PRE-UNIFICATION SCHOOLING AND ITALY’S PRIMARY-EDUCATION SYSTEM (1861–1911)

At the time Italy was unified in 1861, none of the pre-unification polities had yet fully embarked on sustained industrialization and economic growth: a substantial share of its labor force was employed in agriculture, very limited railway and road infrastructure had been built, and extensive poverty and land inequality characterized large areas (Fenoaltea Reference Fenoaltea2011; Toniolo Reference Toniolo2013; Felice Reference Felice2015). Despite its limited economic development, the country was already characterized by regional economic inequality. Available estimates of per-capita GDP show little regional disparities in 1871 (Felice Reference Felice2013), yet a large divide was already evident in many dimensions, such as natural endowment (A’Hearn and Venables Reference A’Hearn, Venables and Toniolo2013), market potential and literacy (Basile and Ciccarelli Reference Basile and Ciccarelli2018), infrastructures (Ciccarelli and Groote Reference Ciccarelli and Groote2017), inventive activity (Nuvolari and Vasta Reference Nuvolari and Vasta2017), and real wages (Federico, Nuvolari, and Vasta Reference Federico, Nuvolari and Vasta2019).

Likewise, the pace and timing of educational development largely varied across territories, for example, concerning literacy rates (A’Hearn and Vecchi Reference A’Hearn, Vecchi and Vecchi2017a) and the gender gap in primary schooling (Bertocchi and Bozzano Reference Bertocchi and Bozzano2016), by and large favoring the North over the South.

Figure 1 shows enrollment rates and adult literacy rates during preunification years (the left-hand-side figure shows the names of pre-unification states).Footnote ⁶ Large regional inequality was evident at the beginning of the nineteenth century and, as we will see, remained a central feature of Italy’s literacy and schooling. While the former Kingdoms of Sardinia and of Lombardy-Venetia—with Lombardy ahead—exhibited enrollment and literacy rates not distant from those characterizing central-European countries (e.g., Prussia), the Center and the South lagged behind.

Figure 1 LITERACY AND ENROLLMENT RATES IN PRE-UNIFICATION STATES

Notes: Literacy rates in 1831 refer to adult population aged 30–40.

Sources: Data on enrollment rates are taken from Vigo (Reference Vigo1971). Literacy rates in 1831 are taken from Ciccarelli and Weisdorf (CW) (2019).

Indeed, Italy’s different pre-unification states pursued varying public-education policies. Most reforms initiated in the eighteenth century marginally addressed primary education, focusing on universities and secondary schooling instead. However, by Italy’s unification in 1861, the Kingdom of Sardinia had already established a primary school system heavily based on centralized norms and public funding by the municipalities (Ricuperati Reference Ricuperati2015). Since the mid-eighteenth century, the Kingdom of the Two Sicilies had also sought to reform education in the context of the Enlightenment (Felice Reference Felice2013). In the second half of the century, a new tax was imposed on monasteries’ income to fund public schools; further attempts at reforming the school system were carried out in the early nineteenth century, but the chronic unavailability of primary-school teachers forced the authorities to keep relying on religious orders. In 1810, primary education was made compulsory, yet school fees kept being charged. Albeit slowly, primary education grew to some extent between 1760 and 1815; yet, with the restoration of the absolutist regime of the Bourbons, and the threats to their rule coming from liberal ideas that might have been spread by mass education, primary schooling was permanently relegated to the control of the clergy and private initiatives (Genovesi Reference Genovesi1998). In the rest of the peninsula, the Duchies of Modena and Parma, the Granduchy of Tuscany, and the Papal States had begun a process of reform to introduce public primary schooling to varying degrees (Ricuperati Reference Ricuperati2015).

Concerning schooling—and literacy—the Kingdom of Lombardy-Venetia was split: even though the two regions of Lombardy and Venetia were part of the same kingdom from 1815 to 1859, Lombardy had forged ahead of the rest of Italy and caught up with Piedmont, constituting a high-literacy cluster in North-Western Italy. First, this happened as a result of enlightened school reforms introduced in the Duchy of Milan under Austrian rule (1714–1797). Even though primary schools were initially organized and managed through private initiatives charging tuition fees, the latter were abolished with the introduction of cost-free public education in 1791 (Genovesi Reference Genovesi1998). Secondly, Lombardy’s initial educational development was reinforced by the Napoleonic occupation of the Duchy of Milan and a few provinces further East, which was called the Cisalpine Republic (1797–1815). Male and female public schools were opened in every district of the Republic (Genovesi Reference Genovesi1998, p. 29). Such trends were not reversed by the restoration of Austrian rule and the foundation of the Kingdom of Lombardy-Venetia in 1815: primary education was compulsory for children aged between 6 and 12 years, even though each municipal school was under the supervision of the local priest. Despite this development, the region of Venetia had long relied on religious education, and, following its annexation to the Kingdom of Italy in 1866, the problem of agrarian elites hampering investments in public schools slowed the development of mass education into the Liberal Age—at least initially.

The effort to harmonize Italy’s education system took place since the very beginning of the unification process (which started during the 1860s and was completed by 1871), when the Kingdom of Italy implemented its first national school act.Footnote ⁷ Figure 2 displays youngsters’ literacy rates in 1871 and 1911, obtained from population censuses.Footnote ⁸ The regional pattern of literacy rates strongly echoes the spatial distribution of early schooling and human capital in the pre-unification states shown in Figure 1.

Figure 2 LITERACY RATES (AGE 15–19) IN 1871 AND 1911

Notes: Literacy rates (age 15–19) in 1871 and 1911 refer to population aged 15–19.

Source: Census data.

A clear picture of the state of the education system in recently unified Italy was depicted by Gerolamo Buonazia in 1868:

The first comprehensive set of norms was contained in the Casati Law, passed by the Kingdom of Sardinia in 1859 and later extended to the newly annexed areas of the unified country. As far as primary education was concerned, the central government set all the formal norms regulating the primary schools. The law stated that primary education had to be provided free of charge for at least two years. Additional two years were compulsory in larger municipalities and where a secondary school had already been established.

Despite a bulk of centralized formal provisions, the funding and management of schooling were fully decentralized: hiring teachers, paying them, building and running schools, and enforcing attendance were responsibilities born by the municipalities.Footnote ¹⁰ No redistributive mechanism was set up to correct the large regional inequality concerning fiscal capacity and educational investments. This issue was addressed through the Coppino Reform (1877), which introduced (weak) sanctions for missing enrollment and brought compulsory education to three years (up from two), introducing very limited subsidies, which had to be demanded by each municipality. This set of norms had little effect on human capital accumulation. Due to this, state intervention became more substantial at the turn of the twentieth century, when the Nasi (1903) and Orlando (1904) Laws improved the salaries of primary-school teachers and strengthened their legal position vis-à-vis the city councils, which up to that point could discretionarily lay them off. The actual step toward centralization was taken in 1911, through the Daneo-Credaro Reform: with it, the state fully funded current expenditure on primary education, supported investments for school buildings, and managed directly the organization of primary schooling through provincial school boards. Recent research shows that this reform brought about a positive increase in the pace of human capital accumulation, as measured by municipal-level literacy rates (Cappelli and Vasta Reference Cappelli and Vasta2020).

DATA AND SOURCES

To study how school inputs contributed to literacy in Liberal Italy, and how this relationship was linked to pre-unification conditions, we must first measure inputs into schooling and educational outcomes. Although collecting provincial data for the period that we study is normally very challenging, Italy’s historical statistics provide detailed figures on schooling and literacy, as well as other contextual variables. We built a new panel dataset on literacy rates and school inputs at the provincial level (roughly today’s NUTS-3) and at ten-year intervals between 1861 and 1911 (mostly corresponding to census years). A detailed description of each variable in our dataset, as well as the related sources consulted to elaborate on it, is included in the Online Data Appendix (Table DA1). We briefly discuss the dataset next. We start with the main education and schooling indices, then describe the geographic, demographic, and socio-economic controls, concluding with the presentation of the early-nineteenth-century variables that we use to explore the legacy of preunification states on human capital accumulation in Liberal Italy.

Descriptive Statistics

Our final dataset covers 69 Italian provinces (at 1871 boundaries) for five benchmark years (1871, 1881, 1891, 1901, and 1911), spanning the initial 50 years of the unified Kingdom of Italy. These figures are uncommonly rich for a nineteenth-century country and give us the opportunity to gain new insights into the performance of the Italian education system before more pervasive and coherent education policies were introduced starting in the first decade of the twentieth century. Selected descriptive statistics are reported in Table 1 (see Tables A1-a and A1-b in the Online Appendix for the full list).

Table 1 SUMMARY STATISTICS

Notes: All school inputs, except for the density of Evening and Sunday schools, are lagged by 10 years. This means that, for example, the GER 1871 refers to pupils aged 6–10 enrolled in 1861 who are aged 15–19 in 1871.

Source: See text.

Relying on the cohort structure of our dataset, we already uncover an important piece of evidence by comparing the primary GER of the age group 6–10 with the literacy rates of the same cohort ten years later—that is, the age group 15–19. The figures show that, on average, out of 53 pupils enrolled in primary schooling in Italy in 1861, only 34 had become literate by 1871. The same figures in 1911 were 95 and 72 percent, respectively. This discrepancy is partly due to the lack of data on repeaters, as well as attendance rates being normally lower than enrollment rates. This gap grows when adding private enrollments to public ones.Footnote ²³ Therefore, the correlation between class size and the GER-LIT Gap should be studied with some caution. The GER can be larger than 100 percent due to a high incidence of non-school-age pupils (less than six and more than ten years old). Instead, the literacy rate concerns the same cohort ten years later but is naturally capped at 100 percent. It is worth stressing that this is a limitation intrinsic to the sources, since they do not allow a calculation of the Net Enrollment Rate (excluding repeaters) for Italy’s provinces in the Liberal Age.Footnote ²⁴

Figure 3 shows the ratio between the entire school-age population and primary-school teachers (panel a), as well as expenditure divided again by the whole school-age population (panel b). This pattern indicates that the supply of schooling compared to the school-aged population was considerably higher in the North than in the South. For example, on average, each one of the teachers in the provinces of Piedmont covered, potentially, no more than 46 school-age children; the same ratio grew along a North-South gradient to reach peaks of 170 school-age children per teacher in Calabria and Sicily.

Figure 3 SUPPLY OF PUBLIC SCHOOLING: CHILD-TEACHER RATIO AND EXPENDITURE PER CHILD, 1871

Notes: The child-teacher ratio is measured as the number of children (aged 6 to 10) per primary-school teacher (including private education), while expenditure per child is calculated as municipal (public) expenditure on education per child aged 6 to 10 (in 1911 lire).

Source: See text.

Yet, in Figure 4, a remarkable result emerges: the South supplied a larger number of teachers and invested more resources, compared to the number of enrolled pupils, than the North. Indeed, while the average class size (pupils per teacher) was as low as 23 in the South in 1871, the same ratio was larger than 45 in all provinces of Piedmont.

Figure 4 SPATIAL DISTRIBUTION OF SCHOOL INPUTS: CLASS SIZE AND EXPENDITURE PER PUPIL, 1871

Notes: Class size is measured by the number of pupils enrolled in primary schools per teacher (including private education), while expenditure per pupil is calculated as the amount of municipal (public) expenditure on education per pupil enrolled in (public) primary schools (in 1911 lire).

Source: See text.

These patterns suggest the existence of at least two models of providing schooling across the country: in the Northern provinces, the authorities sought to provide mass education. Instead, in the South, fewer people had access to schooling, yet they were allocated more educational resources. This is somewhat mediated by demand-side variables—something that we consider when we set out to explore whether the two allegedly different models of providing mass education in pre-unification Italy cast a long shadow into the Liberal Age.

EMPIRICAL STRATEGY

Our empirical strategy first focuses on estimating a linear model where the dependent variable is the GER and the regressors are lagged inputs into schooling—class size, expenditure per pupil, and primary-school density over the territory—to explore the factors associated with the expansion of enrollments, that is, mass schooling. Our models control for the geographic, demographic, and socioeconomic factors presented in the previous sections.Footnote ²⁵

Secondly, we estimate a similar model to explore whether and to what extent inputs into schooling correlate with school efficacy—the capability to produce literacy given enrollment rates—while controlling for social, economic, demographic, and political factors.

We also expand our benchmark specifications concerning both enrollment rates and school efficacy by including pre-unification literacy rates estimated by Ciccarelli and Weisdorf (Reference Ciccarelli and Weisdorf2019) as a proxy for educational attainment in the first half of the nineteenth century (1831). We aim at identifying whether a historical legacy stemming from pre-unification regional states might have affected the development of literacy beyond local school inputs and contextual factors.

The models that we employ draw from the literature on Education Production Functions (EPFs), which are an established methodology to explore the effectiveness of schooling in generating learning outcomes, even though most contributions rely on modern data. For example, the seminal article by Card and Krueger (Reference Card and Krueger1992) has underlined the positive role of school quality on economic returns to education in the United States. It is worth noting that, recently, Hanushek and Woessmann (Reference Hanushek, Woessmann, Bradley and Green2020) have reviewed the economic literature on international educational achievements, given the new availability of international cross-country data. They have confirmed previous findings, which suggest that policies aimed at putting more resources into schooling are unlikely to foster educational outcomes if they are not accompanied by complementary interventions.

By contrast, there is little research investigating the relationship between school inputs and educational outputs from a long-term perspective. Mitch (Reference Mitch1984) has estimated the return to male literacy in Victorian England, while, more recently, Schüler (Reference Schüler2016) has explored the relationship between school inputs and earnings by using original data on Prussian counties between 1886 and 1891. The present paper is therefore the first study from a historical and long-term perspective to estimate school efficacy by exploring the factors that improved the production of literacy given the existing rates of primary-school enrollment.

Following Schwartz and Zabel (Reference Schwartz, Zabel and Steifel2013), our province’s aggregate production function is:

(1) $${Q_{pt}} = {f_{pt}}\left( {S{C_{pt}},{X_{pt}},{\rm{ }}S{E_{pt}}} \right) + \varepsilon ,$$

which produces the output Q _pt by combining educational inputs (SC _pt ), province-specific environmental characteristics (X _pt ), and socio-economic variables (SE _pt ), for province p and year t.

We assume a linear specification, as followsFootnote ²⁶:

(2) $${Q_{pt}} = {\beta _0} + {\beta _{\rm{1}}}S{C_{pt}} + {\beta _{\rm{2}}}{X_{pt}} + {\beta _{\rm{3}}}S{E_{pt}} + {\varepsilon _{pt}}$$

In the first version of our model, Q _pt is represented by the GER. All school inputs, SC _pt , included in the regression analysis are measured in the same year, since current conditions affect enrollment decisions immediately. In a second version, Q _pt represents school efficacy, which is the GER-LIT Gap (percentage points) where the GER is measured in the previous benchmark year—so that school efficacy concerns the same cohort, as it should. Clearly, since we are focusing on the literacy of the cohort that was in primary schooling ten years earlier, we measure all other school inputs as first lags.

Educational inputs (SC _pt ) include class size, expenditure per pupil, primary school density, and the density of Evening and Sunday schools, while demographic and socio-economic variables are in SE _pt and X _pt . In our core estimates, we focus on public-school inputs, since private schooling was rather marginal in Italy during the Liberal Age.Footnote ²⁷

The estimation of an EPF presents several challenges from a technical point of view, but our dataset allows us to tackle such issues. The most important ones concern omitted variable bias, as well as selection bias and potential endogeneity (Hanushek and Woessmann Reference Hanushek, Woessmann, Bradley and Green2020). We address these issues as follows. First, we consider the omitted variable bias by including a rich and comprehensive vector of controls. Second, thanks to the panel structure of our data, we can employ lagged inputs into the EPF regressions to limit potential endogeneity and reverse causation—although the latter cannot be completely ruled out. Third, since we use provincial figures based on the whole population, the data that we use should not be prone to selection and self-selection bias at the individual and school level. By the same token, bias from unobservable differences in individual ability should not be an issue in our regressions since we focus on the aggregate provincial-level measures (Schwartz and Zabel Reference Schwartz, Zabel and Steifel2013).

We start estimating our two models through panel-data regressions with provinces FE as well as Pooled OLS with macro-regional FE and controlling for an extensive set of correlates. We then estimate the same models with repeated cross-sections to highlight the trends in the relationships that we study. Finally, we further expand our specifications by including pre-unification literacy rates estimated by Ciccarelli and Weisdorf (Reference Ciccarelli and Weisdorf2019).Footnote ²⁸ This allows us to identify whether a pre-unification legacy stemming from different models of providing mass schooling can be discerned, even when controlling for the local provision of education by Italy’s municipalities in the Liberal Age and independently of contextual factors influencing the demand for schooling.

DETERMINANTS OF SCHOOLING AND SCHOOL EFFICACY

This section presents the results on the relationship between school inputs and the GER, on the one hand, and school inputs and school efficacy, on the other hand. We first discuss the panel-data analysis in Table 2. We regress the GER (Col. (1)) and the GER-LIT Gap (Col. (2)) on all school inputs and controls. Then, we include macro-regional dummies instead of provincial FE to show that most of the variation in the GER-LIT Gap that is estimated to be due to time-invariant provincial features is actually linked to features that are common within macro-regions (Col. (3)). This last model will be later used to focus on the long-term relationship between educational levels in pre-unification areas and the quantity and quality of education in the Liberal Age (1861–1911). Since early nineteenth-century variables are available as sectional data, it is crucial to employ a model with macro-regional dummies instead of provincial FE.Footnote ²⁹

Table 2 PANEL-DATA REGRESSIONS

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Notes: The benchmark macro-region in Column (3) is the North-West. GER-LIT Gap stands for school efficacy. The controls include population density, the urbanization rate, the dependency ratio, infant mortality rate, height, the share of labor force employed in agriculture, industrial VA per capita, the outward migration rate, electoral franchise (the share of people allowed to vote in local elections), and the migration proxy based on population ratios.

Source: See text.

First, we find that a large average class size (enrolled pupils per teacher) was associated with high enrollment rates (Col. (1)). Increasing the class size by 9.3 units (roughly a standard deviation) was associated with an increase in the GER equal to 13.9 percentage points; therefore, substantial economies of scale were pursued in the attempt to expand basic schooling. Similarly, the density of schools over the territory was an important aspect correlated with the expansion of enrollments: increasing the number of primary schools per 1,000 square kms by 239 units (a standard deviation) was associated with an increase in the GER equal to 16 percentage points. However, the coefficient of the density of Evening and Sunday schools is not statistically different from zero. Likewise, expenditure per pupil was not correlated with enrollment rates. Among the controls, electoral franchise was positively correlated with the expansion of primary-school enrollment rates; population density and the dependency ratio are characterized by a negative coefficient, and those of the other variables are not statistically significant.

Focusing on school efficacy (the GER-LIT Gap), one notices that a reduction in class size equal to a standard deviation (9.3 pupils per teacher) is associated with a reduction in the gap between enrollment rates and literacy equal to 6.6 percentage points (Col. (2)). Once pupils were enrolled, reducing the class size was necessary to improve literacy. The density of primary schools is associated with a larger divide between enrollment and literacy rates; an increase of one standard deviation in the density of schools would increase the GER-LIT Gap by 12.4 percentage points. This suggests that a growing network of schools over the territory was important for a larger number of boys and girls to enroll in primary education, yet the growing quantity of schooling might have compromised its quality—a known issue among policymakers aiming at universal primary education in low-development countries in the last 30 years (Glewwe Reference Glewwe2013). Evening and Sunday schools, similarly, were not linked to educational performance given enrollment rates. When school efficacy is considered, rising expenditure per pupil is associated with a reduction in the GER-LIT Gap: an increase in this index by 13.65 lire (1911 prices, i.e., one standard deviation) implies a decline in the gap between enrollment and literacy rates equal to 2.3 percentage points. Among the controls, population density is associated with a reduction of the GER-LIT Gap in every benchmark year, whereas all other controls—except temperature, heights, and the migration proxy (population ratios)—have seldom coefficients that are statistically significant.

To sum up, expanding the quantity of schooling required economies of scale given limited resources. However, for increasing educational outputs, improving the quality of schooling was crucial, which could be pursued by reducing the class size and increasing expenditure per pupil.

We find a positive coefficient for our migration proxy based on population ratios. Since the receiving provinces were the ones characterized by large cities, it is plausible that more internal migration went hand in hand with fewer literate people as a share of the population aged 15 to 19. However, it is worth noting that, although we control for outward migration rates as well as for a proxy of internal migrations, evidence on the selectivity of migrants is rather limited and still being debated.Footnote ³⁰ If the most educated people emigrated abroad, the inefficacy problem would be magnified—especially if they were replaced by illiterate immigrants from other areas of Italy. Indeed, if illiterate people moved from Southern to Northern Italy, the measure of school efficacy would tend to be low in the North compared to the South. First, it is worth noting that South-North migration was marginal in the period concerned. By looking at information from the early twentieth century, Federico, Nuvolari, and Vasta (Reference Federico, Nuvolari and Vasta2019) show that most migrants either moved for short distances or migrated abroad. Therefore, one should worry about the selectivity of migrants moving abroad. Since studying migrants’ selectivity is a daunting task given that it depends on several factors in both the source regions and destination countries (A’Hearn and Ciccarelli Reference A’Hearn and Ciccarelli2021), as well as the large return flows and seasonal movements of Italian migrants in the period concerned, such an analysis goes beyond the scope of this paper. Yet, our evidence suggests a regional pattern of selectivity. A GER-LIT Gap larger in the North than in the South is consistent with the hypothesis of positive selection in outmigration in Northern provinces, and more negative selection in Southern areas. All in all, the complexity of international migration in the case of Italy calls for further evidence on this very relevant issue.

Another issue worth discussing is students’ absenteeism, which might explain part of the GER-LIT Gap itself. Absenteeism may be captured by attendance rates. Although we do not have consistent provincial data for the whole period under study, evidence exists on the magnitude of the phenomenon in specific periods. Given our historical sources, attendance rates may be defined as the number of pupils regularly attending classes at the end of the school year as a share of all pupils officially enrolled. Based on data for 1869, we find that 25 percent of pupils enrolled did not attend at the end of spring—mostly because they were working in agriculture (the attendance rate was equal to 75 percent at the national level). Although attendance rates did not vary much across Italy’s macro-regions, the lowest attendance rates at that time were found in the North-West (70 percent), where the diffusion of formal enrollment was more rapid yet associated with sporadic participation by the most disadvantaged children.

Finally, we undertake a number of robustness checks to verify whether our baseline panel models in Table 2 are biased by other potential omitted variables, including parental literacy, real wages, high-to-low value added in industry, patents per million inhabitants, and land inequality c. 1871 (see Table A3 in the Online Appendix). All results hold.

Although private primary schooling played a minor role in Italy’s education system—and a declining one toward the eve of the twentieth century—private education might have still mattered for some population groups, like wealthy families and religious communities, and through the potential preference of wealthy families for private schooling to educate their daughters. Moreover, in some provinces, low public investment in education could be associated with high private enrollment rates.

In Table 3, we compare public and private GERs with public-school and private-school inputs in a panel model with provincial FE and time dummies (Cols. (1)–(2)).Footnote ³¹ As expected, two distinct patterns emerge: higher public-school enrollment rates were associated with a larger class size and a denser network of schools over the territory. By contrast, economies of scale were not pursued by private institutes—class size was not a significant correlate of enrollment rates. Likewise, a higher number of private schools per square km is positively associated with private GERs, but the coefficient is much smaller than that of public-school density. The two coefficients of school inputs into private education tell a plausible story. Contrary to what happened in public schools, private education essentially targeted a few groups of selected students.

Table 3 PANEL-DATA REGRESSIONS, CONSIDERING PUBLIC AND PRIVATE EDUCATION, AS WELL AS FEMALES AND MALES, SEPARATELY

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Notes: All models are based on a FE panel-data regression with time dummies. The controls include population density, the urbanization rate, the dependency ratio, infant mortality rate, height, the share of labor force employed in agriculture, industrial VA per capita, the outward migration rate, electoral franchise, and the migration proxy based on population ratios.

Source: See text.

Column (3) tests the same model with the GER-LIT Gap as the dependent variable. Given the nature of this index (both public and private-schooling GERs contribute to literacy and are included in the dependent variable), we consider the different types of education (private vs. public) by computing (i) the ratio of private-education teachers over total teachers and (ii) the ratio of private schools over total schools and including them as regressors (lag 1). The two coefficients are not statistically significant, suggesting that the share of private over total schooling did not affect the capability to produce literacy given enrollment rates—a result that is perhaps to be expected, given that private education was not very common: the national average of private over total enrollments was 6.6 percent in 1871 and 4.3 percent in 1911.

Next, we present results split by gender to investigate whether increasing school inputs was associated with a different change in enrollment rates and school efficacy for boys and girls (Cols. (4)–(7)). The gender educational gap in the Liberal Age is still under-researched, with few exceptions that have shown important differences in educational levels and trends across Italy’s macro-regions (Bertocchi and Bozzano Reference Bertocchi and Bozzano2016; Cappelli and Vasta Reference Cappelli and Vasta2021). Our results indicate that growing enrollment rates for girls were associated with growing class sizes, and this correlation was more prominent than in the case of boys (Cols. (4)–(5)). An increase equal to 10 female pupils per female teacher was associated with an increase in female enrollment rates of about 7 percentage points, whereas the same increase in male schools would be equal to a premium on enrollment rates of about 3 percentage points. A similar difference is found concerning the school-density coefficient. Both results are expected given that female schooling was less common at that time; thus, the same increase in school inputs might have improved access to schooling to a greater extent for females than males (Bloom Reference Bloom, Joel, Bloom and Malin2006). Again, the coefficient of expenditure per pupil is negative, confirming that the provinces that witnessed the largest increase in enrollment rates were the ones that spread available resources to a larger number of enrolled pupils.

At that time, the expansion of primary schooling was sustained by a growing number of female teachers and schools (including mixed ones). Starting in the 1870s, the density of female schools converged toward that of male schools, with the exception of the North-West, where parity had already been reached by the 1860s. Our data show that the average female class size rapidly grew toward the end of the nineteenth century, when the male class size had already picked up and was declining (on average). This is consistent with the differences in the magnitude of the coefficients discussed previously: a more rapid expansion of school inputs for females pushed up female enrollment rates from low values, thus explaining the larger marginal impact in the case of girls compared to boys.

No major divide between females and males is visible in the way that class size and school density affected school efficacy (Cols. (6)–(7)). In contrast, the coefficient of expenditure per pupil is larger for boys than girls. A one-standard-deviation increase in expenditure per pupil (13.65 lire per pupil) was associated with a reduction in the GER-LIT Gap equal to 2 percentage points for females and almost 3 for males, suggesting a bias in resource allocation favoring boys over girls in primary education.

To highlight the changing trends in the relationship among the variables that we study, we present cross-section regressions, always including all control variables, with the GER (Cols. (1)–(5)) and the GER-LIT Gap (Cols. (6)–(10)) as the dependent variables (Table 4). We present the same results for females and males separately in the Online Appendix (Tables A4 and A5, respectively).Footnote ³² The variation of the coefficients over time highlights an inverse U-shaped pattern in the relationship between class size and the GER, on the one hand, and the density of primary schools and the GER, on the other hand. Initially, growing school inputs (increasing economies of scale in education) are associated with large improvements in the GER. Toward the beginning of the twentieth century, improvements prompted by growing school inputs became smaller. Tables A4 and A5 show that this pattern is the result of a composite effect driven by gender differences.

Table 4 CROSS-SECTION REGRESSIONS

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Notes: In Columns (6) to (10), all school inputs excluding the density of Evening and Sunday schools are lagged by one period (10 years). The controls include latitude, longitude, average temperature, average rainfall, population density, the urbanization rate, the dependency ratio, infant mortality rate, height, the share of labor force employed in agriculture, industrial VA per capita, the outward migration rate, electoral franchise, and the migration proxy based on population ratios.

Source: See text.

A PRE-UNIFICATION LEGACY?

So far, our study, stretching from the pre-unification period to the early twentieth century, has highlighted the central role played by local school inputs in expanding enrollments and, at a later stage, improving literacy during Italy’s Liberal Age (1861–1911). This is suggestive of persistent differences stemming from the legacy of pre-unification education policies. Figure 5 highlights the early divide in literacy by relying on the dataset on adult literacy by Ciccarelli and Weisdorf (Reference Ciccarelli and Weisdorf2019). Significant pre-unification differences in literacy existed across the macro-regions of Italy:Footnote ³³ the Kingdom of Sardinia (Piedmont and Liguria) as well as Lombardy in the Lombardy-Venetia Kingdom led the way, while the Center and (even more) the South lagged behind. This pattern seems to be linked to substantial variation in schooling post-1861, which can be fully appreciated by looking at macro-regional averages concerning two school inputs: class size in public schooling (enrolled pupils per available teacher) and public-schooling expenditure per pupil.Footnote ³⁴

Figure 5 ADULT LITERACY IN ITALY’S MACRO-REGIONS, FROM 1831 TO 1911

Note: We rely on literacy rates from Ciccarelli and Weisdorf (CW) (2019). The vertical line marks the unification of Italy in 1861.

Source: See text.

Class size indicates remarkable disparities in the way that primary education was provided in the pre-unification era: in Figure 6, both the North-West and—to a lesser extent—the North-East exhibit a large average class size, whereas the Center as well as the South and Islands show a small class size.

Figure 6 CLASS SIZE AMONG ITALY’S MACRO-REGIONS, FROM THE PRE-UNIFICATION ERA TO THE EARLY TWENTIETH CENTURY

Source: See text.

These patterns are consistent with the hypothesis that the North had already paved the way toward a model of public mass schooling—relying on economies of scale in primary education in an attempt to reach most of the population.Footnote ³⁵ Instead, the Center and the South of the country had not yet abandoned a more elitist view of education based on private institutions, near-zero public education, and a focus on educational grades higher than the primary; as a result, fewer primary-school pupils were taught by more teachers. In the long run, the two areas converged toward a similar class size, the one based on the provision of schooling for all and mass education—under the same national school act—but this process took more than 50 years, holding back convergence in literacy rates across the country.

Figure 7 on educational expenditure per enrolled pupil highlights the same divide. Unfortunately, harmonized data for the pre-unification period are not available, so these results must be taken with a pinch of salt. In 1861, the average expenditure was 15 lire per enrolled pupil in the North-West; all other macro-regions spent more. Two distinct models evolved over time: the Center and the North-East converged with the low expenditure per enrolled pupil that characterized the North-West, a sign of the effort to provide mass education. Instead, the South and Islands are characterized by a higher expenditure per pupil enrolled compared to the North-West.

Figure 7 EXPENDITURE PER PUPIL ACROSS ITALY’S MACRO-REGIONS, 1861 TO 1911

Source: See text.

The intuition that class size and expenditure per pupil can be used as a proxy of preferences or models in the provision of primary schooling is backed up by the relationship between the importance of private education and class size or expenditure per pupil. As Figure 8 shows, a larger share of enrollments in private over total education is associated with a smaller class size (panel a) and a larger expenditure per pupil (panel b).

Figure 8 PRIVATE EDUCATION, CLASS SIZE, AND EXPENDITURE PER PUPIL, FULL SAMPLE (ALL YEARS)

Source: See text.

Figure 9 interestingly shows that early movers that adopted a more inclusive model of educational provision reduced considerably gender inequality in literacy, much more rapidly than the Center and the South. The percentage-point gap was reduced to virtually zero in the North-West, and to about five in the North-East. Instead, in the Center and in the Islands, the average gap decreased slightly, yet it remained larger than 10 percentage points, while in the South the gap remained stable and above 20 percentage points. These trends illustrate that the expansion of mass schooling through economies of scale in the North happened with boys first, thus implying an increasing gender gap—before the rise of female education would reduce the gap below the one characterizing the regions relying on a more elitist model of schooling.

Figure 9 GENDER INEQUALITY IN LITERACY: MALE LITERACY MINUS FEMALE LITERACY, 1831–1911

Source: See text.

We further explore this legacy with a series of regressions with year dummies and macro-regional FE instead of provincial FE, which would otherwise capture any time-invariant effect that varies merely across provinces. We expect to find a link between pre-unification education and post-unification literacy within each macro-region. Since we control for a large set of contextual variables, this would imply that pre-unification states that had adopted an early system of mass education tended to be characterized by higher enrollment rates and, possibly, school efficacy.

Table 5 shows a positive and significant correlation between enrollment rates and pre-unification (1831) literacy (Col. (1)). In Column (2), we interact pre-unification (1831) literacy with the year dummies to explore whether this alleged pre-unification educational legacy persisted over time: in every year considered, the coefficient of the legacy of preunification literacy is given by the sum of the 1831 coefficient and the interaction between the latter and the dummy for the concerned year (the excluded year is 1871). For example, in 1881, the marginal effect of preunification (1831) literacy was 0.950 + (–0.270) = 0.680. Therefore, we find pre-unification literacy to have been relevant until the early twentieth century; this marginal effect further declines over time, to the point of becoming virtually zero in 1911.

Table 5 PANEL-DATA REGRESSIONS: A PRE-UNIFICATION LEGACY?

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Notes: Literacy 1831 is from Ciccarelli and Weisdorf (CW) (2019), who have provided such estimates. Class size, primary-school density, and 1831 literacy refer to females and males in Columns (3) and (4), respectively. In all other specification they refer to the total population. In Columns (3) and (4), the number of observations drops to 276 due to schooling data divided by gender being unavailable for 1911. For a full list of controls, see Table 4.

Source: See text.

This is plausible since the previous decade marked the first era of substantial state intervention in matters related to primary schooling (the aforementioned Nasi and Orlando Laws). We also explore whether this legacy was different for girls and boys (Cols. (3)–(4)). The legacy appears to have been more persistent for girls, which resonates with the more prominent effort by local authorities to provide schooling to boys. As Ciccarelli and Weisdorf (Reference Ciccarelli and Weisdorf2019, fig. 6) have shown, female adult literacy caught up relatively quickly with male literacy in the North-West after unification. The same, though to a lesser extent, happened in the Center and North-East. However, in the South, a substantial gender gap in education remained. This result is fully consistent with our estimates in Column (3): while Central and Southern regions closed the gap in male enrollment rates following unification, the gap in female enrollment rates persisted into the twentieth century.

Finally, we focus on school efficacy and highlight that more literate provinces before unification tended to expand enrollment rates while reducing school efficacy (a larger GER-LIT Gap) (Cols. (5)–(6)). This suggests the existence of two distinct models: an inclusive one aimed at fostering education for all in the North-West and North-East, and a more elitist one elsewhere, particularly in the South. Yet, over time, the net coefficient of 1831 literacy gets smaller, as the areas that expanded enrollments early reduced the GER-LIT Gap starting in the 1890s (the North and Center), whereas the South and Islands improved it in the first decade of the twentieth century.Footnote ³⁶

Our findings resonate with the results offered by Postigliola and Rota (Reference Postigliola and Rota2021), who emphasize the importance of the legacy of French reforms, which improved the quality of schooling and changed “collective preferences” toward more education. Our results suggest that there existed different equilibria of educational provision across Italian macro-regions, even if the French reforms did not necessarily translate into quality of schooling during the Liberal Age. We find that the areas most affected by the French legacy first increased enrollment rates by aiming at economies of scale and, only at a later stage, sought to improve the efficacy of schooling in generating literacy. In 1861, the gap between GER and literacy rates for the same cohort was particularly large in the North-West, the area most affected by French reforms (Figure 10).

Figure 10 THE GER-LIT GAP IN ALL MACRO-REGIONS, 1871–1911

Note: The GER (L.1) is the first lag (–10 years) of the GER variable.

Source: See text.