Landed Elites and Education Provision in England: Evidence from School Boards, 1870-99∗ Marc Go˜ ni† August 22, 2017

Abstract This paper studies the relationship between landownership concentration and state-sponsored education in late-nineteenth century England. Cross-sectional estimates show a negative association between land inequality and a wide range of education measures. To establish causality, I exploit the redistribution of land after the Norman conquest in 1066 and geographical variation in soil texture. The estimated effects are stronger where landlords had political power and weaker for variables related to the demand for education (e.g., attendance). These results imply that the mechanism through which landownership affects state education is political inequality rather than economic inequality or the lack of demand for education. Keywords: Education, land concentration, taxation, political elites, English economic history. JEL Classification Numbers: I25, O43, Q15, N33.

1

Introduction

Inequality can be harmful for economic growth (Galor and Zeira 1993). One reason is that growth-promoting institutions such as state education may be difficult to implement where wealth is concentrated in the hands of a small elite (Galor and Moav 2006). America’s reversal of fortune often serves as an example: countries in South America had an unequal distribution of land, limited the suffrage, underinvested in ∗ I thank the suggestions made by Ran Abramitzky, Sasha Becker, Francesco Cinnirella, Gregory Clark, Oded Galor, Daniel Garcia, Albrecht Glitz, Alfonso Herranz-Loncan, Eeva Mauring, David Mitch, Omer Moav, Joel Mokyr, Giacomo Ponzetto, Juha Tolvanen, Felipe Valencia, Hans-Joachim Voth, and seminar participants at the All-UC Economic History meeting (Chapman), SAEe (Mallorca), Warwick, FRESH (Barcelona), Louvain-la-Neuve, University of Vienna, and GREQAM. † Department of Economics, University of Vienna

state education, and despite being rich in the past, by the nineteenth century fell behind the United States.1 Similarly, nineteenth century Europe experienced a reversal of fortune when Prussia overtook England as the world’s industrial leader. In contrast to America’s case, however, inequality is not in the spotlight as an explanation for Europe’s reversal of fortune (McCloskey and Sandberg 1971). In fact, the view on landownership concentration in Europe is often positive, as it is traditionally associated with the agricultural revolution in England, and hence, with the Industrial Revolution (Allen 1992). Are the negative effects of land inequality restricted to agrarian societies with limited suffrage? This paper studies the introduction of state education in England, and shows that land inequality had perverse effects on human capital formation also in an industrialized economy that considerably extended the franchise. Examining the effect of land inequality on human capital is not straightforward. First, land inequality is potentially correlated with unobservables like geography or culture, which can crucially drive education provision (Easterly 2007). Second, it is difficult to pinpoint the causal channels through which landownership concentration undermines human capital. One possibility is that in less equal societies the poor do not invest in human capital (Banerjee and Newman 1993; Galor and Zeira 1993). Economic inequality may also affect the supply of education (Murphy, Shleifer, and Vishny 1989). On the other hand, the ownership of land is linked to political power. For example, in England the landed elite monpolized public offices (Allen 2009). Hence, land inequality not only reflects economic inequality but also political inequality; in the sense that decisions are taken by a small elite (Acemoglu, Bautista, and Robinson 2008). The landed elite, in turn, may oppose education supply if there is no complementarity between human capital and agrarian work and to reduce the mobility of the rural labor force (Galor and Moav 2006; Galor, Moav, and Vollrath 2009). Finally, where land is more concentrated the private demand for education by peasants can be lower. The demand for education is especially important where elites extended the franchise2 or after the emancipation of serfs (Cinnirella and Hornung 2016; Ashraf et al. 2017). These distinct causal channels have critically different implications when evaluating available policy instruments; for example, the implications of land reform depend on whether landownership affects human capital through economic inequality, through the political process, or through the private demand for 1

Engerman and Sokoloff 2000, Coastworth 1993, Nugent and Robinson 2010, Vollrath 2009. Acemoglu and Robinson (2000), Engerman and Sokoloff (2000), Mariscal and Sokoloff (2000), Gallego (2010), and Go and Lindert (2010). 2

2

education. However, disentangling these mechanisms is difficult. Political and economic inequality are typically correlated and data is often limited to enrollment or literacy rates, measures related both to the supply of and to the demand for education.3 I address these limitations by studying an economy with a well-defined political elite and digitizing high-quality data on education supply and demand. This allows me to disentangle these different mechanisms. This paper shows that landownership concentration had a negative effect on state education in late-nineteenth century England, and that this was the result of the political opposition of landed elites rather than economic inequality or the lack of demand for education. Using a new database on 1,387 School Boards4 and 32 counties between 1871 and 1899, I find that landownership concentration had a strong, negative effect on various dimensions of education. School Boards next to the largest landlords in England set lower property taxes to fund education. As a consequence, where landownership was more concentrated the ratio of state to private schools, teacher’s salaries and expenditures per pupil were lower. Also, fewer teachers and class assistants were hired. These had important consequences for human capital accumulation: children were less likely to pass the reading, writing, and arithmetics’ national exams. To address endogeneity concerns in the relationship between landownership concentration and education, I use an instrumental variables approach exploiting two sources of exogenous variation: one novel and another well-established. The first is a large ‘natural experiment’ that massively redistributed landownership in England: the Norman conquest of 1066. Anglo-Saxon England had been a mosaic of landowners. After the conquest, William took half of the land for himself, gave a quarter to the Church, and divided the rest among 190 Norman nobles—the ancestors of the large landowners in the nineteenth century. Using geo-referenced data from the Domesday Book, I show that there is a strong persistence in land inequality over eight centuries. This finding is interesting in its own right, as it emphasizes the “deep roots” of inequality (Clark et al. 2014). Furthermore, this finding is related to a literature that shows that medieval institutions have persistent economic and political consequences (Angelucci, Meraglia, and Voigtlnder 2017).5 Second, I also use exogenous variation in land inequality from differences in soil texture, which is a well-established instrument in the prior literature (Cinnirella and Hornung 2016; Easterly 2007). Overall, 3

See, for example, Easterly (2007), Acemoglu, Bautista, and Robinson (2008), Gallego (2010), Clark and Gray (2014), or Tapia and Martinez-Galarraga (2015) 4 Geographically, a School Board is equivalent to a district or a borough. 5 Specifically, they show that granting a Charter of Liberties to a borough before 1348 led to inclusive political institutions in the long run.

3

my results strongly suggest that the effect of landownership concentration on state education is causal. I find that reducing land inequality by one standard deviation would, for example, increase property tax rates by 37.5 percent, expenditures per pupil by one third of a standard deviation (16 pence), and the percentage of children passing the national arithmetics exam by half a standard deviation (2.6 points). Next, I examine the mechanisms through which landownership concentration undermined state education in England. State education may be hampered by political or economic inequality (Acemoglu, Bautista, and Robinson 2008). Disentangling these two channels is a major empirical challenge, as economic and political inequality are usually correlated. To address this, I take advantage of the fact that peers retained political influence in the late-nineteenth century, especially at the local level (Allen 2009). Specifically, I code biographical information for 369 “large landlords” (i.e., peers who owned 2,000 acres or more) and match it to local data on state education from 1,387 School Boards. I find that School Boards functioned normally in areas where wealth was concentrated in the hands of landowners who were not politically relevant. In contrast, School Boards exposed to peers who were MPs or appointed to the most important local offices (e.g., Lord Lieutenant) systematically raised less funds for education. In other words, considerable political power is required for transforming economic inequality into unequal education provision.6 Another important distiction is whether landownership concentration reduces the supply of education (Galor, Moav, and Vollrath 2009) or is associated with a low private demand for education. Elsewhere it has been argued that the private demand for education is the main driver of state education, especially where the elite extended the franchise7 or after the emancipation of serfs (Cinnirella and Hornung 2016; Ashraf et al. 2017). I show that this was not the case in England, even though by 1870 two in three males had the vote. To disentangle the supply and demand side, I construct a rich dataset to evaluate measures of education supply (e.g., number of state schools) and of demand for education (e.g., number of pupils attending). Using cross-sectional variation across 32 counties, I find that where land is concentrated in few hands education-supply is lower, but the demand for education is not. In other words, even if the elite extended the franchise, landowners still retained de facto political power, opposed education supply, and hence, reduced human capital accumulation. In contrast, the supply of education is positively associated to the share 6

This result is not specific to my historical setting. Similarly, Alatas et al. (2013) found that local leaders in modern-day Indonesia captured welfare programs only where they held formal positions. 7 Acemoglu and Robinson (2000), Engerman and Sokoloff (2000), Mariscal and Sokoloff (2000), Gallego (2010), and Go and Lindert (2010).

4

of manufacturing workers. This finding suggests that old landed elites and emerging industrialists clashed over the provision of state education, as proposed by Lindert (2004) and Galor and Moav (2006). The empirical setting I examine offers a number of advantages. First, the purpose of introducing state education in England was to provide an educated workforce to the industry. This is in sharp contrast to other contemporary education reforms in Prussia (Cinnirella and Schueler 2016), Austria (Cvrcek and Zajicek 2016), and the United States (Bandiera et al. 2015). These reforms were concerned not only with human capital but also with nation building, something that might be in the interest of landowners. Therefore, England provides a better setting to evaluate whether land inequality is detrimental for human capital acumulation. Second, the data on state education is very rich. The reports of the Committee of Council on Education detail the funds raised for education, the expenditures, and the examination results for all the School Boards operating in 1870–99. This allows me to exploit crosssectional variation in state education across 32 counties as well as local data from 1,387 School Boards. Furthermore, I can evaluate many outcomes, not only literacy and enrollment rates as is typically done in the literature. Particularly, I evaluate measures of education supply (e.g., schools built) and of demand for education (e.g., attendance) separately. And third, the introduction of state education in England was decentralized, which allows for plausible identification using cross-sectional variation. In detail, School Boards were created in the districts and boroughs where there was a shortfall in education. Each Board could: raise funds from a rate (i.e., a property tax), build and run state schools, subsidize private schools, and create by-laws making attendance compulsory. Local landed elites, unwilling to subsidize state education with taxes on their wealth, took over School Boards. The takeover was facilitated by the local nature of School Boards and by the election system, which practically guaranteed landed elites representation in the Board (Stephens 1998). To sum up, my contribution to the literature is threefold. First, I show that the negative effects of landownership concentration on state education are not restricted to developing societies, with limited suffrage and a feeble industrial sector. A frontier economy like late-nineteenth century England also suffered the perverse effects of landownership concentration. This suggests that landownership concentration, by affecting human capital formation, is important for the economic and demographic changes that began after the Industrial Revolution in England (Galor and Moav 2006). Furthermore, land inequality is one reason why state education lagged behind Prussia and the United States, nations that eventually overtook England as the world’s 5

industrial leaders (McCloskey and Sandberg 1971). Second, this paper shows that landownership concentration adversely affected human capital through the political process. Earlier, the literature had argued that the negative effects of land inequality on human capital were a byproduct of economic inequality (Engerman and Sokoloff 2000), limited suffrage, or the lack of private demand for education.8 I show that these explanations played a minor role in England. Third, I identify causal effects in a novel manner, documenting a strong persistence in land inequality from the Norman conquest of England (1066) to the late-nineteenth century. The paper proceeds as follows. Section 2 reviews the literature. Section 3 describes School Boards and the expansion of state education in England. Section 4 describes the data. Section 5 estimates the effect of land concentration on state education. Section 6 examines the mechanisms behind this relation. Section 7 concludes.

2

Related literature

This paper is related to a vast literature on the long-run consequences of inequality for human capital and growth. In a seminal contribution, Engerman and Sokoloff (2000) suggest that the reversal of fortune in America steams from geographically driven differences in inequality. The United States, with a more egalitarian land distribution, extended the franchise and introduced state education already by the nineteenth century. As a concequence, they overtook the Caribbean and Central American economies, which had been the richest by the 1500s.9 Many empirical studies have found a strong relation between landownership concentration and state education in the Americas (Coastworth 1993; Engerman and Sokoloff 2000; Nugent and Robinson 2010; Easterly 2007)10 and in similar pre-industrial economies (Tapia and Martinez-Galarraga 2015). Is the distribution of land important for human capital also in industrialized economies with extended suffrage? Galor, Moav, and Vollrath (2009), Vollrath (2009) and Ramcharan (2010) analyze the US high-school movement and find that land inequality reduced educational expenditures. In Europe, Cinnirella and Hornung (2016) show that landownership concentration reduced enrollment rates in state schools in 8

Acemoglu and Robinson (2000), Mariscal and Sokoloff (2000), Gallego (2010), Go and Lindert (2010), and Cinnirella and Hornung (2016). 9 While the reversal of fortune at the national level is a stylized fact in the literature, Maloney and Valencia Caicedo (2016) show that there is persistence at the sub-national level. 10 Admittedly, land inequality is not the sole driver of state education in the Americas. Dell (2010) finds a positive association between land inequality and literacy rates in Peru’s mining Mita, and argues that large landowners can ensure public goods in the presence of a highly extractive state.

6

Prussia, particularly before the abolition of serfdom. The advantage of my case study is that the purpose of introducing state education in England was to provide an educated workforce to the industry (Hurt 1971). As opposed to Prussia (Cinnirella and Schueler 2016), Austria (Cvrcek and Zajicek 2016), or the United States (Bandiera et al. 2015), education reforms in England had nothing to do with “nation building,” something that might be in the interest of landowners. Studying the introduction of state education in England is important in its own right, as human capital is at the spotlight of the changes initiated there after the Industrial Revolution. In detail, as the Industrial Revolution progressed to its second phase (1870–1914), average human capital contributed to technological progress and to the demographic transition (Galor and Weil 2000; Galor and Moav 2002).11 Since state education played a crucial role in raising human capital, “examining the history of its introduction becomes crucial” (Mokyr and Voth 2009). Few have addressed this question. Galor and Moav (2006) develop a theoretical model where, due to complementarities between human and physical capital, capitalists have a higher incentive to support and subsidize education than landowners. The empirical evidence, however, is elusive. While Galor and Moav (2006) find that MPs from more skilled-intensive districts supported the 1902 Education reform, Clark and Gray (2014) show that literacy rates between 1815–45 varied across England due to culture, not to landownership. At that time, however, the elementary system was not state-sponsored but based on Voluntary schools, ran by the church, and funded with donations and bequests (Mitch 1992: p. 115). In this context, landowners subsidized education because of religious motivations, or because rivalry and emulation among their fellows (Hurt 1968; Thompson 1963). By examining the introduction of state-funded education in 1870, this paper is the first to show that landownership concentration distorted the introduction of state education in England—an industrialized, frontier economy. This finding shows that the distribution of land is an important factor in explaining human capital and, hence, the transition to sustained economic growth. My second contribution is to examine why landownership concentration affects human capital. On the one hand, the negative effects of landownership concentration can be driven by economic inequality. Under imperfect capital markets, economic 11

There is debate on the importance of average human capital for the First Industrial Revolution (1760-1840). While Mokyr (2005) and Squicciarini and Voigtlander (2015) emphasize upper-tail human capital, Franck and Galor (2017) link early industrialization with the demand for average human capital. In contrast, the central role of human capital for the Second Industrial Revolution (1870-1914) and the demographic transition has been established both theoretically (Kogel and Prskawetz 2001; Jones 2001; Galor and Moav 2002; Hansen and Prescott 2002) and quantitatively (Doepke and Zilibotti 2005; Fernandez-Villaverde 2007; Lagerlof 2006; Cinnirella and Streb 2017).

7

inequality can generate a poverty trap in which a large share of the population does not invest in human capital (Banerjee and Newman 1993; Galor and Zeira 1993). Economic inequality may also distort the composition of aggregate demand and affect the supply of state education (Murphy, Shleifer, and Vishny 1989). On the other hand, the ownership of land is linked to political power. Therefore, the negative effects of landownership may reflect the opposition of entrenched, landed elites, to state education. Landed elites oppose education supply if there is no complementarity between human capital and agrarian work and to reduce the mobility of the rural labor force (Galor and Moav 2006; Galor, Moav, and Vollrath 2009).12 This political channel has been emphasized by Acemoglu, Bautista, and Robinson (2008), Ramcharan (2010), and Vollrath (2013). Finally, it may also be that where land is more concentrated the demand for education by the masses is lower. Acemoglu and Robinson (2000), Engerman and Sokoloff (2000), Mariscal and Sokoloff (2000), and Gallego (2010) suggest that the demand for education manifested in Latin America after the extension of the franchise. Go and Lindert (2010) show that the extension of the franchise in the US North led to higher enrollment rates. Cinnirella and Hornung (2016) and Ashraf et al. (2017) show that the emancipation of serfs in Prussia spurred the demand for state education. More generally, many have argued that democratization can constrain elite capture in developing economies (Foster and R. Rosenzweig 2001). My paper compares, in a unified framework, these three mechanisms proposed by the literature. First, I exploit the fact that the political elite in England was was welldefined (Allen 2009) to disentangle the political channel from economic inequality. In detail, I find that School Boards functioned relatively normally in areas with high economic inequality but where landowners were not politically relevant. In contrast, School Boards exposed to landlords who were MPs or who were appointed to important local positions, raised less funds for education. In other words, inequality affected the supply of education especially where the landed elite was sufficiently politically influential to take over School Boards. Similarly, Alatas et al. (2013) find that local leaders in modern-day Indonesia captured welfare programs only where they held formal positions. Second, I construct a rich dataset with measures related to education supply (e.g., schools built) and the private demand for education (e.g., attendance 12 The lack of complementarity between between human capital and agrarian work can also explain the opposition of landed elites to child labor laws (Doepke and Zilibotti 2005) or public services Lizzeri and Persico (2004). Similarly, Ashraf et al. (2017) show that the skill-labor complementarity was crucial for the abolition of serfdom in Prussia. Landed elites emancipated serfs to incentivize their acquisition of human capital where they owned substantial physical capial, that is, where labor and human capital were complementary.

8

rates). In detail, I computerize a source that, to the extent of my knowledge, remains unexplored by economists: the reports of the Committee of Council on Education. My results show that landownership concentration distorted education supply but did not affect the private demand for education. In England, therefore, the private demand for education was not the main driver of state education even though the franchise had been considerably extended. This paper also moves the literature forward by addressing endogeneity concerns in the relationship between landonwership concentration and state education. I use two sources of exogenous variation in landownership: one novel and another wellestablished. The first is the massive redistribution of land after the Norman conquest of England in 1066. Using data from the Domesday Book, I document a strong persistence in landownership over eight centuries, a finding that is interesting in its own right. This result shed new light to a literature that argues that inequality is highly persistent at the top of the wealth distribution (Clark et al. 2014). Similarly, this finding is linked to a literature that shows that medieval institutions can have long-term economic consequences. Most relevant for this paper, (Angelucci, Meraglia, and Voigtlnder 2017) argue that granting a Charter of Liberties to a borough before 1348 led to inclusive political institutions over centuries. Second, this paper also exploits exogenous variation in landownership coming from soil texture. The use of the geological composition of the soil as an instrument goes back to Easterly (2007) and Vollrath (2009), who exploited crop suitability. Here I follow Cinnirella and Hornung (2016) in using soil texture. My first-stage findings provide new evidence to an important hypothesis in agricultural economics; i.e., that poorer soils are associated with larger landownership concentration.13 Finally, this paper is related to the literature studying Britain’s loss of industrial leadership (McCloskey and Sandberg 1971; Coastworth 2004; Nicholas 2004). The classic explanation is that entrepreneurs overinvested in old staple industries and did not understand the growing importance of science (Landes 1960; Aldcroft 1964; Saul 1968). More recently, Britain’s loss of industrial leadership has been related to institutions (Elbaum and Lazonick 1984), a rigid class structure (Weiner 1981), the gentrification of capitalists (Thompson 1994), and Anglicanism (Berghoff 1990). Few have considered education as a possible explanation. Allen (1979) suggests that Public schools (e.g., Eton) excluded science and technology from the curriculum. It is not clear, however, that the German schools were more conductive to industrial progress (Pollard 1989; Berghoff and Moller 1994; Cassis 1997). What is clear is that 13

Bhalla (1988), Bhalla and Roy (1988), Benjamin (1995), Barrett, Bellemare, and Hou (2010).

9

state-sponsored schools had been at place for half a century longer than in England (Sanderson 1995). My results suggest that the British aristocracy weakened the industrial spirit not (only) by encouraging gentrification of emerging capitalists, but (also) by depriving the masses of education. After years of blaming the entrepreneurs, my findings suggest to turn the attention to land rentiers. Relative to the existing literature, I make the following contributions: First, this paper is the first to show that landownership concentration distorted state education in England—a frontier economy with a strong industrial sector. Second, I examine the mechanisms behind the relationship between landownership and state education. I construct a new dataset and I exploit the unique features of my case study to disentangle the three mechanisms highlighted in the literature: political inequality, economic inequality, and the private demand for education. My results show that land inequality distorted education supply through the political process. Pure economic inequality and the private demand for education played a minor role in England, despite the fact that the franchise had been considerably extended. Third, this paper addresses endogeneity concerns in a novel manner, documenting a strong persistence in landownership from the Norman conquest in 1066 to the late-nineteenth century. Finally, my results suggest that entrenched landed elites were at least as responsible as failed entrepreneurs for Britain’s loss of industrial leadership.

3

Historical background

In the second phase of the Industrial Revolution (1870–1914), the demand for skilled labor increased. For example, job advertisements mentioned literacy as a desired characteristic as of the 1850s (Mitch 1993: 292). At that time, however, England lacked an education system that could meet this demand. In 1851, thirty percent of the adult population could not read or write. Contemporaries like Alonzo Potter claimed that “England has neglected the education of her laboring population, and the consequence is that the land swarms with paupers and vagabonds” (Potter and Emerson 1989: 116). In fact, state education in England lagged behind west Europe and the United States by fifty years (Sanderson 1995). For example, while Prussia introduced state education in the eighteenth century, the first state schools were not built in England until 1870. Before 1870, the system was based on Voluntary schools run by the Church and fee-charging Public Schools (Green 1990). This system was damned, especially after the 1867 Interntional Exposition in Paris. England only won ten prizes out of 90. According to Lyon Playfair, member of the jury, England 10

fell behind because “France, Prussia, Austria, Belgium and Switzerland possess good systems of industrial education and England possesses none” (Green 1990: 296). The introduction of state education in the 1870s14 was the response to the demands for an educated workforce. William Edward Forster, the cabinet member in charge of the Education Act of 1870, told the House of Commons: Upon the speedy provision of elementary education depends our industrial prosperity ... if we leave our work-folk any longer unskilled ... they will become overmatched in the competition of the world. Hurt (1971): 223–4 In contrast to other contemporary reforms in Prussia (Cinnirella and Schueler 2016), Austria (Cvrcek and Zajicek 2016), or the United States (Bandiera et al. 2015), the purpose of Education Act of 1870 was not nation-building, something that might be in the interest of landowners.15 Specifically, the Education Act of 1870 declared that the ratepayers of each Poor Law Union (i.e., district) or borough could petition the creation of a School Board. Board powers included: raising funds from a rate (i.e., a property tax), build and run state schools, subsidize Voluntary schools, pay the fees of the poorest children, and create by-laws making attendance compulsory (Stephens 1998). School Boards were also eligible for grants from the central Committee of Education on the basis of their performance in the national reading, writing, and arithmetics exams (Green 1990: p.7). Finally, School Boards also charged school fees and sold books to children. The election system of Board members suggests that landowners could take over School Boards. First, Board members were elected by ratepayers; i.e., those paying an annual rent of £10 or holding land valued at £10.16 In addition, the voting system was cumulative voting. Each voter could choose three (or more) Board members from a list of candidates. This ensured landed elites some representation on the Board (Stephens 1998). According to Thompson (1963), the opposition of landed elites: proceeded a little sporadically and lazily until galvanized into a sudden fury of action the 1870 Education Act. Then [...] there was a flurry of activity to ward off the dread intrusion of a School Board. (p. 208). Between 1870–1902, School Boards created 5,700 state schools, providing education for 2.6 million pupils (Stephens 1998). Several Education Acts enforced and 14

A similar act was passed in 1872 for Scotland. In opposition to England, attendance was compulsory from the start and churches handed over their schools without charge (Tod 1873). 15 Addmittedly, landowners were not the only to object universal education. Some industrialists feared that education would spur revolution by the laboring classes (Stephens 1998). 16 The franchise was different from national elections; e.g., women could vote and stand for office.

11

extended the principals of the Education Act of 1870 (see Table A.1 in the Appendix). For example, attendance of children under 11-12 became compulsory in 1880. School Boards were finally abolished by the Balfour Education Act of 1902.

4

Data

I use five data sources, three of which are newly computerized. To study state education, I computerize the reports of the Committee of Council on Education. Next, I construct measures of landownership concentration from Bateman (1883), a list of all owners of 3,000 acres or more. I complement this evidence with the country seats of peers from Burke (1826) and with biographical information from thepeerage.com. Finally, I use evidence from the Domesday Book computerized by Palmer (2010) and the British Geological Survey (2014) on soil texture in modern-England.

4.1

Reports from the Committee of Council on Education

To study state education, I computerize the reports of the Committee of Council on Education. These annual reports stand as “the most significant single source in existence for the study of elementary education, particularly on State interest in education” (Stephens 1997). As the data is broken down by counties and districts, this source can be used for analysis at the local level. Figure 1 shows parts of the 1883-84 report for School Boards in Berkshire. The three main sources of income were grants from the Committee (col. 1), funds raised from property taxes (col. 2), and school fees (col. 3). The reports also detail expenditures: investements in state schools, the number of state schools, teachers hired, etc. Finally, the reports assess whether these investments actually helped the children to learn something. From 1879–95, the reports state the percentage of children passing the reading, writing, and arithmetic national exams. Committee grants were sometimes given as a function of exam results. This was known as Payment by Results. Admittedly, there was an incentive to limit education to the three Rs (reading, writing, and arithmetics) which may explain the high success rates (Green 1990: 7). The reports not only provide information on the supply of education (e.g., number of state schools). They also list measures reflecting the demand for education, such as the number of pupils in attendance and examinees. The number of examinees is broken down by age and by standards, from simple additions to mastering fractions.17 17

See Table A.2 in the appendix for a detailed description of the standards.

12

[FIGURE 1 HERE] I computerized the information on funding, expenditures, and outcomes for all counties in England between 1871 and 1895;18 and the information on funding for all School Boards between 1873 and 1878. In the empirical analysis, I will analyze the data at two levels: I will exploit (i) cross-sectional variation across 32 counties and (ii) local data from 1,387 School Boards.

4.2

Data sources on landownership

Data on landownership in late-nineteenth century England is available in Bateman (1883). The book lists of all owners of 3,000 acres or more by 1876, worth £3,000 a year. It also includes 1,300 owners of 2,000 acres or more. Furthermore, the book provides a table for each county showing the number of landowners and cumulative acreage, all divided into eight classes: peers or peers’ sons (3,000 acres or more), commoner great landowners (3,000 acres or more), squires (1,000 to 3,000 acres), greater yeomen (300 to 1,000 acres), lesser yeomen (100 to 300 acres), small proprietors (one to 100 acres), cottagers (less than one acre), and public bodies.19 I define landownership concentration in two ways: For the analysis using crosscounty variation, I define landownership concentration as the percentage of land in each county in the hands of large landowners (i.e., owners of 3,000 acres or more). Given that political power was linked to the ownership of land (Allen 2009), this measure captures political inequality better than alternative measures such as the Gini index. Moreover, the Gini index measures between-group and within-group inequality, while my measure considers only the former. This is imporatant because within-group inequality does not necessarily affect redistributive policies like state education. For the analysis using local data from 1,387 School Boards, I define land inequality as the acreage of large landlords (i.e., peers who owned 2,000 acres or more). Specifically, I computerize individual-level information on acreage, annual land rents, and the family seat for all 487 peers listed in Bateman (1883) (for the sake of illustration, Figure A.1 in the Appendix shows the entry for Lord Lyttelton). I compliment this dataset with information from two sources: First, I include 265 18

Unfortunately, some variables are not available in all reports. See Table A.3 for details. Bateman (1883) is based in the Return of Owners of Land ; the first complete survey of landownership in Britain since the Domesday Book. Bateman corrected many inaccuracies (Bateman 1883: preface). Several peers, however, complained that the book underestimated their possessions (Bateman 1883: 348). Bateman’s estimates, therefore, should be taken as a lower bound. 19

13

additional family seats listed in Burke (1826). Adding this is important as Bateman (1883) lists only one seat per lord, while many peers owned several houses. Second, I assess the political power of these landowners. I scrape information on the public offices held by peers from thepeerage.com. This website provides a short biography for members of the peerage. Figure A.2 in the appendix shows the biography of Henry Herbert, 4th Earl Carnavon: he was Secretary of State for the Colonies, Privy Counsellor, Deputy Lieutenant of Nottinghamshire, and LordLieutenant of Ireland and Hampshire—where his principal estates were located. Using regular expressions, I identify the public offices held by 369 peers in possession of 2,000 acres or more. Table A.4 in the appendix summarizes the appointments considered. Finally, note that Bateman (1883) is a cross-sectional survey, so I do not exploit time variation in land concentration. However, Britain’s landownership was remarkably stable from about 1750 (Beckett 1977: 567). This justifies a multi-level approach were education variables and other covariates vary both in the cross-section and over time but landownership concentration only varies in the cross-section.

4.3

Domesday Book

To address endogeneity concerns in the relation between land inequality and state education, I exploit the redistribution of land after the Norman conquest of England. I reconstruct the distribution of land after 1066 using the Domesday Book, a survey of all English landholdings commissioned by William the Conqueror. It is the oldest public record in England; no survey approaching the extent of Domesday was attempted until 1873. It names 13,418 places covering most of the territory, except for lands in the north and tax-exempt cities like London and Winchester. All the information in the Domesday Book is available in electronic format in Palmer (2010).20 The database includes an entry for each fief (i.e., manor) stating the landlord before and after the conquest, the value of the land, the population, and other resources (e.g., number of ploughs or the presence of a mill). Section 5.3.1 and Appendix B describe in detail the data and the construction of the instrument.

4.4

British Geological Survey

To address endogeneity concerns, I also exploit geographical variation in soil texture. Soil texture does not change over time and cannot be altered by human intervention. 20

The digitization of the Domesday Book was undertaken by typists in an ESRC project, Dr Natasha Hodgson, Dr and Mrs. Thorn, and Phillimore and Co Ltd. See Palmer (2010) for details.

14

Under this premise, I use modern-day data from the British Geological Survey. The data is a vector grid with 1km by 1km cell dimensions. Soils are classified in 40 categories according to the relative proportions of sand, silt, and clay. The classification also takes into account the presence of chalk and peat fragments. Section 5.3.2 and Appendix B describe the different soil types and the construction of the instrument.

4.5

Data descriptives

Table 1 provides summary statistics of the main variables used in the analysis. Investments in education were not large. The average county only raised 92.9 pence per child every year for education. Most of these funds come from property taxes, although the average School Board set a tax rate of only 2.5 percent. Funds from property taxes also present larger standard errors than grants, fees, and books sold. In other words, not all counties were willing to fund state education with taxes on property. For example, in 1894–95, Essex raised 87 pence per child from property taxes while Rutland only raised 12 pence per child. School Boards spent most of their funds in state schools. That said, in the average county there were only 35 state schools for every 100 Voluntary schools (i.e., private schools) and only around a thousand were employed as teachers and class assistants.21 On average, £22,689 was spent in teachers’ salaries. The percentage of children passing the national exams was very high. This reflects the fact that some grants were allocated as a function of exam results. However, there is meaningful variation across counties, especially in arithmetics. My dataset also includes measures that reflect the private demand for education. The average county had 76,000 pupils in attendance. Although education was made compulsory for children aged 5–11 in 1880, half of the children presented for examination were above 10. Many of those took high level exams (standards IV to VII). Overall, my dataset on state education allows me to evaluate many outcomes beyond literacy and enrollment rates, the measures traditionally used in the literature. Panel B presents the variables related to landownership. It is hard to exaggerate the extent to which landownership was concentrated in England. In the average county, 41 percent of the land was owned by large landowners (i.e., owners of 3,000 acres or more). Most of them were peers. On average, peers owned estates of 7,840 acres in the counties where their seats were built. Peers also controlled public offices. Fourty percent sat on the Parliament and 75 percent were appointed to the most 21

Class assistants are female class assistants in the reports from the Committee.

15

important local offices: Lord Lieutenant, Deputy Lieutenant, High Sheriff, or Sheriff. [TABLE 1 HERE] Figure 2 shows the geographical distribution of landownership concentration and education funds from property taxes. Information is presented both at the county level and using local data from 487 lord’s seats and 1,387 School Boards. Landownership was heavily concentrated in the West Midlands region: Fourty to 60 percent of the land was owned by large landowners (panel (a)). The largest estates, denoted by larger circles in panel (b), were mostly north of Cambridge. In Rutland, for example, more than 30 percent of the land was owned by the Duke of Rutland! In contrast, in the South-East and in the North-West—the cradle of the Industrial Revolution—such large estates were rare. There, the distribution of land was relatively more equal. The spatial distribution of education funds from property taxes is diametrical to that of landownership. Education funds were larger in the counties in the North-West and, especially, the South-East, where landownership was less concentrated (panel (c)). Similarly, panel (d) shows that most of the School Boards imposing the largest tax rates (i.e., above 7 percent) were in the South-East. In contrast, very few School Board set taxes above 4.5 percent in the West Midlands. For example, in Rutland there was only one School Board: Essendine. Between 1873 and 1878, the average tax rate in Essendine was 0.57 percent. [FIGURE 2 HERE]

5

Land concentration and state education

Did landownership concentration distort state education in England? To answer this question, I examine the data at two levels: First, I exploit cross-county variation in landownership and a wide range of educational measures between 1871 and 1899. Second, I use local data from 1,387 School Boards (equivalent to 1,387 districts and boroughs). In detail, I match each School Board to the landlords living nearby and evaluate whether the size of their landholdings affected tax rates. Finally, I estimate an instrumental variables model using the Norman conquest in 1066 and soil texture as instruments.

5.1

OLS results

Here I examine the relation between land inequality and a broad range of education measures by exploiting cross-county variation. I construct a panel of 32 counties and 3 decades (1870s, 1880s, and 1890s) in which the measures of state education and the 16

county controls vary over time but landownership only varies in the cross-section.22 This multi-level approach is justified because landownership in Britain was stable from 1750 onward (Beckett 1977: 567). Formally, I estimate an OLS model educ,t = α + β landc + V0c,t δ + c,t ,

(1)

where educ,t is an education measure in county c at decade t and landc is landownership concentration. In detail, it is the share of county c in the hands of large landowners (i.e., owners of 3,000 acres or more). The coefficient β captures the association between land concentration and state education, which I expect to be negative. Vc,t is a vector of covariates including the share of employment in manufacturing, income, ideology,23 the percentage of non-conformists, and religiosity. Since landc only varies in the cross-section, I cluster standard errors by county. Table 2, panel A shows the OLS estimates for education funding. Counties in which land was more concentrated raised less funds for education. Every percentage point increase in land inequality is associated with 2 fewer pence per child raised from property taxes. Grants from the Committee of Education and funds raised from books and fees were also scarcer where land was more concentrated.24 Counties raising less funds consequently under-provided education. Panel B shows that where land was more concentrated, fewer School Boards were established and the system relied on existing Voluntary schools (i.e., private schools) rather than on state schools. Increasing land inequality by one percentage point is associated to a decrease by 0.59 in the ratio of state to private schools. Expenditures in these schools and expenditure per pupil were also lower. Land inequality also affected the number of teachers hired: every percentage point increase is associated with 12 fewer teachers and a reduction of almost 100 pence in their salaries. Finally, land inequality is strongly negatively associated to the number of classroom assistants. This suggests that the opposition of landowners to state education was more effective over discretionary expenditures. Panel C shows that land inequality had important consequences for human capital accumulation: Children in counties where land was concentrated were less likely to 22

For consistency, I only include the 32 counties fully surveyed in the Domesday book. Similarly, to measure state education I use decade averages rather than annual values because the relevant county-level covariates only vary by decade. 23 Ideology is measured by the percentage voting conservative. Hechter (1976) does not provide the evidence for the 1871-80 decade, so I take the values of the following decade. 24 Other covariates have the expected signs. Income and the percentage of non-conformists are positively associated with state education, as suggested by Galor and Moav (2006). In contrast, the coefficient for religiosity is negative.

17

pass the reading, writing and, especially, the arithmetics’ national exam. Every 10 percentage points increase in land concentration is associated to a 14-points drop in the probability of passing arithmetics’. Given the high success rates, this marginal effect is considerable. Finally, more industrial counties raised more funds, had higher expenditures, and higher success rates in the national exams. The fact that landownership concentration and manufacturing employment have opposite effects is consistent with two theories: on the one hand, it could reflect a clash between landed and industrial elites for the provision of state education (Lindert 2004; Galor and Moav 2006). On the other hand, it could be a byproduct of a higher demand for education in more industrial aeras. In Section 6 I disentangle these two views. [TABLE 2 HERE]

5.2

Local data

In this section I examine the relation between landownership and state education using local data. I consider 1,387 School Boards operating in 1873–78 and 487 seats of large landlords (i.e., peers who owned 2,000 acres more). In detail, I draw a 25mile radius around each peer’s seat and identify all the School Boards in it. If land inequality is negatively associated to state education, I expect School Boards next to larger landowners to raise less funds for education. To illustrate my strategy consider Washingborough, a village in Lincolnshire (Figure 3). Washingborough is only 2.7 miles away from Burton House, the magnificent country house of William Monson, 1st Viscount Oxenbridge. He was a large landlord: he owned 8,100 acres in Lincolnshire. How did the School Board in Washingborough fare with so much wealth to levy taxes on? Between 1873 and 1878, they only taxed 0.8 percent of the rateable property. This was no exception. Figure 3 shows that School Boards within ten miles from Blankney Hall (e.g., Ingham, North Scarle, and Stow by Gainsborough) set tax rates below 1 percent. In contrast, only School Boards 18 miles further away from William Monson’s seat (e.g., West Stickwith, Hibaldstow, Belchford, and Stickford) managed to impose tax rates above 4 percent. [FIGURE 3 HERE] Formally, I regress the acreage of a large landlord on the property tax rate set by each School Board in a 25-mile radius: edub,s = α + β lord acreages,c + V0c δ + b,s,c ,

18

(2)

where edub,s is the average tax rate in 1873–78 set by School Board b, which is located in a 25-mile radius of the seat s. The variable lord acreages,c is the acreage by the large landlord living in seat s. I consider his acreage only in the county c where seat s is located rather than his total acreage, which may include estates scattered all over Britain and Ireland. The coefficient β captures the association between a landlord’s acreage and the education provided by the nearby School Boards, which I expect to be negative. Note that, in the baseline specification, I treat each School Board and landlord pair (b, s) as an independent observation.25 Hence, I cluster the standard errors by landlord’s seat s. To control for potential confounders I include covariates at the School Board level (e.g., distance to an industrial center) and county-level controls from Hechter (1976). Finally, note that this specification resembles a gravity equation where the influence that a landlord has over a School Board depends on his wealth and on the distance between his seat and the School Board. Table 3 presents the results. In the baseline specification, I find a negative relationship between the acreage of a landlord and the taxes set by School Boards nearby. Increasing the acreage of a landlord by one standard deviation (i.e., by 9,799 acres) would reduce tax rates by 0.1 percentage points. Results are robust to the inclusion of covariates that are potentially correlated with state education. Column 2 includes the set of Hechter (1976) county-level controls: employment in manufacturing, income per capita, ideology, the percentage of nonconformists, and religiosity. Column 4 considers two covariates at the School Board level: the distance to an industrial center and to the closest cathedral.26 The former captures the private demand for education of the working population, as the benefits from schooling were larger for families who could easily migrate to an industrial center. For a detailed analysis of these covariates, see Section 6. Finally, columns 3-5 examine the robustness of the results to alternative specifications. I relax the assumption that each School Board and landlord pair (b, s) is an independent observation in two ways. First, I collapse the data at the seat level. That is, I regress each lord’s acreage on the average tax rate raised within 25 miles from 25

For example, the School Board in Brotton, Yorkshire is within 25 miles of three different seats: Duncombe Park, Helmsley, and Scutterskelfe. Since each of these lords may have influenced the School Board differently, I treat each pair, Brotton-Duncombe, Brotton-Helmsley, and BrottonScutterskelfe, as a different observation. 26 Industrial cities: Birmingham, Bradford, Bristol, Coventry, Derby, Leeds, Liverpool, London, Manchester, Newcastle, Nottingham, Prescot, Preston, Sheffield, and Southampton. Cathedrals: Bristol, Canterbury, Carlisle, Chester, Chichester, Durham, Ely, Exeter, Gloucester, Hereford, Lichfield, Lincoln, London, Manchester, Newcastle, Norwich, Oxford, Peterborough, Ripon, Rochester, Salisbury, Southwell, St Albans, Truro, Wakefield, Wells, Winchester, Worcester, and York.

19

his seat. This stricter specification yields very similar results despite the fact that the number of observations is only 487. Second, I weight each observation using the distance between School Boards and seats. This is based on the assumption that the influence of a landlord over a School Board decreases with the distance to his seat. In ds,b , where ds,b is the detail, I weight each seat–School Board pair (s, b) with ws,b = dminb distance between s and b and dminb is the shortest distance between School Board b and any seat in England. This gives larger weights to closer seat–School Board pairs (see Figure A.3 in the Appendix). Finally, column 5 includes fixed effects for each landlord. Results do not change across specifications. [TABLE 3 HERE]

5.3

Identification

The previous section suggests that there was a negative association between landownership concentration and state education in late-nineteenth century England. The coefficients estimated by OLS, however, cannot be interpreted as causal. Although reverse causality is not a major issue—British landownership had been stable from 1750 onward (Beckett 1977)—unobserved heterogeneity may bias the estimates. It might be that areas where land was more concentrated are intrinsically different in terms of culture or geography, and that these differences explain state education. I address endogeneity concerns using an instrumental variables approach that exploits a novel and a well-established source of exogenous variation in the distribution of land: the Norman conquest of England in 1066 and soil texture. 5.3.1

The Norman conquest of England (1066)

In 1066, William the Conqueror crossed the Channel from Normandy, defeated the Anglo-Saxons in Hastings, and was proclaimed King of England. One of his first acts was to redistribute land ownership. Anglo-Saxon England had been a mosaic of landowners. In contrast, William took one half of all the lands for himself, gave a quarter to the Church, and divided the rest among 190 Normans—the ancestors of the large landowners in the nineteenth century. My instrument exploits the fact that much of the land inequality in the nineteenth century can be traced back to 1066. To construct this instrument I use data from the Domesday Book, a survey of all landholdings c.1066. Since the Domesday lists land values (not acreage), I define my instrument as the share of the total land value in each region in the hands of the

20

top five Norman landowners.27 For the analysis using county data, I consider the five largest landowners for every county surveyed in the Domesday Book. For the analysis using local data, I identify the top five landowners in each 25-mile radius around the 487 seats of the nineteenth-century large landlords. Figure 4 displays the geographical distribution of this instrument. Remarkably, the distribution of land in England did not change much from 1066 to the late-nineteenth century. In the counties in the West Midlands, between 32 and 84 percent of the land value was in the hands of the top five Norman landowners in 1066 (panel (a)). Eight centuries later, around half of the land was owned by large landowners (see Figure 2, panel (a)). In contrast, in the South (excluding Sussex) the land grab by the Normans was smaller and so was landownership concentration in the late-nineteenth century. Local data suggests that there is also spatial correlation between the location of the largest estates in nineteenth-century England (Figure 2, panel (b)) and land inequality in 1066 (Figure 4, panel (b)). [FIGURE 4 HERE] As with every instrumental variables approach, the identifying assumption is that the instrument is relevant and that the exclusion restriction is satisfied. Figure 4 shows that the instrument is relevant: there is a strong association between landownership concentration in 1066 and eight centuries later. In other words, land inequality has deep roots in England. However, the relevance condition would be violated if William’s conquest did not lead to more land inequality or if landownership was already concentrated before 1066. The evidence suggests that this is note the case: Anglo-Saxon England had been a mosaic of landowners (Cahill 2001). To show this, I identified the 4,690 farms given to 31 Norman nobles that appear in the Bayeux Tapestry—a Norman embroidery which depicts the conquest of England. Table 4 presents the number of owners in these lands before and after the conquest. The Norman conquest was a massive shock to the distribution of land. For example, in Buckinghamshire, 285 farms which used to have 181 different owners became the possession of only 10 Norman lords after 1066. Overall, the surveyed farms had 1,516 different owners before the conquest and 100 after it. [TABLE 4 HERE] As for the exclusion restriction, it seems obvious that the motives that drove William to redistribute the land in 1066 had nothing to do with state education eight 27

Landowners were the immediate lords of the peasantry; i.e., either the tenant-in-chief himself or a tenant to whom he had granted the estate.

21

centuries later. However, the exclusion restriction would be violated if the Norman conquest triggered persistent local differences in institutions or religion in addition to landownership. This scenario is unlikely. According to Angelucci, Meraglia, and Voigtlnder (2017), the Norman conquest “resulted in largely homogenous formal institutions across England” (p. 1). Addmitedly, the fact that William gave a fourth of the land to the Church can be problematic. Most of this land was later confiscated by Henry VIII during the reformation, and was sold to the most fervent Anglican noblemen. Including these landholdings in the analysis could violate the exclusion restriction, as religion plays an important role for education (Green 1990; Clark and Gray 2014). To address this issue, I do not consider the half of the lands that William took for himself and the fourth given to the Church. In other words, my instrument is based on the top five landowners in 1066, excluding the King and the Church. Another valid concern is whether William gave richer lands to his companions, and whether land quality affected state education later on or not. To address this, I consider an additional source of exogenous variation in land inequality: soil texture. 5.3.2

Soil texture

Using the geological composition of the soil as a source of exogenous variation in landownership goes back to Easterly (2007) and Vollrath (2009), who exploited crop suitability. Although it is certainly true that some crops have economies of scale and hence are associated with higher inequality, it is also true that farmers can to some extent choose the crops they cultivate. Here I follow Cinnirella and Hornung (2016) in using a truly exogenous source of variation: soil texture. The texture of a soil is determined by the percentage of sand, silt, and clay and by the presence of chalk and peat fragments. Soil texture satisfies the exclusion restriction, as it is truly exogenous: it does not change over time and cannot be altered by human intervention (Cinnirella and Hornung 2016). In addition, soil texture is arguably unrelated to local institutions that may affect education provision. For example, Angelucci, Meraglia, and Voigtlnder (2017) show that the quality of the soil did not affect whether a borough received a Charter of Liberties or not, which in the long run crucially led to inclusive political institutions and the representation of merchants’ interests in Parliament. Furthermore, soil texture is a relevant instrument. Differences in soil texture are strongly related to differences in landownership concentration. In particular, soils with a larger sand component and soils with chalk fragments do not retain water well, are drought-prone, and hence, worse for agriculture (Leeper and Uren 1993). At the 22

same time, areas with a less productive agriculture are subject to lower population pressure and a weaker demand for land. As a consequence, landownership ends up more concentrated than in areas where the soil is rich.28 To construct the soil-texture instrument I use data from the British Geological Survey. The survey classifies English soils in 40 categories according to the relative proportions of sand, silt, clay, chalk, and peat fragments. These categories, in turn, can be classified into nine broad groups (see Table 5). I define the soil-texture instrument as the percentage of land with soils where sand is the largest component (henceforth: sandy soils). Sandy soils include sandy loam soils (50-80% sand), loam soils (30-55% sand), and clayey loam soils (20-50% sand). The instrument also includes the percentage of chalky soils, which, as explained above, are also drought-prone and hence less suited for agriculture. Conversely, the reference group includes non-sandy soils: silty loam (0–50% sand), clay (0–40% sand), silt (0–40% sand), and soils with peat fragments, which are usually very fertile. Note that I also include pure sand (90– 100% sand) in the reference group. The reason is that pure sand is typically found next to rivers, canals, or the coastline. While very bad for agriculture, these lands may have been very profitable for other economic activities, especially for trade. It is reasonable to conjecture that these areas also experienced a high population pressure and a strong demand for land, and therefore, that landownership might have ended up more fragmented than what their agricultural productivity suggests. [TABLE 5 HERE] In sum, I define the instrument as the percentage of sandy and chalky soils in each county. For the analysis using local data, I consider the percentage of sandy and chalky soils in a 25-mile radius around each of the 487 seats of nineteenth-century large landlords. Appendix B describes the construction of the instrument in detail. Figure 4 suggests that the instrument is relevant, that is, that there is a negative association between soil quality and landownership concentration in England. In detail, it displays the raw data from the British Geological Survey (panel (d)) and the percentage of sandy and chalky soils in every county (panel (c)). Counties where sandy and chalky soils prevail are in the West Midlands and in the North-East, where land was more concentrated and where the largest estates were located (see Figure 2, panel (a) and (b)). In contrast, in the South East soil quality is higher and landownership was more fragmented in the nineteenth century. 28

Bhalla 1988; Bhalla and Roy 1988; Benjamin 1995; Barrett, Bellemare, and Hou 2010; Cinnirella and Hornung 2016.

23

5.4

Instrumental variables’ results

In this subsection, I estimate the effects of landownership concentration on state education using the Norman conquest of 1066 and soil texture as sources of exogenous variation. Formally, landownership concentration in the late-nineteenth century is treated as an endogenous variable and modeled as: landr = κ + λ land1066r + σ soil texturer + νr ,

(3)

where landr captures land inequality in the nineteenth century. For my county level analysis, landr is the percentage of land in county r in the hands of large landowners (i.e., owners of 3,000 acres or more). For my analysis using local data from 1,387 School Boards and 487 seats, landr is the acreage of a large landlord (i.e., a peer who owns 2,000 acres or more) living in seat r. Similarly, land1066r is the percentage of the total land value in region r that was given to the top five Norman noblemen after 1066; and soil texturer is the percentage of sandy and chalky soils in region r. For my county-level analysis, r is a county. For my analysis using local data, r is a 25-mile circle around each seat. The second stage takes the form of equation (1) for the county-level analysis and equation (2) for the estimation using local data. Note that I use a triangular IV model in which the treatment (landr ) and the instruments (land1066r and soil texturer ) vary by geographical region r, whereas education measures also vary over other dimensions (e.g., time). To fit this model, I estimate the recursive equation system defined by the first and the second stage by maximum likelihood. Table 6 presents the first-stage results. I find a strong persistance in land inequality over eight centuries. A one percentage point increase in the land owned by the top five landowners in 1066 is associated to an increase of 0.3 percentage points in the land owned by large landowners in the nineteenth century (panel A, cols. 1 and 3). Furthermore, the largest estates in nineteenth century England arose in areas where land was more concentrated in Norman times (panel B, cols. 1 and 3). In detail, increasing land concentration in 1066 by one percentage point is associated with an increase by 120 of the acreage of large landlords in the nineteenth century. The share of sandy and chalky soils (i.e., soils of poorer quality) is also associated with more landownership concentration (cols. 2 and 3). In areas where the soil exhibits poorer quality, land becomes more concentrated (panel A) and larger estates emerge (panel B). This relation is consistent with previous findings for India29 and 29

Bhalla 1988; Bhalla and Roy 1988; Benjamin 1995; Barrett, Bellemare, and Hou 2010

24

late-nineteenth century Prussia (Cinnirella and Hornung 2016). These results do not hinge on an arbitrary definition of landownership concentration: choosing the share of land in the hands of the top five, top three, or top ten landowners yields similar estimates (cols. 4 and 5). Finally, note that the F-statistic is small but above the standard threshold of 10 in the full specification (col. 3).30 [TABLE 6 HERE] Table 7 presents second-stage results using county-level data. Landownership concentration had a negative effect on state education in late-nineteenth century England. Counties in which landownership was more concentrated raised fewer funds from taxes on property and received fewer grants from the Committee of Education (panel A). In these counties, fewer School Boards were created, less money was spent per pupil, and investments in state schools were lower. Also, fewer teachers and class assistants were hired and their salaries were lower (panel B). As a result of this under-investment, children’s human capital deteriorated: they were less likely to pass the national reading, writing, and especially, arithmetics’ exam (panel C). The magnitudes are big. For example, decrasing landownership concentration by one standard deviation (i.e., by 10 percentage points) would increase the funds raised from property taxes, teacher’s salaries, and the percentage of children passing arithmetics by half a standard deviation (30 pence per child, £5, and 2.6 percentage points, respectively) and the expenditure per pupil by one third of a standard deviation (16 pence). Note that more industrial counties raised more funds for education, invested more in state schools, and had better education outcomes. Again, this suggests that manufacturers pushed for education supply and/or that the demand for education by the working population was higher in more industrial counties. [TABLE 7 HERE] Table 8 presents second-stage estimates using local data from 1,387 School Boards and 487 seats of large landlords (i.e., peers who owned 2,000 acres or more). Results confirm my previous findings: the size of the landholdings decreased School Boards’ capacity to raise funds for education. Increasing by one standard deviation 30

Note that the first stage excludes county covariates, namely: log income, % voting conservative, % non-conformists, and religiosity. The reason is that these variables are uncorrelated with land concentration in the nineteenth century. In fact, they are realized after the distribution of landownership, which justifies the recusive approach used above. Including these covariates in the first stage would only reduce its strength, inducing a weak instruments’ problem. In Appendix A, Table A.5, I estimate the same IV model including all covariates in the first stage and correcting for weak instruments. Table A.6 does the same exercise using local data. Results are robust.

25

the landholdings of a nearby lord (i.e., by 9,799 acres) would reduce tax rates by 1.5 percentage points. Given that the average tax rate was only 4 percent, the estimated effects amount to a decrease of 37.5 percent. Note that these effects are larger than the OLS results. This suggests that the endogeneity bias in the OLS specification goes against my hypothesis. One possibility is that ommitted variables such as culture are positiviely correlated both with land inequality and with state education. Results are robust to including covariates that are potentially correlated with state education: employment in manufacturing, income per capita, ideology, percentage of non-conformists, and religiosity (col. 2) and the distance to the closest industrial center and the closest cathedral (col. 3). Furthermore, alternative specifications relaxing the assumption that each School Board and landlord pair (b, s) is an independent observation also yield very similar results. In detail, estimates are robust to collapsing the data at the seat level (col. 4) and to weighting each observation using the distance between School Boards and seats (col. 5). Finally, I include fixed-effects for each landlord (col. 6). The effect remains, but the fixed effects are highly significant. This suggests that individual characteristics of lords had strong effects on state education. In the next Section, I will explore these characteristics and show that the landlords that opposed education supply more effectively were those with more political power. [TABLE 8 HERE]

6

Mechanisms

This section investigates the mechanisms through which landownership concentration undermined human capital in late-nineteenth century England. In detail, I evaluate whether landownership concentration affects human capital through economic inequality, through the political process, or through the private demand for education. On the one hand, in societies in which landownership is concentrated economic inequality is high. If capital markets are imperfect, the poor may not invest in human capital (Banerjee and Newman 1993; Galor and Zeira 1993). Economic inequality may also distort the supply of state education (Murphy, Shleifer, and Vishny 1989). On the other hand, the ownership of land is linked to political power (Allen 2009). Therefore, land inequality not only reflects economic inequality but also political inequality, in the sense that decisions are taken by a small elite (Acemoglu, Bautista, and Robinson 2008). In this case, by a landed elite. Landed elites typically opposed 26

education supply, as human capital and agricultural work are not compementary and also to reduce the mobility of the rural labor force (Galor and Moav 2006; Galor, Moav, and Vollrath 2009). In addition, since state education was funded with property taxes, the opposition by landed elites should be stronger where land is more concentrated (Thompson 1963: 208). For example, in Lincolnshire sixty-eight landowners owned 40 percent of the land. Any state school, therefore, would have to be paid by them. It is reasonable to conjecture that these 68 landowners used their political influence to oppose the supply of state education in Lincolnshire. Finally, where land is more concentrated the private demand for education by peasants can be lower. The demand for education can be an important factor in latenineteenth century England, as by that time the elites had extended the franchise considerably. In the United States (Go and Lindert 2010) and in South America (Acemoglu and Robinson 2000; Engerman and Sokoloff 2000; Mariscal and Sokoloff 2000; Gallego 2010) the extension of the franchise was associated with an increase in the demand for education. Similarly, Cinnirella and Hornung (2016, Ashraf et al. (2017) show that the emancipation of serfs spurred education demand in Prussia. Here I disentangle these three mechanisms. First, I exploit the fact that England had a well-defined political elite (Allen 2009) to disentangle political inequality from economic inequality. Second, I exploit my rich dataset to disentangle the supply and demand channels. In detail, I evaluate measures of education supply (e.g., number of state schools) and of demand for education (e.g., number of examinees).

6.1

Political inequality vs. economic inequality

Disentangling political and economic inequality is a major empirical challenge, since they usually come hand in hand. To address this issue, I use the fact that England had a well-defined political elite—the peerage. According to Douglas Allen, It is hard to exaggerate the extent to which the aristocracy ruled Britain through its control over what we now call public offices. Both houses of Parliament were controlled by them until the turn of the twentieth century. The King’s household, which evolved into the executive arm of the government, was the domain of the aristocracy, as were the great offices and tenures of state. The army and navy officers were drawn from the aristocracy, as were the judges, justices of the peace, and other local administrators. (Allen 2009: p. 301) Thus, if landownership concentration affects the supply of education through the political process, the political relevance of landowners should matter. In detail, the 27

supply of education should be lower in areas where the landowners were politically relevant peers than in areas where landowners did not control public offices. In contrast, if the effect of landownership concentration on state education is driven by economic inequality per se, the political power of landowners should not matter. Specifically, I disentangle political and economic inequality by coding the biographies in thepeerage.com of 369 large landlords (i.e., peers who owned 2,000 acres or more) and matching them to local data from 1,387 School Boards. I then regress the (instrumented) acreage of a large landlord on the property tax rate set by School Boards nearby. That is, the equation system (2) and (3). I estimate the effects for two subsamples. On the one hand, I consider School Boards in a 25-mile radius of the seat of a politically relevant landlord; i.e., who ever held an important political office. On the other hand, I consider School Boards nearby politically irrelevant landlords. Table 9 presents the results. In the baseline specification, I find a strong negative effect of landownership on the tax rates set by School Boards (col. 1). The estimated effect disappears when the sample is restricted to landlords who were not elected Member of Parliament (MP), that is, to landlords who had no political power. In other words, School Boards operated normally in areas with high economic inequality but where landowners were not politically relevant. Similarly, the political ideology of a large landlord significantly affected the supply of education near his domains. The size of a landlord’s estate has a negative, significant effect on the tax rates set by School Boards near liberal large landlords.31 However, the effect is three times stronger for School Boards near Conservative landlords. Specifically, increasing the landholdings of a nearby lord by one standard deviation (i.e., by 9,799 acres) would reduce tax rates by 0.5 percentage points for School Boards under the influence of a liberal landlord (col. 2), and by 1.6 percentage points for School Boards near a conservative landowner (col. 3). These results have to be taken with a grain of salt for two reasons: First, because splitting the sample diminishes the strength of the instruments. In fact, for these subsamples, the F-statistic in the first stage is below 10. To correct for weak instruments, Table 9 reports confidence intervals based on Moreira’s (2003) conditional likelihood ratio (CLR) approach.32 The confidence intervals overlap, that is, the estimated effects are not statistically different across subsamples. 31

Liberal landlords were members of a liberal party (Libs and Whigs) or a liberal political club (Brooks, Reform, and Devonshire). Conservatives were members of a conservative party (Tories and Unionists) or a conservative club (Carlton, Junior Carlton, Conservative, and St. Stephen’s). 32 In detail, I estimate an instrumental variables’ model including all the covariates in the first stage and then calculate the corresponding CLR confidence intervals.

28

The second reason why these results are only suggestive is that, while being elected MP reflects political power, it mostly carried political influence at the national level. One may expect that the landlords who took over School Boards were those who held local political power (see Stephens 1998 for historical evidence). In panel B I address these issues by focusing on local political power. I identify the large landlords who were appointed, at some point in their life, to the most important local offices: Lord Lieutenant, Deputy Lieutenant, High Sheriff, or Sheriff. The baseline specification in column 1 reports, again, the main result of the paper: a strong negative effect of landownership on the property tax rates set by School Boards. The estimated effect is much smaller when the sample is restricted to Scholboards near landlords who were not appointed to any of the aforementioned local posts (col. 2). A Wald test confirms that the two estimated coefficeints are different. In other words, in areas where economic inequality was high but landowners did not control local political offices, the work of School Boards was not heavily distorted. In contrast, School Boards exposed to landowners who were appointed to local offices systematically raised less funds for education. Finally, note that splitting the sample by the local political power does not lead to weak instruments’ problems. Admittedly, landlords who were and were not appointed to local offices were intrinsically different. To address this issue, I consider landlords who were appointed to local posts, but these were posts in counties far away from their estates. These landlords were politically relevant, but lacked the necessary local political power to take over School Boards near their estates. In detail, column 3 considers landlords who were not appointed to a local post in England, even though some of them might have been Lord Lieutenants, Sheriffs, etc. in Ireland or Scotland. Results are similar to those reported in column 2. [TABLE 9 HERE] These results suggest that the channel through which landownership affected state education was political rather than economic inequality. While the landed elite in England opposed education reforms, they could only do so effectively where they held sufficient political power. This result is in line with previous findings of Galor, Moav, and Vollrath (2009), Acemoglu, Bautista, and Robinson (2008), Ramcharan (2010), and Vollrath (2013).

6.2

Demand vs. supply

So far, I have documented a strong, negative effect of landownership concentration on state education, especially in areas where landlords were politically relevant. I 29

have argued that this result reflects the opposition of landed elites to the supply of education. The estimated effects, however, could also be explained by a lower private demand for education in areas where land was more concentrated. This demand channel has been emphasized in other settings. Cinnirella and Hornung (2016, Ashraf et al. (2017) show that the emancipation of serfs spurred the demand for education in Prussia. Acemoglu and Robinson (2000), Engerman and Sokoloff (2000), Mariscal and Sokoloff (2000), Gallego (2010), and Go and Lindert (2010) suggest that the demand for education grew in South America and in the United States after the extension of the franchise. At first sight, late-nineteenth century England seems to fit well into this story. After the Reform Acts of 1832, 1867, and 1884, two in three adult males had the vote. How relevant was the demand channel in England? Here I exploit my rich dataset to gauge the importance of the demand for education and to disentangle it from the opposition of landed elites to education supply. Specifically, I estimate the IV model in equations (1) and (3) with education-demand measures (e.g., attendance rates) and education-supply measures (e.g., numer of state schools) as dependant variables. Since most of these variables depend on the number of children in the county, all regressions include this covariate.33 Table 10 presents the results. Enrolment rates, a measure traditionally associated with the private demand for education, were not significantly affected by landownership concentration (cols. 1 and 2). The estimates are negative, but not statistically different from zero. This result has to be taken with a grain of salt. In 1880, education was made compulsory for all children aged 5–11. Enrolment rates, therefore, may not reflect the private demand for education but just the change in the law. To correct for this, columns 3 to 6 look at the number of students presented for examination. In particular, examinees over age 10 and examinees in the highest levels of education (standards IV to VII)34 were not binded by law to attend school and, hence, should reflect well the demand for education. Results suggest that the number of students presented for examination was larger where landownership was more concentrated, although the magnitudes are small (cols. 3 and 4). In the average county, increasing land inequality by one standard deviation (i.e., by ten percentage points) would increase the number of examinees over age 10 by 0.04 standard deviations (i.e., by 1,339), and the number of students taking the highest level exams by 0.06 standard deviations (i.e., by 1,528). Nevertheless, these results suggests that, in England, there 33 Results are robust to defining the dependant variables relative to the number of children aged 5 to 15 in each county. This robustness check is available upon request. 34 See Table A.3 in the Appendix for details on each Standard.

30

was not a lack of demand for education where landownership was more concentrated. Next, I compare the importance of the demand and the supply channels. To do so, I define two indexes. These indexes are the first principal component from measures that reflect education demand (the number of pupils attending and registered, the examinees under age 10, over age 10, presented for standards I to III, and for standards IV to VII) and education supply (the number of School Boards, the ratio of state to private schools, teachers and class assistants hired, and expenditures per pupil and per state school) respectively. To ease comparisions, both indexes are standardized to have zero mean and standard deviation equal to one. Columns 7 and 8 regress the education demand and supply indexes on landownership concentration, instrumented with landownership in 1066 and with soil texture. Results suggest that the demand for education is not affected by landownership concentration: the magnitude of the effect is small and not statistically different from zero. In contrast, land concentration has a strong, negative effect on education supply. In the average county, increasing land inequality by one standard deviation (i.e., by ten percentage points) would decrease by 0.4 standard deviations the index of education supply. Wald tests confirm that the effect of landownership on education supply is larger than on education demand. [TABLE 10 HERE] In sum, while landownership concentration seriously affected the supply of education (e.g., number of state schools and teachers hired), there does not seem to be a lack of demand for education in areas where landowership is more concentrated. This strongly suggest that the opposition of entrenched, landed elites was the reason why England lagged behind in state education in the late-nineteenth century.

7

Conclusion

At the dawn of the Second Industrial Revolution (1870–1914), England failed to educate their workforce (Sanderson 1995). I show that this was the result of the opposition of entrenched, landed elites to the supply of state education. Using a new database on 1,387 School Boards and 32 counties between 1871 and 1899, I find that landownership concentration had a strong, negative effect on various dimensions of state education. School Boards next to the largest landlords in England set lower property taxes to fund education. As a consequence, where landownership was more concentrated the ratio of state to private schools, teacher’s salaries, and expenditures per pupil were lower. Also, fewer teachers and class assistants were hired. These had 31

important consequences for human capital accumulation: children were less likely to pass the reading, writing, and arithmetics’ national exams. I argue that the relation between land inequality and state education is causal by exploiting two sources of plausibly exogenous variation in landownership: soil texture and the massive redistribution of land after the Norman conquest of England in 1066. This instrumental variables’ approach provides two sets of interesting results: On the one hand, first-stage estimates document a strong persistence in land inequality over eight centuries, from 1066 to the late-nineteenth century. On the other hand, second-stage results suggest that the effect of landownership concentration on state education is causal. These findings contribute to a vast literature that studies the long-run consequences of inequality on human capital. Previous research has shown that inequality in the form of landownership concentration slowed down the introduction of state education in agrarian, developing economies like the US South (Vollrath 2009), the plantation economies in the Caribbean (Engerman and Sokoloff 2000), and South America (Coastworth 1993; Nugent and Robinson 2010). Here I show that landownership concentration can also distort state education in an industrialized economy with considerable extension of the franchise. These results suggest that landownership concentration, by affecting human capital formation, is important for the economic and demographic changes that began after the Industrial Revolution in England. This paper also examines the causal channels through which land inequality undermines state education. Coding biographical information on 369 peer landowners and matching it to local data on state education from 1,387 School Boards, I find that the negative effects of land inequality are restricted to areas where landowners were prominent political figures. This suggests that the detrimental effects of land inequality on human capital were engendered by the political opposition of landed elites to education supply. Furthermore, the percentage of workers employed in manufacturing was positively associated with state education, which is consistent with the idea that old landed elites opposed and emerging industrialists supported the supply of state education (Galor and Moav 2006; Galor, Moav, and Vollrath 2009). In contrast, economic inequality per se is not central. School Boards in areas with high economic inequality but where landowners were not politically relevant set higher tax rates to fund education. I also find that in areas where land was more concentrated, the private demand for education was not lower. Exploiting my rich dataset, I show that landownership had a negative effect on education-supply variables (e.g., number of state schools). In contrast, variables related to the private demand for education were 32

not significantly affected (e.g., attendance rates) or where even positively affected by landownership concentration (e.g., number of examinees). These findings shed new light on the mechanisms through which landownership concentration undermines human capital. Elsewhere it has been argued that the private demand for education was the main driver of state education in countries where the elite had extended the franchise35 or after the emancipation of serfs (Cinnirella and Hornung 2016; Ashraf et al. 2017). My results suggests that this was not the case in late-nineteenth century England. Even though two in three males had the vote, entrenched, landed elites still retained de facto political power to oppose the supply of state education. In other words, extended suffrage did not act as a constraint on elite capture. Admittedly, this paper looks at a specific historical setting. Nevertheless, inequality today is on the rise. The top one percent income share, for example, has more than doubled over the last decades. Several raised concerns on whether this will distort policies like state education. My findings suggest that the political power of those at the top of the distribution and their willigness to provide state education are important elements to take into account when designing policies aimed redistribution and human capital formation.

References Acemoglu, Daron, Maria Angelica Bautista, and James A. Robinson. 2008. “Economic and Political Inequality in Development: the case of Cundinamarca, Colombia.” In Institutions and Economic Performance, edited by Elhanan Helpman. Cambridge: Harvard University Press. Acemoglu, Daron, and James A. Robinson. 2000. “Why Did the West Extend the Franchise? Democracy, Inequality, and Growth in Historical Perspective*.” The Quarterly Journal of Economics 115 (4): pp. 1167–1199. Alatas, Vivi, Abhijit Banerjee, Rema Hanna, Benjamin A. Olken, Ririn Purnamasari, and Matthew Wai-Poi. 2013, February. “Does Elite Capture Matter? Local Elites and Targeted Welfare Programs in Indonesia.” Working paper 18798, NBER. Aldcroft, D. H. 1964. “The Entrepreneur and the British Economy, 1870-1914.” Economic History Review 17 (1): pp. 113–134. 35

Acemoglu and Robinson (2000), Engerman and Sokoloff (2000), Mariscal and Sokoloff (2000), Gallego (2010), and Go and Lindert (2010).

33

Allen, Douglas W. 2009. “A Theory of the Pre-Modern British Aristocracy.” Explorations in Economic History 46 (3): pp. 299–313. Allen, Robert C. 1992. Enclosure and the Yeoman. Oxford: Clarendon Press. Allen, Robert Carson. 1979. “International Competition in Iron and Steel, 1850– 1913.” Journal of Economic History 39 (04): pp. 911–937. Angelucci, Charles, Simone Meraglia, and Nico Voigtlnder. 2017, July. “The Medieval Roots of Inclusive Institutions: From the Norman Conquest of England to the Great Reform Act.” Working paper 23606, NBER. Ashraf, Quamrul, Francesco Cinnirella, Oded Galor, Boris Gershman, and Erik Hornung. 2017. “Capital-Skill Complementarity and the Emergence of Labor Emancipation.” Cesifo working paper series 6423, CESifo Group Munich. Bandiera, Oriana, Myra Mohnen, Imran Rasul, and Martina Viarengo. 2015, April. “Nation-Building Through Compulsory Schooling During the Age of Mass Migration.” Economic Organisation and Public Policy Discussion Papers 57, LSE. Banerjee, Abhijit V, and Andrew F Newman. 1993. “Occupational Choice and the Process of Development.” Journal of Political Economy 101 (2): pp. 274–98. Barrett, Christopher B., Marc F. Bellemare, and Janet Y. Hou. 2010. “Reconsidering Conventional Explanations of the Inverse Productivity-Size Relationship.” World Development 38 (1): 88 – 97. Bateman, John. 1883. The Great Landowners of Great Britain and Ireland. London: Harrison and Sons. Beckett, J. V. 1977. “English Landownership in the Later Seventeenth and Eighteenth Centuries: The Debate and the Problems.” Economic History Review 30 (4): pp. 567–581. Benjamin, Dwayne. 1995. “Can Unobserved Land Quality Explain the Inverse Productivity Relationship?” Journal of Development Economics 46 (1): pp. 51– 84. Berghoff, H., and R. Moller. 1994. “Tired Pioneers and Dynamic Newcomers? A Comparative Essay on English and German Entrepreneurial History, 1870-1914.” Economic History Review 47 (2): pp. 262–287. Berghoff, Hartmut. 1990. “Public Schools and the Decline of the British Economy 1870–1914.” Past & Present 129 (1): pp. 148–167.

34

Bhalla, Surjit S. 1988. “Does Land Quality Matter?: Theory and Measurement.” Journal of Development Economics 29 (1): pp. 45–62. Bhalla, Surjit S, and Prannoy L Roy. 1988. “Mis-specification in Farm Productivity Analysis: The Role of Land Quality.” Oxford Economic Papers 40 (1): pp. 55–73. British Geological Survey. 2014. Soil Parent Material 1kilometre dataset. [data collection]. British Geological Survey. Burke, John. 1826. A General and Heraldic Dictionary of the Peerage and Baronetage of the United Kingdom. London: Henry Colburn. Cahill, Kevin. 2001. Who Owns Britain. UK: Bath Press. Cassis, Y. 1997. Big Business. The European Experience in the Twentieth Century. Oxford: Oxford University Press. Cinnirella, Francesco, and Erik Hornung. 2016. “Landownership concentration and the expansion of education.” Journal of Development Economics 121:135 – 152. Cinnirella, Francesco, and Ruth Maria Schueler. 2016. “Nation Building: The Role of Central Spending in Education.” CESifo Working Paper Series 6103. Cinnirella, Francesco, and Jochen Streb. 2017. “The role of human capital and innovation in economic development: evidence from post-Malthusian Prussia.” Journal of Economic Growth 22 (2): 193–227 (Jun). Clark, Gregory, Neil Cummins, Yu Hao, Daniel Diaz Vidal, Tatsuya Ishii, Zach Landes, Daniel Marcin, Firas Abu-Sneneh, Wilfred Chow, Kuk Mo Jung, Ariel M. Marek, and Kevin M. Williams. 2014. The Son also Rises: Surnames and the History of Social Mobility. Princeton: Princeton University Press. Clark, Gregory, and Rowena Gray. 2014. “Geography is not destiny: geography, institutions and literacy in England, 1837–63.” Oxford Economic Papers. Coastworth, J.H. 1993. “Notes on the Comparative Economic History of Latin America and the United States.” In Development and Underdevelopment in America, edited by W. Bernecker and H.W. Tobler. New York: Walter de Gruyter. . 2004. “Long-run Growth.” In The Cambridge Economic History of Modern Britain, volume 2, edited by R. Floud and P. Johnson. Cambridge: Cambridge University Press. Cvrcek, Tomas, and Miroslav Zajicek. 2016. “The Making of a Liberal Education: Political Economy of the Austrian School Reform, 1865 1875.” Technical Report.

35

Dell, M. 2010. “The Persistent Effects of Peru’s Mining Mita.” Econometrica 78 (6): 1863–1903. Doepke, Matthias, and Fabrizio Zilibotti. 2005. “The Macroeconomics of Child Labor Regulation.” American Economic Review 95 (5): pp. 1492–1524. Easterly, William. 2007. “Inequality Does Cause Underdevelopment: Insights from a New Instrument.” Journal of Development Economics 84 (2): pp. 755–776. Elbaum, Bernard, and William Lazonick. 1984. “The Decline of the British Economy: An Institutional Perspective.” Journal of Economic History 44 (2):pp. 567–583. Engerman, Stanley L., and Kenneth L. Sokoloff. 2000. “Institutions, Factor Endowments, and Paths of Development in the New World.” Journal of Economic Perspectives 14 (3): pp. 217–232. Fernandez-Villaverde, J. 2007. “Was Malthus Right? Economic Growth and Population Dynamics.” Working paper 13208, NBER. Foster, Andrew, and Mark R. Rosenzweig. 2001. “Democratization, Decentralization and the Distribution of Local Public Goods in a Poor Rural Economy.” 11. Franck, Raphael, and Oded Galor. 2017, February. “Technology-Skill Complementarity in Early Phases of Industrialization.” Working paper 23197, National Bureau of Economic Research. Gallego, Francisco A. 2010. “Historical Origins of Schooling: The Role of Democracy and Political Decentralization.” The Review of Economics and Statistics 92 (2): pp. 228–234. Galor, Oded, and Omer Moav. 2002. “Natural Selection and the Origin of Economic Growth.” Quarterly Journal of Economics 117 (4): pp. 1133–1191. . 2006. “Das Human-Kapital: A Theory of the Demise of the Class Structure.” Review of Economic Studies 73 (1): pp. 85–117. Galor, Oded, Omer Moav, and Dietrich Vollrath. 2009. “Inequality in Landownership, the Emergence of Human-Capital Promoting Institutions, and the Great Divergence.” Review of Economic Studies 76 (1): pp. 143–179. Galor, Oded, and David N. Weil. 2000. “Population, Technology, and Growth: From the Malthusian Regime to the Demographic Transition and Beyond.” American Economic Review 90:pp. 806–838. Galor, Oded, and Joseph Zeira. 1993. “Income Distribution and Macroeconomics.” Review of Economic Studies 60 (1): pp. 35–52. 36

Go, Sun, and Peter Lindert. 2010. “The Uneven Rise of American Public Schools to 1850.” The Journal of Economic History 70 (1): 1–26. Green, A. 1990. Education and State Formation. Hampshire: MacMillan Palgrave. Hansen, Gary D., and Edward C. Prescott. 2002. “Malthus to Solow.” American Economic Review 92 (4): pp. 1205–1217. Hechter, M. 1976. U.K. County Data, 1851–1966. [data collection]. Colchester, Essex: UK Data Archive. SN: 430, http://doi.org/10.5255/UKDA-SN-430-1. Hurt, J. 1968. “Landowners, Farmers and Clergy and the Financing of Rural Education before 1870.” Journal of Educational Administration and History 1:pp. 6–16. Hurt, John. 1971. Education in Evolution. London: Paladin. Jones, Charles I. 2001. “Was an Industrial Revolution Inevitable? Economic Growth Over the Very Long Run.” Advances in Macroeconomics. Kogel, Tomas, and Alexia Prskawetz. 2001. “Agricultural Productivity Growth and Escape from the Malthusian Trap.” Journal of Economic Growth 6 (4): pp. 337–357. Lagerlof, Nils-Petter. 2006. “The Galor-Weil Model Revisited: A Quantitative Exercise.” Review of Economic Dynamics 9 (1): pp. 116–142. Landes, D. 1960. “The Structure of Enterprise in the Nineteenth Century: Great Britain and Germany.” In The Rise of Capitalism, edited by D. Landes. New York: MacMillan. Leeper, Geoffrey W., and Nicholas C. Uren. 1993. Soil Science: an Introduction. Melbourne University Press. Lindert, P.H. 2004. Growing Public: Social Spending and Economic Growth since the Eighteenth Century. Cambridge: Cambridge University Press. Lizzeri, Alessandro, and Nicola Persico. 2004. “Why did the Elites Extend the Suffrage? Democracy and the Scope of Government, with an Application to Britain’s Age of Reform.” Quarterly Journal of Economics 119 (2): pp. 707–65. Maloney, William F., and Felipe Valencia Caicedo. 2016. “The Persistence of (Subnational) Fortune.” The Economic Journal 126 (598): 2363–2401. Mariscal, Elisa, and Kenneth L. Sokoloff. 2000. “Schooling, suffrage, and inequality in the Americas, 1800-1945.” In Political Institutions and Economic Growth in 37

Latin America. Essay in Policy, History, and Political Economy, edited by S.H. Haber, pp. 159–218. Stanford: Hoover Institution Press. McCloskey, Donald N., and Lars G. Sandberg. 1971. “From Damnation to Redemption: Judgments on the Late Victorian Entrepreneur.” Explorations in Economic History 9 (1): pp. 89–108. Mitch, D. 1992. The Rise of Popular Literacy in Victorian England: The Influence of Private Choice and Public Policy. Philadelphia: University of Pennsylvania Press. Mitch, David. 1993. “The role of human capital in the First Industrial Revolution.” In British Industrial Revolution: An Economic Perspective, edited by Joel Mokyr, pp. 267–307. Boulder: The Westview Press. Mokyr, J., and H.J. Voth. 2009. “Understanding Growth in Europe, 1700–1870: Theory and Evidence.” In The Cambridge Economic History of Modern Europe Volume 1 - 1700–1870, edited by S. Broadberry and K.H. O’Rourke. Cambridge: Cambridge University Press. Mokyr, Joel. 2005. “Long-Term Economic Growth and the History of Technology.” Chapter 17 of Handbook of Economic Growth, edited by Philippe Aghion and Steven Durlauf, Volume 1, Part B, 1, 1113–1180. Elsevier. Moreira, Marcelo J. 2003. “A Conditional Likelihood Ratio Test for Structural Models.” Econometrica 71 (4): 1027–1048. Murphy, Kevin M., Andrei Shleifer, and Robert Vishny. 1989. “Income Distribution, Market Size, and Industrialization.” Quarterly Journal of Economics 104:pp. 537–64. Nicholas, T. 2004. “Enterprise and Management.” In The Cambridge Economic History of Modern Britain, volume 2, edited by R. Floud and P. Johnson. Cambridge: Cambridge University Press. Nugent, Jeffrey B., and James A. Robinson. 2010. “Are Factor Endowments Fate?” Revista de Historia Econ´omica / Journal of Iberian and Latin American Economic History (Second Series) 28:pp. 45–82. Palmer, John. 2010. Electronic Edition of Domesday Book: Translation, Databases and Scholarly Commentary, 1086. [data collection]. 2nd Edition. UK Data Service. SN: 5694, http://doi.org/10.5255/UKDA-SN-5694-1. Pollard, S. 1989. Britain’s Prime and Britain’s Decline: The British Economy 1870–1914. London: Edward Arnold. 38

Potter, A., and G. Emerson. 1989. The School and the School Master. New York: Harper and Brothers. Ramcharan, Rodney. 2010. “Inequality and Redistribution: Evidence from U.S. Counties and States, 1890-1930.” The Review of Economics and Statistics 92 (4): 729–744 (November). Sanderson, M. 1995. Education, Economic Change, and Society in England, 1780– 1870. Cambridge: Cambridge University Press. Saul, S.B. 1968. “The Engineering Industry.” In The Cambridge Economic History of Modern Europe Volume 1 - 1700–1870, edited by D. H. Aldcroft. London: Allen and Unwin. Squicciarini, Mara P., and Nico Voigtlander. 2015. “Human Capital and Industrialization: Evidence from the Age of Enlightenment.” The Quarterly Journal of Economics 130 (4): 1825. Stephens, W.B. 1997. “Minutes and Reports from the Committee of Council on Education.” Microform academic publishers. . 1998. Education in Britain, 1750–1914 (Social History in Perspective). New York: St. Martin’s Press. Tapia, Francisco J.Beltran, and Julio Martinez-Galarraga. 2015. “Land Access Inequality and Education in Pre-industrial Spain.” Oxford university economic and social history series 137, Nuffield College, University of Oxford. Thompson, F.M.L. 1963. English Landed Society in the Nineteenth Century. London: Routledge and Kegan Paul. . 1994. “Business and Landed Elites in the Nineteenth Century.” In Landowners, Capitalists, and Entrepreneurs, edited by F.M.L. Thompson. Oxford: Clarendon Press. Tod, J. 1873. Handbook of the Education (Scotland) Act, 1872, 5th ed. Edinburgh: Edmonston and Douglas. Vollrath, D. 2009. “Geographical Elements of Farm Size Inequality in the United States.” Manuscript, University of Houston. Vollrath, Dietrich. 2013. “Inequality and school funding in the rural United States, 1890.” Explorations in Economic History 50 (2): 267 – 284. Weiner, M.J. 1981. English Culture and the Decline of the Industrial Spirit, 1850– 1980. Cambridge: Cambridge University Press. 39

8

Figures and tables Figure 1: Report from the Committee of Council in Education, 1883–84 Panel A: Funding

Panel B: Expenditures

Panel C: Outcomes

40

Figure 2: Concentration of landownership and state education in late-nineteenth century England

spacespace spacespace spacespace spacespace

(a) Landownership; county data

(b) Landownership; local data (487 seats)

(c) Education; county data

(d) Education; local data (1,387 School Boards)

Notes: In panels (a), (c), and (d) values are arranged into different classes using Jenks natural breaks classification method. In panel (b), classes are based on Bateman (1883) appendix classifications. In panel (a), land concentration is the percentage of land of a county in the hands of large landowners (i.e., owners of 3,000 acres or more).

41

Figure 3: Burton House and the School Boards in a 25 miles radius.

spacespacespac Notes: Labels report the name of the School Board and the average tax rate (%) spacespacespac set between 1873 and 1878 in parenthesis.

42

Figure 4: Instruments (a) Landownership after 1066; county data

(b) Landownership after 1066; local data

43

(c) Soil texture; county data

(d) Soil texture; local data

44

Table 1: Descriptive statistics. mean

sd

N

variation

year

unit

2.54 72.2 43.8 16.0 176.3

2.20 65.3 49.1 14.7 188.0

1,387 120 120 120 120

district county×decade county×decade county×decade county×decade

1873-78 1871-94 1871–94 1871–94 1871–94

average % pence per child pence per child pence per child pence per child

34.41 35.22 824.2 167.3 22,689 467.0 147.6

38.01 20.75 913.2 149.4 52,457 46.4 83.0

120 120 120 120 40 120 80

county×decade county×decade county×decade county×decade county county×decade county×decade

1871–94 1879–98 1879–98 1884–98 1878–79 1879–89 1879–94

number ratio number number pounds pence per child pence per child

90.67 82.29 78.08

3.75 3.80 4.72

80 80 80

county×decade county×decade county×decade

1879–90 1879–90 1879–90

% % %

76,642 22,989 25,107 31,674 16,938

99,375 30,566 34,296 39,503 24,026

120 80 80 80 80

county×decade county×decade county×decade county×decade county×decade

1879–96 1879–90 1879–90 1879–90 1879–90

number number number number number

% in county acres in county % %

Panel A: Education. A1. Funding: Tax rate Funds from property taxes Grants from the Committee School fees and books sold Total incomes A2. Expenditures: School Boards State to Voluntary schools Certificate teachers Class assistants Salaries teachers (total) Av. expenditure per pupil Expenditure on state schools A3. Outcomes: Reading exam (% passing) Writting exam (% passing) Arithmetics exam (% passing) A4. Demand: Pupils attending Examinees (U10) Examinees (O10) Examinees (Standard I-III) Examinees (Standard VI-VII)

Panel B: Land concentration and political power. Land concentration Acreage of large landlord Large landlord is MP Large landlord has local power

40.99 7,840 41.2 74.3

10.68 9,799 49.3 81.5

40 487 369 369

county lord’s seat lord lord

1883 1883 lifetime lifetime

36.85 62.58 33.07 60.68

17.18 14.04 15.77 11.90

32 39† 487 487

county county lord’s seat lord’s seat

1086 2014 1086 2014

Panel C: Instruments. Land concentration after 1066 Sandy and chalky soils Land concentration after 1066 Sandy and chalky soils

% in county % in county % in 25 mi. radius % in 25 mi. radius

Note: The sample of counties are all counties in England. To construct per child measures I consider children aged 5 to 10. Large landlords are peers who own 2,000 acres or more. Acreage of large landlord are the acres in the county where his seat is located; i.e., each seat is treated as a different observation. Land concentration is the percentage of a county’s land in the hands of large landowners (i.e., owners of 3,000 acres or more). A large landlord has local power if he is appointed, at some point in his life, to the most important local offices: Lord Lieutenant, Deputy Lieutenant, High Sheriff, or Sheriff. Land concentration after 1066 is the percentage of the total land value in the hands of the top five landowners, excluding the King and the Church, as reported in the Domesday Book. This measure is calculated over each county and over a 25-mile radius surrounding each lord’s seat. Examination standards are in Table A.3. † Excludes Monmouthshire.

45

Table 2: OLS estimates for the relation between landownership and education, by county. Panel A. Funding (pence per child) Rates

Grants

Fees

Other

Total

Land concentration (%)

-1.98*** (0.69)

-1.39*** (0.45)

-0.23 (0.15)

-0.05*** (0.02)

-4.50** (2.14)

Employed in manufacturing (%)

2.50*** (0.41)

1.86*** (0.32)

0.57*** (0.10)

0.07*** (0.02)

6.88*** (1.05)

Observations Adjusted-R2 County controls Available reports

96 0.362 YES 1871–95

96 0.336 YES 1871–95

96 0.344 YES 1871–95

96 0.241 YES 1871–95

96 0.320 YES 1871–95

Panel B. Expenditures Schools

Teachers

pence

School Boards

State to private

Cert. teacher

Class assist.

Teacher salary

per pupil

for State school (pc)

Land concentration (%)

-0.90** (0.38)

-0.59*** (0.20)

-12.10* (6.28)

-4.01*** (1.09)

-97.8** (36.0)

-0.73** (0.30)

-2.71** (1.15)

Employed in manu. (%)

0.74 (0.59)

0.03 (0.15)

28.05** (12.75)

3.37* (1.74)

68.87* (39.2)

0.21 (0.28)

2.92*** (0.78)

96 0.216 YES 1871–95†

96 0.377 YES 1879–99

96 0.228 YES 1879–99

96 0.243 YES 1879–99

32 0.455 YES 1878–79

96 0.394 YES 1879–99

64 0.302 YES 1871–95

Observations Adjusted-R2 County controls Available reports Panel C. Outcomes

% passes in Reading

Writing

Arith.

Total

Land concentration (%)

-0.10** (0.04)

-0.08** (0.04)

-0.14*** (0.03)

-0.10*** (0.02)

Employed in manufacturing (%)

0.05 (0.05)

0.15*** (0.03)

0.19*** (0.03)

0.11*** (0.02)

64 0.076 YES 1879–91

64 0.410 YES 1879–91

64 0.415 YES 1879–91

64 0.483 YES 1879–91

Observations Adjusted-R2 County controls Available reports †

Excluding the reports from 1887-88, 1889-90, 1892-93. Note: The sample consists of a panel of 32 counties and 3 decades (1870s, 1880s, and 1890s). For consistency, I exclude the counties not fully surveyed in the Domesday book ; i.e., Cumber, Durham, Lancaster, Monmouth, Northumber, Westmore, Middlesex, and Hants. Land concentration is the % of land in a county owned by large landowners (i.e., owners of 3,000 acres or more). Education measures are decade averages rather than annual values because county controls only vary by decade. County controls are log income, % voting conservative, % non-conformists, religiosity. Constants not reported. Standard errors clustered by county; *** p<0.01, ** p<0.05, * p<0.1

46

47

24,705 0.003 NO NO NO seat

-0.001*** (0.000) 24,705 0.033 YES NO NO seat

-0.001*** (0.000)

OLS [2]

24,705 0.006 NO YES NO seat

-0.001*** (0.000)

OLS [3]

487 0.058 NO YES NO -

-0.001*** (0.000)

collapsed [4]

24,705 0.006 NO YES NO seat

-0.001*** (0.000)

weighted [5]

24,705 0.050 NO YES Lord Lord

-0.002*** (0.000)

FE [6]

Note: The sample consists of 1,387 School Boards operating in 1873–78 and 487 seats of large landlords (i.e., peers who owned 2,000 acres more). Each observation is a seat–School Board pair {s, b}, where School Board b is within a 25-mile radius of seat s. In col. 4 observations are collapsed by seat and in col. 5 are weighted by the distance between School Board and seat. County controls are log income, % voting conservative, % non-conformists, religiosity (Hechter 1976). Local controls are the distance from each School Board to the closest industrial city and to the closest cathedral. Industrial cities are Birmingham, Bradford, Bristol, Coventry, Derby, Leeds, Liverpool, London, Manchester, Newcastle, Nottingham, Prescot, Preston, Sheffield, and Southampton. Cathedrals are in Bristol, Canterbury, Carlisle, Chester, Chichester, Durham, Ely, Exeter, Gloucester, Hereford, Lichfield, Lincoln, London, Manchester, Newcastle, Norwich, Oxford, Peterborough, Ripon, Rochester, Salisbury, Southwell, St Albans, Truro, Wakefield, Wells, Winchester, Worcester, and York. Constants not reported. Standard errors clustered by seat (except in cols. 4 and 6.); *** p<0.01, ** p<0.05, * p<0.1

Observations R-squared County controls Local controls FE Cluster s.e.

Acreage of large landlord (100s)

OLS [1]

Dep. Variable: Tax rates (%)

Table 3: OLS estimates for the relation between landownership and education, local data.

Table 4: The redistribution of land after 1066. Landowners Sample farms

Before the conquest

After the conquest

Buckinghamshire Cambridgeshire Essex Lincolnshire Norfolk Northamptonshire Oxfordshire Somerset Suffolk Warwickshire

285 207 389 421 1,032 186 113 193 1,724 140

181 101 187 106 224 71 16 90 468 72

10 9 12 9 11 9 9 8 13 10

Total

4,690

1,516

100

County

Note: The sample consists of 4,690 farms given to 39 Norman noblemen that appear in the Bayeux Tapestry: Robert de Beaumont, 1st Earl of Leicester; Humphrey de Bohun; Eustace II, Count of Boulogne; Hugh d’Avranches, 1st Earl of Chester; William, Count of Evreux; Robert, Count of d’Eu; Walter Giffard, Lord of Longueville; Henry de Ferrers; Hugh de Montfort; Baldwin Fitz Gilbert; Hugh de Grandmesnil; Richard Fitz Gilbert; William de Warenne, 1st Earl of Surrey; Hugh II de Gournay; William Malet, Lord of Graville; Walter de Lacy; Odo, Bishop of Bayeux, later Earl of Kent ; Ilbert de Lacy; Turstin FitzRolf; William Malet; Geoffrey de Mowbray, Bishop of Coutances; Sir Geoffrey de Mandeville, 1st Earl of Essex; Robert, Count of Mortain; Roger de Montgomerie, 1st Earl of Shrewsbury; Hugues de Beauchamp, Viscount of Stafford; William de Percy, 1st Baron Percy; William de Bertram, Lord of Briquebec; Alan Rufus, 1st Earl of Richmond; Richard Bigod; Urse d’Abetot le Spencer; and Robert Bigod.

48

Table 5: Soil-texture categories in the British Geological Survey (2014). Soil group

IV group

% sand

% clay

% silt

soils in British Geological Survey (2014)

Sandy loam

instrument

50–85

0–35

0–45

Sandy loam; Sandy loam to clayey loam; Sandy loam to loam; Sandy loam to sand; Sandy loam to silty loam.

Loams

instrument

30–55

5–25

30–50

Loam; Loam to clay; Loam to clayey loam; Loam to sandy; Loam to sandy loam; Loam to silty; Loam to silty loam.

Clayey loam

instrument

20–50

25–40

20–50

Clayey loam to sandy loam; Clayey loam to silty loam.

Chalk

instrument

-

-

-

Chalky clay to chalky loam; Chalky sandy loam; Chalky silty loam; Locally chalky.

Silty loam

reference

0–50

0–40

50–100

Silty loam; Silty loam to sandy loam; Silty loam to silt.

Clay

reference

0–40

40–100

0–40

Clay to clayey loam; Clay to loam; Clay to sandy loam; Clay to silt; Clayey.

Silt

reference

0–40

0–40

40–100

Silt to sand; Silt to silty loam.

Peat

reference

-

-

-

Peat; Peat and peaty clay or silt; Peaty clay; Peaty clay or silt; Peaty silt; Locally peaty.

‘Pure’ sand

reference†

90–100

0–15

0–15

Sand; Sand to loam; Sand to sandy loam; Sand to silt.

Notes: The % of sand, clay and silt is inferred from the USDA textural triangle. I include ‘Pure’ Sand in the reference group because these soils are typically found next to rivers, canals, or the coastline, and hence, may have been well-suited for trade and other economic activities. †

49

Table 6: First-stage results. (1)

(2)

(3)

(4)

(5)

Panel A. County-level analysis Dep. Var.: Land concentration in 19C (%) Land concentration in 1066 (%) top five

0.26*** (0.09)

-

0.26*** (0.08)

top three

0.26*** (0.08)

top ten Sandy and chalky soils (%) Observations R-squared F-stat

0.26*** (0.08) -

0.40** (0.15)

0.40*** (0.13)

0.39*** (0.13)

0.40*** (0.13)

32 0.23 8.8

32 0.19 7.0

32 0.41 10.1

32 0.39 9.3

32 0.40 9.8

Panel B. Local data analysis Dep. Var.: Acreage of large landlords in 19C (100s) Land concentration in 1066 (%) top five

1.22*** (0.28)

-

1.15*** (0.28)

top three

1.09*** (0.29)

top ten Sandy and chalky soils (%) Observations R-squared F-stat

1.18*** (0.27) -

1.25*** (0.37)

1.13*** (0.37)

1.11*** (0.37)

1.22*** (0.36)

487 0.038 19.4

487 0.023 11.5

487 0.057 14.7

487 0.051 13.0

487 0.06 15.4

Note: In panel A, the sample is all counties in England fully surveyed in the Domesday Book. In panel B, the sample comprises 25-mile radius around each of the 487 country seats of large landlords (i.e., peers who owned 2,000 acres or more in the 1880s). Land concentration in 1066 is the share of land value in each geographical region that is owned by the top five, top three, and top ten Norman noble landlords respectively. *** p<0.01, ** p<0.05, * p<0.1.

50

Table 7: IV estimates for the effect of landownership on state education, by county. Panel A. Funding (pence p.c.) Rates

Grants

Fees

Other

Total

Land concentration (%)

-3.09** (1.33)

-2.14** (0.96)

-0.40 (0.28)

-0.05* (0.03)

-7.21* (3.81)

Employed in manufacturing (%)

2.52*** (0.41) 10.1 96 YES 1871–95

1.88*** (0.32) 10.1 96 YES 1871–95

0.57*** (0.10) 10.1 96 YES 1871–95

0.07*** (0.02) 10.1 96 YES 1871–95

6.93*** (1.05) 10.1 96 YES 1871–95

F-stat from first-stage Observations County controls Available reports Panel B. Expenditures Schools

Teachers

pence

School Boards

State to private

Cert. teacher

Class assist.

Teacher salary

per pupil

for State school (pc)

Land concentration (%)

-1.68** (0.74)

-0.43 (0.45)

-18.52 (13.13)

-8.66*** (3.07)

123.4* (68.8)

-1.60** (0.80)

-4.23* (2.56)

Employed in manu. (%)

0.75 (0.56) 10.1 96 YES 1871–95†

0.04 (0.15) 10.1 96 YES 1879–99

27.99** (12.45) 10.1 96 YES 1879–99

3.33* (1.73) 10.1 96 YES 1879–99

67.28* (35.7) 10.1 32 YES 1878–79

0.20 (0.29) 10.1 96 YES 1879–99

2.92*** (0.78) 10.1 64 YES 1871–95

F-stat from first-stage Observations County controls Available reports Panel C. Outcomes

% passes in

Land concentration (%) Employed in manufacturing (%) F-stat from first-stage Observations County controls Available reports

Reading

Writing

Arith.

Total

-0.17* (0.10)

-0.15* (0.08)

-0.26*** (0.09)

-0.16** (0.07)

0.05 (0.05) 10.1 64 YES 1879–91

0.15*** (0.03) 10.1 64 YES 1879–91

0.19*** (0.03) 10.1 64 YES 1879–91

0.11*** (0.02) 10.1 64 YES 1879–91

†

Excluding the reports from 1887-88, 1889-90, 1892-93. Note: The sample consists of a panel of 32 counties and 3 decades (1870s, 1880s, and 1890s). I exclude the counties not fully surveyed in the Domesday book ; i.e., Cumber, Durham, Lancaster, Monmouth, Northumber, Westmore, Middlesex, and Hants. Land concentration is the % of land in a county owned by large landowners (i.e., owners of 3,000 acres or more). This variable is instrumented with landownership in 1066 and soil texture (see Table 6, panel A, col. 3 for details). Education measures are decade averages. County controls are log income, % voting conservative, % non-conformists, religiosity. Constants not reported. Standard errors clustered by county; *** p<0.01, ** p<0.05, * p<0.1

51

52

14.7 24,705 NO NO NO seat

-0.015*** (0.004) 14.7 24,705 YES NO NO seat

-0.006*** (0.002)

IV [2]

14.7 24,705 NO YES NO seat

-0.015*** (0.004)

IV [3]

14.7 487 NO YES NO -

-0.017*** (0.003)

collapsed [4]

14.7 24,705 NO YES NO seat

-0.012*** (0.004)

weighted [5]

14.7 24,705 NO YES Lord Lord

-0.015** (0.004)

FE [6]

Note: The sample consists of 1,387 School Boards operating in 1873–78 and 487 seats of large landlords (i.e., peers who owned 2,000 acres more in the 1880s). Each observation is a seat–School Board pair {s, b}, where School Board b is within a 25-mile radius of seat s. In col. 4 observations are collapsed by seat and in col. 5 are weighted by the distance between School Board and seat. Acreage of grat lord (100s of acres) is instrumented with landownership in 1066 and soil texture (see Table 6, panel B, col. 3 for details). County controls are log income, % voting conservative, % non-conformists, religiosity (Hechter 1976). Local controls are the distance from each School Board to the closest industrial city and to the closest cathedral. Industrial cities are Birmingham, Bradford, Bristol, Coventry, Derby, Leeds, Liverpool, London, Manchester, Newcastle, Nottingham, Prescot, Preston, Sheffield, and Southampton. Cathedrals are in Bristol, Canterbury, Carlisle, Chester, Chichester, Durham, Ely, Exeter, Gloucester, Hereford, Lichfield, Lincoln, London, Manchester, Newcastle, Norwich, Oxford, Peterborough, Ripon, Rochester, Salisbury, Southwell, St Albans, Truro, Wakefield, Wells, Winchester, Worcester, and York. Constants not reported. Standard errors clustered by seat (except in cols. 4 and 6.); *** p<0.01, ** p<0.05, * p<0.1.

F-stat from first-stage Observations County controls Local controls FE Cluster s.e.

Acreage of large landlord (100s)

IV [1]

Dep. Variable: Tax rates (%)

Table 8: IV estimates for the effect of landownership on state education, local data.

Table 9: Political mechanism, local data IV estimates. (1) Panel A. Parliament

Acreage of large landlord (100s) Ho: prob > χ2 95% c.i. (β) CLR 95% c.i. (β) F-stat from first-stage Observations Controls

(2)

(3)

(4)

Dep. Variable: property tax rate (%) baseline

no MP

Liberal MP

Tory MP

-0.015*** (0.004)

-0.011 (0.009)

-0.005*** (0.002)

-0.016*** (0.004)

[-0.02,-0.01] 14.7 24,705 local

β(1) = β(2) 0.61 [-∞, +∞] 3.6 13,668 local

[-0.03,-0.00] 8.8 3,506 local

β(4) = β(5) 0.01** [-0.04,-0.01] 5.7 5,785 local

Panel B. Local power

Dep. Variable: property tax rate (%) local appointments

Acreage of large landlord (100s) Ho: prob > χ2 CLR 95% c.i. (β) F-stat from first-stage Observations Controls

baseline

none

none in England

-0.015*** (0.004)

-0.009*** (0.003)

-0.010*** (0.003)

14.7 24,705 local

β(1) = β(2) 0.03** 12.9 10,754 local

β(1) = β(3) 0.05** 14.6 12,188 local

The sample consists of 1,387 School Boards operating in 1873–78 and 487 seats of large landlords (i.e., peers who owned 2,000 acres more in the 1880s). Each observation is a seat– School Board pair (s, b), where School Board b is within a 25-mile radius of seat s. The sample is split according to the political relevance of landlords: Member of Parliament (MP) and local appointments (Lord Lieutenant, Deputy Lieutenant, High Sheriff, and Sheriff). Cols. 2 to 4 are restricted to landlords who were older than 21 in 1882, when Bateman’s (1883) landownership data was collected. Acreage of large landlord (100s) is instrumented with landownership in 1066 and soil texture. Local controls are the distance from each School Board to the closest industrial city and to the closest cathedral. CLR 95% c.i. are reported when the first-stage instruments are weak. Standard errors clustered by seat; *** p<0.01, ** p<0.05, * p<0.1

53

54

10.1 96 YES YES 1879–97

F-stat from first-stage Observations Children aged 5 to 15 County controls Available reports

10.1 96 YES YES 1879–97

410.3*** (144.5)

-

-74.9 (328.3)

Pupils registered (2)

10.1 64 YES YES 1879–91

23.6 (35.5)

-

141.4 (87.2)

under age 10 (3)

10.1 64 YES YES 1879–91

77.6** (34.4)

-

133.9* (69.4)

over age 10 (4)

10.1 64 YES YES 1879–91

23.9 (46.9)

-

152.8* (84.2)

Standard I to III (5)

Presented for exam

10.1 64 YES YES 1883–91

77.0*** (21.4)

-

110.9** (50.5)

Standard IV to VII (6)

10.1 64 YES -

0.003** (0.001)

0.004 (0.003)

demand (7)

10.1 64 YES -

-0.001 (0.007)

β(7) = β(8) 0.01***

-0.044*** (0.017)

supply (8)

PCA

Excluding the report from 1883-84. Note: The sample consists of a panel of 32 counties and 3 decades (1870s, 1880s, and 1890s). In col. 7, the dependent variable is the first component of a principal components analysis (PCA) including all the variables reflecting demand for education (i.e., the variables in cols. 1 to 6). In col. 8, the dependent variable is the corresponding component of all variables reflecting education supply: the number of School Boards, teachers, and class assistants hired, the ratio of state to private schools, and the average expenditure per child and in state schools. Both measures are standardized to have a mean of 0 and standard deviation of 1. Exam Standards are defined in Table A.3 in the Appendix. As in Table 7, I exclude the counties not fully surveyed in the Domesday book. Land concentration is the % of land in a county owned by large landowners (i.e., owners of 3,000 acres or more). This variable is instrumented with landownership in 1066 and soil texture (see Table 6, panel A, col. 3 for details). Education measures in cols 1–5 are decade averages. All regressions include the number of children aged 5 to 15 as a covariate. County controls are log income, % voting conservative, % non-conformists, religiosity. Constants not reported. Standard errors clustered by county; *** p<0.01, ** p<0.05, * p<0.1

†

432.3*** (116.1)

-

-224.0 (297.6)

Employed in manu. (%)

Ho: prob > χ2 :

Land concentration (%)

Pupils attending (1)

Enrolment

Table 10: Demand vs. supply mechanism, cross-county IV estimates.

Appendix A

Additional figures and tables Figure A.1: Lord Lyttelton, Bateman’s Great Landowners

Notes: The last two columns correspond to acreage and annual land rents respectively. Figure A.2: Henry Herbert, 4th Earl Carnavon, thepeerage.com

55

Figure A.3: Example for the weights used in Table 3, col. 5 25 miles 25 miles

A

w

=

1

w

w<1

w

=

1 Seat 1

=

1

Seat 2

C

B

Notes: In this example I consider three School Boards (A, B, and C) and two seats. In the baseline specification, I treat each pair A-Seat1, B-Seat1, C-Seat1, and C-Seat2 as a different observation. Note that, as the 25-mile radius around Seat 1 and Seat 2 overlap, C is paired twice, once with Seat 1 and once with Seat 2. In contrast, in Table ds,b , where ds,b is the 3, col. 5, I weight each School Board-seat pair with ws,b = dminb distance between School Board b and seat s and dminb is the shortest distance between b and any seat in the sample. In this example, the pairs A-Seat1, B-Seat1, and CSeat2 have a weight equal to one. In contrast, since distSeat 1,C > dminC = distSeat 2,C the pair C-Seat1 receives a lower weight. The underlying assumption is that the influence of a lord over a School Board decreases with distance, so C-Seat1 should weight less than C-Seat2.

56

Table A.1: Elementary Education Acts, 1870 to 1902. Year

Act

Description

1870

Education Act

State to provide education

1873

Education Act

School attendance condition for outdoor relief

1876

Sandon’s Act

Creates School Attendance Committees

1879

Industrial School

School Boards to manage Industrial Schools

1880

Mundella’s Act

Attendance compulsory for children aged 5–10

1890

Education Code

Standards of education

1891

Free Grant

Virtually establishes free elementary schooling

1893

Blind and Deaf

Special schools for blind and deaf children

1893

School Attendance

Attendance compulsory for children aged 5–11

1899

School Attendance

Attendance compulsory for children aged 5–12

1902

Balfour’s Act

Abolishes School Boards

Source: Stephens (1998).

57

Table A.2: Examination standards Standard I Reading

One of the narratives next in order after monosyllables in an elementary reading book used in the school.

Writing

Copy in manuscript character a line of print, and write from dictating a few common words.

Arithmetic

Simple addition and subtraction of numbers of not more than four figures, and the multiplication table to multiplication by six. Standard II

Reading

A short paragraph from an elementary reading book.

Writing

A sentence from the same book, slowly read once, and then dictated in single words.

Arithmetic

The multiplication table, and any simple rule as far as short division (inclusive). Standard III

Reading

A short paragraph from a more advanced reading book.

Writing

A sentence slowly dictated once by a few words at a time, from the same book.

Arithmetic

Long division and compound rules (money). Standard IV

Reading

A few lines of poetry or prose, at the choice of the inspector.

Writing

A sentence slowly dictated once, by a few words at a time, from a reading book, such as is used in the first class of the school.

Arithmetic

Compound rules (common weights and measures). Standard V

Reading

A short ordinary paragraph in a newspaper, or other modern narrative.

Writing

Another short ordinary paragraph in a newspaper, or other modern narrative, slowly dictated once by a few words at a time.

Arithmetic

Practice and bills of parcels. Standard VI

Reading

To read with fluency and expression.

Writing

A short theme or letter, or an easy paraphrase.

Source: Revised code of Regulations, 1872

58

59

rates, grants, and fees

yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes not available (N.A.) N.A. N.A. N.A.

report

1871-72 1872-73 1873-74 1874-75 1875-76 1876-77 1877-78 1878-79 1879-80 1880-81 1881-82 1882-83 1883-84 1884-85 1885-86 1886-87 1887-88 1888-89 1889-90 1890-91 1891-92 1892-93 1893-94 1894-95 1895-96 1896-97 1897-98 1898-99

Funding

yes no no no no no no yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes N.A. N.A. N.A. N.A.

other and total yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes no yes no yes yes no yes yes N.A. N.A. N.A. N.A.

number of School Boards N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes

teachers, schools, expendit. per pupil yes no no no no no no yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes N.A. N.A. N.A. N.A.

expenditure for State school

Expenditures

county-level data

N.A. N.A. N.A. N.A. N.A. N.A. N.A. yes N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A.

teacher salary N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. yes yes yes yes yes yes yes yes yes yes yes yes N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A.

% passing and examinees N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. yes yes yes yes yes yes yes yes N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A.

examinees standards IV to VII N.A. N.A. N.A. N.A. N.A. N.A. N.A. N.A. yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes yes N.A. N.A.

pupils attending, registered

Outcomes and Demand

no yes yes yes yes yes yes no no no no no no no no no no no no no no no no no N.A. N.A. N.A. N.A.

tax rates

school boards

Table A.3: Information computerized from the Reports of the Committee of Council on Education

Table A.4: Political appointments Panel A. Member of Parliament MP Tory

Liberal

Member of Parliament. MP for the Conservative party or the Unionists and/ or member of Carlton, Jun. Carlton, Conservative, St. Stephen’s Club. MP for the Liberal party, the Whigs, or the Liberal Union and/or member of Brook’s, Devonshire, Reform Club.

Panel B. Local appointments (local ) Lord Lieutenant Deputy Lieutenant (D.L.) High Sheriff Sheriff

Monarch’s representative in a county. Assistant to the Lord Lieutenant. Monarch’s judicial representative in the county. Monarch’s judicial representative in cities/boroughs.

Source: thepeerage.com

60

Table A.5: IV with all covariates in first-stage and weak instruments, county data Panel A. Funding (pence p.c.) Land concentration (%) CLR 95% c.i. K p-value K-J p-value AR p-value Observations Available reports

Rates

Grants

Fees

Other

Total

-3.54

-1.90

-0.61

-0.06

-10.4

[-6.8, -1.7] 0.00 *** 0.00 *** 0.01 *** 32 1871–95

[-3.9, -0.7] 0.01 *** 0.01 ** 0.01 ** 32 1871–95

[-1.3, -0.3] 0.05 * 0.07 * 0.11 32 1871–95

[-0.1, 0.0] 0.13 0.16 0.26 32 1871–95

[-20.9, -4.0] 0.01 *** 0.01 *** 0.02 ** 32 1871–95

Panel B. Expenditures Schools

pence

School Boards

State to private

Cert. teacher

Class assist.

Teacher salary

per pupil

for State school (pc)

-1.92

-0.55

-33.11

-6.58

150.5

-1.06

-5.35

[-4.7, -0.0] 0.06 * 0.07 * 0.15 32 1871–95†

[-2.0, 0.9] 0.58 0.02 ** 0.01 ** 32 1879–99

[-85.8, 9.4] 0.13 0.15 0.30 32 1879–99

[-14.5, -1.3] 0.02 ** 0.03 ** 0.07 * 32 1879–99

[-311, -48] 0.02 ** 0.03 ** 0.02 ** 32 1878-79

[-2.2, -0.2] 0.02 ** 0.03 ** 0.06 * 32 1879–99

[-11.0, -2.0] 0.01 ** 0.01 ** 0.01 ** 32 1871–95

Land conc. (%) CLR 95% c.i. K p-value K-J p-value AR p-value Observations Av. reports

Teachers

Panel C. Outcomes % passes in

Land concentration (%) CLR 95% c.i. K p-value K-J p-value AR p-value Observations Available reports

Reading

Writing

Arithmetics

Total

-0.12

-0.10

-0.19

-0.15

[-0.3, 0.0] 0.16 0.19 0.14 32 1879–91

[-0.3, 0.0] 0.11 0.14 0.22 32 1879–91

[-0.4, -0.1] 0.00 *** 0.00 *** 0.01 *** 32 1879–91

[-0.3,-0.1] 0.01 *** 0.01 *** 0.00 *** 32 1879–91

†

Excluding the reports from 1887-88, 1889-90, 1892-93. Note: This table presents the results from an IV specification estimated using LIML. In contrast to Table 7, here I include all the covariates in the first stage and correct for weak IV. The table reports p-values based on: Moreira’s (2003) conditional likelihood ratio (CLR); the Lagrange multiplier K; a combination of the K and J overidentification tests; and Anderson-Rubin’s test (AR). The sample consists of a cross-section of 32 counties (i.e., I do not exploit time variation over decades). Education measures are averages between 1871 and 1899. I exclude the counties not fully surveyed in the Domesday book. Land concentration is the % of land in a county by large landowners (i.e., owners of 3,000 acres or more). This is instrumented with landownership in 1066 and soil texture. County controls are log income, % voting conservative, % non-conformists, and religiosity. To fit the first stage, controls take their 1871 values; *** p<0.01, ** p<0.05, * p<0.1

61

62

24,705 NO NO NO seat

[-0.033, -0.012] 0.00 *** 0.00 *** 0.00 ***

-0.015

24,705 YES NO NO seat

[-0.042, -0.005] 0.00 *** 0.00 *** 0.00 ***

-0.006

weak IV [2]

24,705 NO YES NO seat

[-0.040, -0.013] 0.00 *** 0.00 *** 0.00 ***

-0.015

weak IV [3]

487 NO YES NO -

[-0.040, -0.014] 0.00 *** 0.00 *** 0.00 ***

-0.018

collapsed [4]

24,705 NO YES NO seat

[-0.037, -0.009] 0.00 *** 0.00 *** 0.00 ***

-0.012

weighted [5]

24,705 NO YES Lord Lord

[-0.038,-0.006] 0.00 *** 0.00 *** 0.00 ***

-0.008

FE [6]

Note: This table presents the results from an IV specification estimated using LIML. In contrast to Table 7, here I include all the covariates in the first stage and correct for weak IV. The table reports 95% confidence intervals based on Moreira’s (2003) conditional likelihood ratio (CLR) and p-values based on the Lagrange multiplier K; a combination of the K and J overidentification tests; and Anderson-Rubin’s test (AR). As before, the sample consists of 1,387 School Boards operating in 1873–78 and 487 seats of large landlords (i.e., peers who owned of 2,000 acres more in the 1880s). Each observation is a seat–School Board pair {s, b}, where School Board b is within a 25-mile radius of seat s. In col. 4 observations are collapsed by seat and in col. 5 are weighted by the distance between School Board and seat. Acreage of grat lord (100s of acres) is instrumented with landownership in 1066 and soil texture. County controls are log income, % voting conservative, % non-conformists, religiosity (Hechter 1976). Local controls are the distance from each School Board to the closest industrial city and to the closest cathedral. Constants not reported. Standard errors clustered as indicated; *** p<0.01, ** p<0.05, * p<0.1

Observations County controls Local controls FE Cluster s.e.

CLR 95% c.i. K p-value K-J p-value AR p-value

Acreage of large landlord (100s)

weak IV [1]

Dep. Variable: Tax rates (%)

Table A.6: IV with all covariates in first-stage and correcting for weak instruments, local data

B

Construction of the instruments

B.1 Landownership concentration in 1066 I use an instrumental variables approach that exploits two sources of plausibly exogenous variation in the distribution of land: the first is the Norman conquest of England in 1066. Anglo-Saxon England had been a mosaic of landowners. After the Norman conquest in 1066, however, William the conqueror took one half of all the lands for himself, gave a quarter to the Church, and divided the rest among 190 Norman—some of them ancestors of the large landowners in the nineteenth century. In other words, much of the inequality in landownership in the nineteenth century can be traced back to 1066. My instrument is the share of the total land value in a particular territory that was concentrated in the hands of top landowners after 1066. To construct this instrument, I use data from the Domesday Book. The Domesday is a survey of all landholdings in England commisioned in 1086 by William the Conqueror. Its main purpose was to determine what taxes had been owed during the reign of Edward the Confessor. This explains why London and Winchester—taxexempt cities—were not surveyed. The lands in the north, yet to be conquered, were also excluded. In total, the Domesday names 13,418 places. All the information is available in electronic format in Palmer (2010). The digitization of the Domesday Book was a monumental effort undertaken by various contributors over a long period, including Dr. Natasha Hodgson, Dr and Mrs. Thorn, the Phillimore and Co Ltd., and government typists under an ESRC-funded research project during the 1980s. The database includes an entry for each fief (i.e., manor) stating the landlord before and after the conquest, the value of the land, the population, and other resources (e.g., number of ploughs or the presence of a mill). For the sake of illustration, Figure B.1 shows the entry in the Domesday book corresponding to Leckhampsted, a small village in Berkshire. Gilber Maminot, a trusted collaborator of William,36 was the lord in 1086. He had replaced Earl Leofwin, the Anglo-Saxon lord before the conquest. The land in Leckhampsted was worth £6, it was populated by 32 households (including two slaves), and counted 12 ploughs and 400 pigs. I define my instrument as the share of the total land value in a particular territory owned by the top five landowners after 1066. For the sake of illustration, Figure B.2 shows how I construct this instrument for the county of Berkshire. Panel (a) shows the geographical location of all the farms surveyed in the Domesday. Larger circles correspond to more valuable farms. First, I remove from the analysis all the landholdings that William took for himself and the “fourth” of the lands given to the Church—see panel (b). The reason is that most of this land was later confiscated by Henry VIII during the reformation, and was sold to the most fervent Anglican noblemen. Including these landholdings in the analysis would violate the exclusion restriction, as religion plays an important role for education (Green 1990; Clark and 36

William the Conqueror sent Gilber Maminot to Rome to obtain the Pope’s blessing after the battle of Hastings. Charles Carlton, 1986, Royal childhoods, Taylor & Francis p. 24.

63

Gray 2014). Next, I identify the top landowners. In Berkshire, the top five landowners after 1066 were Henry of Ferrers (£52), Geoffrey de Mandeville (£49), Robert d’Oilly (£47), Walter Giffard (£39), and William, son of Ansculf (£34). In total, 19 percent of the total land value in Berkshire was owned by these top five landowners. Figure B.1: Domesday Book (1066), entry for Leckhampsted

Hundred: Stotfold. Value in 1086: £6. Population: 32 households (2 slaves). Other resources: Meadow 12 ploughs. Woodland 400 pigs. Lord in 1066: Earl Leofwin. Lord in 1086: Gilbert Maminot. Figure B.2: Construction of the Domesday instrument, Berkshire example (a) All surveyed farms

64

(b) Excluding the King and the Church

Top five share (instrument): 19.1 %

B.2 Soil texture To address endogeneity concerns in the relation between land inequality and education, I also exploit exogenous variation in soil texture. The texture of a soil crucially affects how well it retains storm water (Leeper and Uren 1993), and hence, how suited it is for agriculture. At the same time, areas where the soil is less suited for agriculture are subject to lower population pressure and a weaker demand for land, which leads to a more concentrated landownership.37 My instrument is the percentage of land in a particular area under soil textures associated with worse storm water retention. Which are these soil textures? Soil texture is determined by the percentage of sand, silt, and clay. Sand particles are relatively round as compared to silt and clay particles. Since the space between particles is larger, storm water is not retained well in soils with a relatively large sand component (Leeper and Uren 1993). This can lead, for example, to frequent droughts. In addition, the presence of chalk and peat fragments in the soil also affects infiltration rates of storm water. Peaty soils are usually very fertile and hold much moisture,38 while chalky soils are alkaline, droughtprone, and poorer in nutrients.39 In sum, soils with a larger sand component and with chalk fragments are less suited for agriculture, and hence, should be associated with 37

Bhalla (1988), Bhalla and Roy (1988), Benjamin (1995), Barrett, Bellemare, and Hou (2010), and Cinnirella and Hornung (2016). 38 Robert E. Stewart. 2017. “Agricultural technology.” Encyclopaedia Britannica. https://www. britannica.com/technology/agricultural-technology. 39 Brady, Nyle C. and Ray R. Weil. 2010. Elements of the nature and properties of soils (3rd Ed.) N.J.: Pearson Prentice Hall.

65

more landownership concentration. In contrast, silty, clayey, or peaty soils are less drought-prone, experienced a stronger demand for land, and therefore, should be associated with a more fragmented distribution of landownership. To construct the soil-texture instrument I use data from the British Geological Survey (2014). This survey classifies English soils in 40 categories according to the relative proportions of sand, silt, clay, chalk, and peat fragments. I construct my instrument as the percentage of land in a particular area with soils where sand is the largest component. That is, the percentage of sandy loam soils (50–85% sand), loam soils (30–55% sand), and clayey loam soils (20–50% sand). The instrument also includes the percentage of chalky soils, which as explained above are also droughtprone and hence less suited for agriculture. On the other hand, the reference group includes silty loam soils (0–50% sand), clay soils (0–40% sand), silt soils (0–40 % sand), and soils with peat fragments, which are usually very fertile. Note that “pure” sand soils (90–100% sand) are included in the reference group. The reason is that these soils are typically found next to rivers or the coastline which, while very bad for agriculture, may have been very profitable for trade for example. It is reasonable to conjecture that these areas also experienced a high population pressure and a strong demand for land, and therefore, that landownership might have ended up more fragmented. For the sake of illustration, Figure B.3 shows how I construct the soil-texture instrument for the county of Berkshire. Panel (a) displays the raw data from the British Geological Survey. The dataset is a vector grid with 1km x 1km cell dimensions. Each cell presents information on soil texture, classified in the 40 categories described above. To construct the instrument, I calculate the percentage of territory under chalky soils and soils with a larger sand component (i.e., sandy loam soils, loam soils, and clayey loam soils). In the case of Berkshire, this amounts to 59.9 percent. These are mostly lands in the west and north of the county, as illustrated in panel (b). In detail, the soils in the west of Berkshire are predominantly chalky, and hence, are included in the instrument. In contrast, the lands in the fertile south east are dominated by clay to silt, and therefore, are included in the reference group. Finally, panel (c) illustrates why I include “pure” sand soils (90–100% sand) in the reference group. Note that the areas where sand is predominant are next to rivers Loddon and Pang, the Holy Brook channel, and the Kennet and Avon Canal. Although these soils may not be ideal for agriculture, owning a plot of land next to a river or a canal may have been very profitable. It is reasonable to conjecture that these areas experienced a strong demand for land, and therefore, that landownership might have ended up more fragmented than what its agricultural productivity may suggest.

66

Figure B.3: Construction of the soil-texture instrument, Berkshire example (a) British Geological Survey

(b) Soil-texture instrument

Finally, panel (c) illustrates why I include “pure” sand soils (90–100% sand) in the reference group. Note that the areas where sand is predominant are next to rivers Loddon and Pang, the Holy Brook channel, and the Kennet and Avon Canal. Although these soils may not be ideal for agriculture, owning a plot of land next to 67

(c) Sand soils and rivers

a river or a canal may have been very profitable. It is reasonable to conjecture that these areas experienced a strong demand for land, and therefore, that landownership might have ended up more fragmented than what its agricultural productivity may suggest.

68