variable takes a value of one if there was at least one bookseller in town during 1813–1820, and a value of zero if there was no bookseller. The table 8 results are based on a Probit model, so the table also includes a set of marginal effects: these numbers indicate the average change in the probability of there being a bookseller in town when there is a unit increase in the explanatory variable. Table 8. Determinants of the Probability of a Bookseller in Town in a Town: Probit Model Estimates   (i)  (ii)    Full sample (348 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  m.e.  Coeff.  t-ratio  m.e.  Town in c14  0.285  0.62  0.07  0.225  0.59  0.05  Logarithm of population in c14  −0.123  −0.24  −0.03  −0.145  −0.29  −0.03  Logarithm of population in c17  −0.059  −0.31  −0.01  −0.162  −0.82  −0.03  Logarithm of population in 1841  0.709  4.74***  0.16  0.968  5.29***  0.18  Medieval school  0.958  2.46***  0.31  1.055  2.53***  0.31  Medieval library  −0.297  −0.79  −0.06  −0.596  −1.47  −0.08  Archa  0.993  1.65*  0.33  1.076  1.83*  0.32  Saracen  0.273  0.93  0.07  0.410  1.41  0.09  Laborers        6.227  2.29**    Laborers2        −15.480  −3.44***    Port town        0.140  0.46  0.03  Inns        −0.020  −0.07    Inns2        −0.017  −0.28      (i)  (ii)    Full sample (348 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  m.e.  Coeff.  t-ratio  m.e.  Town in c14  0.285  0.62  0.07  0.225  0.59  0.05  Logarithm of population in c14  −0.123  −0.24  −0.03  −0.145  −0.29  −0.03  Logarithm of population in c17  −0.059  −0.31  −0.01  −0.162  −0.82  −0.03  Logarithm of population in 1841  0.709  4.74***  0.16  0.968  5.29***  0.18  Medieval school  0.958  2.46***  0.31  1.055  2.53***  0.31  Medieval library  −0.297  −0.79  −0.06  −0.596  −1.47  −0.08  Archa  0.993  1.65*  0.33  1.076  1.83*  0.32  Saracen  0.273  0.93  0.07  0.410  1.41  0.09  Laborers        6.227  2.29**    Laborers2        −15.480  −3.44***    Port town        0.140  0.46  0.03  Inns        −0.020  −0.07    Inns2        −0.017  −0.28    Note: The sample is the same as in table 7 (column (ii)). M.e. stands for the marginal effect of the variable; marginal effects for laborers and inns are not shown because these values are not constant: more information is available on request. See table 7 for other notes. Table 8. Determinants of the Probability of a Bookseller in Town in a Town: Probit Model Estimates   (i)  (ii)    Full sample (348 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  m.e.  Coeff.  t-ratio  m.e.  Town in c14  0.285  0.62  0.07  0.225  0.59  0.05  Logarithm of population in c14  −0.123  −0.24  −0.03  −0.145  −0.29  −0.03  Logarithm of population in c17  −0.059  −0.31  −0.01  −0.162  −0.82  −0.03  Logarithm of population in 1841  0.709  4.74***  0.16  0.968  5.29***  0.18  Medieval school  0.958  2.46***  0.31  1.055  2.53***  0.31  Medieval library  −0.297  −0.79  −0.06  −0.596  −1.47  −0.08  Archa  0.993  1.65*  0.33  1.076  1.83*  0.32  Saracen  0.273  0.93  0.07  0.410  1.41  0.09  Laborers        6.227  2.29**    Laborers2        −15.480  −3.44***    Port town        0.140  0.46  0.03  Inns        −0.020  −0.07    Inns2        −0.017  −0.28      (i)  (ii)    Full sample (348 towns)  Restricted sample with all controls (348 towns)    Coeff.  t-ratio  m.e.  Coeff.  t-ratio  m.e.  Town in c14  0.285  0.62  0.07  0.225  0.59  0.05  Logarithm of population in c14  −0.123  −0.24  −0.03  −0.145  −0.29  −0.03  Logarithm of population in c17  −0.059  −0.31  −0.01  −0.162  −0.82  −0.03  Logarithm of population in 1841  0.709  4.74***  0.16  0.968  5.29***  0.18  Medieval school  0.958  2.46***  0.31  1.055  2.53***  0.31  Medieval library  −0.297  −0.79  −0.06  −0.596  −1.47  −0.08  Archa  0.993  1.65*  0.33  1.076  1.83*  0.32  Saracen  0.273  0.93  0.07  0.410  1.41  0.09  Laborers        6.227  2.29**    Laborers2        −15.480  −3.44***    Port town        0.140  0.46  0.03  Inns        −0.020  −0.07    Inns2        −0.017  −0.28    Note: The sample is the same as in table 7 (column (ii)). M.e. stands for the marginal effect of the variable; marginal effects for laborers and inns are not shown because these values are not constant: more information is available on request. See table 7 for other notes. Table 8 shows that the probability of there being a bookseller in town is significantly associated with its nineteenth-century population and with the fraction of the population who are unskilled laborers. Conditional on these effects, there is a significant association with both the presence of a medieval school, which is estimated to raise the bookseller probability by 31 percentage points in column (ii), and with the presence of an archa, which is estimated to raise the bookseller probability by 32 percentage points. However, neither the presence of a medieval library nor the presence of a Saracen’s Head inn has any significant effect. Taken together, tables 7 and 8 show evidence that the exposure of a town to medieval ethnic and religious diversity is significantly associated with a greater density of educational institutions at a later date. Towns with a Saracen’s Head inn, which suggests the salience of the crusades to the local community, had more premodern libraries, while towns with a Jewish community in the twelfth and thirteenth century were more likely to have a bookseller in the nineteenth century.16 The effects are large: the mean probability of a town having a bookstore is 0.17, and holding all other factors constant, the presence of a Jewish community is estimated to increase this probability to 0.49. The mean number of pre-1851 libraries is seven, and holding all other factors constant, the presence of a Saracen’s Head inn is estimated to increase this number by 30 percent, giving the town two more libraries. As we have seen, a larger number of pre-1851 libraries implies a larger number of students in the twenty-first century, and this implies a local population with more positive attitudes toward immigration and toward equal rights for gays and lesbians, ethnic minorities, and women. Conclusion In this paper, we have traced a chain of connections between (i) English attitudes toward minority rights, immigration, and European integration in twenty-first-century opinion polls and (ii) exposure to ethnic and religious diversity in the twelfth and thirteenth centuries. Attitudes are significantly more positive in locations that have larger universities (measured by student numbers), university size depends on the local density of earlier educational institutions, and this density is higher in places showing signs of medieval exposure to diversity. The persistence of geographical variation in the density of educational institutions seems to have been a vehicle for the persistence of geographical variation in attitudes. Is it the only vehicle? In appendix D, we present evidence suggesting that the answer to this question is “probably not.” When our indicators of medieval exposure to diversity (archa and saracen) are added to the model of attitudes summarized in table 3, their effect is found to be jointly statistically significant in five out of seven cases, the two exceptions being attitudes toward the European Union in 2010 and 2015. That is, even when we control for modern university size, there is still a significant association between twenty-first-century attitudes and medieval exposure to diversity. This conditional association seems to be largely through the archa effect (reflecting a local Jewish heritage) rather than through the saracen effect (reflecting local salience of the crusades). The other mechanisms responsible for the inter-generational persistence of regional variation in attitudes have yet to be identified, but they could include, for example, informal social interaction that reinforces conformity to local norms in a cultural evolutionary process of the kind proposed by Cavalli-Sforza and Feldman (1973). Further research into these mechanisms will be required for a full understanding of the reasons why social attitudes display such persistent geographical variation. While our results regarding the historical drivers of variation in the density of English educational institutions are consistent with existing evidence regarding the effect of international linkages on university expansion (Schofer and Meyer 2005), this effect is only half of the story. The presence (or absence) of a university has itself contributed to a high degree of local persistence in attitudes toward foreigners and minority groups. This second effect suggests that modern social and economic forces that cause universities to expand or contract could also lead to substantial changes in attitudes, with long-lasting political consequences. Notes 1 Throughout this paper, the term “English” refers specifically to England, excluding Wales and Scotland. Some of the data described in subsequent sections of the paper are not available for Wales or Scotland. 2 Results disaggregating men and women are available on request. Unfortunately, it is not possible to disaggregate the sample by sexual orientation. 3 In terms of census terminology, a university student is any person in full-time tertiary education. 4 We use a logarithmic transformation because the distribution of student ratio is highly skewed, with a few constituencies that have very high student numbers. 5 The immigrant population density is potentially endogenous to attitudes. Following Dustmann and Preston (2001), it is possible to use county-level population density as an instrument for constituency-level population density. Results using such an instrument are very similar to the ones reported in table 3. 6 This immigration variable has four possible values. In this case, the marginal effect measures the impact of a unit increase in the logarithm of student ratio on the probability that the value will be greater than one, that is, that immigration will be ranked as one of the top three issues. 7 Some constituencies do contain more than one town, and in these cases we select the town with the largest population in the 1841 census. The within-constituency ranking of towns by size has changed very little since this time, so our results are very similar if we use population numbers for a later date. Other constituencies comprise just part of one large city, and in these cases the library number is ascribed to all of the constituencies in the city. 8 An alternative approach to dealing with the skewness is to use a set of indicator variables for constituencies with over 30 libraries, constituencies with over 20 libraries, and constituencies with over 10 libraries. This produces results that are very similar to those in table 5. 9 The regression equation also includes the instrumental variables in the IV model of attitudes discussed in the second part of appendix A: these variables are uncorrelated with libraries (r < 0.01), but they do explain some of the variation in the logarithm of student ratio, and their inclusion does increase the precision of the estimates of the effect of other explanatory variables. 10 Huguenot communities assimilated very quickly to the local English culture, and the location of Huguenot settlements turns out not to be a significant factor in the models discussed below. 11 The one attack in which locals were clearly directly implicated was at York; this attack seems to have been organized by members of the petty nobility who had run up large debts to local Jews. 12 In most cases, there is no direct archaeological evidence with which to establish a precise date for the foundation of these inns. However, Saracen’s Head inn sites that have been excavated have produced evidence for a foundation date contemporary with or not much later than the crusading period (Andrews et al. 2003; Bowsher et al. 2007), and in some cases there is documentary evidence for a medieval foundation date (English Heritage 2010). 13 It is also possible to include in our dataset the location of late medieval schools. However, the rapid expansion of schools after 1300 meant that by the end of the Middle Ages almost all towns of moderate size had a school, and the location of early medieval schools is more useful in explaining the location of premodern libraries and bookstores. That is, the popularity of the later types of institution (libraries and bookstores) is predicted by early adoption of the original type of institution (schools). 14 The list includes all towns of at least 480 men. For towns not on the list, population in c14 is set equal to 480. The inclusion of the binary variable town in c14 in the model means that this arbitrary choice does not affect the results. 15 These figures incorporate all inns, public houses, and taverns. 16 Why are Saracen’s Head inns more important for library location and archae more important for bookseller location? Both libraries and bookstores reflect a demand for new ideas, but libraries are the more communal institution. Crusading was a group activity that exposed the individual to foreign culture at the same time as promoting an esprit de corps, and perhaps it created a taste for communal activity as well as openness to new ideas. Another possibility is that the presence of a bookstore indicates a town with an entrepreneurial spirit and a large financial sector that is favorable to small business. The presence of a medieval Jewish community could be associated with this type of business environment, in which case the archa effect is explained by an economic mechanism rather than a cultural one. We can test this conjecture by using the Shaw-Taylor, Wrigley, and Kitson (2006) dataset to identify those towns that were home to businesses selling luxury items other than books (for example, towns that were home to silversmiths), and then fitting a model of the presence of these businesses. If there is an economic mechanism at work, then the location of archae should predict the location of silversmiths as well as the location of booksellers. When we do fit a silversmith model, the estimated archa coefficient is insignificantly different from zero (p > 0.25 in all versions of the model). This suggests that the mechanism is not an economic one. About the Author David Fielding is Donald Reid Chair of Economics at the University of Otago, New Zealand. His current research examines the social and economic consequences of ethnic diversity, and drivers of the salience of ethnic difference, both in Europe and in Africa. He also researches attitudes toward foreign aid and aid allocation. Recent publications include papers in the British Journal of Political Science (2018), Journal of Development Studies (2017), and Oxford Bulletin of Economics and Statistics (2017). Supplementary Material Supplementary material is available at Social Forces online. References Alesina, Alberto, and Paola Giuliano. 2015. “ Culture and Institutions.” Journal of Economic Literature  53( 4): 898– 944. Google Scholar CrossRef Search ADS   Alesina, Alberto, Paola Giuliano, and Nathan Nunn. 2013. “ On the Origins of Gender Roles: Women and the Plough.” Quarterly Journal of Economics  128( 2): 469– 530. Google Scholar CrossRef Search ADS   Allport, Gordon W. 1954. The Nature of Prejudice . Reading, MA: Addison-Wesley. Alston, Robin. 2011. Libraries in England to 1850. www.scribd.com/doc/63097781/Robin-Alston-Library-History-England. Andrews, Philip, J. Lowe, K. Nichols, C. Phillpotts, and A. Powell. 2003. Charter Quay—The Spirit of Change: The Archaeology of Kingston’s Riverside . Salisbury, UK: Wessex Archaeology. Attiya, Hussein M. 1999. “ Knowledge of Arabic in the Crusader States in the Twelfth and Thirteenth Centuries.” Journal of Medieval History  25( 3): 203– 13. Google Scholar CrossRef Search ADS   Banaszak, Lee Ann, and Eric Plutzer. 1993. “ Contextual Determinants of Feminist Attitudes: National and Subnational Influences in Western Europe.” American Political Science Review  87( 1): 145– 57. Google Scholar CrossRef Search ADS   Benjamin, David J. 2015. Support Structures in Crusading Armies, 1095–1241. Doctoral dissertation, Leicester University. Bennett, Robert. J. 2011. Local Business Voice: The History of Chambers of Commerce in Britain, Ireland and Revolutionary America, 1760–2011 . Oxford: Oxford University Press. Google Scholar CrossRef Search ADS   Bobo, Lawrence, and Frederick C. Licari. 1989. “ Education and Political Tolerance: Testing the Effects of Cognitive Sophistication and Target Group Affect.” Public Opinion Quarterly  53( 3): 285– 308. Google Scholar CrossRef Search ADS   Bowsher, David, Tony Dyson, Nick Holder, and Isca Howell. 2007. The London Guildhall: An Archaeological History of a Neighbourhood from Early Medieval to Modern Times . London: MoLAS. Brand, Paul. 2003. “The Jewish Community of England in the Records of English Royal Government.” In The Jews in Medieval Britain: Historical, Literary and Archaeological Perspectives , edited by Patricia Skinner, 73– 85. Woodbridge, UK: Boydell Press. Brown, Reva B., and Sean McCartney. 2005. “ The Exchequer of the Jews Revisited: The Operation and Effect of the Scaccarium Judeorum.” Medieval History Journal  8( 2): 303– 22. Google Scholar CrossRef Search ADS   Bruter, Michael. 2005. Citizens of Europe? The Emergence of a Mass European Identity . Basingstoke, UK: Palgrave Macmillan. Google Scholar CrossRef Search ADS   Burke, Bernard. 1884. The General Armory of England, Scotland, Ireland and Wales . London: Harrison. Calkin, Siobhain B. 2012. “Saracens.” In Heroes and Anti-Heroes in Medieval Romance , edited by Neil Cartlidge, 185– 200. Cambridge: Brewer. Cavalli-Sforza, Luca, and Marcus W. Feldman. 1973. “ Models for Cultural Inheritance I: Group Mean and Within-Group Variation.” Theoretical Population Biology  4( 1): 42– 55. Google Scholar CrossRef Search ADS PubMed  Coenders, Marcel, and Peer Scheepers. 2003. “ The Effect of Education on Nationalism and Ethnic Exclusionism: An International Comparison.” Political Psychology  24( 2): 313– 43. Google Scholar CrossRef Search ADS   Dittmar, Jeremiah E. 2011. “ Information Technology and Economic Change: The Impact of the Printing Press.” Quarterly Journal of Economics  126( 3): 1133– 72. Google Scholar CrossRef Search ADS   Dovidio, John F., Peter E. Glick, and Laurie A. Rudman. 2005. On the Nature of Prejudice: Fifty Years after Allport . Oxford: Blackwell Publishing. Dustmann, Christian, and Ian Preston. 2001. “ Attitudes to Ethnic Minorities, Ethnic Context and Location Decisions.” Economic Journal  111( 470): 353– 73. Google Scholar CrossRef Search ADS   Dyer, Alan. 2000. “Ranking Lists of English Medieval Towns.” In The Cambridge Urban History of Britain (Volume I: 600–1540) , edited by David M. Palliser, Peter Clark, and Martin J. Daunton, 747– 70. Cambridge: Cambridge University Press. Google Scholar CrossRef Search ADS   English Heritage. 2010. Newport Pagnell: Historic Town Assessment Report. Milton Keynes, UK: Milton Keynes Council/Buckinghamshire County Council. Feather, John. 2006. “The Book Trade and Libraries.” In The Cambridge History of Libraries in Britain and Ireland (Volume 2: 1640–1850) , edited by Giles Mandelbrote and K. A. Manley, 301– 12. Cambridge: Cambridge University Press. Fielding, David. 2018. “ Traditions of Tolerance: The Long-Run Persistence of Regional Variation in Attitudes towards English Immigrants.” British Journal of Political Science  48( 1): 167– 88. Google Scholar CrossRef Search ADS   Hamilton, Bernard. 1997. “ Knowing the Enemy: Western Understanding of Islam at the Time of the Crusades.” Journal of the Royal Asiatic Society (Third Series)  7( 3): 373– 87. Google Scholar CrossRef Search ADS   Hello, Evelyn, Peer Scheepers, and Peter Sleegers. 2006. “ Why the More Educated Are Less Inclined to Keep Ethnic Distance: An Empirical Test of Four Explanations.” Ethnic and Racial Studies  29( 5): 959– 85. Google Scholar CrossRef Search ADS   Hello, Evelyn, Peer Scheepers, Ad Vermulst, and Jan R M Gerris. 2004. “ Association between Educational Attainment and Ethnic Distance in Young Adults: Socialization by Schools or Parents?” Acta Sociologica  47( 3): 253– 75. Google Scholar CrossRef Search ADS   Hillaby, Joe. 2003. “Jewish Colonisation in the Twelfth Century.” In The Jews in Medieval Britain: Historical, Literary and Archaeological Perspectives , edited by Patricia Skinner, 15– 41. Woodbridge, UK: Boydell Press. Hillaby, Joe, and Caroline Hillaby. 2013. The Palgrave Dictionary of Medieval Anglo-Jewish History . Basingstoke, UK: Palgrave Macmillan. Google Scholar CrossRef Search ADS   Hooghe, Marc, Cecil Meeusen, and Ellen Quintelier. 2013. “ The Impact of Education and Intergroup Friendship on the Development of Ethnocentrism.” European Sociological Review  29( 6): 1109– 21. Google Scholar CrossRef Search ADS   Jha, Saumitra. 2013. “ Trade, Institutions and Ethnic Tolerance: Evidence from South Asia.” American Political Science Review  107: 806– 32. Google Scholar CrossRef Search ADS   Khanmohamadi, Shirin. 2010. “ Casting a ‘Sideways Glance’ at the Crusades: The Voice of the Other in Joinville’s Vie de Saint Louis.” Exemplaria  22( 3): 177– 99. Google Scholar CrossRef Search ADS   Kibre, Pearl. 1946. “ The Intellectual Interests Reflected in Libraries of the Fourteenth and Fifteenth Centuries.” Journal of the History of Ideas  7( 3): 257– 97. Google Scholar CrossRef Search ADS   Langmuir, Gavin I. 1963. “ The Jews and the Archives of Angevin England: Reflections on Medieval Anti-Semitism.” Traditio  19: 183– 244. Google Scholar CrossRef Search ADS   Mayda, Anna-Maria. 2006. “ Who Is Against Immigration? A Cross-Country Investigation of Individual Attitudes toward Immigrants.” Review of Economics and Statistics  88( 3): 510– 30. Google Scholar CrossRef Search ADS   McVeigh, Rory, and Maria-Elena D. Diaz. 2009. “ Voting to Ban Same-Sex Marriage: Interests, Values, and Communities.” American Sociological Review  74( 6): 891– 915. Google Scholar CrossRef Search ADS   Meyer, John W., Francisco O. Ramirez, David J. Frank, and Evan Schofer. 2008. “Higher Education as an Institution.” In Sociology of Higher Education: Contributions and Their Contexts , edited by Patricia J. Gumport, 187– 221. Baltimore, MD: Johns Hopkins University Press. Mundill, Robert R. 2010. The King’s Jews: Money, Massacre and Exodus in Medieval England . London: Continuum Books. Ohlander, Julianne, Jeanne Batalova, and Judith Treas. 2005. “ Explaining Educational Influences on Attitudes toward Homosexual Relations.” Social Science Research  34( 4): 781– 99. Google Scholar CrossRef Search ADS   Orme, Nicholas. 2006. Medieval Schools: From Roman Britain to Renaissance England . New Haven, CT: Yale University Press. Pettigrew, Thomas F., and Linda R. Tropp. 2006. “ A Meta-Analytic Test of Intergroup Contact Theory.” Journal of Personality and Social Psychology  90: 751– 83. Google Scholar CrossRef Search ADS PubMed  Pettigrew, Thomas F., and Linda R. Tropp. 2012. When Groups Meet: The Dynamics of Intergroup Contact . New York: Psychology Press. Pierotti, Rachael S. 2013. “ Increasing Rejection of Intimate Partner Violence: Evidence of Global Cultural Diffusion.” American Sociological Review  78( 2): 240– 65. Google Scholar CrossRef Search ADS   Raven, James. 2006. “Libraries for Sociability: The Advance of the Subscription Library.” In The Cambridge History of Libraries in Britain and Ireland (Volume 2: 1640–1850) , edited by Giles Mandelbrote and K. A. Manley, 239– 63. Cambridge: Cambridge University Press. Rouleau, Benoit. 2005. L’Islam et les Musulmans dans l’Oeuvre de Jean de Joinville (1224–1318) [Islam and Muslims in the Work of Jean de Joinville (1224–1318)]. Master’s dissertation, Sherbrooke University. Schofer, Evan, and John W. Meyer. 2005. “ The Worldwide Expansion of Higher Education in the Twentieth Century.” American Sociological Review  70( 6): 898– 920. Google Scholar CrossRef Search ADS   Semyonov, Moshe, and Rebeca Raijman. 2006. ‘ The Rise of Anti-Foreigner Sentiment in European Societies, 1988–2000.” American Sociological Review  71( 3): 426– 49. Google Scholar CrossRef Search ADS   Shaw-Taylor, Leigh, Edward A. Wrigley, and Peter M. Kitson. 2006. Male Occupational Data from English Parish Registers (Processed Summary Data), c. 1700–1820. http://dx.doi.org/10.5255/UKDA-SN-5398-1. Stacey, Robert C. 2003. “The English Jews under Henry III.” In The Jews in Medieval Britain: Historical, Literary and Archaeological Perspectives , edited by Patricia Skinner, 41– 55. Woodbridge, UK: Boydell Press. Suarez, Michael F., S.J. 2009. “Towards a Bibliometric Analysis of the Surviving Record, 1701–1800.” In The Cambridge History of the Book in Britain , edited by Michael F. Suarez, S.J., and Michael L. Turner, 37– 65. Cambridge: Cambridge University Press. Google Scholar CrossRef Search ADS   Voigtländer, Nico, and Hans-Joachim Voth. 2013. “ Persecution Perpetuated: The Medieval Origins of Anti-Semitic Violence in Nazi Germany.” Quarterly Journal of Economics  127: 1339– 92. Google Scholar CrossRef Search ADS   © The Author(s) 2018. Published by Oxford University Press on behalf of the University of North Carolina at Chapel Hill. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Social ForcesOxford University Press

Published: Feb 21, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off